International Journal of Peptide Research and Therapeutics https://doi.org/10.1007/s10989-018-9730-9

Design of Tat-Activated Cdk9 Inhibitor Yunjie Zhao1,2 · Hao Chen2,3 · Chenghang Du3 · Yiren Jian3 · Haotian Li4 · Yi Xiao4 · Mohammed Saifuddin5 · Fatah Kashanchi5 · Chen Zeng3 Accepted: 18 June 2018 © Springer Nature B.V. 2018

Abstract Virus mutates quickly and develops drug resistance over time if targeted directly. A possible alternative for drug design is to target cellular proteins that the virus needs. To be viable, the drug ought to inhibit the cellular protein only when bound by viral protein. We use a particular virus–host complex to illustrate how to identify such an inhibitor. HIV-1 uses its viral protein Tat to hijack a cellular protein complex Cdk9/Cyclin T1, termed positive elongation factor b (p-TEFb), to enhance its viral transcription. Tat binds to both Cdk9 and Cyclin T1 and thus creates a second pathway between Cdk9 and Cyclin T1 that is distinct from the Cdk9/Cyclin T1 interface. Dynamical network analysis based on molecular dynamics simulations reveals a pocket on Cdk9 that becomes allosterically correlated with the interface via the second pathway. Computational docking simulations indicate a noticeable weakening of the interface formation of Cdk9 and Cyclin T1 upon binding of a small molecule F07#13 to the pocket and Tat to the complex. We verified experimentally via site mutagenesis that F07#13 indeed targets this pocket and diminishes the kinase activity of Cdk9 in the presence of Tat. Together with previous experiments that showed little effect of F07#13 in the absence of Tat, the proposed computational framework therefore provides some insights on how to design antiviral drugs with reduced risk of drug resistance. Keywords  MD simulation · Network · Correlation · Docking · Community

Introduction

Electronic supplementary material  The online version of this article (https​://doi.org/10.1007/s1098​9-018-9730-9) contains supplementary material, which is available to authorized users. * Fatah Kashanchi [email protected] * Chen Zeng [email protected] 1



Institute of Biophysics and Department of Physics, Central China Normal University, Wuhan, China

2



School of Chemistry and Environmental Engineering, Wuhan Polytechnic University, Wuhan, China

3

Department of Physics, The George Washington University, Washington, DC, USA

4

Department of Physics, Huazhong University of Science and Technology, Wuhan, Hubei, China

5

Laboratory of Molecular Virology, George Mason University, Manassas, VA, USA



Approximately 37  million people infected with HIV-1 are living in the world, and only 40% of them are receiving Antiretroviral Therapy (Bigna et al. 2016). Majority of the standard anti-HIV drugs directly target virus particles. Although direct acting drugs have proven to be highly successful in mitigating the disease, permanently eliminating HIV-1 from the infected individuals has been challenging due to the emergence of drug resistance. As obligatory parasites, viruses are dependent on the host cell machinery for their lifecycle. As such, besides viral receptor/co-receptors, HIV-1 interacts with numerous intracellular and extracellular proteins, and many of them are critical for mRNA transcription, protein expression, and assembly of progeny viruses (Ahlquist et al. 2003; Prussia et al. 2011). Since the appearance of HIV-1 resistant mutations resulted from constant drug pressure on the viral component(s), targeting host proteins that are essential for HIV lifecycle have been considered important alternatives for preventing drug resistance. Cellular proteins often possess higher genetic barrier than

13

Vol.:(0123456789)



the viral proteins to development of resistance under drug selection pressure (Heredia et al. 2015). Computational structure-based drug design (CSBDD) is proven to be an effective method for accelerating the drug development process. CSBDD relies on the ability to determine and analyze 3D structures of relevant biological targets. Indeed, these structure-based computer techniques have been a powerful tool to unravel HIV biology and to analyze and understand their structural and functional influence in drug resistance. With the combination of increasing computational power (Zhao et al. 2011, 2012, 2013, 2015; Wang et al. 2015; Xing et al. 2016; Chen et al. 2014; Li et al. 2012, 2017) and the availability of structural and biological data of protein complexes, in silico experiments have played a major role in exploiting the interactions between drugs and viral targets, such as HIV protease, reverse transcriptase, and integrase. Furthermore, the identification and validation of new targets have also significantly benefited from several such techniques including computer modeling for protein–ligand interactions and simulation, molecular docking and visualization, 3-D database techniques, and free-energy perturbation (Reynolds 2014; Sliwoski et al. 2014; Forli and Olson 2015). In particular, molecular docking has been widely used for drug target identification. The traditional strategy seeks and predicts all possible binding proteins as well as the drug binding sites on the given structures (Lengauer and Rarey 1996; Ou-Yang et al. 2012). In contrast to the above traditional computational approach for targeting viral protein directly, we screen host protein complex that a virus needs to identify suitable drug target. For instance, HIV-1 must recruit the host protein Cdk9 for its viral transcription. Cdk9 is such a critical enzyme for stimulating transcription elongation of most protein coding genes and is also associated with many cancer types (Morgan 1995; Krystof et al. 2012; Krystof and Uldrijan 2010). The positive transcription elongation factor, P-TEFb, which is comprised of Cdk9 paired with Cyclin T1, plays a key role in regulating the elongation phase of transcription by RNA polymerase II. HIV-1 Tat protein recruits the host protein complex Cdk9/Cyclin T1 onto a TAR RNA stem-loop structure on the HIV-1 long terminal repeat promoter to promote processive elongation of HIV-1 transcription. In addition to Cdk9/Cyclin T1, HIV-1 Tat recruits transcription factor NFkB and also stimulates phosphorylation of SP1 transcription factor to induce basal transcription of HIV-1. Inhibition of Cdk9 by drugs or peptides will selectively interfere with Tat-mediated HIV-1 transactivation (Kumari et al. 2014). Since P-TEFb is also associated with many cancer types (Morgan 1995; Krystof et al. 2012; Krystof and Uldrijan 2010). There has been intensive search for Cdk9 inhibitors for therapeutic applications (Canduri et al. 2008; Nemeth et al. 2011). Currently, there are two approaches to inhibit

