Computer-aided design of PVR mutants with enhanced binding affinity to TIGIT

Background TIGIT, as a novel immune checkpoint molecule involved in T cell and NK cell anergy, could induce the immune tolerance and escape through binding with its ligand PVR. Blockade of TIGIT/PVR is considered as a promising strategy in cancer immunotherapy. However, to facilitate the design of inhibitors targeting TIGIT/PVR, the structural characteristics and binding mechanism still need to be further studied. Methods In this study, molecular dynamics (MD) simulations and in silico mutagenesis were used to analyze the interaction between TIGIT and its ligand PVR. Then, PVR mutants were designed and their activities were determined by using TIGIT overexpressed Jurkat cells. Results The results suggested that the loops of PVR (CC′ loop, C′C″ loop, and FG loop) underwent a large intra-molecular rearrangement, and more hydrogen bond crosslinking between PVR and TIGIT were formed during MD simulations. The potential residues for PVR to interact with TIGIT were identified and utilized to predict high affinity PVR mutants. Through the biological activity evaluation, four PVR mutants (PVRS72W, PVRS72R, PVRG131V and PVRS132Q) with enhanced affinity to TIGIT were discovered, which could elicit more potent inhibitory effects compared with the wild type PVR. Conclusions The MD simulations analysis provided new insights into the TIGIT/PVR interaction model, and the identified PVR mutants (PVRS72W, PVRS72R, PVRG131V and PVRS132Q) could serve as new candidates for immunotherapy to block TIGIT/PVR. Video Abstract

T cells and Tregs cells [13][14][15][16][17]. Until now, the identified ligands of TIGIT contain CD155 (also known as PVR or nectin-like 5), CD112 and CD113 [15,18,19]. Among these ligands, PVR exhibited a higher affinity with TIGIT compared with CD112 and CD113 [19,20]. PVR is mainly expressed on dendritic cells (DCs), T cells, B cells, macrophages and all kinds of tumor cells [14,[21][22][23][24]. Engagement of TIGIT with PVR has been involved in modulating the cytokine production of DCs and facilitating the polarization of pro-inflammatory M1 macrophages into anti-inflammatory M2 macrophages, which in turn lead to the inhibition of effector T cells and NK cells activation [13,[25][26][27]. The roles of PVR on normal cells are to prevent excessive immune cells activation and sustain immune homeostasis [25,28]. However, tumor cells evade immune surveillance and induce T cells and NK cells exhaustion by overexpressing PVR [29][30][31]. Blocking TIGIT-PVR interaction can restore immune exhaustion of T cells and NK cells [27,32]. Therefore, the development of TIGIT/PVR inhibitors may have potentials in cancer immunotherapy. Antibodies targeting TIGIT/PVR pathway have achieved good clinical results in cancer treatment, as so far six TIGIT-targeting antibodies were under pre-clinical or clinical trials [33,34].
However, some intrinsic adverse effects of antibody drugs, such as off-target effects, poor tissue penetration and Fc-effector functions, could deplete lymphocytes, which limit the application of antibodies in cancer treatment [35,36]. Therefore, the development of other types of drugs targeting TIGIT/PVR pathway is essential for cancer intervention and treatment. Studies have shown that engineered protein drugs targeting immune checkpoint molecules can avoid the disadvantages of antibodies and exhibit better antitumor efficacy [37,38]. We reported that blockade of TIGIT/PVR by peptide could elicit strong tumor tissue penetration ability and antitumor immune response, even in anti-programmed cell death protein 1 (PD-1) resistant tumor model [39]. As the important role of TIGIT/PVR pathway in regulating the function of the immune cells, it's essential to develop alternative protein drugs targeting the TIGIT/PVR pathway to use individually and in combination with other treatment methods to improve the therapeutic efficacy.
The structure of TIGIT in complex with its ligand PVR has been resolved. It has been shown that both TIGIT and PVR possess three domains: extracellular domain, transmembrane region and cytoplasmic domain, and the extracellular domain has a series of glycosylation sites (Additional file 1: Fig. S1a-d). The extracellular domain of PVR is composed of D1, D2, and D3 domain, and D1 domain of PVR plays an important role in interacting with TIGIT to deliver the inhibitory signals [40]. A series of crystal structures of the extracellular domain of TIGIT and PVR was reported by using X-ray diffraction and electron microscopy (Table 1). These findings facilitate computer-aided drug design using molecular dynamics simulations for molecules which can regulate the TIGIT pathway.
In this study, we used molecular dynamics (MD) simulations to study the interaction between TIGIT and PVR based on the resolved crystal structure (PDBID: 3UDW). We analyzed the structural moment, atomic dynamics of PVR and variations of residual pairs, which helps to identify several novel residues of PVR that potentially bind to TIGIT. Through In silico mutagenesis and cell assay with flow cytometry, we successfully predicted high affinity PVR mutants, and measured the binding affinity of the mutants and the effects of mutants on TIGIT overexpressed Jurkat cells. Finally, we identified several high affinity PVR mutants ( PVR S72W, PVR S72R, PVR G131V, and PVR S132Q) with more potent inhibitory effects on TIGIT overexpressed Jurkat cells. The MD simulations of TIGIT/PVR complex may help us to understand the binding mechanism of PVR with TIGIT, and the high affinity mutants may serve as start points for the development of TIGIT-PVR inhibitors.

