Cyclic peptides with a distinct arginine-fork motif recognize the HIV trans -activation response RNA in vitro and in cells

RNA represents a potential target for new antiviral therapies, which are urgently needed to address public health threats such as the human immunode ﬁ ciency virus (HIV). We showed previously that the interaction between the viral Tat protein and the HIV-1 trans -activation response (TAR) RNA was blocked by TB-CP-6.9 a . This cyclic peptide was derived from a TAR-binding loop that emerged during lab evolution of a TAR-binding protein (TBP) family. Here we synthesized and characterized a next-generation, cyclic-peptide library based on the TBP scaffold. We sought to identify conserved RNA-binding interactions and the in ﬂ uence of cyclization linkers on RNA binding and antiviral activity. A diverse group of cyclization linkers, encompassing disul ﬁ de bonds to bicyclic aromatic staples, was used to restrain the cyclic peptide geometry. Thermodynamic pro ﬁ ling revealed speci ﬁ c arginine-rich sequences with low to submicromolar af ﬁ nity driven by enthalpic and entropic contributions. The best compounds exhibited no appreciable off-target binding to related molecules, such as BIV TAR and human 7SK RNAs. A speci ﬁ c arginine-to-lysine report the analysis of an expanded library of TBP-derived cyclic peptides to explore the in ﬂ uence of arginine composition, spacing, and the intramolecular cross-link on their TAR recognition properties and antiviral activity. Alkyl linkers diminish binding but acetone-based linkers enhance af ﬁ nity of TB-CP-6.9 variants

RNA represents a potential target for new antiviral therapies, which are urgently needed to address public health threats such as the human immunodeficiency virus (HIV). We showed previously that the interaction between the viral Tat protein and the HIV-1 trans-activation response (TAR) RNA was blocked by TB-CP-6.9a. This cyclic peptide was derived from a TAR-binding loop that emerged during lab evolution of a TARbinding protein (TBP) family. Here we synthesized and characterized a next-generation, cyclic-peptide library based on the TBP scaffold. We sought to identify conserved RNA-binding interactions and the influence of cyclization linkers on RNA binding and antiviral activity. A diverse group of cyclization linkers, encompassing disulfide bonds to bicyclic aromatic staples, was used to restrain the cyclic peptide geometry. Thermodynamic profiling revealed specific arginine-rich sequences with low to submicromolar affinity driven by enthalpic and entropic contributions. The best compounds exhibited no appreciable off-target binding to related molecules, such as BIV TAR and human 7SK RNAs. A specific arginine-to-lysine change in the highest affinity cyclic peptide reduced TAR binding by tenfold, suggesting that TBP-derived cyclic peptides use an arginine-fork motif to recognize the TAR major groove while differentiating the mode of binding from other TARtargeting molecules. Finally, we showed that HIV infectivity in cell culture was reduced in the presence of cyclic peptides constrained by methylene or naphthalene-based linkers. Our findings provide insight into the molecular determinants required for HIV-1 TAR recognition and antiviral activity. These findings are broadly relevant to the development of antivirals that target RNA molecules.
Targeting RNA with small molecules to disrupt crucial pathways in the microbial lifecycle has gained significant attention as a drug-development strategy (1)(2)(3)(4). Progress in this area is due in part to the propensity of some structured RNAs to form clefts, pockets, or deep grooves that are receptive to ligand binding (5)(6)(7)(8)(9)(10). In this respect, promising steps have been made to target the internal bulged loop of the HIV-1 trans-activation response (TAR) element RNA, which represents a model system for RNA drug development (11)(12)(13)(14)(15).
The significance of TAR in the viral life cycle is underscored by its presence in every HIV-1 transcript (16,17), where the TAR hairpin is highly conserved in terms of its sequence and structure (18). Following integration of proviral DNA into the host genome, the virus initiates transcription using host RNA polymerase II, but stalls subsequently due to negative elongation factors (19)(20)(21). To activate viral transcription, the polymerase must interact with the positive transcription elongation factor (pTEFb) complex, which contains cyclin T1 and CDK9 (22) (Fig. 1A). HIV has evolved the trans-activator of transcription (Tat) protein to hijack this host complex, freeing it from inactivation by the host 7SK small nuclear ribonucleoprotein assembly, which includes the host protein HEXIM (22)(23)(24)(25)(26)(27). By mimicking the RNA-binding domain of HEXIM, the Tat arginine-rich motif (ARM) binds 7SK RNA; simultaneously, Tat releases HEXIM and recruits the pTEFb complex to TAR where CDK9 phosphorylates the CTD of RNA polymerase II, leading to transcription elongation (26)(27)(28)(29). Because Tat is not produced in appreciable amounts in latent viral reservoirs (30,31), inhibitors that block the Tat-TAR interaction could ostensibly abrogate viral mRNA transcription, promoting a functional cure (32,33).
Rationally designed TAR-binding peptides that mimic the Tat ARM have been derived from structures of the bovine immunodeficiency virus (BIV) TAR-Tat complex. Therein, the arginine-rich BIV Tat peptide adopts a sharp type V 0 turn that supports major-groove recognition at a bulged UU internal loop (34). This observation provided a foundation to produce a series of cyclic β-hairpin peptides that have been iteratively refined and include natural and unnatural positively charged amino acids that mimic the HIV Tat ARM (33,35,36). An ingenious aspect of the underlying scaffold is that peptides are closed by D-and L-proline to enforce cyclization. This feature also allows the peptide to adopt a well-packed core similar to a mini β-sheet, which likely ‡ These authors contributed equally to this work. * For correspondence: Joseph E. Wedekind, joseph.wedekind@rochester. edu, Rudi Fasan, rudi.fasan@rochester.edu. Present address for Sachitanand M. Mali: Protomer Technologies Inc., Pasadena CA 91105, USA reduces conformational entropy. This strategy has yielded tight binding TAR binders with affinities that range from nanomolar to picomolar and IC 50 values that range from 4 μM to 120 μM in antiviral assays tested using laboratory viral strains in cell culture (35,37). In contrast to structure-guided design principles garnered from the BIV TAR-Tat complex, we employed a fundamentally different strategy based on the chemical features observed in lab-evolved TAR-binding proteins (TBPs) selected to recognize HIV-1 TAR (38)(39)(40). A key observation is that the arginine composition and spacing within the lab-evolved β2-β3 loops of TBPs are primary factors in high-affinity TAR recognition (Fig. 1B), as demonstrated by the binding series: TBP6.9 (K D of 3.0 ± 0.3 nM) > TBP6.7 (K D of 5.3 ± 0.9 nM) > TBP6.R (K D of 34.3 ± 2.7 nM) > TBP6.3 (K D of 45.2 ± 3.2 nM) (40). Moreover, only three arginines of the β2-β3 loop engaged in recognition of conserved guanines, even when β2-β3 loops with as many as five arginines were tested. This finding parallels the observation that the Tat ARM naturally recognizes both host 7SK RNA and HIV TAR RNA by using different subsets of arginine residues wherein only 2 to 4 guanidinium groups were engaged at one time (28).
Significantly, our previous structural and biochemical analyses of individual TAR-TBP complexes also revealed that each lab-evolved protein recognizes TAR using analogous structural features on a global level (Fig. 1C, inset). Significantly, each TBP uses a specific arginine fork motif wherein R47 recognizes TAR at the Hoogsteen edge of Gua26, as well as the phosphate backbone of Uri23. This mode of readout entails highly specific geometric features of the target RNA (38,40) (Fig. 1C) that appear integral to TAR recognition by HIV Tat (28,40). Notably, the latter arginine fork motif is absent in existing labderived peptides and small molecules to our knowledge (33,40).