13

International Journal of Peptide Research and Therapeutics

active Cdk9. One is targeting ATP-binding pockets. Several small molecule inhibitors, such as flavopiridol (Baumli et al. 2008), CR8 (Bettayeb et al. 2010), and DRB (Baumli et al. 2010), have been developed towards applications in cancer therapy. Unfortunately, the ATP-binding pocket of Cdk9 shares sequence homology with a family of Cdk proteins (Malumbres et al. 2009). These Cdk9 inhibitors are unselective, targeting several Cdks and other kinases, thus causing significant off-target associated toxicity. The other approach is breaking up the Cdk9/Cyclin complex by targeting Cdk9/ Cyclin interface. While it is difficult to use small molecule inhibitors to compete directly with the much larger interface area, it has been shown that there also exist non-catalytic pockets on Cdks away from the interface for small molecules or peptides to bind, which can at least partially break up the Cdk/Cyclin complex. These allosteric binding pockets are far less conserved than that of the ATP-binding pockets and may thus lead to more specific inhibitors against different Cdks (Betzi et al. 2011). It is crucial for an inhibitor to be Cdk9 specific to reduce its off-target effect. More importantly, the inhibitor should also be infected-cell specific to reduce its impact on healthy and uninfected cells. A recently solved crystal structure of the Tat-P-TEFb complex shows that Tat interacts predominantly with Cyclin T1 but also with the T-loop of Cdk9. Thus, Tat further stabilizes the Cdk9/Cyclin T1 complex by forming a new pathway between Cdk9 and Cyclin T1 distinct from the Cdk9/Cyclin T1 interface (Van Duyne et al. 2013). The hijacking of a host protein complex alters the dynamics. We exploited the changes to search for allosteric inhibitors that break up the complex only when this new pathway is formed. Allosteric drug discovery has been an active area in drug design (Lewis et al. 2008; Ivetac and McCammon 2010) that targets sites far away from the known active sites. The allosteric pockets are typically not obvious from the static structures. Since molecular dynamics (MD) simulations can capture the correlated dynamics of up to millions of atoms in the biological molecules, MD approach are thus widely used for allosteric drug discovery (Dror et al. 2012; Karplus and McCammon 2002; Ivetac and McCammon 2012). Our methodology is inspired by dynamic network analysis of MD simulations (Sethi et al. 2009), which has been instrumental in identifying allosteric pathways and explaining the structural dynamics (del Sol et al. 2009; Rivalta et al. 2012; Vanwart et al. 2012; Sethi et al. 2013; Fuglestad et al. 2013). In this work, we first construct the dynamic contact network from the MD simulations of Cdk9/Cyclin and Cdk9/ Cyclin/Tat complexes to probe their correlated motions. We then compare and contrast the two networks to identify clusters of non-interface residues on Cdk9 whose correlations with the Cdk9/Cyclin interface are nonetheless greatly enhanced upon Tat binding. The clusters, if any, thus

International Journal of Peptide Research and Therapeutics

become potential binding pockets to screen for inhibitors that may impact the interface formation only in the presence of Tat binding. A small molecule inhibitor F07#13, obtained previously by automated docking screening simulations (Van Duyne et al. 2013), are shown here to bind to the identified pocket and disrupt the Cdk9/Cyclin complex formation and its kinase activity in the infected cells. This study demonstrates a novel allosteric drug discovery approach that targets the host protein complex only in infected cells.

Results In order to promote productive of HIV mRNAs, the viral protein Tat needs to hijack the host protein complex P-TEFb that consist of Cdk9 and human Cyclin T1. The previous experimental studies have shown that Tat requires the interactions not only with Cyclin T1 subunit but also with the T-loop of the Cdk9 subunit (Tahirov et al. 2010). Therefore, there are some conformational changes of the P-TEFb protein structure induced by Tat hijacking. In order to observe this, we have done MD simulations using Cdk9/Cyclin and Cdk9/Cyclin/Tat as starting structures in all-atom level (see “Materials and Methods” section for details). MD simulations show differences of the dynamical behavior when comparing Cdk9/Cyclin and Cdk9/Cyclin/Tat complexes. We observed that some subunits are more ordered in Cdk9/ Cyclin/Tat than those of in Cdk9/Cyclin. Tat binding alters fewer fluctuations in three regions (Fig. 1 and Supplementary Fig. 1). Box 1 (colored in blue) is nearby the wellknown ATP-binding pocket. Box 2 (colored in red) is the T-loop of Cdk9, which contacts Tat directly. Box 3 (colored in orange) is the region nearby the T-loop. As expected, Tat binding does not change much of the fluctuations of

Fig. 1  a Ribbon representation of the Cdk9/Cyclin/Tat structure. Cdk9 is colored in green, blue, red, and orange. Cyclin T1 is colored in light blue. Tat is colored in magenta. b Root-mean-square deviation (RMSD) fluctuations of Cdk9/Cyclin (red line) and Cdk9/Cyclin/Tat (blue line) structures, calculated with respect to the average

the Cdk9/Cyclin interface residues. Activity of CDK9 is dependent on binding to a regulatory Cyclin subunit and is further regulated through association with other macromolecules. The stability of Cdk9/Cyclin interface is critical for the functions, and, Tat needs the functional Cdk9/ Cyclin complex to produce HIV mRNAs, which means Tat still needs the stability and strong interactions between the Cdk9 and Cyclin structures. These observations corroborate the following questions. Do these residues of different box regions play any crucial roles upon Tat binding? Can we use the information of the induced conformational changes for solving drug design problem?

Correlated Motions in Cdk9 The analysis of the correlations was calculated from the final 20 ns MD simulations (see “Materials and Methods” section for details). If any two heavy atoms of the residues were < 4.5 Å for 75% of the snapshots during 20 ns trajectories, the correlation values were kept, otherwise the correlation values were considered to be zero. If the residues move in the same directions in most frames, the motions are considered as correlated, as positive values. If the residues move in the opposite directions in most frames, the motions are defined as anti-correlated, as negative values. If the correlation values between the residues are close to zero, then the motions are defined to be uncorrelated. Figure 2 shows our results for correlation analysis of Cdk9 (Fig. 2a) and differences with respect to the corresponding values for the Tat bound complex (Fig. 2b). Large values ­(Cij > 0.7) are obtained for the highly coupled in the Cdk9/Cyclin MD simulation, including residues with common secondary structure elements (near off-diagonal elements, Fig. 2a). Some secondary structure elements are

structures during the MD simulations. Box  1 (colored in blue) is nearby the well-known ATP-binding pocket. Box  2 (colored in red) is the T-loop of Cdk9, which contacts Tat directly. Box 3 (colored in orange) is the region nearby the T-loop