Molecular dynamics simulations
The 3D structures of hPVR alone and the PVR/TIGIT complex were retrieved from crystal structures with a PDBID of 3UDW, respectively. The 3D structures were processed by GROMACS (Version 4.6) by using OPLS/ AA force field. The processed proteins were then separately solvated in two simulation systems, which were made by water cubic box with SPC water model and periodic boundary conditions (PBC = 1.0 nm) were selected. The solvated systems were neutralized with sodium ions at the physiological conditions. The molecular dynamics simulations in our study were performed in three steps. The structures were firstly relaxed through energy minimization (EM). The potential energy (E pot ) was negative on the order of 10 -5 -10 -6 and the maximum force (F max ) was less than 1000 kJ/mol/nm. Next, the solvent and ions around the protein were equilibrated via two simulation phases. At the first phase, the systems were restrained under an NVT ensemble (constant Number of particles, Volume, and Temperature) and the Verlet scheme was selected for 1 ns simulation with the temperature of the system reaching a plateau at 310 K. At the second phase, equilibration of pressure was conducted under an NPT ensemble (constant Number of particles, Pressure, and Temperature) for 1 ns simulation with the pressure set to 1 bar. The Parrinello-Rahman baroslat was used to control the pressure and all bonds involving hydrogen atoms were constrained with the LINCS algorithm during the equilibration procedure. Finally, the position restraints were then released and a production run of 50 ns were performed, where the time step were set to 2 fs.

Virtual alanine and residue scanning mutagenesis
The alanine scanning mutagenesis sequentially mutated each residue of PVR at a time to alanine and the binding energy for each mutant was calculated. Residue Scanning, also known as site-directed mutagenesis, was used to successively mutate important residues of PVR to 20 natural amino acids. In this study, Molecular operating environment (MOE) software (version 2018) was used to performed all the calculation and one site mutation was taken out for the mutagenesis. The calculations were performed by using Amber10 force field and low-mode simulation was used to generate mutants' conformation with a maximum of 30. MM/GBVI scoring function was used to calculate the binding energy between TIGIT and PVR mutants.

Expression of TIGIT, PVR and the mutants
The full-length human PVR and human TIGIT was cloned into the pLVX-Puro vector for constructing PVR mutants and functional experiments. All mutants were acquired under site-directed mutagenesis by using a QuickChange mutagenesis kit (Thermo Fisher, USA) and ensured via DNA sequencing. Transfections of plasmids into CHO-K1 or Jurkat cell line were carried out with PowerTrans293 ™ (Sixiang Biological Inc., China) according to the manufacturer's instructions. The cells stably expressing PVR, PVR mutants or TIGIT were ascertained by fluorescent-activated cell sorting (FACS) Caliber flow cytometry, and the cells stably expressing PVR and PVR mutants were next identified using PVR antibody (A5753, ABclonal) by western blotting.

