Identification of small molecule modulators of class II transactivator-I using computational approaches

Abstract Major histocompatibility complex II (MHCII), a mediator of the innate and adaptive immune system, plays a central role in regulating inflammation and its progression. Class II transactivator (CIITA) is a master regulator of MHCII expression and controls antigen presentation followed by T-cell activation. Regulation of inflammation by modulation of CIITA has been suggested as a promising intervention for several disorders, including neuroinflammation, rheumatoid arthritis and other autoimmune diseases. This study aimed to (i) identify possible pharmacological agents which could bind to and inhibit isoform I of CIITA (CIITA-I) and (ii) determine their strength of interactions. The structure of CIITA-I isoform was predicted using phyre2 and refined via 3D refine. Loops were refined using ModBase, followed by quality assessment based on ERRAT value. The refined 3D structure was subjected to docking via Maestro (from Schrodinger) using glide module against small molecule databases. Molecules having the least glide score and favorable ADME properties were subjected to molecular simulation by GROMACS. We used the 3D refined structure of CIITA-I, with a score of 83.4% in ERRAT for docking studies. The ligand 4-(2-((6-oxo-4-phenyl-1,6-dihydropyrimidin-2-yl) thio) acetamido) benzamide (ZINC5154833), showed maximum glide score (−6.591) followed by N-[4-(3-oxo-3-{4-[3-(trifluoromethyl) phenyl] piperazin-1-yl} propyl)-1,3-thiazol-2-yl] benzamide (F5254-0161, glide score −6.41). Simulation studies using GROMACS showed F5254-0161 to have a more stable interaction with CIITA-I. Based on our analysis, we propose ZINC5154833 and F5254-0161 as potential modulators for CIITA-I. Communicated by Ramaswamy H. Sarma