13



International Journal of Peptide Research and Therapeutics

Fig. 2  a Correlation analysis of the motions of Cdk9 during the Cdk9/Cyclin MD simulation. b Comparison of generalized correlation values for Cdk9 between Cdk9/Cyclin/Tat and Cdk9/Cyclin MD simulations. The blue, red, orange boxes are the same regions those are described in Fig. 1

also correlated with strong interactions (­ Cij around 0.5–0.6, the secondary structure elements are described in Supplementary Fig. 2). For example, the residues at 𝛼E (residues 122–143) and 𝛼J (residues 305–311) also show relatively high correlations in the MD simulation (left dash box in Fig. 2a). Similar results are found for residues at 𝛼H (residues 274–284) with 𝛼F (residues 208–225) (middle dash box in Fig. 2a), and 𝛼G (residues 234–245) (right dash box in Fig. 2a). As expected, the residues of the colored boxes described in Fig. 1 are poorly correlated with other residues including the Cdk9/Cyclin interface residues (blue, red, and orange boxes in Fig. 2a). However, the MD simulations show that these subunits are much more ordered upon Tat binding. Therefore, we asked what are the consequences of the correlation changes after binding the Tat? The correlation matrix differences between Cdk9/Cyclin and Cdk9/Cyclin/Tat are shown in Fig. 2b. Comparison of correlated motions of Cdk9 residues between Cdk9/Cyclin and Cdk9/Cyclin/Tat MD simulations reveals increased of correlations, induced by Tat binding. In particular, large increases of correlations are observed between the residues of 𝛽4 (residues 81–87), 𝛽5 (residues 98–104), T-loop, 𝛼F (residues 208–225), 𝛼G (residues 234–245) and the Cdk9/Cyclin interface residues. The residues in 𝛽4 (residues 81–87), and 𝛽5 (residues 98–104) are close to ATP-binding pocket. Note that the ATP-binding pockets are highly conserved. These residues of Cdk9 share sequence homology with other Cdk family members. The drugs those targeting ATP-binding pocket will shut down all normal and infected Cdk family members. Therefore, we focused on the residues of T-loop and nearby T-loop those with increased correlations upon Tat binding.

the detailed analysis and how the residues of Cdk9 affect the interface of Cyclin T1 residues. Figure 3 shows results for correlation analysis of Cdk9/ Cyclin and differences with respect to the corresponding values for the Tat bound complex. Here, we focused on the T-loop (red box) and residues nearby T-loop (orange box) to the interface residues of Cyclin T1. As we expected, there are some increased correlations in the colored boxes regions. The T-loop (red residues in Fig. 1a) and residues nearby the T-loop (orange residues in Fig. 1a) show higher correlations with the residues at H3 (residues 79–95), H4 (residues 100–113), and H5 (residues 123–144) in Cyclin T1 structure, most of which are the Cdk9/Cyclin interface residues (the secondary structure elements of Cyclin are described in Supplementary Fig. 3).

Correlated Motions to Cdk9/Cyclin Interface Previous studies show that the stability of the Cdk9/Cyclin interface is required for Tat hijacking. In order to understand the motion changes upon Tat binding, we also characterized

13

Fig. 3  Comparison of generalized correlation values for Cdk9 and Cyclin residues between Cdk9/Cyclin/Tat and Cdk9/Cyclin MD simulations. The red and orange boxes are the same regions that are described in Fig. 1

International Journal of Peptide Research and Therapeutics

In summary, the results of correlation analysis indicate that Tat binding induces changes in the correlation motions of some residues at Cdk9. Especially in the T-loop and the residues nearby T-loop, significant increases of correlations are observed to the Cdk9/Cyclin interface. More interestingly, the nearby T-loop residues are more than 25 Å from the Cdk9/Cyclin interface but correlated to the Cdk9/Cyclin interface upon the Tat binding. These suggest that there are long-distance coupling motions between residues nearby T-loop and Cdk9/Cyclin interface upon the Tat binding.

Community Networks The community analysis shows the allosteric pathways changes between Cdk9 and Cyclin upon Tat binding. As described in the method of community analysis, the residues belonging to the same community are much more strongly connected than those belonging to other communities in the network. This means the nodes in the same community can ‘traffic’ to each other much easier through multiple ‘highways’. The Girvan–Newman algorithm splits the network of the Cdk9/Cyclin and Cdk9/Cyclin/Tat into small communities (Fig. 4). The width of the bonds connecting the communities is the betweenness describing the information flow. A wider link means more communications between the two communities.

In the Cdk9/Cyclin complex, C9 located around 𝛼C (residues 61–72 in Cdk9) contains the Cdk9/Cyclin interface residues, acting like a hub, which involves most of the information flow ‘traveling’ from Cdk9 (C3 and C2) to Cyclin (C7). However, in the Cdk9/Cyclin/Tat network, the community C9 is merged into C′5 and C′4. The community C′4 contains most of the information flow those ‘traveling’ from Cdk9 (C′3 and C′2) to Cyclin (C′7). More interestingly, while the community C′3 is mostly preserved as C3, there is a much larger information flow from C′3 to C′2 than that of C3 to C2. The community C3 is kind of isolated in the Cdk9/Cyclin network. However, the C′3 becomes correlated with others, especially C′4 and C′7 that contain residues at Cdk9/Cyclin interface. Tat binding builds a bridge connecting the T-loop and Cyclin, stabilizes the flexible T-loop regions, alters the community network of C3 into C′3, which now is much more correlated to Cdk9/Cyclin interface residues. Taken together, the results reveal that this area may be a potential drug target will break up the host–protein or weaken the Cdk9/Cyclin interface upon binding to the HIV viral protein, Tat, to halt the transcription elongation and thus the eventual replication.