Cell staining analysis
For cell staining, the Fc-fused protein of human TIGIT (Sino Biological Inc., China) were serially diluted from 120 to 1.875 nM. CHO-K1 cells expressing WT PVR or PVR mutants were suspended in phosphate buffered saline (PBS) and incubated with hTIGIT-Fc protein for 30 min at 4 ℃. After incubating with hTIGIT-Fc protein, cells were stained with APC conjugated anti-human IgG Fc (HP6017, Biolegend, USA) for 30 min at 4 ℃. Then, CHO-K1-PVR and CHO-K1-mutants was washed in FACS buffer (PBS with 2% FBS) and analyzed by FACS Caliber flow cytometry (BD Bioscience, USA). The cells incubated only with PE conjugated human Fc antibody were regarded as the negative controls. The mean fluorescence intensity (MFI) was used to analyze binding affinity of WT and mutant PVR to its ligand TIGIT.
The protein transport inhibitor (eBioscience) were added at the last 4 h before the end of co-culture. For cell staining, Jurkat-hTIGIT cells collected after co-culture were incubated with anti-human TIGIT PE (MBSA43, eBioscience) for 30 min at 4 ℃ and then fixated and permeabilized. Afterwards, permeabilized cells were stained with anti-human IL-2 APC (MQ1-17H12, Belegend) or isotype control for 30 min at 4 ℃. Jurkat cells was washed in FACS buffer (PBS with 2% FBS) for flow cytometry analysis.

Statistical analysis
For binding and coculture assay, statistical differences were analyzed by GraphPad Prism 7.04. Statistical significance among different groups was calculated by Student's t-test. For correlation analysis, the data were calculated by R language. The Pearson product-moment correlation coefficient represented correlation between protein expression and binding affinity. The statistically significant values were considered as follows: *P < 0.05, **P < 0.01, and ***P < 0.001. Values of P < 0.05, P < 0.01, and P < 0.001 were considered as statistically significant.

Structural dynamics and hydrogen bond crosslinking of PVR
The structural properties of TIGIT/PVR complex and PVR bound to poliovirus were summarized (Table 1), and crystal structure of 3UDW was chosen based on the resolved resolution, R-value, and mutation sites introduced in the crystal structure to study the binding of PVR and TIGIT. To further elucidate the interaction between human PVR and human TIGIT as well as structural dynamics at the atomic level, 50-ns MD simulations were performed for two systems (hPVR apo and hPVR bound states) under the physiological conditions in which the effects of force field, water, temperature and pressure were well considered. The root mean square deviation (RMSD) calculations were monitored during MD simulations. The RMSD curves of the two trajectories gradually reached to the equilibrium state, indicating that the PVR molecules attained a structurally stable state (Fig. 1a). Therein, the PVR in the bound state showed few fluctuations than PVR in the apo state, indicating that the conformation of PVR in complexed with TIGIT was more stable than that in apo state (Fig. 1a). The PVR in the apo state fluctuated greatly in 30-40 ns and then reached a stable state, which implied that the PVR in the apo state went through momentous structural rearrangements during the MD simulations (Fig. 1a). The averaged structure after 50-ns MD simulations was superimposed to the relevant crystal structure (PDBID: 3UDW) (Fig. 1b), and the results indicated that there were more obvious changes on the loops near the TIGIT binding interface, but not the beta-sheets at the interface between PVR and TIGIT (Fig. 1b). Hereafter, the interaction networks between crystal structure and the averaged structure were analyzed to study whether the MD simulations could indicate more potential residues for TIGIT interaction (Fig. 1c, d). Residues S62, Q63, S74, H79, Q80, P84, S85, T127, P129, S132 in PVR protein were involved in TIGIT binding based on the crystal structure, while the contact residues in PVR were H60, S62, Q63, G73, S74, Q82, P84, V126, P129, G131, S132 after the MD simulations (Fig. 1c, d). The MD averaged conformation showed that residues H60, G73, Q82, V126 and G131 in PVR generated hydrophobic interaction to TIGIT, which were not observed from the crystal structure.