Introduction
Interaction between major histocompatibility complex II (MHCII) and T cell receptors plays a significant role in the adaptive immune system. Expression of MHCII by antigenpresenting cells (APCs) is a highly regulated process. Class II transactivator (CIITA) is a master regulator of MHCII expression following the activation of helper T cells (Th cells) (Devaiah & Singer, 2013). Apart from MHCII, CIITA is known to regulate several other inflammatory and immunologically active genes, such as interleukin-4 (IL-4), IL-5 and IL-13 (Devaiah & Singer, 2013;Mori-Aoki et al., 2000).
CIITA belongs to the nucleotide-binding oligomerization domain (NOD) like receptor (NLR) protein family (Meissner et al., 2010). This family includes NOD1, NOD2, NLRC5 and NLRP3 (Nakamura, 2014). Four isoforms of CIITA (I to IV) are known in humans (Muhlethaler-Mottet et al., 1997). These isoforms are differentially expressed depending on the cell types and signaling molecules involved. Isoform I of CIITA (CIITA-I) is expressed by conventional dendritic cells. While CIITA-III is expressed by plasmacytoid dendritic cells, activated T cells, and B cells, CIITA-IV is expressed by interferongamma (IFN-c) stimulated cells (Monie, 2017). The expression of CIITA-II has not been described so far. CIITA proteins interact with other transcription factors such as regulatory factor X (RFX), cyclic AMP response element binding protein (CREB), and nuclear factor (NF-Y) to regulate the expression of MHC II (Zhu et al., 2000). Antigen presentation acts as a toll-gate between innate and acquired immunity, wherein CIITA plays a vital role. Constitutive expression of CIITA leads to overexpression of MHCII (Chou & Tomasi, 2008). A patient with a deficiency in MHCII expression was reported to have a homozygous intronic splice site variant in the CIITA gene (Hsieh et al., 2018). Also, patients with Type II bare lymphocyte syndrome, a hereditary immunodeficiency condition characterized by deficient MHCII expression, showed a loss of functional CIITA (Reith & Mach, 2001;Steimle et al., 2007). Additionally, experimental autoimmune encephalitis (EAE) rats fed with vitamin D resulted in reduced expression of CIITA and MHC II as well as a dramatic decrease in neuroinflammation and demyelination (Hochmeister et al., 2020). These studies suggest a functional correlation between CIITA and MHCII expression in various disease conditions. Several studies have also implicated CIITA in inflammatory conditions and infections. Inhibition of CIITA expression reduced neuroinflammation and neurodegeneration in a mouse model of Parkinson's disease (Williams et al., 2018). On the other hand, the activation of CIITA has been proposed to play an essential role in cancer immunotherapy (Le� on Machado & Steimle, 2021). CIITA has also been implicated in myocardial infarction, multiple sclerosis, and rheumatoid arthritis (Mart� ınez et al., 2007). Polymorphism in CIITA-IV has been associated with persistent hepatitis B infection (Zhang et al., 2007). A recent report suggests that strategies to increase the expression of CIITA may help prevent the spread of SARS-CoV-2 disease (Butowt et al., 2020). Hence, targeting CIITA could be a novel strategy to regulate inflammation.
So far, only a few studies have focused on the modulation of CIITA activity. Various approaches have been employed in this regard, including dietary vitamin D (Hochmeister et al., 2020), genetic manipulations (St€ uve et al., 2002;Williams et al., 2018), and indirect inhibition using cannabinoid receptor agonists (Cankara et al., 2021). However, pharmacological modulation of CIITA has not been reported yet. Therefore, targeting CIITA via small molecules would be relevant from a therapeutic point of view. To this end, analyzing the protein structure would be necessary to identify putative small molecule interactors.