Mandates to produce next-generation HIV antivirals emphasize a need for drugs that are long acting, have fewer side effects, and promote compliance and ease of use (https:// grants.nih.gov/grants/guide/notice-files/NOT-OD-15-137.html). Hence, although TBPs represent a promising route to block infectivity, treatment would likely require injections similar to Tat interactions with HIV TAR leading to RNA polymerase II elongation and the TAR binding protein (TBP) consensus sequence with a representative conformation from the TAR-TBP6.9 complex. A, left, cartoon model of the inactive pTEFb complex sequestered by the host 7SK smallnuclear ribonucleoprotein (7SK RNP) complex. The arginine-rich motif (ARM) of Tat displaces its counterpart from host HEXIM, producing a free Tat-pTEFb complex. Right, Tat conveys the pTEFb complex to TAR, leading to formation of a Tat-TAR elongation complex. Transcription of viral mRNA is activated by CDK9-mediated phosphorylation of host RNA polymerase II (105,106). B, amino acids in the β2-β3 loop of U1A that were diversified and selected for TAR binding (39). β2-β3 loop amino acids are depicted for TBPs of known structure (38,40) that were used as a basis to produce the cyclic peptides of this investigation. The Weblogo depicts the consensus of known β2-β3 loop sequences resulting from lab evolution (39). C, inset, global view of the TAR-TBP6.9 complex highlighting the β2-β3 loop (yellow). The main cartoon shows ribbon and ball-and-stick depictions of key interactions required for molecular recognition, including the arginine fork motif (highlighted in green) (38,40,65). enfuvirtide (41), which can erode patient adherence (42). Moreover, recombinant biologics can lose efficacy over time due to immunogenicity (43,44). In contrast, cyclic peptides have gained traction as therapeutics in the past decade with some showing oral availability and others attaining plasma half-lives of 1 day (45).
Accordingly, we leveraged both global and local structural details of TBPs to design a small cyclic peptide that mimics the lab-evolved β2-β3 loop of TBP6.9 and its interaction with the TAR major groove (Fig. 1C). Our efforts led to the synthesis and characterization of TAR-binding cyclic-peptide-6.9a (or TB-CP-6.9a; Fig. 2, A and B), which binds HIV-1 TAR with a K D of 3.6 ± 0.4 μM and blocks association between TAR and a Tat ARM peptide (40). In contrast, a related TBP6.7-derived cyclic peptide (TB-CP-6.7a), which contains three arginines and differs only by an R-to-Q change (Fig. 1B), binds TAR with a K D of 20.0 ± 0.1 μM and did not obstruct binding of the Tat ARM to TAR (40). Altogether this work supported the feasibility of producing short cyclic peptide mimics of TBPs to target HIV TAR.
Herein, we report the analysis of an expanded library of TBP-derived cyclic peptides to explore the influence of arginine composition, spacing, and the intramolecular cross-link on their TAR recognition properties and antiviral activity. Thermodynamic profiling by isothermal titration calorimetry (ITC) revealed a range of affinities from 24.2 ± 3.4 μM to 0.8 ± 0.1 μM. Importantly, the most promising cyclic peptides show specificity for TAR while avoiding interactions with off-target RNAs. A cyclic peptide variant predicted to lack the argininefork elicited poor TAR affinity, suggesting that this motif is operative in binding. Significantly, a handful of cyclic peptides showed antiviral activity against HIV in cell culture without substantial cytotoxicity. Overall, this work establishes key molecular recognition determinants and cyclization chemistries that show promise for development of new HIV TARtargeting antivirals.

Results
Choice of peptide sequence and strategy for optimization of TAR-binding cyclic peptides Our previous work on TBPs demonstrated that specific amino acids are important not only for RNA binding but also to stabilize a distinct β2-β3 loop conformation conducive to TAR recognition (Fig. 1, B and C) (38,40). In particular, we noted that T50 fulfills an important role in stabilizing the R52 guanidinium group. Similarly, R48 hydrogen bonds to the R47 carbonyl oxygen while simultaneously salt-bridging to the TAR backbone (Fig. 1C). Prolines 46 and 51 also favor a distinct backbone conformation. Accordingly, we sought to design a series of small-sized cyclic peptides that would mimic the β2-β3 loop of TBPs and retain conformations conducive to cognate RNA readout. To stabilize these conformations and promote serum stability (46,47), we chose to prepare cyclic peptides by restraining the peptide sequence encompassing positions 46 through 54 of the β2-β3 loop region via cysteine Figure 2. Schematic diagrams of cyclic peptide sequences and linkers. A, cyclic peptide sequences produced in this investigation. The sequences were derived from the β2-β3 loops of TBP6.3, TBP6.9, and TBP6.7(Q48R/T51R) as described (40). Cysteines were added to each terminus to promote cyclization with various linking chemical groups. B, the linker groups shown were used to cyclize peptides in panel A at the site of the R-group.
Targeting HIV TAR with a focused library of cyclic peptides bis-alkylation (48). Residues V45 and A55 appeared well suited as cyclization sites because their side chains reside on the same face of neighboring β-strands in all TAR-TBP crystal structures (e.g., Fig. 1C). Although cysteines will form a disulfide bond, we employed them in a workflow outlined in Figure S1 to generate stable alkane and aromatic thioacetal linkers (49,50). To empirically optimize the size, we synthesized and tested peptides containing a variety of linkers including methyl, ethyl, butyl, acetone, pyridine, xylene, and naphthalene groups ( Fig. 2 and Table S1). Arene substitution and cysteine stereochemistry were also treated as variables. The choice of the peptide sequences was based on three β2-β3 loops identified in TBPs that were found to bind TAR with high affinity in previous studies (40), namely TBP6.3, TBP6.9, and TBP6.R (Fig. 1B). Each peptide sequence contains a residue equivalent to R47 ( Fig. 2A) that likely recognizes TAR by an HηηP arginine fork (38,40). To optimize these compounds as β2-β3 loop mimics and TAR-targeting agents (Fig. S1), these peptides were further diversified by incorporating different linkers including methyl, ethyl, butyl, acetone, pyridine, xylene, and naphthalene groups (Fig. 2B) and through variation of other structural elements (i.e., cysteine configuration).
Terminal cysteine chirality alters affinities of methylene-linked cyclic peptides derived from TBP6.9 As a first step in identifying specific cyclic peptides for antiviral analysis, we characterized the TAR-binding profiles of various cyclic peptides by ITC ( Fig. 3 and Table 1). In this manner, we identified specific chemical attributes that support high-affinity binding (Fig. S1). As a starting point, we chose to modify cyclic peptide TB-CP-6.9a, which we found previously to have a 5.6-fold affinity improvement over TBP6.7-derived cyclic peptide TB-CP-6.7a (40). Consistent with previous studies, ITC analysis herein on TB-CP-6.9a showed a K D of 5.2 ± 0.2 μM compared with 23 ± 1 μM for TB-CP-6.7a (Table 1).