Detecting Cavities Using Docking Simulations Using virtual screening, we previously used Cdk2 modeling of a Tat inhibitor, F07#13, which is a close homolog of

Fig. 4  a, b MD representative snapshots of Cdk9/Cyclin complex (a) and Cdk9/Cyclin/ Tat complex (b). c, d Colored community networks of the Cdk9-Cyclin (c) and the Cdk9/ Cyclin/Tat (d) complexes

13



Cdk9, and only focused on the interface-binding pocket of the Cdk. In order to observe whether the F07#13 could target some residues around the community C′3 described above, we ran global docking simulations using F07#13 and Cdk9 structures to observe whether there are cavities available around C′3 position. The docking protocol is described in “Materials and Methods”. A cluster analysis was performed and the threshold was set to 3.0 Å. We then focused on the top 10 largest clusters and located the binding pocket for the lowest docking energy conformation within each cluster by manual inspection. One well describe pocket is the well-known ATP-binding pocket. However, as we described above, the ATP-binding pocket is not suitable for the drug binding. Therefore, we focused on the potential pockets around community C′3. According to the docking results, there are three structures shown in Fig. 5.

Comparison Correlation Analysis upon Binding F07#13 In the proposed hypothesis, the drug F07#13 could weaken or break up the host–protein interface using allosteric interactions upon Tat binding. Breaking up the host–protein interface needed by host–protein/virus complex could possibly reduce the risk of HIV-1 drug resistance. In order to verify this hypothesis, we have done MD simulations using the docking results as the starting structures. The network representing the structures after the ligand F07#13 binding at Pocket 1, Pocket 2, and Pocket 3 are obtained from the final 20 ns MD simulations. The comparison of the generalized correlation values between Cdk9/Cyclin/Tat and those binding with F07#13 at Pocket 1 (Supplementary Fig. 4A), Pocket 2 (Supplementary

Fig. 5  Ribbon representation of the Cdk9/Cyclin/Tat structure. Cdk9 is colored in green. Cyclin T1 is colored in light blue. Tat is colored in magenta. The region that colored in orange is the community C′3. F07#13 (colored in red) can target three different cavities labeled as Pocket 1, Pocket 2, and Pocket 3 in the main text

13

International Journal of Peptide Research and Therapeutics

Fig. 4B), and Pocket 3 (Supplementary Fig. 4C) are shown in Supplementary Fig. 4. Note that the features in the black rectangle boxes represent the regions for Cdk9/Cyclin interface residues. The light blue color indicates a decrease of correlation values from − 0.3 to − 0.1. As we described above, large values (> 0.7) are highly coupled residues at the same secondary structure elements. The correlation values of functional Cdk9/Cyclin interface are relatively high around 0.5. These decreases of correlations at interface regions imply that the F07#13 have weakened and have the potential ability of breaking up the host–protein (Cdk9/Cyclin) interface when binding at Pocket 1, Pocket 2, and Pocket 3. We also have performed MD simulations using Cdk9/ Cyclin with F07#13 binding at cavities 1, 2, and 3 as starting structures to observe whether this drug would weaken the host–protein interface without Tat binding. The yellow and light blue color indicates correlations that vary from − 0.1 to 0.1 (Supplementary Fig. 5). These results indicate that the drug could weaken and potentially dissociate the host–protein interface when binding to HIV Tat.

Dissociation of Cdk9 Away from the HIV‑1 Transcription Complex We next asked whether F07#13 was able to dissociate Cdk9 away from the HIV-1 transcription complex. We have previously performed similar experiments with a series of Tat peptide analogs containing various amino acid substitutions in the core domain and could effectively inhibit Tat transactivation of the HIV-1 promoter (Van Duyne et al. 2013). We searched for synthetic small molecules (1st and 2nd generation drugs) that could mimic the inhibitory peptide effects and using 3-pharmacophore models, we identified inhibitors that effectively bind to Cdk9 and inhibit HIV-1 transcription both in vitro and in vivo (Van Duyne et al. 2013). Here we looked at whether epitope-tagged Cdk9 (either wild type of the S175A mutant) was able to phosphorylate either of the two natural substrates including RNA Pol II-CTD or histone H1. We chose J1-1 cells as they are infected T-cells and produce live virus as previously described (Butera et al. 1994; Ryckman et al. 2002; Fernandez and Zeichner 2010; Narayanan et al. 2013). These cells also contain wild type Tat as well as, p-TEFb complex that are kinase active (Narayanan et al. 2012). We transfected J1-1 cells with Flag-Tag Cdk9 DNA (20 µg; either wild type or S175 A mutant; generous gift of Dr. S. Nekhai at Howard University) and incubated at 37 °C for 24 h. Subsequently, cells were treated with F07#13 at various concentrations (0.1, 1.0, and 10 µM). Samples were further incubated for 48 h and subsequently processed for kinase assay. Nuclear extracts were prepared and IPed with anti-flag antibody overnight; washed and used for in vitro kinase assay. IgG immunoprecipitation serves as a control.

International Journal of Peptide Research and Therapeutics

Results in Fig. 6 indicate that F07#13 is capable of inhibiting kinase activity of the wild type Cdk9 more efficiently compared to the mutant S175A Cdk9. A pronounced effect was observed at 1 µM with either CTD or histone H1 as a substrate, indicating that p-TEFb/Tat complexes that contain wild type S175 is the target of the F07#13 inhibitor. Collectively, this data further emphasizes the significance of the Pocket 3 for Tat or F07#13 binding.

Discussion The computation framework proposed in the present study aims to target the cellular proteins needed by virus as an alternative means to design antiviral drugs with reduced risk of drug resistance. In fact, high throughput techniques have given us the complete landscape of virus–host interactomes (Jäger et al. 2011) and thus many more new potential cellular targets to explore the current framework. To demonstrate the feasibility in our case study, we focused on one particular virus—HIV1, one particular viral protein—Tat, one particular host–protein complex—Cdk9/Cyclin T1 that the virus needs, and one particular function of the complex—the interface stability. It is tantalizing to expand any of these particular choices. There is yet another potential benefit of the present approach. Viral protein recruits the cellular machinery for its biological functions, so it avoids the catalytic sites in order to preserve the function and binds instead to non-catalytic sites. These non-catalytic sites are normally far less conserved within the protein family. In the case of Cdk9, we analyzed the conservation score of each amino acid using the ConSurf program with results shown in Supplementary

Fig. 6  Effect of F07#13 inhibitor on wild type and mutant Cdk9 complexes in  vivo. HIV-1 infected J1-1 cells were electroporated with either Tagged wild type flag-Cdk9 or mutant flag-Cdk9 plasmids (20 µg each). Samples were kept at 37 °C for 24 h and subsequently treated with varying concentrations of F07#13 (0.1, 1, and 10  µM). Cells were further incubated for 48 h at 37 °C and subsequently processed for kinase assay. One hundred micrograms of nuclear extracts were used for immunoprecipitation using anti-flag antibody or IgG (10 µg) overnight at 4 °C. The next day protein A &G were added for 2  h, washed with kinase buffer and used for in  vitro kinase reaction using [γ-32P] ATP and GST-CTD (2 µg) or purified histone H1 (1 µg) as substrates