CIITA possesses the NACHT domain and leucine-rich repeats (LRR) domains. Among CIITA isoforms, domains of CIITA-I have been studied in detail. In addition to NACHT and LRR, CIITA-I has a caspase activation and recruitment domain (CARD), activation domain, acidic domain and proline/serine/threonine (PST) domain (Le� on Machado & Steimle, 2021;Tyler et al., 2000). These domains regulate the transcription of genes involved in inflammation. Among them, 125 amino acid residues in the N-terminal region are acidic and are involved in binding transcription factors (Fontes et al., 1997). Up to 331 amino acid residues in the N terminal region form the PST domain are involved in IL-4 expression. Residues 408-857 (especially 405-414) constitute the GTP binding domain (GBD), involved in transport across nuclear pores to regulate the transcription of target genes (Cressman et al., 2001;Tyler et al., 2000). In addition, it also possesses an acetyltransferase domain, which plays a role in the expression of MHC II and IL-4 (Devaiah & Singer, 2013). Gene organization of only CIITA-I has been studied experimentally (Drozina et al., 2006;Lohsen et al., 2014). In the present study, computational approach were used to study the protein structure of CIITA-I. This was followed by identification of putative small molecules which could modulate its activity.
Here, the structure of CIITA-I was predicted and refined using computational tools. Further, pockets for binding of putative small molecules were identified. Next, molecular docking studies were performed to screen ligand libraries for putative modulators of CIITA-I, and druggability was predicted, resulting in two lead molecules. Finally, molecular dynamics simulations were performed with these two molecules to determine their biophysical interactions. The small molecules reported here could serve as putative pharmacological agents to target hyperinflammation by inhibiting CIITA-I.

Structure prediction
We retrieved the amino acid sequence of CIITA-I from Uniprot (human CIITA-I, Uniprot ID P33076-1). Since CIITA-I is a large protein containing 1130 amino acids, its 3D structure was predicted using Phyre2 v 2.0 (Kelley et al., 2015). Phyre2 predicts protein structure using the advanced remote homology detection method (Kelley et al., 2015). The structure was further refined using 3D refine (Bhattacharya et al., 2016), and the optimized structure was selected based on the 3 D refinescore. The server refines the predicted structure based on hydrogen bonding and atomic level energy minimization (Bhattacharya et al., 2016). The error in the residues (loops) of the refined structure was resolved using the ModBase server (Fiser et al., 2000;Fiser & Sali, 2003). The quality of the protein structure was assessed by SAVES server v 6.0 (https://saves.mbi.ucla.edu/). ERRAT quality score (assuming > 80% to be an acceptable score; Colovos & Yeates, 1993) and Ramachandran plot server (Anderson et al., 2005) were used for structure validation. ERRAT values are used for detecting errors in the 3 D structures of protein based on non-covalently bonded interactions between atoms, which is more reliable in modern protein structure validation methods along with considerations of Ramachandran plots (Colovos & Yeates, 1993). The refined and raw structure from phyre2 was visualized using UCSF chimera (Pettersen et al., 2004).