To test the importance of the stereochemical configuration of the cysteine residues for TAR binding, a series of TB-CP-6.9a analogs was first prepared in which each of the terminal L-cysteine residues in this peptide was replaced by a D-cysteine (Fig. 2). Compared with TB-CP-6.9a, the resulting cyclic peptides TB-CP-6.9aLD and TB-CP-6.9aDD proved to be poorer binders, with K rel values of 4.7 and 3.2 ( Table 1). Each peptide showed a favorable enthalpy (i.e., ΔH of −12.9 ± 7.6 and −7.2 ± 0.5 kcal mol −1 ) but unfavorable entropies (i.e., -TΔS of 7.0 ± 6.2 and 0.7 ± 0.6 kcal mol −1 ) (Fig. 3). In particular for TB-CP-6.9aLD, the large entropic cost of binding could imply a significant distortion of the peptide conformation from the bioactive one, as induced by inversion of the α-carbon configuration of only one of the two cross-linked cysteines. This effect was less pronounced in the case of TB-CP-6.9aDD, where the configuration of both cysteines was inverted. Notably, the side chain of the corresponding residues (i.e., V45 and A55) in the TBP protein does not interact with TAR in parental RNA-TBP complexes (e.g., Fig. 1C). These results highlight the sensitivity of TAR binding to the conformation of the cyclic peptide as influenced by the chirality of the terminal cysteines. In addition, the findings define the use of L-cysteines as optimal choices for subsequent compound optimization. Figure 3. Scatter bar graph representation of thermodynamic parameters for TAR binding by cyclic peptides. TAR recognition by cyclic peptides is enthalpically (ΔH) driven in the case of most cyclic peptides, whereas 6.9ss, 6.9d, 6.9h, and 6.9i display an apparent favorable entropic (-TΔS) contribution. Methylene (6.9a), 1,3 di-bromo methylene (6.9i) and xylene (6.9f-m) linkers display an increased enthalpic contribution that contributes to higher affinity (improved K D ) and favorable binding free energies (improved ΔG). The methylene and xylene linkers display comparable thermodynamic signatures in the context of the TBP6.3 β2-β3 loop sequence (6.3a and 6.3f) but display tighter binding and higher enthalpic favorability in the context of the TBP6.R β2-β3 loop sequence (6.Ra and 6.Rf). The naphthalene linkers (6.9l and 6.9m) display comparable thermodynamic signatures in the presence of the 6.9a methylene linker. Representative thermograms and isotherm fits are provided in Figure S2. Each experiment was replicated two or more times. Errors represent the standard deviation taken from separate replicate measurements.
Targeting HIV TAR with a focused library of cyclic peptides Alkyl linkers diminish binding but acetone-based linkers enhance affinity of TB-CP-6.9 variants We next examined the effect of varying the length of the linker in the TB-CP-6.9-based cyclic peptides. Accordingly, we prepared and tested TB-CP-6.9SS, wherein the methylene linker is replaced by a disulfide bond (Fig. 2B). This variant produced comparable affinity to its methylene-linked counterpart, as indicated by a K D of 7.5 ± 0.3 μM (K rel of 1.4). TB-CP-6.9SS showed favorable entropic and enthalpic contributions to binding (i.e., ΔH of −3.4 kcal mol −1 and -TΔS of −3.4 kcal mol −1 ) (Fig. 3). A series of peptides with longer cross-links were also prepared, including TB-CP-6.9b and TB-CP-6.9c, in which the methylene linker was replaced by an ethyl and a propyl linker (Fig. 2). The respective K D values were 6.9 ± 0.3 μM (TB-CP-6.9b) and 12.1 ± 2.0 μM (TB-CP-6.9c) with enthalpically driven binding (ΔH of −8.6 ± 0.3 kcal mol −1 and −8.6 ± 0.9 kcal mol −1 ) and unfavorable entropies (+1.8 ± 0.02 kcal mol −1 and +2.0 ± 1.0 kcal mol −1 ) ( Fig. 3 and Table 1). Interestingly, an acetone-based linker (51) resulted in a cyclic peptide (TB-CP-6.9d) with a K D of 4.0 ± 0.7 μM and a K rel of 0.8 compared with TB-CP-6.9a. Like the disulfide-bridged peptide, TB-CP-6.9d exhibited a favorable entropy, which largely drives the binding process (-TΔS of −6.6 ± 0.3 kcal mol −1 compared with a ΔH of −0.7 ± 0.4 kcal mol −1 ). Although the chemical basis of the underlying interactions is unknown, the favorable entropic contribution is desirable for binding (52) and suggests that the acetone-based linkage is able to induce a peptide conformation conducive to TAR recognition. This finding prompted us to examine a series of conformationally rigid linkers, including xylene and dimethylpyridine groups.
Xylene enhances binding of TB-CP-6.9 peptides but pyridine does not Accordingly, we next produced a series of cyclic peptides via cysteine bis-alkylation using ortho, meta, and para xylene (Fig. 2B). Peptides TB-CP-6.9f-o (ortho) and TB-CP-6.9f-p (para) had no detectable binding heats (data not shown). In contrast, TB-CP-6.9f-m (meta) showed strong TAR binding, as indicated by a K D of 1.7 ± 0.4 μM and a K rel of 0.3 compared with starting peptide TB-CP-6.9a (Table 1). This value is notably tighter in its binding than our starting peptide TB-CP-6.9a (40), and like the methylene-linked parental peptide, TB-CP-6.9f-m elicited a favorable enthalpy (ΔH of −8.9 ± 0.8 kcal mol −1 ) but an unfavorable entropy (-TΔS of 1.1 ± 0.7 kcal mol −1 ) ( Fig. 3 and Table 1). We next explored the impact of substitutions at the level of the meta-xylene linker. We synthesized and tested TB-CP-6.9i, which is cross-linked by a dibromo-substituted meta-xylene unit (Fig. 2B). This peptide showed improved TAR affinity compared with TB-CP-6.9a (K rel of 0.6) but no improvement over TB-CP-6.9f-m, indicating that substitution of the aryl ring did not improve binding. One notable difference in the thermodynamic profile of TB-CP-6.9i is that the enthalpic and entropic contributions to binding were both favorable (ΔH of −7.3 ± 1.5 kcal mol −1 and -TΔS of −0.5 ± 0.9 kcal mol −1 ) (Fig. 3). Replacement of the meta-xylene linker with its pyridine-based counterpart (TB-CP-6.9h) produced a similar thermodynamic signature (ΔH of −5.1 ± 0.8 kcal mol −1 and -TΔS of −1.6 ± 0.6 kcal mol −1 ) but resulted in a 2-and 5.7-fold lower affinity for TAR (K D of 9.8 ± 2.5 μM) compared with TB-CP-6.9a and TB-CP-6.9f-m, respectively ( Fig. 3C and Table 1). Altogether, these observations pinpointed m-xylene as one of the best linkers for improving TAR-binding affinity among this group of peptides.

Xylene-linked peptides with varied β2-β3 loop sequences show enhanced TAR binding
To test the effects of various β2-β3-loop sequences from previously characterized TBPs, we synthesized peptides based on TBP6.3 and a TBP6.9 mutant with a T50R variation called TBP6.R ( Fig. 2A). Our rationale with the former sequence was to evaluate other high-affinity TBP sequences; our motivation with the latter sequence was to augment the arginine content to improve cell penetration (53,54). The resulting cyclic Table 1 Thermodynamic parameters for TAR binding by cyclic peptides The ortho and para versions of this peptide revealed no detectable heats of binding to HIV TAR. c K rel calculated relative to TB-CP-6.Rf-m.