Fig. 6. For example, the catalytic ATP-binding pocket sites have scores of 7–9 on the scale of 1–9 with 9 being the most conserved. This is why most ATP-pocket inhibitors lack specificity toward a particular Cdk. In contrast, the binding pocket we identified for F07#13 contains sites of smaller conservation scores of 4.5–5.6. This more variable pocket permits further fine tuning of F07#13 to target Cdk9 specifically and to minimize the side-effect of cross-toxicity for other Cdks. Finally, results using in vitro kinase inhibition assays show that both RNA Pol II CTD and histone H1 are inhibited by F07#13. These substrates are important proteins for both transcription initiation elongation as well as unwinding the chromatin DNA. Importantly the SI75A mutant form of Cdk9 still phosphorylated both CTD and H1, but F07#13 did not significantly inhibit this mutant, further implying that this position was important for binding to F07#13. Collectively these data imply that Cdk9/Cyclin T1 binds to Tat and F07#13 can dissociate this complex through an unknown mechanism. However, the mutant form of Cdk9 may not be stable and it only is able to show kinase activity that is not regulated by Tat (Fig. 7). Future experiments will determine whether there are structural conformational changes that Tat binding contributes to, make the complex stable, yet analogs such as F07#13 are able to dissociate the Cdk9/Cyclin complex. In conclusion, computational structure-based drug design will continue to play major role in discovering novel cellular targets and to develop effective new HIV therapies with reduced risk of drug resistance.

Materials and Methods Molecular Dynamics Simulations We carried out MD simulations using GROMACS package (Van Der Spoel et al. 2005) with Gromas53a6 force field (Oostenbrink et al. 2004) and SPC water salvation model. All of the structures were simulated with periodic boundary conditions and four systems were investigated: (1) Cdk9 and Cyclin, (2) Cdk9, Cyclin, and Tat, (3) Cdk9, Cyclin, and drug, (4) Cdk9, Cyclin, Tat, and drug, respectively. The structure of Cdk9, Cyclin and Tat were extracted from the experimental structure (PDB ID: 3MI9) (Tahirov et al. 2010). The protein was solvated with 0.9% NaCl solution in truncated octahedron boxes. The long-range electrostatic interactions were treated with Particle Mesh Ewald method (Darden et al. 1993) and the VDW interactions were cut-off at 14 Å. The LINCS algorithm was used to constrain all bond lengths (Hess et al. 1997) and a time-step of 2 fs was used for numerical integration. After the energy minimization with the steepest descent method, a 100 ps simulation in NVT ensemble with position restrains of all heavy atoms

13



International Journal of Peptide Research and Therapeutics

Fig. 7  The diagram depicts how Cdk9/Cyclin T/Tat complex is dissociated in the presence of F07#13 decreasing the overall kinase activity. The Cdk9 mutant S17A however is not able to bind to Tat, therefore F07#13 is not able to down regulate its kinase activity

of proteins were performed to equilibrate the simulated system. After that, the product simulations were run in the NPT ensemble, which temperature was coupled at 300 K with V-rescale algorithm and pressure was coupled at 1 atm. with Parrinello–Rahman algorithm. For each system, we have done two 30 ns-trajectories. The total simulation time was up to 240 ns. The structures were visualized and analyzed by VMD and PyMOL (Humphrey et al. 1996).

Docking of Ligand F07#13 onto Cdk9 The docking simulations on Cdk9 were performed using AutoDock software package version 3.05 (Sousa et al. 2006; Trott and Olson 2010). We followed the docking procedure exactly as previously described by Chen et al. (2009). For the small molecule ligand F07#13, the crystal structure of HIV-1 Tat complexed with human P-TEFb (Cdk9/Cyclin T1, PDB ID: 3MI9) was used for docking simulations (Tahirov et al. 2010). After docking simulations for the ligand (F07#13) and receptors (one of the three Cdk9 structures), 600–2400 docking conformations were obtained. For the optimal binding mode (i.e., the lowest energy docking conformation of highly populated clusters based on a cluster analysis of all resulting docking conformations), the energy contribution toward binding from each residue of Cdk9 was computed and ranked using such interactions as van der Waals, hydrogen bond, salvation, and electrostatic terms.

Correlation Analysis We analyzed the correlation motions between different residues based on the coarse-grained model of network. The first step is to define the coarse-grained nodes. In the network, one single amino acid is defined as a node. Thus, the nodes are located at the C𝛼 atoms for different amino acids.

13

The second step is to define the edges between the coarsegrained nodes. Protein or protein complex folds into a specific functional structure depending on the interactions between the residues. What is the proper distance cutoff for representing the protein structures as a network? In the earlier studies, the edges of the network between residues were built within a cutoff in the range from 7 to 9 Å. Recently, much higher resolution protein structures can be obtained from X-ray experiments. It is more common to define the edges of the network between residues in a range from 4 to 5 Å now (Sun et al. 2013; Greene and Higman 2003). Another question is that how to define the edges from MD simulations. In order to capture the dynamical behavior, we define the edges using a probability cutoff. Therefore, if the distance of any two heavy atoms of the different nodes is < 4.5 Å for at least 75% in the frames of the MD simulations, then, we define it as an edge. The neighboring nodes in sequence are not considered to be in contact. We have done two 30 ns MD trajectories for each different state. The dynamical network is constructed with the final 20 ns from the 30 ns trajectories sampled every 100 ps using Carma package (Glykos 2006). The next step is to define the correlation calculations. The nanoseconds MD simulations could provide much more information than static structure of proteins. The Pearson correlation coefficient is a method to make use of the atomic positional fluctuations for identifying the collective motions in proteins from MD trajectories (Ichiye and Karplus 1991; Hunenberger et al. 1995). According to the established network, correlations between different residues are calculated using Pearson correlation coefficient, ⟨ ⟩ Δ⃗ri (t) ⋅ Δ⃗rj (t) Cij = (⟨ (1) ⟩⟨ ⟩)1∕2 Δ⃗ri (t)2 Δ⃗rj (t)2 where,

International Journal of Peptide Research and Therapeutics

(2) and the ⃗ri (t) is the fluctuation vector of the atom ( C𝛼 ) corresponding to the ith node. The values vary from − 1 to 1.