Identification of ligand binding pockets
The refined 3D structure (.pdb format) of CIITA-I was used to identify ligand binding pockets on the CASTp (Computer Atlas of Surface Topography of proteins, v 3.0) server (Tian et al., 2018). CASTp characterizes interior voids and surface pockets of proteins based on their computational geometry (Tian et al., 2018). Further, domains were identified using Interproscan (Apweiler et al., 2000).

Small molecule databases
Small molecule databases are a vital source for the virtual screening of lead molecules. Molecular databases from Life Chemicals (https://lifechemicals.com, 12,800 molecules), Apexbio (https://www.apexbt.com, 1449 molecules), and Selleckchem compound library (https://www.selleckchem. com, 4744 molecules) were used for high throughput screening. The molecules count and categorizations were retrieved as SDF files.

Molecular docking
Docking of small molecules with CIITA-I was performed using the glide module of Schr€ odinger Maestro v 12. Using the receptor grid generation module, a grid was generated for specified amino acids. Glide (Schr€ odinger Release 2022-1: Glide, Schr€ odinger, LLC, New York, NY, 2021;Friesner et al., 2004;Halgren et al., 2004) provides a platform for an efficient high-throughput virtual screening (HTVS) of millions of compounds (Friesner et al., 2006). Here, we used HTVS for approximately 18,000 molecules. The dimension of the grid was maintained at 20 Å. The interactive residues were identified using a 2D interaction diagram in Schr€ odinger.

ADME property
Prediction of absorption, distribution, metabolism, and elimination (ADME) properties would be of prime importance in understanding the bio-compatibility of potential drug molecules. Therefore, ADME properties were analyzed using Qikprop (Schr€ odinger Release 2022-1: QikProp, Schr€ odinger, LLC, New York, NY, 2021), and the reference values were obtained for the candidate ligands.

Simulation of native protein
The modeled three-dimensional structure of CIITA-I was subjected to simulation using GROMACS v 5.0.2, implementing the OPLS all-atom force field to understand the molecular stability. GROMACS is a high-performance software to calculate the dynamics of biomolecules. The OPLS4 implementation causes high accuracy in terms of small molecules solvation and protein-ligand interaction studies (Lu et al., 2021). Here, the protein was included in a cubic box, and a , and (c) changes observed following refinement. Black box -alpha helix from ARG941 to ALA953 in raw; from ARG941 to GLU950 in refined structure. Red boxes-extended loop at GLU897, SER781 and SER924 observed after refinement which is not present in the raw structure. The structure was predicted using Phyre2 and refined using ModBase web servers.
periodic body was applied with a dimension extending to 1.50 nm. The water molecules were neutralized with sodium ions. The system was subjected to energy minimization of 50,000 steps using the steepest descent algorithm followed by minimization through the conjugate gradient method. The minimized systems were then equilibrated at both NVT (constant number of particles, volume, and temperature) and NPT (constant number of particles, pressure, and temperature) for 100 picoseconds (ps) for position restraining. All the bond angles were constrained using the LINCS algorithm, whereas the SETTLE algorithm was used to constrain the geometry of water molecules. The temperature (at 310 K) of the system was regulated by using a V-rescale in which velocity is rescaled by a random factor appropriately chosen, and performance excels independent of thermostat and dynamics of properties (Bussi et al., 2007). V-rescale is a weak coupling method, and the pressure (at 1 atm) was set by using Parrinello-Rahman method, which is the pressure-induced structural transition that elaborates the metadynamics of pressure-induced changes (Marto� n� ak et al., 2003). The equilibrated systems were then set up for a 100 nanosecond (ns) of the simulation run with a time step of 2 femtoseconds (fs). Structural coordinates for every 2 ps were saved and analyzed using suitable tools in the GROMACS package. The minimum energy CIITA-I structure was retrieved.