Targeting HIV TAR with a focused library of cyclic peptides peptides were TB-CP-6.3f-m and TB-CP-6.Rf-m (Fig. 2, B and C). We also produced methylene-linked versions of these peptides to compare their TAR binding properties to TB-CP-6.9a. ITC revealed that cyclic peptides containing the TBP6.3 sequence in the context of methylene (K D of 3.1 ± 1.4 μM and K rel of 0.6) or meta-xylene (K D of 2.0 ± 0.5 μM and K rel of 0.4) linkers have higher affinity for TAR than the corresponding TBP6.9-derived cyclic peptides ( Fig. 3 and Table 1). Notably, comparable cyclic peptides containing five arginines show even higher affinity for TAR, as revealed by K D values of 1.0 ± 0.1 μM (K rel of 0.2) and 0.8 ± 0.1 μM (K rel of 0.15) ( Fig. 3 and Table 1). Like the parental TB-CP-6.9a peptide, each new peptide had favorable enthalpies (i.e., ΔH ranging from of −8.3 ± 1.0 kcal mol −1 to −9.9 ± 0.05 kcal mol −1 ) with unfavorable entropies (i.e., -TΔS ranging from +0.8 ± 1.5 kcal mol −1 to +1.9 ± 0.1 kcal mol −1 ). The results further demonstrated the importance of both the β2-β3 loop sequence and the linker structure for high-affinity TAR binding.
Naphthalene linkers enhance affinity and possess cell penetrating properties As a preface to antiviral analysis, we sought to use an alternative strategy to promote cell penetration by our arginine-rich cyclic peptides. This goal led us to employ naphthalene-based cysteine cross-linkers based on reports that naphthyl-based amino acids promote cell entry of cationic cyclic peptides (55)(56)(57). Accordingly, we synthesized and tested two TBP6.9-derived cyclic peptides constrained by a 1,4-and a 2,3-dimethyl-naphthalene linker (i.e., TB-CP-6.9l and TB-CP-6.9m, Fig. 2B).
Structure-activity analysis of cyclic peptide TB-CP-6.9a for HIV TAR variants While the studies above revealed the importance of the peptide sequence and cross-link for TAR recognition, the mode of binding of the cyclic peptides to TAR remains unknown and can be inferred only from cocrystal structures of parental HIV TAR-TBP complexes (e.g., Fig. 1C) (38,40). To gain structure-activity insights in this regard, we tested binding of TB-CP-6.9a to TAR variants with deleted or substituted nucleobases. Variants included: TARΔ25, in which internal bulge-loop base U25 was deleted; GNRA TAR, wherein the apical loop was replaced by a GAAA tetraloop stabilized by a CG closing pair (58); GNRA TARΔ35, in which the apical loop and bulged nucleotide Ade35 were replaced and removed; and TARΔbulge, in which the conserved central UCU bulged loop was deleted (Fig. 4A).
As expected, TB-CP-6.9a retains binding to TARΔ25 (K D of 6.0 ± 0.5 μM and K rel of 1.2; Fig. 4, A and B and Table 2). This finding suggests that our cyclic peptides will bind this TAR variant, which is present in 9% of circulating recombinant forms of the virus (38,59). Replacement of the apical loop with a GNRA tetraloop is not a naturally occurring TAR variation, but this mutant was tested to identify whether TB-CP-6.9a recognizes the conserved apical loop of TAR, which is integral to binding of the host super-elongation complex (Fig. 1A) (22,29,60,61). TAR binding analysis showed that the loop variant binds with slightly less affinity than wild-type TAR, as indicated by a K D of 6.7 ± 0.8 μM and K rel of 1.3 ( Fig. 4C and Table 2). This observation was not entirely surprising since the TAR apical loop does not contact TBP6.9 in the cocrystal structure (Fig. 1C). Furthermore, a recent analysis of TBP6.9 revealed that the apical loop can be replaced by GAAA without significant reduction to TAR binding (62). In contrast, changing the apical loop to a GAAA tetraloop accompanied by deletion of the conserved Ade35 nucleotide causes substantial affinity loss (i.e., K D of 17.4 ± 2.7 μM and K rel of 3.4) ( Fig. 4D and Table 2). This finding is consistent with the close proximity of the bulged Ade35 base to the β2-β3 loop in the TBP6.9 cocrystal structure (Fig. 1C). Importantly, when the conserved UCU bulged loop was eliminated, TB-CP-6.9a showed no detectable binding (Fig. 4E). This outcome is consistent with the essentiality of the internal bulged loop for β2-β3 loop recognition (38). Overall, the results suggest that TB-CP-6.9a binds to TAR in a manner analogous to the parental TAR-TBP6.9 complex (Fig. 1C).

TB-CP-6.9a resists binding to off-target RNAs
We next assessed the specificity of the TB-CP-6.9a cyclic peptide for TAR compared with other RNAs. Accordingly, we tested binding of this cyclic peptide to human U1 snRNA hairpin II, BIV TAR, and human 7SK RNA (Fig. S3A). Each of these RNAs exhibits multiple major-groove guanines for arginine recognition. Whereas U1 hpII possesses a loop on top of a stem, BIV and 7SK exhibit pyrimidine-rich internal loops with major-groove base triples that share sequence features with HIV TAR (28,34,63,64).
We first titrated TB-CP-6.9a into solutions of U1 snRNA hp II. This RNA is the natural target of the parental U1A protein that was subjected to yeast display maturation as a means to select the family of TBPs (Fig. 1B) (39). No heats of binding were detected (Fig. S3B), indicating that the cyclic peptide does not interact appreciably with the hpII stemloop. This observation parallels a previous analysis that revealed the absence of a hpII interaction with proteins TBP6.6 and TBP6.7 by surface plasmon resonance (39).
The BIV TAR RNA was tested next and also showed no heat of binding (Fig. S3C). BIV TAR adopts a bulged loop conformation analogous to HIV TAR but assumes a different internal loop fold that necessitates a distinct mode of recognition by the BIV Tat ARM compared with that of HIV Tat (34, 64). The Targeting HIV TAR with a focused library of cyclic peptides  Table 2. C, representative thermogram of TB-CP-6.9a titrated into TAR with a single deletion at Uri25 (Δ25). D, representative thermogram of TB-CP-6.9a titrated into a TAR variant in which the apical loop was replaced by a stable GNRA tetraloop (GNRA). E, representative thermogram of TB-CP-6.9a titrated into a TAR variant in which the apical loop was replaced by a GNRA tetraloop and bulged nucleotide Ade35 was deleted (GNRA Δ35). F, representative thermogram of TB-CP-6.9a titrated into a TAR variant in which residues Uri23, Cyt24 and Uri25 of the internal bulged loop were deleted (Δbulge). Errors represent the standard deviation taken from separate replicate measurements. Targeting HIV TAR with a focused library of cyclic peptides ability to discriminate between HIV TAR and BIV TAR proved to be a shortcoming of prior cyclic peptides that were produced by structure-guided design using the BIV TAR-Tat structure (33,35,36). Finally, we titrated TB-CP-6.9a into human 7SK RNA. We chose 7SK because it is the target of HIV Tat binding (Fig. 1A), which utilizes a series of arginines to read specific guanines in the RNA major groove (28). This mode of RNA binding has commonalities with TBP6.9 (Fig. 1C) (38,40). Although the Tat ARM peptide showed a strong binding response for the 7SK RNA (average K D of 6.4 ± 0.2 μM), we observed no detectable heat of binding for cyclic peptides (Fig. S3, D and E). Titration of the methylene-and meta-xylene-linked cyclic peptides TB-CP-6.Ra and TB-CP-6.Rf-m similarly indicated no evidence of specific binding, although small interaction heats were visible that implied traces of off-target interactions (Fig. S3, F and G). This effect was likely due to the higher arginine content of the peptide sequence, which was derived from site-directed mutagenesis of TBP6.9 rather than labbased selection (40). Collectively, the results suggest that these TAR-targeting cyclic peptides-and in particular TB-CP-6.9a-are highly selective toward TAR and resist offtarget binding interactions.