Δ⃗ri (t) = ⃗ri (t) − ⃗ri (t)

Community Analysis The community network analysis was used to further simplify the coarse-grained network. In the first step, we defined the dynamical network as a weighted network. The weight of an edge describes the ‘distance’ between nodes i and j, ( ) | | Wij = − log |Cij | (3) | | The betweenness of an edge is defined as the number of shortest paths between pairs of vertices. It describes the ‘traffic’ information passing through the edges. With the information of edge betweenness, then, we use the Girvan–Newman algorithm to identify and optimize the community substructures of the dynamical network (Girvan and Newman 2002; Newman 2006). Different communities are the subnetworks from the entire coarse-grained network. Based on the correlation weight, the nodes in a community have much stronger interactions/connections than those in the other communities.

Sequence Conservation Analysis The protein sequences of Cdk9 and Cyclin T1 (PDB ID: 3MI9) (Tahirov et al. 2010) were obtained from the National Center for Biotechnology Information (NCBI). Sequence redundancy was removed by identifying clusters of sequences at 99% identity. Then, a cluster was replaced with a representative sequence using CD-HIT program (Huang et al. 2010). The non-redundant sequences were subjected to multiple alignments using MUSCLE program (Edgar 2004). The conservation scores of the amino acid residues were calculated using ConSurf server (Landau et al. 2005).

Kinase Assays Kinase assays were performed with immunoprecipitates from transfected cells (either wild type or S175A mutant Cdk9). Plasmids were electroporated into J1-1 cells followed by F07#13 treatment. Nuclear extracts were used for immunoprecipitation using anti-flag antibody. The next day immune complexes were bound to protein A &G, and washed with kinase buffer in a 20-µl reaction volume containing 0.1 nM ATP, 1 µCi of [γ-32P] ATP (6000 Ci/mmol; Amersham Biosciences), and GST-CTD (2 µg) or purified histone H1 (1 µg) as substrates in TTK kinase buffer containing 50 mM HEPES (pH 7.9), 10 mM ­MgCl2, 6 mM EGTA, and 2.5  mM dithiothreitol for 30  min at 30  °C.

Phosphorylated GST-CTD or histone H1 was resolved on 4–20% SDS–PAGE, dried and subjected to autoradiography and quantitation with a PhosphorImager. Acknowledgements  This work was supported by National Science Foundation (NSF) Grant 0941228 to CZ; NIH grants AI078859, AI074410, and AI043894 to FK; partially supported by Scientific Research Foundation of Central China Normal University 20205170045 to YZ. Author Contributions  YZ performed most computational analysis; HC performed MD simulations; CD and YJ helped with the network analysis; HL and YX helped with the conservation analysis; YZ, FK and CZ supervised the overall study and wrote the paper. All authors edited the manuscript.

Compliance with Ethical Standards  Conflict of interest  The authors declare no competing financial interests.

References Ahlquist P, Noueiry AO, Lee WM, Kushner DB, Dye BT (2003) Host factors in positive-strand RNA virus genome replication. J Virol 77(15):8181–8186 Baumli S, Lolli G, Lowe ED, Troiani S, Rusconi L, Bullock AN et al (2008) The structure of P-TEFb (CDK9/cyclin T1), its complex with flavopiridol and regulation by phosphorylation. EMBO J 27(13):1907–1918. https​://doi.org/10.1038/emboj​.2008.121 Baumli S, Endicott JA, Johnson LN (2010) Halogen bonds form the basis for selective P-TEFb inhibition by DRB. Chem Biol 17(9):931–936. https​://doi.org/10.1016/j.chemb​iol.2010.07.012 Bettayeb K, Baunbaek D, Delehouze C, Loaec N, Hole AJ, Baumli S et al (2010) CDK inhibitors roscovitine and CR8 trigger Mcl-1 down-regulation and apoptotic cell death in neuroblastoma cells. Genes Cancer 1(4):369–380. https:​ //doi.org/10.1177/194760​ 1910​ 36981​7 Betzi S, Alam R, Martin M, Lubbers DJ, Han H, Jakkaraj SR et al (2011) Discovery of a potential allosteric ligand binding site in CDK2. ACS Chem Biol 6(5):492–501. https​://doi.org/10.1021/ cb100​410m Bigna JJ, Plottel CS, Koulla-Shiro S (2016) Challenges in initiating antiretroviral therapy for all HIV-infected people regardless of CD4 cell count. Infect Dis Poverty 5(1):85. https​://doi. org/10.1186/s4024​9-016-0179-9 Butera ST, Roberts BD, Lam L, Hodge T, Folks TM (1994) Human immunodeficiency virus type 1 RNA expression by four chronically infected cell lines indicates multiple mechanisms of latency. J Virol 68(4):2726–2730 Canduri F, Perez PC, Caceres RA, de Azevedo WF (2008) Jr. CDK9 a potential target for drug development. Med Chem 4(3):210–218 Chen H, Van Duyne R, Zhang N, Kashanchi F, Zeng C (2009) A novel binding pocket of cyclin-dependent kinase 2. Proteins 74(1):122– 132. https​://doi.org/10.1002/prot.22136​ Chen H, Zhao Y, Li H, Zhang D, Huang Y, Shen Q et  al (2014) Break CDK2/Cyclin E1 interface allosterically with small peptides. PLoS ONE 9(10):e109154. https​://doi.org/10.1371/journ​ al.pone.01091​54 Darden T, York D, Pedersen L (1993) Particle mesh Ewald: an N⋅log(N) method for Ewald sums in large systems. J Chem Phys 98:10089–10092