Simulation of best two docked complexes
GROMACS v 5.0.2 was used for molecular dynamics simulations using the all-atom force field to understand the molecular stability of modeled three-dimensional structures of complexes. First, the protein and ligand topology were constructed. Then the LigParGen server was used to generate ligand topology (Dodda et al., 2017b(Dodda et al., , 2017aJorgensen & Tirado-Rives, 2005). The protein-ligand complexes were soaked in a cubic box with a dimension extended to 1.50 nm. A periodic boundary system was applied in all directions. The solvated systems were neutralized by replacing the water molecules with sodium ions. The remaining procedure was the same as described in section 2.6.1. Structural coordinates for every 2 ps were saved and analyzed using suitable tools in the GROMACS package. The minimum energy complex structures were compared using RMSD, RMSF, H-bond, Radius of gyration, clustering, binding energy, Free Energy Landscape (FEL) analysis, Principal Component Analysis (PCA), Solvent Accessible Area (SASA), and Molecular Mechanics Poisson-Boltzmann Surface Area (MMPBSA). FEL graphs were generated using Wolfram Mathematica software (https://www.wolfram.com), and dumped structure binding affinity was calculated using the prodigy server (Vangone et al., 2019).

Structure prediction
The tertiary structure of CIITA-I obtained from Phyre2 had an ERRAT score of 33% (Figure 1a), indicating a low confidence level. Hence, 3D refine was used to analyze all possible structures, and the highest score of 67% was obtained. The Ramachandran plot for these structures showed more disallowed regions than the raw structure. Since these scores are not considered suitable for docking, specific portions of the protein were further refined using the ModBase server until an ERRAT score of 83% (Figure 1b) was obtained. Superimposing the raw and refined structure revealed a root mean square deviation (RMSD) of 0.3. Positions of alpha helix and beta sheets differed between the two structures. In the refined structure, alpha helix was observed from ARG941 to GLU950, while it was present from ARG941 to ALA953 in the raw structure. Additionally, ALA972 to ILE979 formed beta sheets in the raw structure while it showed a simple loop followed by an alpha helix in the refined structure (Figure 1c).

Druggable site identification
The drug targeting site can be an active or allosteric site that influences protein function. Putative pockets for ligand interaction were identified by CASTp analysis (Figure 2) which identified six pockets (Figure 2b). Pocket 1 was chosen for docking as it possessed the largest surface area and volume. Interproscan revealed that a portion (residue 414-724) of this pocket contains the NACHT NTPase domain. Interestingly, this domain has been implicated in programmed cell death (Koonin & Aravind, 2000).

Molecular docking
Libraries containing anti-inflammatory, anti-apoptotic, and central nervous system (CNS) penetrating molecules (totally 18,993) were selected for docking studies due to their relevance for CIITA function.

ADME properties
ADME properties analyzed using Maestro's QikProp revealed a star value of 8 for ZINC5154833 and 1 for F5254-0161 (Table 2). Oral absorption values for these molecules were 85 and 100%, respectively. Values for the rule of five were zero for both the molecules. CNS penetrability values for these molecules were À 2 and À 1, respectively (Table 2), indicating their inability to cross the blood-brain barrier. Further, these molecules were analyzed to understand the dynamics of their interactions with CIITA-I.