Cyclic peptides block the arginine-rich TAT ARM from TAR binding
We next investigated whether representative cyclic peptides could compete for binding to HIV TAR RNA against a peptide derived from HIV Tat. In this experiment, the cyclic peptide to be tested is titrated into the target TAR RNA and monitored for binding by ITC (Fig. 5A). Next the HIV Tat ARM peptide is injected into the preformed TAR-cyclic peptide complex and the heat of binding is recorded. As a control, we showed previously that the Tat ARM peptide binds to TAR in the absence of cyclic peptide with a K D of 135 ± 31 nM and a 2:1 ratio of peptide to RNA (40). Moreover, TB-CP-6.9a was observed to block binding of the Tat ARM when a fivefold molar ratio of the HIV peptide was titrated into the preformed TAR-TB-CP-6.9a complex (40). In contrast, linear peptide TB-LP-6.9a did not block TAR binding to the Tat ARM. Figure 5. TAR-binding competition assay using the Tat ARM peptide to displace cyclic peptides and vice versa. A, schematic diagram of a TAR competition experiment in which a cyclic peptide is titrated into TAR RNA; the complex is then probed for binding by the Tat ARM peptide, which is titrated subsequently into the ITC sample. In a successful assay, the presence of the cyclic peptide prohibits binding of the Tat ARM to TAR. The far-right panel shows a reverse assay in which a preformed TAR-Tat complex exists in the ITC sample cell and is probed by titration of a cyclic peptide. B, titration of the HIV Tat ARM peptide into a preformed complex between TAR and TB-CP-6.9l. Here and elsewhere, the absence of Tat ARM binding is based on the lack of an appreciable heat change with increasing peptide that is expected for this interaction (40). C, titration of the HIV Tat ARM peptide into a preformed complex between TAR and TB-CP-6.9m. D, titration of the HIV Tat ARM peptide into a preformed complex between TAR and TB-CP-6.Rf-m. E, reverse titration in which TB-CP-6.9a was titrated into the preformed TAR-Tat ARM complex. The heats of titration reveal cyclic peptide binding and saturation consistent with an apparent K D of 18.3 ± 2.1 μM and 1:1 binding stoichiometry; the average values from replicate experiments were: K D = 17.5 ± 1.1 μM, n = 0.96 and c = 2.2. Each experiment in panels B-E was performed twice. Errors represent the standard deviation taken from separate replicate measurements.
Here, we tested TB-CP-6.9l, TB-CP-6.9m, and TB-CP-6.Rfm, due to their high-affinity binding to TAR (Table 1). Like TB-CP-6.9a, each cyclic peptide blocked the TAR interaction with the HIV Tat ARM peptide as indicated by the absence of titration heats while the Tat ARM was added to each preformed TAR-cyclic peptide complex (Fig. 5, B and D). As an additional control experiment, we preformed the HIV TAR-Tat ARM complex and then titrated TB-CP-6.9a into the mixture, which was held in the ITC sample cell (i.e., the backward arrow extending from the TAR-Tat ARM complex in Fig. 5A). The results demonstrated that the cyclic peptide displaced the Tat ARM peptide (Fig. 5E). Although the apparent K D of TB-CP-6.9a for TAR was weaker in the presence of the Tat ARM (average K D of 17.5 ± 1.1 μM), the cyclic peptide still appeared to bind TAR with 1:1 stoichiometry. The ability of our cyclic peptides to block binding by the Tat ARM and to release the Tat ARM peptide from a bound state suggests that there is overlap between the site of cyclic peptide binding to TAR and an RNA site recognized by the Tat ARM. This observation, along with TAR-TBP cocrystal structures and TAR-variant structure-activity analysis (above), localizes cyclic peptide binding to the major groove of the TAR internal bulged loop.

TBP-derived cyclic peptides use an arginine fork motif for TAR binding
A key prediction from the structures of HIV TAR in complex with TBPs is that our designer cyclic peptides, such as TB-CP-6.9a, will exploit a specific arginine fork motif for TAR recognition (38,40). The arginine fork is exemplified by the TBP6.9 interaction with TAR (Fig. 1C). Here, the Hoogsteen edge of conserved base Gua26 accepts hydrogen bonds from the Nη1 and Nη2 moieties of the R47 guanidinium group. R47 also makes a salt bridge and a hydrogen bond to the pro-R p oxygen and O5 0 of Uri23, respectively. Cation-π interactions also occur between the R47 guanidinium group and nucleobases from Ade22 and Uri23 above and below the side chain. These interactions give rise to a type HηηP fork (65) (Fig. 6A). Previously we showed that the R47K mutation to TBP6.7 reduced HIV TAR binding by 327-fold (38). Accordingly, we synthesized a cyclic peptide variant of TB-CP-6.Rf-m in which the R47-equivalent position was changed to lysine. We chose TB-CP-6.Rf-m because this cyclic peptide binds TAR with 0.8 ± 0.1 μM affinity ( Fig. 6B and Table 1). In contrast, the R47K variant produced an average K D of 8.2 ± 1.0 μM, yielding a K rel of 10 ( Fig. 6C and Table 1). Like the parental TB-CP-6.9f-m peptide, the R47K variant shows binding driven by enthalpy (ΔH of −9.5 ± 1.0 kcal mol −1 ) countered by unfavorable entropy (-TΔS of +2.8 ± 0.9 kcal mol −1 ). The ΔΔG of +1.5 kcal mol −1 is consistent with loss of two or more hydrogen bonds or a saltbridge interaction.
We next investigated the ability of the R47K cyclic peptide to block the Tat-ARM from TAR binding. By analogy to TBPs, we hypothesized that R47 is a crucial determinant of our cyclic peptides in blocking TAR recognition by the Tat-ARM. As expected, a competition assay revealed that the R47K variant was unable to compete with the Tat-ARM for TAR binding. Specifically, the Tat-ARM showed binding to a preformed complex of TAR-(TB-CP-6.Rf R47K), which yielded an apparent K D of 2.8 ± 1.1 μM (Fig. 6D).
The Tat-ARM was shown previously to recognize TAR at two binding sites (40,66), but binding here was consistent with a one-site binding during titration of the Tat ARM peptide into the preformed TAR-TB-CP-6.Rf R47K complex. This suggests that the R47K variant incompletely blocks Tat-ARM recognition of TAR, supporting a case for arginine-forkbased TAR recognition. In contrast, cyclic peptides that contain R47, such as TB-CP-6.9a, TB-CP-6.9l, TB-CP-6.9m, and TB-CP-6.Rf, bind TAR with 1:1 stoichiometry while completely abrogating Tat-ARM association with the TARpeptide complex (Fig. 5, B-D). Notably, replacement of other arginine positions in cyclic peptides, such as the R48T variation that differentiates TB-CP-6.9a and TB-CP-6.3a, led to a slight increase in affinity (K rel of 0.6) ( Table 1). This comparison shows that not all arginine changes in cyclic peptides are deleterious. The results underscore the importance of arginine at a position equivalent to R47 in the cyclic peptide, which is expected if cyclic peptides engage TAR via an arginine fork, as observed in all known TAR-TBP cocrystal structures (38,40).