Atomic dynamics of PVR in different states
The residual fluctuations of PVR in apo and ligand bound states were analyzed. During the 50-ns MD simulations, root mean square fluctuation (RMSF) values of each residue were calculated and the results indicated that the residues in apo PVR was more flexible than that in the TIGIT bound PVR (Fig. 2a). The RMSF values of the residues such as G42, N55, G70, S72, and S89 were significantly different between the apo and bound state, and these residues, especially N72, were more flexible in the apo state (RMSF > 2 Å) than in bound state (RMSF < 0.8 Å) (Fig. 2a). The residues at the interface were rigid and the residues at PVR BC loop, CC′ loop, CC″ loop, C″D loop, and FG loop which were adjacent to the interface were relatively more flexible (Fig. 2a and Additional file 1: Fig. S1d).
The comparison between apo PVR and TIGIT bound PVR indicated that the orientations of five residues N55, G70, Q82, S89, and F128 in PVR changed at different states (Fig. 2b). MD simulations revealed that the regions including these five residues were in loop motifs (such as CC′ loop, C′C″ loop, and FG loop) and the five regions which were relatively close to the binding interface underwent a large intra-molecular rearrangement to facilitate the TIGIT binding process. Meanwhile, the hydrogen bonds formed in these regions, such as BC loop and C″D loop, were less stable and might be broken to allosterically regulate the conformation of hPVR to interact with TIGIT (Fig. 2c). Therefore, the residues H60, S72, S74, Q82, P84, and T127 in PVR that are involved in forming the interactions with TIGIT in the loop regions including BC, CC′, C′C″, and FG loop which were adjacent to the binding interface were selected as potential sites for in silico mutagenesis. The residues V61, G73, H79, S85, G131, and S132 which located close to the TIGIT binding residues were also selected as potential sites for in silico mutagenesis. We proposed that mutagenesis at these residual sites were most probably able to improve the binding affinity to TIGIT.

Distance variations of residue pairs during simulations
In order to study the residues that are involved in the process of TIGIT/PVR interactions, we measured the distance changes of the amino acid pairs which either contributed to hydrogen bond formation or Van der Waals interactions during MD simulations. The residue in the hPVR was labelled in the front of the residue pair, and the residue in the hTIGIT was marked in the back. During 50-ns MD simulations, the distance of a few amino acid pairs changed slightly and stayed in a relatively stable state, and large fluctuations have been detected in some amino acid pairs (Fig. 3). The residue pairs for the interactions which were crucial for TIGIT binding had a stable distance during the MD simulations (such as PVR S74-TIGIT P114/D115/G116, PVR S85-TIGIT Y113/P114 and PVR F128-TIGIT Q56/I68/N70/ L73) (Fig. 3). The distances between the residual pairs  (such as PVR H60-TIGIT D72, PVR S62-TIGIT Q56, PVR T65-TIGIT T117, PVR S72-TIGIT T24, PVR Q82-TIGIT Q53, PVR P84-TIGIT Y113,  PVR S87-TIGIT D115,  PVR G131-TIGIT L65, PVR S132-TIGIT N58, and PVR S132-TIGIT H111) exhibited large fluctuations. Considering that the residues with small fluctuations may be crucial for the protein-protein interaction, we decided to mutate the residues with relatively larger fluctuations in PVR in order to facilitate the formation of more stable amino acid pairs with TIGIT (Fig. 3). The residues S72 and S87 on PVR only possessed one interaction to TIGIT and the distance for these interactions varied greatly during MD simulations, therefore we proposed that mutations of the amino acids S72 and S87 of PVR could increase the binding affinity between PVR and TIGIT (Fig. 3). We finally selected the residues with large distance variations for designing the high affinity PVR mutants which may interrupt the formation of wild type of PVR and TIGIT.