Molecular dynamics and ligand interactions
Simulation studies with a 100 ns interval were performed using GROMACS to understand the dynamics of the ligand-protein interactions. ZINC5154833 and F5254-0161 bound to CIITA-I were labeled as complex 1 and 2, respectively.
Stability and structural changes in ligand-protein interactions were analyzed using RMSD (root mean square deviation) and RMSF (root mean square fluctuations). RMSD and RMSF are first-class analyses of the stability of ligand-protein complexes. In RMSD, deviations in the backbone of proteins from their initial time to the end of the given simulation time are calculated. A lesser difference in RMSD shows higher stability (Aier et al., 2016). Here, deviations in the backbone with respect to time were analyzed for the two complexes and the native protein. Our results showed �1.5 nm RMSD for all three entities (Figure 4a). Notably, RMSD showed many deviations across time for protein alone. Though the RMSD was less in complex 1 than complex 2, the average displacement was not constant, with an initial peak that lowered after 25 ns and remained unstable ( Figure  4a) for the rest of the duration of simulation. RMSD for these ligands also showed that ligand 2 (F5254-0161) has lower deviations than ligand 1 (ZINC5154833) (Supplementary    1). In complex 2, the RMSD stabilized between 25 and 50 ns and peaked at 1.63 nm between 51 and 60 ns. A stable value followed this for the rest of the period, indicating a relatively stable configuration. RMSF graph denotes fluctuations in the displacement of residues participating in the interaction with the ligand with respect to the native protein (Mart� ınez, 2015). Maximum fluctuations were observed for residues 408, 440, 471, 487, 586 and 760 in complex 1 compared to complex 2 and the native protein (Figure 4b). Least fluctuations were observed for residues 502, 697, 951 and 969 in complex 2 (Figure 4b). For complex 2, residues towards the C-terminal (816 À 1130) showed higher fluctuations compared to complex 1. Notably, residues 423 and 425 showed fewer fluctuations over time in both complexes. Overall, complex 2 showed lower fluctuations among the three entities. The radius of gyration indicates the structure compactness and size of protein molecules (Arnittali et al., 2019). Specifically, our analysis revealed the compactness of complexes across time. The interaction of ligands resulted in the convergence of the protein in both complexes. In contrast, the native protein showed divergence (Figure 4c). Complex 2 exhibited profound convergence, which remained intact across the analysis period. For complex 2, the radius of gyration started around 3.5 nm and fell below 3.2 nm, much less than complex 1 or the native protein (Figure 4c). This showed that the interaction of F5254-0161 with CIITA-I could be more stable than ZINC5154833.
Solvent accessible surface area (SASA) enumerates the bimolecular surface accessible to solvents (Kamaraj & Purohit, 2013). With time, a decrease in SASA was observed among the complexes and the native protein (Figure 4d). However, the reduction in SASA of complex 1 was more than complex 2 and native protein. For complex 2, SASA initially peaked at 480 nm 2 (higher than complex 1 and native protein), then stabilized after 25 ns and decreased again after 40 ns. Overall, complex 2 had a higher SASA across the 100 ns duration than complex 1 and native protein.
The number of hydrogen bonds and GROMACS energies indicate the strength of ligand-protein interactions over time. Figure 5a represents the accountable hydrogen bonds of complexes 1 and 2. For complex 1, the number of hydrogen bonds varied over the 100 ns period-seven at 80 ns; five at 0, 24, 46 ns, and several times between 51 and 100 ns; none at 10, 16, 25, 34, 42.5, 45 and 58 ns. Complex 2 showed less variation in the number of hydrogen bonds over time.  The number of hydrogen bonds for complex 2 varied between 1 and 4 for most of the 100 ns duration. However, no hydrogen bond was observed at 31, 39 42.5, 68, 69, and 81 ns for complex 2 (Figure 5a). GROMACS interaction energies revealed relatively stable energies for complex 2 (between À 2.985e 6 and À 2.995e 6 kJ/mol) over the period of analysis (Figure 5b). In contrast, complex 1 exhibited variation in the interaction energies. Though complex 1 showed overall lower energy than complex 2, the distribution of energies was more compact for complex 2 (around �2.99e 6 kJ/mol).
Principal component analysis (PCA) allows us to analyze the motion of proteins and their complexes in solution (David & Jacobs, 2014). Figure 6 shows the PCA of the complexes and native protein. The motion of the native protein in 2D trajectories was found to be less, especially at eigenvector 2. The native structure reached positive eigenvector 2 for a brief period in the middle of the analysis, but remained at a negative value for most of the time (Figure 6). Both complexes showed relatively higher motion compared to the native protein. In particular, complex 1 attained a value of �20 nm in both eigenvectors. Complex 2 attained a value of �25 nm in eigenvector 1 and �10 nm in eigenvector 2. These results suggest that interaction causes higher protein motion in 2D trajectories than protein alone for both complexes.
Free energy landscapes (FEL) are used to analyze protein-ligand binding affinities. Overall, complex 1 showed a higher number of hydrogen bonds compared to complex 2 (Figure 7). The binding affinity was observed to lie between À 8.3 and À 8.6 kcal/mol (Supplementary Table 2) for complex 1 (Figure 7a, column b), and between À 9.6 and À 9.7 kcal/mol (Supplementary Table 2) for complex 2 (Figure 7b, column  b). 2D interaction analysis of complex 1 showed hydrogen bonds with GLY423, GLU502 and GLN619, along with a few hydrophobic interactions (Figure 7a column b). On the other hand, complex 2 showed mostly hydrophobic and electrostatic interactions with fewer hydrogen bonds. Notably, both complexes showed hydrogen bonds with GLY423 in all three conformations (Figure 7a, b). These results suggest that the ligand F5254-0161 has a higher affinity for CIITA-I compared to ZINC5154833.
Decomposition analysis on per-residue affinity with the ligand enumerates the contribution of residues to the total binding energies (Chaudhary & Aparoy, 2020). The MMPBSA highlighted the amino acids and their Delta values of normal generalized born (GB) model energies-total decomposition contribution (TDC), side-chain decomposition contribution (SDC) and backbone decomposition contribution (BDC) were calculated at the binding sites ( Figure 8). Similar energy values were observed for TDC, SDC and BDC for both complexes indicating that side chain energies are the ones mainly contributing to the total energies, as BDC energies were zero for both complexes. The least energies were observed for residues 428,454,502,503,548,615,711 and 742 for both complexes. In addition, for complex 2, the least energies were observed for residues 619, 641 and 692. However, the standard deviation for these residues was higher compared to residues 422, 423, 425 and 612, which showed higher energies.