Methylene and naphthalene-linked cyclic peptides exhibit antiviral activity
We next evaluated the most promising TAR-binding cyclic peptides for antiviral activity, namely TB-LP-6.9, TB-CP-6.9a, TB-CP-6.Rl, TB-CP-6.Ra, TB-CP-6.Rf-m, TB-CP-6.Rm, TB-CP-6.9i, TB-CP-6.9l, TB-CP-6.9f-m, and TB-CP-6.3f-m. Linear peptide TB-LP-6.9-derived from TBP6.9 and described previously (40)-was included as reference. For these studies, a single-round infectivity assay was employed in which pseudotyped HIV-1 was produced by HEK293T cells (67,68). The virus is made from a modified proviral HIV-1 plasmid lacking the env gene that is replaced by the vesicular stomatitis virus (VSV) G-protein. A gene encoding green fluorescent protein replaces the nef gene, allowing virion production to be tracked in producer cells. A range of concentrations of each cyclic peptide was incubated for up to 1 h with TZM-bl cells prior to infection with pseudotyped virus. Viral infection was monitored after 48 h by detection of luciferase activity resulting from TAR-mediated expression that depends on viral infection and Tat binding (69). As a control for antiviral activity, we used temacrazine, which selectively inhibits viral transcription without affecting cellular genes (70). Of the cyclic peptides tested, no appreciable antiviral activity was observed for TB-CP-6.Ra, TB-CP-6.Rl, TB-CP-6.Rf-m, TB-CP-6.Rm, TB-CP-6.9i, TB-CP-6.9f-m, and TB-CP-6.3f-m compared with control levels of infectivity measured in the absence of cyclic peptide (Fig. S4). Indeed, peptides such as TB-CP-6.Ra and TB-CP-6.9i enhanced infectivity by 300 and 150% at 200 μM concentrations. The positive charge of the arginine residues within the cyclic Targeting HIV TAR with a focused library of cyclic peptides peptides could cause cells to become more susceptible to virus uptake-a phenomenon similar to that reported for polybrene (hexadimethrine bromide) whose polycationic character enhances pseudotyped virus capsid binding to the cell membrane (71)(72)(73).
In contrast, we observed a dose-dependent decrease in viral infectivity for TB-CP-6.9a (Fig. 7A) and TB-LP-6.9 (Fig. S4). Whereas TB-LP-6.9 showed only a modest 19% reduction of viral infectivity at 200 μM concentrations-consistent with its somewhat poorer K D for TAR of 13.8 ± 5.6 μM (40), its methylene-linked counterpart TB-CP-6.9a reduced infectivity by 47% at 250 μM. Fitting a dose-response curve for this cyclic peptide yielded an IC 50 of 410 ± 18 μM (Fig. 7A). A cellular toxicity assay was then used to derive a CC 50 , which was estimated to be 2200 μM (Fig. 7A). The selectivity index (SI) of this compound, as given by the ratio of CC 50 to IC 50 , is 5.5, indicating that TB-CP-6.9a exhibits good selectivity in the cellbased activity assay albeit with modest potency.  (Table 1 and Fig. S1P). B, representative thermogram of the TB-CP-6.Rf cyclic peptide titrated in TAR. C, representative thermogram of the TB-CP-6.Rf R47K mutant peptide titrated into TAR. The lysine substitution diminishes TAR affinity by tenfold, indicating the importance of arginine at a position equivalent to 47. Average thermodynamic parameters are reported in Table 1. D, representative thermogram of the Tat-ARM titrated into the preformed complex of TAR-TB-CP-6.Rf R47K. The Tat-ARM binds to the complex with an average K D of 2.8 ± 1.1 μM, displaying a 1:1 stoichiometry. Each ITC experiment was performed twice. Errors represent the standard deviation taken from separate replicate measurements.
The goal of this investigation was to find new compounds that disrupt the HIV TAR RNA interaction with the viral protein Tat-an essential complex for viral replication (Fig. 1A) (16,(74)(75)(76). HIV TAR is appropriate as a therapeutic target because of its high sequence conservation and low mutation frequency (77). Nonetheless, TAR has eluded thus far the development of inhibitors suitable for clinical applications (16,17,78,79). Toward this goal, we used a nonconventional approach in which lab-based evolution was applied to the human U1A RNA-binding protein to repurpose it for highaffinity TAR binding (Fig. 1B) (39). Cocrystal structures of TAR-TBP complexes revealed that the majority of binding arises from the β2-β3 loop amino acids of TBPs (Fig. 1C). This finding suggested that the loop could function as a standalone TAR-recognition motif (38,40). Here, we reported that the design of a library of cyclic peptide mimics of this TAR recognition motif and analyzed the influence of peptide cyclization linkers and arginine composition on their affinity and selectivity toward TAR, TAR mutants, and off-target RNAs, as well as their ability to interfere with Tat ARM binding to TAR and their antiviral properties in cell culture. Our studies demonstrate that our "TBP β2-β3 loop mimics" can provide promising agents for selective recognition of TAR and inhibition of the TAR-Tat interaction in vitro. Importantly, some of these cyclic peptides exhibit antiviral activity in cells, while displaying relatively low cytotoxicity.
Peptides continue to play key roles in the therapeutic armamentarium with >60 FDA-approved peptide drugs available, including the anti-HIV peptide Fuzeon (80,81). As part of our design strategy, the TAR-binding β2-β3 loop motif of TBPs was translated into macrocyclic peptides constrained by an inter-side-chain cross-link via cysteine bis-alkylation. Cognizant of the impact of backbone modifications on the conformational-and thus functional-properties of cyclic peptides (82)(83)(84), variation of the linker structure was exploited to tune the affinity of these TBP mimics toward TAR, resulting in TAR affinities ranging from 24.2 to 0.8 μM (Table 1). Of note, subtle structural modifications at the level of the linker, including some as small as single-atom substitutions (e.g., TB-CP-6.9h versus TB-CP-6.9f-m or TB-CP-6.9c versus TB-CP-6.9d), were found to impart important changes in TAR affinity (3-6-fold difference in K D ), Figure 7. Antiviral and cell viability assays of cyclic peptides in HIV-1 infectivity assays. A, TB-CP-6.9a reduces HIV-1 infectivity in a dosedependent manner, displaying a half-maximal inhibition (IC 50 ) of infectivity at 410 ± 18 μM and CC 50 (cytotoxicity) of 2200 μM. The selectivity index (SI) is the ratio of cytotoxicity to antiviral activity, which was calculated to be 5.5. Here and elsewhere, values denoted by an asterisk are considered estimates due to extrapolations of CC 50 values. B, TB-CP-6.9l reduces HIV-1 infectivity in a dose-dependent manner, displaying an IC 50 of 461 ± 26 μM and a CC 50 of 819 ± 30 μM. The SI was 1.8. C, TB-CP-6.9m reduces HIV-1 infectivity in a dose-dependent manner, displaying an IC 50 of 233 ± 5 μM and a CC 50 of 327 ± 7 μM. The SI was calculated to be 1.4. Each data point is the result of three biological replicate measurements where error bars represent standard deviations. Standard errors for IC 50 and CC 50 are defined in GraphPad.
Targeting HIV TAR with a focused library of cyclic peptides highlighting the sensitivity of these cyclic peptides to conformational changes imposed by the intramolecular crosslink. A general trend emerging from structure-activity analysis of our library of 16 TAR-targeting cyclic peptides is that shorter and/ or more rigid linkers are associated with tighter TAR binding (e.g., CH 2 > CH 2 CH 2 > CH 2 CH 2 CH 2 ; CH 2 C(O)CH 2 > CH 2 CH 2 CH 2 ; m-xylene >> p-xylene). This may reflect the fact that these linkers allow for the cyclic peptide to adopt a conformation that better resembles that of the β2-β3 loop in the TAR-TBP complexes (Fig. 1C).