hPVR mutants generated by the in silico mutagenesis
The residues of PVR with potential values obtained by different analysis methods described above were summarized and used to perform virtual alanine scanning and residue mutagenesis. Almost all residues that mutated to alanine (Additional file 2: Table S1) reduced the affinity of PVR with TIGIT and decreased the stability of the protein, which proved that these residues were beneficial to the binding of PVR to TIGIT. Then, these residues were mutated into 20 natural amino acids and 3600 PVR mutants were obtained. The binding energies of PVR mutants to TIGIT were calculated by MM/GBVI scoring function in MOE package. Residues substitution of PVR in the "key" region of PVR 127 TFP 129 [41] were not conducive to the improvement of binding affinity and protein stability. However, amino acids replacement of G131 and S132, near the "key" region of PVR 127 TFP 129 , S72, S74, S87 with large distance variations (Fig. 3) and residual dynamics (Fig. 2a) had greater potential to enhance the binding affinity. According to the binding affinity, protein stability and residual fluctuations during MD simulations, 10 mutants S87W, S72W, S72R, T65E, G131W, S74W, S132R, G131M, G131V and S132Q (Table 2) were selected for subsequent measurement of biological functions.

Binding analysis between hPVR mutants and hTIGIT
Cell-based flow cytometry analysis was carried out to measure the binding affinity of PVR mutants with TIGIT. Ten single-point mutations of PVR including S87W, S72W, S72R, T65E, G131W, S74W, S132R, G131M, G131V, and S132Q were constructed and the full-length proteins were expressed in CHO-K1 cells, respectively. We performed monoclonal screening after constructing stable cell lines of mutants. Cell lines expressing basically consistent PVR mutants were selected for this study (Additional file 1: Fig. S2a). It was found that the mutant T65E merely expressed in a small amount on CHO-K1 cells (Additional file 1: Fig. S2a). Total proteins of the cells stably expressing PVR mutants were extracted, and western blotting was used to detect the expression of PVR protein (Additional file 1: Fig. S2b), which were comparative to those obtained by flow cytometry (Additional file 1: Fig. S2a). The PVR antibody used for western blotting is recombinant fusion protein containing a sequence combining with amino acids 220-345 of human PVR, which indicated that mutations in the residues of PVR D1 domain (corresponding to amino acids 30-143) do not affect the detection of PVR expression by PVR antibodies. These results implied that mutant T65E affected PVR expression.
After the successful construction of the PVR mutant cell lines, we used flow cytometry to measure the binding affinity of PVR mutants (S87W, S72W, S72R, T65E, G131W, S74W, S132R, G131M, G131V or S132Q) to TIGIT-Fc eukaryotic protein. The concentrations of eukaryotic human-TIGIT-Fc were 1.875 nM, 3.75 nM, Fig. 3 The distance variations of key residue pairs for hTIGIT/hPVR complex. Graphs showed the distance changes of residue pairs forming hydrogen bonds and Van der Waals distance interactions in TIGIT/PVR complex during MD simulations. The residues in the hPVR are written in the front of the residue pair, the residues in the hTIGIT are written in the back  (Fig. 4a, c). The mutations of S72R, S72W, G131V, S132Q and S74W significantly enhanced the binding of PVR-ecto mutants to TIGIT (Fig. 4a, b). The mutant G131V significantly enhanced the binding affinity to TIGIT because of the site was replaced by hydrophobic side chain (Fig. 4b). The mutant S132Q presented higher binding affinity with TIGIT and Glutamine (Gln) had a longer polar uncharged side chain than