13

del Sol A, Tsai CJ, Ma B, Nussinov R (2009) The origin of allosteric functional modulation: multiple pre-existing pathways. Structure 17(8):1042–1050. https​://doi.org/10.1016/j.str.2009.06.008 Dror RO, Dirks RM, Grossman JP, Xu H, Shaw DE (2012) Biomolecular simulation: a computational microscope for molecular biology. Annu Rev Biophys 41:429–452. https​://doi.org/10.1146/annur​evbioph​ys-04291​0-15524​5 Edgar RC (2004) MUSCLE: multiple sequence alignment with high accuracy and high throughput. Nucleic Acids Res 32(5):1792– 1797. https​://doi.org/10.1093/nar/gkh34​0 Fernandez G, Zeichner SL (2010) Cell line-dependent variability in HIV activation employing DNMT inhibitors. Virol J 7:266. https​ ://doi.org/10.1186/1743-422X-7-266 Forli S, Olson AJ (2015) Computational challenges of structure-based approaches applied to HIV. Curr Top Microbiol Immunol 389:31– 51. https​://doi.org/10.1007/82_2015_432 Fuglestad B, Gasper PM, McCammon JA, Markwick PR, Komives EA (2013) Correlated motions and residual frustration in thrombin. J Phys Chem B 117(42):12857–12863. https​://doi.org/10.1021/ jp402​107u Girvan M, Newman ME (2002) Community structure in social and biological networks. Proc Natl Acad Sci USA 99(12):7821–7826. https​://doi.org/10.1073/pnas.12265​3799 Glykos NM (2006) Software news and updates. Carma: a molecular dynamics analysis program. J Comput Chem 27(14):1765–1768. https​://doi.org/10.1002/jcc.20482​ Greene LH, Higman VA (2003) Uncovering network systems within protein structures. J Mol Biol 334(4):781–791 Heredia A, Le N, Gartenhaus RB, Sausville E, Medina-Moreno S, Zapata JC et  al (2015) Targeting of mTOR catalytic site inhibits multiple steps of the HIV-1 lifecycle and suppresses HIV-1 viremia in humanized mice. Proc Natl Acad Sci USA 112(30):9412–9417. https​://doi.org/10.1073/pnas.15111​44112​ Hess B, Bekker H, Berendsen HJC, Fraaije JGEM (1997) LINCS: a linear constraint solver for molecular simulations. J Comput Chem 8(12):1463–1472. https​://doi.org/10.1002/(SICI)1096987X(19970​9)18:12%3C146​3::AID-JCC4%3E3.0.CO;2-H. Huang Y, Niu B, Gao Y, Fu L, Li W (2010) CD-HIT suite: a web server for clustering and comparing biological sequences. Bioinformatics 26(5):680–682. https:​ //doi.org/10.1093/bioin​forma​tics/btq003​ Humphrey W, Dalke A, Schulten K (1996) VMD: visual molecular dynamics. J Mol Graph 14(1):33–38 (27–28). Hunenberger PH, Mark AE, van Gunsteren WF (1995) Fluctuation and cross-correlation analysis of protein motions observed in nanosecond molecular dynamics simulations. J Mol Biol 252(4):492–503. https​://doi.org/10.1006/jmbi.1995.0514 Ichiye T, Karplus M (1991) Collective motions in proteins: a covariance analysis of atomic fluctuations in molecular dynamics and normal mode simulations. Proteins 11(3):205–217. https​://doi. org/10.1002/prot.34011​0305 Ivetac A, McCammon JA (2010) Mapping the druggable allosteric space of G-protein coupled receptors: a fragment-based molecular dynamics approach. Chem Biol Drug Des 76(3):201–217. https​:// doi.org/10.1111/j.1747-0285.2010.01012​.x Ivetac A, McCammon JA (2012) A molecular dynamics ensemblebased approach for the mapping of druggable binding sites. Methods Mol Biol 819:3–12. https​://doi.org/10.1007/978-1-61779​ -465-0_1 Jäger S, Cimermancic P, Gulbahce N, Johnson JR, Mcgovern KE, Clarke SC et al (2011) Global landscape of HIV–human protein complexes. Nature 481(7381):365 Karplus M, McCammon JA (2002) Molecular dynamics simulations of biomolecules. Nat Struct Biol 9(9):646–652. https​://doi. org/10.1038/nsb09​02-646 Krystof V, Uldrijan S (2010) Cyclin-dependent kinase inhibitors as anticancer drugs. Curr Drug Targets 11(3):291–302

13

International Journal of Peptide Research and Therapeutics Krystof V, Baumli S, Furst R (2012) Perspective of cyclindependent kinase 9 (CDK9) as a drug target. Curr Pharm Des 18(20):2883–2890 Kumari N, Iordanskiy S, Kovalskyy D, Breuer D, Niu X, Lin X et al (2014) Phenyl-1-Pyridin-2yl-ethanone-based iron chelators increase IkappaB-alpha expression, modulate CDK2 and CDK9 activities, and inhibit HIV-1 transcription. Antimicrob Agents Chemother 58(11):6558–6571. https​://doi.org/10.1128/ AAC.02918​-14 Landau M, Mayrose I, Rosenberg Y, Glaser F, Martz E, Pupko T et al (2005) ConSurf 2005: the projection of evolutionary conservation scores of residues on protein structures. Nucleic Acids Res 33(Web Server issue):W299–W302. https​://doi.org/10.1093/nar/ gki37​0 Lengauer T, Rarey M (1996) Computational methods for biomolecular docking. Curr Opin Struct Biol 6(3):402–406 Lewis JA, Lebois EP, Lindsley CW (2008) Allosteric modulation of kinases and GPCRs: design principles and structural diversity. Curr Opin Chem Biol 12(3):269–280. https​://doi.org/10.1016/j. cbpa.2008.02.014 Li L, Li C, Sarkar S, Zhang J, Witham S, Zhang Z et al (2012) DelPhi: a comprehensive suite for DelPhi software and associated resources. BMC Biophys 5:9. https​://doi.org/10.1186/2046-1682-5-9. Li L, Jia Z, Peng Y, Chakravorty A, Sun L, Alexov E (2017) DelPhiForce web server: electrostatic forces and energy calculations and visualization. Bioinformatics 33(22):3661–3663. https​://doi. org/10.1093/bioin​forma​tics/btx49​5 Malumbres M, Harlow E, Hunt T, Hunter T, Lahti JM, Manning G et al (2009) Cyclin-dependent kinases: a family portrait. Nat Cell Biol 11(11):1275–1276. https​://doi.org/10.1038/ncb11​09-1275 Morgan DO (1995) Principles of CDK regulation. Nature 374(6518):131–134. https​://doi.org/10.1038/37413​1a0 Narayanan A, Sampey G, Van Duyne R, Guendel I, Kehn-Hall K, Roman J et al (2012) Use of ATP analogs to inhibit HIV-1 transcription. Virology 432(1):219–231. https​://doi.org/10.1016/j. virol​.2012.06.007 Narayanan A, Iordanskiy S, Das R, Van Duyne R, Santos S, Jaworski E et al (2013) Exosomes derived from HIV-1-infected cells contain trans-activation response element RNA. J Biol Chem 288(27):20014–20033. https​://doi.org/10.1074/jbc.M112.43889​5 Nemeth G, Varga Z, Greff Z, Bencze G, Sipos A, Szantai-Kis C et al (2011) Novel, selective CDK9 inhibitors for the treatment of HIV infection. Curr Med Chem 18(3):342–358 Newman ME (2006) Modularity and community structure in networks. Proc Natl Acad Sci USA 103(23):8577–8582. https​://doi. org/10.1073/pnas.06016​02103​ Oostenbrink C, Villa A, Mark AE, van Gunsteren WF (2004) A biomolecular force field based on the free enthalpy of hydration and solvation: the GROMOS force-field parameter sets 53A5 and 53A6. J Comput Chem 25(13):1656–1676. https​://doi.org/10.1002/ jcc.20090​ Ou-Yang SS, Lu JY, Kong XQ, Liang ZJ, Luo C, Jiang H (2012) Computational drug discovery. Acta Pharmacol Sin 33(9):1131–1140. https​://doi.org/10.1038/aps.2012.109 Prussia A, Thepchatri P, Snyder JP, Plemper RK (2011) Systematic approaches towards the development of host-directed antiviral therapeutics. Int J Mol Sci 12(6):4027–4052. https​://doi. org/10.3390/ijms1​20640​27 Reynolds CH (2014) Impact of computational structure-based methods on drug discovery. Curr Pharm Des 20(20):3380–3386 Rivalta I, Sultan MM, Lee NS, Manley GA, Loria JP, Batista VS (2012) Allosteric pathways in imidazole glycerol phosphate synthase. Proc Natl Acad Sci USA 109(22):E1428–E1436. https​://doi. org/10.1073/pnas.11205​36109​ Ryckman C, Robichaud GA, Roy J, Cantin R, Tremblay MJ, Tessier PA (2002) HIV-1 transcription and virus production are both