Indeed, although cocrystals structures of our cyclic peptides in complex with TAR are not yet available, our results suggest that TB-CP-6.9a-and related peptides of this study-interacts with TAR in a manner comparable to TBP6.9 recognition of TAR (Fig. 1C) (40). For example, the Δ25 mutant, which eliminated this nucleotide from the central bulge, had little effect on TAR binding (Fig. 4C). This finding is consistent with cocrystal structures wherein Cyt25 bulges outside the core fold (38,40). Similarly, replacement of the apical loop with a GNRA tetraloop had no appreciable effect on TB-CP-6.9a affinity (Fig. 4D). This finding concurs with cocrystal structures wherein no contacts occur between the TAR apical loop and the β2-β3 loop (Fig. 1C). In contrast, the Δ35 mutant indicates a key role for this nucleotide in high-affinity TB-CP-6.9a binding to TAR, in accord with the close proximity of this bulged base to the β2-β3 loop (e.g., Fig. 1C) (40). These findings pinpoint cyclic peptide binding near the TAR Uri23-⋅Ade27-Uri37 base triple, which implies arginine-mediated readout at the Hoogsteen edges of Gus26, Gua28, and Gua36.
Additional evidence of major groove recognition by cyclic peptides is based on competition experiments using the Tat ARM. This peptide was unable to bind TAR when specific cyclic peptides were present in preformed complexes (Fig. 5, B-D). Previously, TB-CP-6.9a produced an analogous result, whereas its linear counterpart, TB-LP-6.9, did not (40). Conversely, we found that preformed TAR-Tat complexes did not impede binding of TB-CP-6.9a (Fig. 5E). One plausible explanation of these results is that cyclic peptides compete with the Tat ARM for overlapping binding sites. Indeed, a solution structure of the TAR-Tat ARM complex shows direct overlap of the polypeptide with the site of TBP β2-β3 loop recognition (28,33,40). Although kinetic measurements of Tat ARM binding to TAR have been unsuccessful, it is plausible that the viral peptide dissociates more rapidly than cyclic peptides. Indeed, the NMR ensemble of the HIV TAR-Tat complex reveals substantial dynamics and flexibility (28).
Another explanation that could account for cyclic peptide obstruction of Tat binding to TAR comes from Evrysdi. This FDA-approved drug is used to treat spinal muscular atrophy (SMA) caused by SMN1 gene mutations. Although this drug binds with a modest affinity of 28 ± 9 μM, it targets the E7 5 0 -splice site of SMN2 pre-mRNA (85). The resulting conformational changes render the message accessible to the U1-C zinc-finger (86), which promotes E7 inclusion in the spliced transcript leading to a functional compensatory SMN2 protein (87,88). This result illustrates that a small molecule can influence RNA conformation to induce an alternative function.
Our cyclic peptides could conceivably alter the TAR RNA conformation as well, preventing Tat from recognizing the TAR major groove.
To compete with the Tat ARM for TAR binding, our cyclic peptides were designed to use an arginine fork motif. This motif is a key determinant of major groove recognition that is observed in all TAR-TBP cocrystal structures (e.g., Fig. 6A) (38,40). Specifically, R47 stabilizes the Uri23⋅Ade27-Uri37 base triple, which is a hallmark of most ligand-bound TAR structures (33). Solution NMR pinpoints arginine binding to Gua26 of TAR (28,89) and these observations collectively explain the finding that at least one arginine in the Tat ARM is required to promote TAR binding and viral transactivation (28,90). Evidence that our cyclic peptides use arginine-fork readout is based on TAR affinity shown by the TB-CP-6.Rfm R47K variant. This mutation reduced binding by tenfold and rendered the RNA susceptible to Tat ARM binding (Fig. 6, C and D). These findings allow us to infer two important properties regarding cyclic peptide binding to TAR. First, loss of TAR affinity by the R47K mutation in TB-CP-6.Rf-m is consistent with disruption of arginine fork recognition of the major groove. We hypothesize that fork ablation promotes TAR major groove expansion-as observed in the unliganded state (33)-as a preface to Tat-ARM binding. Second, the R47K variant along with other peptides, such as TB-CP-6.9a, TB-CP-6.9f-m, TB-CP-6.9l, and TB-CP-6.9m that effectively compete with the Tat-ARM, each possesses four arginines. Our ITC analysis suggests that the ability to block the Tat-ARM varies based on arginine positioning in the sequence to recognize a specific geometry-not on the number of arginines. These results concur with our previous observations about arginine composition and placement in the context of TBP β2-β3 loop recognition of HIV TAR (40).
Significantly, some cyclic peptides of this investigation exhibit noticeable, albeit modest, antiviral activity, as measured by a cell-based infectivity assay (IC 50 values in a range of 230-460 μM). Among these, TB-CP-6.9a and TB-CP-6.9m show the most promising activity, while exhibiting comparatively low and modest cytotoxicity, respectively. Interestinglyand in contrast to the clear SAR trends observable for these compounds with respect to TAR affinity-both the linker structure and peptide composition were found to elicit idiosyncratic effects in terms of antiviral activity. In addition, a strict correlation between in vitro TAR affinity and antiviral activity was not observed, which might be attributable to the differential ability of these compounds to penetrate cells and reach the cytoplasm. For example, while the meta-xylene linker produced relatively tight TAR binding across multiple peptide sequences (e.g., K D of 0.8-2.0 for TB-CP-6.9f-m, TB-CP-6.3f-m, and TB-CP-6.Rf-m), these compounds showed low to no antiviral activity (Fig. S4). In contrast, while showing slightly weaker binding to TAR (K D values of 3.7-5.0 μM), naphthalene-containing peptides proved effective as antivirals in the context of the TBP6.9 sequence (Fig. 7, B and C). This difference may be the result of improved cytosolic localization of these compounds, as observed for other naphthalenecontaining peptides (91-93).
Arginine-rich linear peptides, such as the HIV Tat ARM, readily enter cells (94)(95)(96)(97)(98)(99)(100), and arginine-rich cyclic peptides -with molecular weights and arginine compositions analogous to those described here-have been reported to undergo facile cell entry with no toxicity (56,91,93,(101)(102)(103). Surprisingly, we found that our penta-arginine cyclic peptides (i.e., TB-CP-6.Ra and TB-CP-6.Rf-m) failed to elicit antiviral activity (Fig. S4), despite their relatively high affinity for TAR (K D = 0.8-1.0 μM). Similarly, both naphthalene-linked peptides based on the TBP6.R sequence showed no antiviral activity (Fig. S4), while their TBP6.9 counterparts with one less arginine showed dose-dependent antiviral activity (Fig. 7, B and C). Overall, these results indicate that while beneficial for the TAR interaction, an increase in the arginine content in these cyclic peptides is not necessary or sufficient to elicit antiviral activity in cells.
RNA-targeting cyclic peptides are still at nascent stage of development. In this work, we demonstrated the design and development of a new class of macrocyclic peptides capable of targeting HIV-1 TAR RNA with high affinity and selectivity at the binding interface between TAR RNA and the Tat ARM peptide. Furthermore, some of these compounds were also found to interfere with the viral lifecycle in cells, albeit with modest activity. We envision that both the in vitro affinity and cellular activity of these compounds can be further optimized by leveraging the structure-activity insights gained from the present investigation. As such, this work provides a foundation for further development of cyclic peptide-based anti-HIV agents as well as a framework for a better understanding of RNA recognition by small macrocyclic peptides.