serine (Ser), which indicated that lower steric hindrance performed a crucial role in the interaction between PVR and TIGIT (Fig. 4b). It was surprising that the mutants S72R and S72W which changed the polar uncharged Ser (serine) to positively charged Arg (arginine) or hydrophobic Trp (tryptophan) with aromatic group side chains slightly increased their affinity with TIGIT (Fig. 4b). However, the corresponding PVR-ecto mutants which introduced S87W, T65E, G131W, S132R or G131M mutation decreased the binding affinity to TIGIT (Fig. 4c, d). However, mutant T65E exhibited relatively weak binding capacity to TIGIT (Fig. 4d), which might be caused by the low expression level. The relative binding affinity of each mutants was standardized by the binding affinity of the wild type PVR at the concentration of 120 nM (Fig. 4e). The correlation between the binding affinity and the protein expression level of PVR mutants was studied and the P values were greater than 0.05 and R 2 were less than 0.35 under different TIGIT protein concentrations (Additional file 1: Fig. S3). The results showed no correlation between the binding affinity and the expression level Fig. 4 The binding affinity of hPVR mutants with hTIGIT. a, c Parental CHOK1 cells as well as CHOK1 cells overexpressing WT hPVR and mutant hPVR were used in binding assays. The binding of WT and mutant hPVR with hTIGIT-Fc were assessed by flow cytometry using an anti-human Fc antibody. Representative curves of three independent measurements were shown. b, d The binding affinity between hPVR mutants and hTIGIT-Fc at various concentrations from 120 to 1.875 nM. Graphs showed mean ± standard error of the mean (SEM) of three independent experiments. e The binding affinity of hPVR mutants were normalized versus WT hPVR (dashed line) at the concentration of 120 nM, which was defined as relative hTIGIT binding potency (RP) values of the hPVR mutants. *P < 0.05, **P < 0.01 and ***P < 0.001 by Student's t-test of PVR mutants, which indicated that the binding affinity of the PVR mutants were mainly depended on the actions of the mutations.
To further verify that there was no correlation between the binding affinity and protein expression, additional experiment was conducted. Firstly, the membrane protein of PVR mutants fused with EGFP were extracted and quantified by the fluorescence value of EGFP to exclude the influence of protein expression on binding affinity. Then the binding affinity of PVR mutants with TIGIT-His eukaryotic protein was identified by microscale thermophoresis (MST) to calculate the K D values of PVR mutants. It was shown that the binding affinities of mutants PVR G131V (K D = 0.016 μM) and PVR S132Q (K D = 0.058 μM) to TIGIT-His were enhanced about 72-fold and 20-fold, respectively (Additional file 1: Fig.  S4a, d-e) compared with wild-type PVR (K D = 1.146 μM). The mutants PVR S72W (K D = 0.328 μM) and PVR S72R (K D = 0.549 μM) increased their binding affinity to hTIGIT by approximately 4 and 2 times (Additional file 1: Fig. S4a-c). The results were consistent with that obtained through flow cytometry methods (Additional file 1: Fig.  S4a-e, Fig. 4a, c). Also, the mutants PVR S74W, PVR S87W, PVR T65E, PVR G131W, PVR S132R, and PVR G131M with lower binding affinity consistently impaired the binding (Additional file 1: Fig. S4a, f-k). It was worth mentioned that the fluorescence value of the extracted membrane protein of mutant PVR T65E was almost undetectable which was consistent with the previous results of protein expression (Additional file 1: Fig. S2) by flow cytometry and western blotting and it was difficult to detect the binding affinity between these two proteins through MST.