International Journal of Peptide Research and Therapeutics accentuated by the proinflammatory myeloid-related proteins in human CD4+ T lymphocytes. J Immunol 169(6):3307–3313 Sethi A, Eargle J, Black AA, Luthey-Schulten Z (2009) Dynamical networks in tRNA: protein complexes. Proc Natl Acad Sci USA 106(16):6620–6625. https​://doi.org/10.1073/pnas.08109​61106​ Sethi A, Tian J, Derdeyn CA, Korber B, Gnanakaran S (2013) A mechanistic understanding of allosteric immune escape pathways in the HIV-1 envelope glycoprotein. PLoS Comput Biol 9(5):e1003046. https​://doi.org/10.1371/journ​al.pcbi.10030​46 Sliwoski G, Kothiwale S, Meiler J, Lowe EW (2014) Jr. Computational methods in drug discovery. Pharmacol Rev 66(1):334–395. https​ ://doi.org/10.1124/pr.112.00733​6 Sousa SF, Fernandes PA, Ramos MJ (2006) Protein-ligand docking: current status and future challenges. Proteins 65(1):15–26. https​ ://doi.org/10.1002/prot.21082​ Sun J, Jing R, Wu D, Zhu T, Li M, Li Y (2013) The effect of edge definition of complex networks on protein structure identification. Comput Math Methods Med 2013:365410. https​://doi. org/10.1155/2013/36541​0 Tahirov TH, Babayeva ND, Varzavand K, Cooper JJ, Sedore SC, Price DH (2010) Crystal structure of HIV-1 Tat complexed with human P-TEFb. Nature 465(7299):747–751. https:​ //doi.org/10.1038/natur​ e0913​1. Trott O, Olson AJ (2010) AutoDock Vina: improving the speed and accuracy of docking with a new scoring function, efficient optimization, and multithreading. J Comput Chem 31(2):455–461. https​ ://doi.org/10.1002/jcc.21334​ Van Duyne R, Guendel I, Jaworski E, Sampey G, Klase Z, Chen H et al (2013) Effect of mimetic CDK9 inhibitors on HIV-1-activated transcription. J Mol Biol 425(4):812–829. https​://doi. org/10.1016/j.jmb.2012.12.005

Van Der Spoel D, Lindahl E, Hess B, Groenhof G, Mark AE, Berendsen HJ (2005) GROMACS: fast, flexible, and free. J Comput Chem 26(16):1701–1718. https​://doi.org/10.1002/jcc.20291​ Vanwart AT, Eargle J, Luthey-Schulten Z, Amaro RE (2012) Exploring residue component contributions to dynamical network models of allostery. J Chem Theory Comput 8(8):2949–2961. https​://doi. org/10.1021/ct300​377a Wang J, Zhao Y, Zhu C, Xiao Y (2015) 3dRNAscore: a distance and torsion angle dependent evaluation function of 3D RNA structures. Nucleic Acids Res 43(10):e63. https​://doi.org/10.1093/nar/ gkv14​1 Xing S, Li F, Zeng Z, Zhao Y, Yu S, Shan Q et al (2016) Tcf1 and Lef1 transcription factors establish CD8(+) T cell identity through intrinsic HDAC activity. Nat Immunol 17(6):695–703. https:​ //doi. org/10.1038/ni.3456 Zhao Y, Gong Z, Xiao Y (2011) Improvements of the hierarchical approach for predicting RNA tertiary structure. J Biomol Struct Dyn 28(5):815–826. https ​ : //doi.org/10.1080/07391​ 102.2011.10508​609 Zhao Y, Huang Y, Gong Z, Wang Y, Man J, Xiao Y (2012) Automated and fast building of three-dimensional RNA structures. Sci Rep 2:734. https​://doi.org/10.1038/srep0​0734 Zhao Y, Zeng C, Tarasova NI, Chasovskikh S, Dritschilo A, Timofeeva OA (2013) A new role for STAT3 as a regulator of chromatin topology. Transcription 4(5):227–231. https​://doi.org/10.4161/ trns.27368​ Zhao YJ, Zeng C, Massiah MA (2015) Molecular dynamics simulation reveals insights into the mechanism of unfolding by the A130T/V mutations within the MID1 zinc-binding Bbox1 domain. PLoS ONE. https​://doi.org/10.1371/journ​al.pone.01243​77

13