RNA synthesis and purification
HIV-1 TAR 27-mer was used for all ITC experiments. The RNA was in vitro transcribed and purified by denaturing gel electrophoresis (104). The product was desalted, lyophilized, and stored at −20 C until needed. RNAs used for specificity and off-target analysis such as BIV TAR, TAR Δbulge, TAR-GNRA, TAR-GNRAΔ35, TAR ΔU25, U1hpII RNA, and 7SK RNA were synthesized (Horizon Discovery Group) and purified as described (38).

Peptide library synthesis and purification
Synthesis of sequences provided in Table S1 was carried out by conventional Fmoc-solid-phase peptide synthesis on Knorr amide resin (0.4 mmol g −1 on a 0.1-mmol scale). Peptides were generated manually in syringes equipped with a Teflon frit. Initially, the resin was subjected to swelling for 30 min in 5 ml of a 1:1 mixture of dichloromethane/N,N-dimethylformamide (DCM/DMF). Swollen resin was then treated with 20% piperidine in DMF containing 0.05 M 1-hydroxybenzotriazole (3 cycles of 3 min, 5 min, and 3 min) to achieve the deprotection of Fmoc. After washing with DMF (5×), the resin was loaded with four equivalents of amino acid, four equivalents of HCTU, and eight equivalents of N,N-diisopropylethylamine (DIPEA) with respect to the initial loading of the resin. The coupling reaction was left at room temperature for 40 min with shaking. This process was repeated to attain the desired peptide sequence. Upon final Fmoc-deprotection, the peptide N-terminus was acetylated using ten equivalents of acetic anhydride and 20 equivalents of DIPEA in DMF for 30 min.
The polypeptide product was cleaved from resin as follows: the resin was washed with DMF (3 × 5 ml), DCM (3 × 5 ml) and dried in vacuo. A 10 ml volume of cleavage mixture comprising 95% (v/v) TFA, 2.5% (v/v) triisopropylsilane and 2.5% (v/v) water was added to the resin and shaken 2 h at room temperature. The filtrate was collected and concentrated under reduced pressure and was mixed with a tenfold excess volume of cold diethyl ether and centrifuged. The crude precipitated peptide was dissolved in 10 ml of 1:1 acetonitrile:water and lyophilized.
Peptide cyclization was performed as follows: lyophilized peptide (4 mmol) was dissolved in 20 ml of 0.05 M ammonium bicarbonate buffer in acetonitrile: water (1:1), pH 8.5. Tris(2-carboxyethyl)phosphine (TCEP) (8 mmol, 2 equivalents) was added, and the reaction mixture was shaken at 37 C for 1 h. The respective bis-halide linker (24 mmol, six equivalents) dissolved in DMF (0.20 ml) was added, and the reaction mixture was shaken further overnight. Cyclization progress was monitored by MALDI-MS. Upon quantitative cyclization, the solution was lyophilized and the cyclic peptide was purified by semipreparative HPLC using a gradient of 5 to 30% acetonitrile over 30 min. The average overall yield was 10%. The pure and lyophilized peptide was treated further with 0.10 M HCl and lyophilized to remove TFA.

Isothermal titration calorimetry
Lyophilized TAR RNA was dissolved in an RNA dilution buffer of 0.01 M HEPES pH 7.5 and heated to 65 C for 3 min. TAR folding buffer containing 0.01 M Na-HEPES pH 7.5, 0.05 M NaCl and 0.002 M MgCl 2 was added and incubated at 65 C for 3 min. The RNA was cooled to room temperature over 5 to 6 h. The RNA was dialyzed overnight against 4 l of ITC buffer (0.05 M Na-HEPES pH 7.5, 0.05 M NaCl, 0.05 M KCl, and 0.002 M β-ME). All experiments were performed using a PEAQ ITC (Malvern Panalytical) with peptide in the syringe and RNA in the cell. RNAs used for structure-activity analysis were handled similarly. All ITC thermograms were measured at 20 C unless noted. In each experiment, 200 or 300 μM of the cyclic peptide in the syringe was titrated into the cell containing 15 or 20 μM TAR RNA. Data were analyzed using PEAQ-ITC software (Malvern Panalytical) and all thermograms fit best to one-site binding models based on criteria described (62). Average thermodynamic parameters from duplicate measurements are reported in Figure 3 and Table 1. Representative thermograms are provided in Figure S2. Errors represent the standard deviation derived from replicate measurements.

Competition ITC assays
Tat competition experiments were performed as described elsewhere (40). Briefly, pure Tat-ARM peptide was prepared Targeting HIV TAR with a focused library of cyclic peptides from Genscript, Inc; the sequence (GISYGRKKRRQRRRAHQ) (28) contains an acetylated N-terminus and an amidated C-terminus. In each competition assay, 400 μM of cyclic peptide was titrated into 20 μM TAR in the sample cell. All competition and reverse competition experiments were conducted at 20 C. Thermograms were analyzed using PEAQ-ITC analysis software and fit best to a one-site binding model. Subsequently, the Tat-ARM at a concentration of 500 μM was titrated from the syringe into the TAR-cyclic peptide complex preformed in the sample cell during a preceding ITC experiment. Most resulting thermograms showed no substantial heats of binding and data failed to fit available binding models (Fig. 5, B-D). One exception was TB-CP-6.Rf R47K, which showed discernible binding (Fig. 6D). For reverse competition assays, the Tat-ARM peptide at 2.6 μM was titrated dropwise into 40 μM HIV TAR RNA and incubated on the bench for 1 h at room temperature. TB-CP-6.9a at a concentration of 620 μM was then titrated into the preformed Tat-TAR complex in the ITC sample cell. The resultant thermogram showed substantial heats of binding and fit best to a one-site binding model.

Pseudotyped HIV-1 infectivity and cell toxicity assays
The antiviral activity of TAR-binding cyclic peptides was measured by single-round infectivity assays using pseudotyped HIV-1 produced by HEK293T cells (67). A single round of infectivity was achieved by the transient cotransfection of the viral vector and a plasmid expressing VSV-G envelope. After 24 h, the p24 viral capsid protein was quantified by ELISA to normalize the viral load. Individual cyclic peptides or temacrazine was added 1 h prior to the infection of TZM-bl reporter cells with p24 normalized virus. Each well contains 10 4 cells and equal viral loads. The reporter cells express luciferase in a TAR-dependent manner following successful HIV infection. Luminescence was used as a direct measure of viral infectivity in the presence of cyclic peptide. In parallel with infectivity, cytotoxicity was tested across the same dose ranges using temacrazine or peptide-treated mock-exposed cells. At 48 h after treatment, the CellTiter-Glo Luminescent Cell Viability Assay (Promega Corp.) was performed according to the manufacturer's instructions. The infectivity and cytotoxicity experiments were performed in triplicate in 96well plates. IC 50 and CC 50 values were calculated from the Richard's five-parameter dose-response curve of values plotted by percent of control infectivity and cytotoxicity in Prism v9.0 (GraphPad Software Inc).

Data availability
All data produced in this investigation are provided in the manuscript. Individual reagents can be obtained by contacting the corresponding authors.
Supporting information-This article contains supporting information.
Acknowledgments-We thank members of the Wedekind and Fasan labs for advice during the preparation of this manuscript. We appreciate the technical assistance of Dr Jermaine L. Jenkins from the University of Rochester Structural Biology and Biophysics Facility.
Funding and additional information-This work was supported in part by National Institutes of Health grant R01 AI150463 (J. E. W.) and grant R01 GM134076 (R. F.). R. B. was supported in part from National Institutes of Health training grant T32 GM135134 and an Elon Huntington Hooker graduate fellowship. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
Conflict of interest-Corresponding author Joseph E. Wedekind has a pending patent application (USPTO #16/723164) for work disclosed in this article.