High affinity hPVR mutants decreased IL-2 production
To further explore whether PVR mutants affect the biological function, we transfected human TIGIT into Jurkat cells which lacked endogenous TIGIT expression and co-cultured Jurkat-TIGIT cells with CHOK1 cells that expressing PVR Wild-Type or mutants. The expression of wild type PVR on CHOK1 cells reduced the proportion of Jurkat-hTIGIT cells which produced IL-2 compared to the parental CHOK1 cell line by using flow cytometry analysis (Fig. 5a, b). The mutants PVR S72W, PVR S72R, PVR G131V, and PVR S132Q induced more potent inhibitory effects of PVR on TIGIT, resulting in a remarkable reduction in the proportion of Jurkat-hTIGIT cells producing IL-2 compared to Jurkat-TIGIT cells co-cultured with wild-type PVR (Fig. 5b), which was consistent with previous results that these mutants showed higher affinity with TIGIT (Fig. 4a, b). However, PVR S87W, PVR T65E, PVR G131W, PVR S74W, PVR S132R, and PVR G131M exhibited a reduced inhibitory effect on IL-2 secretion, and the proportion of Jurkat-TIGIT cells that produced Fig. 5 IL-2 production of Jurkat-hTIGIT cells co-cultured with hPVR mutants. a Jurkat cells overexpressing hTIGIT were co-cultured with CHOK1, CHOK1-hPVR, CHOK1-mutants for 48 h, which was stimulated with 1 μg/mL human anti-CD3 and 0.5 μg/mL human anti-CD28. Protein transport inhibitor was added in the last 4 h. The frequency of IL-2-secreting Jurkat-hTIGIT cells were detected by flow cytometry. Data were representatively independent of three measurements. b Analysis of the frequency of IL-2-secreting was shown. Graphs showed mean ± standard error of the mean (SEM) of three independent experiments higher level of IL-2 compared to wild-type PVR (Fig. 5b). Among them, PVR S87W, PVR T65E, PVR G131W, PVR S132R, and PVR G131M had a weakened inhibitory effect on TIGIT, which were consistent with the results of the aforementioned binding affinity on TIGIT (Fig. 4a, b). The mutant PVR S74W slightly reduced inhibitory effect on TIGIT compared with wild-type PVR, but there was no significant difference (Fig. 5b). The previous affinity results showed that the binding affinity of PVR S74W with TIGIT was lower than that of wild-type PVR at low TIGIT-Fc concentrations, but the affinity of PVR S74W with TIGIT was slightly higher than wild-type PVR when TIGIT-Fc was almost reached the saturated concentration (120 nM) (Fig. 4b). Therefore, the effect of PVR S74W on TIGIT was not different from that of wild-type PVR in the co-culture experiment (Fig. 5b), which might due to the reason that the concentration of PVR interacting with TIGIT did not reach the saturated state. Surprisingly, the co-culture of Jurkat-hTIGIT and CHOK1 cells secreted higher level of IL-2 than that secreted by Jurkat-hTIGIT cells alone (Fig. 5a, b), which might be caused by some stimulatory molecules expressing on CHOK1 cells.