Discussion
In this study, we analyzed the structure of CIITA-I, a master regulator of the adaptive immune response, and identified two ligands that could potentially modulate its functions. To the best of our knowledge, this is the first report on small molecule modulators of CIITA.
The crystal structure of CIITA has not been determined yet. A model predicted using Alpha Fold is available on Uniprot (Uniprot.org, AF-P33076-F1). However, it contains several unstructured regions. Also, there are unstructured isolated residues between several domains. Notably, the regions corresponding to the activation domain of CIITA-I (MET1 to LYS330) were associated with a low confidence score per residue. Owing to these observations, we performed a structural prediction of CIITA-I, followed by further refinement. Notably, the refined structure proposed here has no unstructured regions and also shows reduced errors in the Ramachandran plot. However, X-ray crystallography will be required for further confirmation of this structure.
Subsequently, putative pockets on the predicted structure were identified. The pocket having the highest surface area and volume was chosen for further docking studies. This pocket lies within the GTP binding domain (GBD) of CIITA-I (Figure 2), which is vital for its transport across the nuclear pore (Tyler et al., 2000). Small molecules targeted to this pocket could potentially inhibit the binding of GTP, prevent translocation of CIITA-I to the nucleus and thereby modulate the expression of its target genes, including MHC II. Interestingly, both the ligands identified here show an affinity for GBD. A previous study suggested that inhibition of CIITA translocation across the nuclear pore by leptomycin could enhance MHCII expression by sequestering it in the nucleus (Cressman et al., 1999). Thus, these ligands could potentially inhibit as well as induce MHC II expression. However, this would need further experimental confirmation.
Following structure prediction and pocket identification, docking analysis was performed using Maestro (Schrodinger), considered the most accurate molecular docking software (Lather et al., 2018). Further validation was also performed by redocking with other programs mentioned in Supplementary Table 1. As a result, two potential smallmolecule modulators of CIITA-I were identified. Furthermore, ADME properties suggested these molecules to be highly permeable through the gut-blood barrier, indicating their suitability for oral administration in humans.
A similar workflow has been carried out in other studies to identify modulators of protein function. For example, Pang et al. (2022) targeted selective glucocorticoid receptor modulators (SGRM) by HT-15 to downregulate the expression of inflammatory genes. Notably, this study also used glide modules and Qikprop to screen ligands for SGRM.
The ligand ZINC5154833 is a known anti-inflammatory molecule (life chemicals database). This molecule inhibits leukotriene A4 (LTA4) hydrolase, an enzyme implicated in hyperinflammation, by activating leukotriene B4 (Stsiapanava et al., 2017). LTA4 hydrolase contains a metal-binding site (especially for zinc) and can form hydrogen bonds and hydrophobic interactions. Our results also suggest that the ligand forms extensive hydrophobic interactions. Further, it was found to interact with CIITA-I via pi-pi stacking interaction at TYR428 (Figure 3a). With regard to drug-likeness, ZINC5154833 satisfies Lipinski's rule of five, rule of three, and intestinal absorbability, highlighting its translational value. However, this ligand seems to lack permeability through the blood-brain barrier. Thus, further chemical modifications would be required if one aims to target CNS inflammation using this ligand.
The other potential ligand of CIITA-I, F5254-0161, is an inhibitor of phosphatidylinositide 3-kinases (PI3K), which is involved in several cellular functions, including inflammation and immune responses (Stark et al., 2015). Our results suggest a novel mechanism of action (via CIITA-I) for the modulation of inflammatory processes by F5254-0161. Furthermore, this ligand also showed a good absorptive score in ADME analysis, indicating its suitability for clinical use.
CIITA is an intracellular protein whose activity depends on translocation across the nuclear membrane. Therefore, drugs targeting such molecules need to cross the plasma membrane. The potential CIITA-interactors identified in the present study are known to inhibit LTA4 hydrolase and PI3K, both of which are intracellular proteins (Brock et al., 2001). Hence, these ligands would be suitable for targeting CIITA intracellularly and modulating its functions. However, further experimental validation will be required in this regard.
The present study aimed to identify ligands that could modulate inflammation by targeting CIITA, a regulator of MHC-II expression. Regulation of MHC II-mediated signaling is an important strategy for treating hyperinflammation. Since CIITA is known to activate MHC II and other pro-inflammatory genes, targeting CIITA would be an effective clinical strategy to treat hyper-inflammation. In line with this, simvastatin, currently used for treating multiple sclerosis (Mach, 2004;Markovic-Plese et al., 2008), is known to down-regulate CIITA expression. Other studies targeting CIITA have employed dietary supplements (Hochmeister et al., 2020) or siRNA (Williams et al., 2018). Here, we propose two potential drug molecules to manage hyperinflammation by targeting CIITA. Future in vitro and in vivo studies could determine their efficacy against hyperinflammatory conditions.

Summary
Regulating inflammation is a major goal in managing several diseases. As CIITA plays an important role in inflammatory processes, various approaches have been used to target this protein. Here, we used in silico tools to predict the structure and potential small-molecule interactors of CIITA-I. Our analysis identified two molecules, ZINC5154833 and F5254-0161, which could potentially target CIITA-I and help modulate inflammation. In addition, these molecules showed pharmacological properties suitable for clinical applications. Future experimental studies would be valuable to determine the safety and efficacy of these molecules for regulating inflammatory conditions.