Discussion
TIGIT is emerging as a critical immune checkpoint that has been involved in regulating immune effector function. Overexpression of PVR on cancer cells can restrain T cell and NK cell responses because PVR can negatively manipulate TIGIT functions [26,27,42]. The structural properties of TIGIT/PVR complex were well studied so far and several antibody drugs targeting TIGIT/PVR pathway have been developed [33,34,41]. Studying the dynamic characteristics and binding mechanism at the atomic level between TIGIT and PVR will contribute to the rational discovery and design of PVR inhibitors such as high affinity PVR protein, peptides and small molecules.
The crystal structures of TIGIT and PVR (Table 1) provide atomic details to understand the coordinates of the PVR and the binding mode of the complex, but they lack the dynamic information about the protein. In order to deeply study the flexibility and fluctuation of PVR in different states, we analyzed the overall and atomic level fluctuations of PVR by using MD simulations. The distance variations of important residue pairs for TIGIT/ PVR complex revealed that the residue pairs formed by H60 and S62 on the C sheet (near the BC loop) of PVR, Q82 on the C loop of PVR, S87 on the C″D loop of PVR, and G131 and S132 on the FG loop of PVR underwent large conformational movements (Fig. 3). This is consistent with the results of atomic fluctuations for the corresponding residues (Fig. 2a, c) during MD simulations. Most of these residues with the distance between 4.5 and 6 Å to hTIGIT molecule are located in the loop region and fluctuated greatly, which suggested that the residues in the loop region play a crucial role for TIGIT/PVR interaction.
The crystal structure of TIGIT bound to PVR revealed that two motifs ( 112 TYP 114 in TIGIT and 127 TFP 129 in PVR) played a crucial role in the interaction of TIGIT and PVR [41]. The "key" residues Y113 in TIGIT and F128 in PVR were respectively inserted into a hydrophobic pocket formed by the engaging molecule. The mutation analysis of the TIGIT/PVR interaction revealed that substitution of Q63 in the (V/I) (S/T) Q motif and F128 in the 127 TFP 129 motif with alanine in PVR reduced the binding affinity with TIGIT [41]. To obtain the energy contribution of the residues in PVR to the TIGIT/PVR interaction, we analyzed some hotspot residues in PVR through molecular dynamics simulation analysis and then performed in silico mutagenesis on hotspot residues. Among them, mutation of Q63 and F128 in PVR to Ala (alanine) affected the binding of TIGIT to PVR and the structural stability of TIGIT/PVR complex, which was consistent with the results of mutation analysis of the TIGIT/PVR interaction. In addition, H60 and V126 on PVR were important to the interaction between TIGIT and PVR, while other hot spot residues but not G131 showed little effects on TIGIT/PVR interaction. The G131A mutant in PVR was not conducive to the binding of TIGIT to PVR, which might be owing to that the steric hindrance generated by the side chain of alanine which affected the function of the "key" region of PVR 127 TFP 129 . In this study, hotspot residues were considered as potential candidates for designing high affinity PVR which required an in silico virtual screening and in vitro binding affinity verification.
The conformation of PVR was dynamically changed to facilitate TIGIT interaction. The loops in the protein played critical roles in complex formation. We identified several residues on the TIGIT binding interface or adjacent TIGIT/PVR interface which formed new interface during the MD simulations. These residues mainly located in the loop region near the binding interface and can be identified as potential candidates for mutagenesis in the design of high affinity PVR mutants. To prove our hypothesis, we obtained PVR mutants at these sites and then expressed the mutants on CHOK1 cells. Some mutants ( PVR S72W, PVR S72R, PVR G131V, and PVR S132Q) in the CC′ and FG loop located close to the interface enhanced the binding of PVR to TIGIT. Surprisingly, it was found that PVR T65E affected the expression of PVR, which might be due to the fact that charged Glu (glutamate) affected the stability of the molecular structure of PVR. These results are in line with our hypothesis that substitution of residues in the loop region closer to the interface contribute to TIGIT/ PVR interaction, and mutations at the interface are not beneficial to obtain high affinity mutants.
Antibodies targeting TIGIT/PVR pathway have achieved good effects in cancer treatment. The antibody targeting TIGIT using alone or combing other antibodies has achieved significant anti-tumor effects [27]. However, the poor tissue penetration and Fceffector functions limit the development of antibody drugs in cancer treatment. With relatively small molecular weight, PVR mutants without Fc segment might be utilized as protein drugs for cancer treatment, which needs further researches to test the anti-tumor effects of the mutant proteins both in vitro and in vivo. Also, high affinity PVR mutants might be used in combination with multispecific drugs targeting multiple immune pathways to improve the anti-tumor effects.
In summary, we provided a dynamic model for PVR at two different states by applying MD simulations. We also identified several high affinity PVR mutants by using in silico mutagenesis and cell assays. Four PVR mutants ( PVR S72W, PVR S72R, PVR G131V and PVR S132Q) enhanced binding affinity of PVR to TIGIT and induced more potent inhibitory effects on TIGIT overexpressed Jurkat cells.

Conclusion
Molecular dynamics simulations and in silico mutagenesis were applied to study the dynamics of PVR in both states and screen a list of high affinity mutants. The binding affinity of the mutants were measured by cell assay with FACS. Four mutants ( PVR S72W, PVR S72R, PVR G131V and PVR S132Q) with enhanced affinity for TIGIT could induce more potent inhibitory effects on TIGIT overexpressed Jurkat cells, which could contribute to design new candidates for TIGIT-targeting therapies.