Trisubstituted 1,3,5-Triazines: The First Ligands of the sY12-Binding Pocket on Chemokine CXCL12.

CXCL12, a CXC-type chemokine, binds its receptor CXCR4, and the resulting signaling cascade is essential during development and subsequently in immune function. Pathologically, the CXCL12-CXCR4 signaling axis is involved in many cancers and inflammatory diseases and thus has sparked continued interest in the development of therapeutics. Small molecules targeting CXCR4 have had mixed results in clinical trials. Alternatively, small molecules targeting the chemokine instead of the receptor provide a largely unexplored space for therapeutic development. Here we report that trisubstituted 1,3,5-triazines are competent ligands for the sY12-binding pocket of CXCL12. The initial hit was optimized to be more synthetically tractable. Fifty unique triazines were synthesized, and the structure-activity relationship was probed. Using computational modeling, we suggest key structural interactions that are responsible for ligand-chemokine binding. The lipophilic ligand efficiency was improved, resulting in more soluble, drug-like molecules with chemical handles for future development and structural studies.

Chemokines are small, chemotactic, proinflammatory proteins.[1] Through G-protein-coupled receptor (GPCR) signaling, chemokines are responsible for physiological processes such as cell trafficking, immune surveillance, organogenesis, angiogenesis, and embryogenesis.[2] Under pathological conditions, this same chemoattractant property implicates chemokines in many diseases, including inflammatory and autoimmune disorders, cardiovascular disease, and cancer. CXCL12 is a constitutively expressed CXC-type chemokine that binds to chemokine receptors CXCR4 and ACKR3 and is essential during embryonic development.[3] After development, the main function of CXCL12 is to mediate the inflammatory response, participate in immune surveillance, and maintain tissue homeostasis by trafficking lymphocytes to tissues such as the lymph nodes, lungs, and bones. Human cancers hijack this process by upregulating chemokine receptors; CXCR4 is upregulated in over 20 human cancers, thereby allowing metastasis to areas of the body producing CXCL12.[4] Small molecules capable of disrupting the CXCL12–CXCR4 signaling axis have demonstrated value,[5] and extensive research has been dedicated to this end, as discussed below.

Traditionally, efforts have been directed at CXCR4 antagonism.[6] However, most clinical trials targeting this GPCR have failed because of toxicity. Though originally halted in Phase II clinical trials for HIV because of cardiotoxicity but now approved for the treatment of non-Hodgkin’s lymphoma and multiple myeloma, AMD3100 (Plerixafor) is the only FDA-approved CXCR4 antagonist to date.[7] Recently, the peptide BL-8040 (Motixafortide) entered Phase II clinical trials, demonstrating promise in the treatment of AML[8,9] and pancreatic cancer.[10] Finally, AMD-070, an orally available analogue of Plerixafor, is in Phase III clinical trials[11] for WHIM syndrome.[12] Despite the tremendous clinical potential for inhibiting CXCL12–CXCR4 signaling, the low success rate and potential toxicity of CXCR4-targeting strategies emphasizes the need for alternative strategies to disrupt this signaling pathway.

An alternative solution is to target the chemokine directly instead of the receptor. Through a “two-step, two-site” binding and activation process,[13] CXCL12 is involved in extensive protein–protein interactions (PPIs) with CXCR4, and these interactions provide an attractive avenue for small-molecule development.[14] For example, a chalcone[15] and its derivatives[16] bind CXCL12, prevent CXCR4 activation, and are active in vivo in models of allergic airway diseases and pulmonary hypertension.[17] However, additional structural studies of the binding of these molecules to CXCL12 is needed. Additionally, NOX-A12, a non-orally bioavailable RNA oligonucleotide that neutralizes CXCL12,[18] is in Phase I/II clinical trials for neurological, colorectal, and pancreatic cancer treatment, and NOX-A12 was recently approved for increased dosing and further studies in humans.[19] Nevertheless, there are no approved therapies targeting CXCL12, and a small molecule targeting CXCL12 has yet to enter any human trial.

The initial step in the formation of the active CXCL12–CXCR4 signaling complex is the binding of sulfotyrosine (sY) residues on the extracellular N-terminus of CXCR4 to the conserved core of CXCL12. CXCL12 has three unique “hot spots” for binding of sY residues on CXCR4: sY7, sY12, and sY21.[20] Of these, the interactions between CXCL12 and sY12 and sY21 of CXCR4 are responsible for most of the binding energy of the complex and confer most of the specificity of the interaction. Therefore, small molecules that bind in these sY-binding pockets are predicted to interrupt the CXCL12–CXCR4 signaling axis by inhibiting the initial formation of the complex.

In 2010, we demonstrated this proof of principle using a structure-based in silico/NMR approach leading to a small molecule that bound to CXCL12.[21] Lead optimization resulted in a set of tetrazole-containing compounds that bind with micromolar affinity to the sY21 binding pocket of CXCL12 and interrupt CXCL12–CXCR4-mediated chemotaxis in vitro.[20]

Achieving this success led us to more broadly investigate the feasibility of individually targeting each of the sY binding sites, resulting in the identification of compound 1 (Table ) through in silico screening of the ZINC library.[22] Compound 1 binds the sY12 pocket of CXCL12.[23,24] Here we investigated the structure–activity relationships (SARs) of 1 with the goal of decreasing the lipophilicity while maintaining or increasing the potency. We anticipate that these studies will enable future fragment-linking campaigns with fragments that bind adjacent sY-binding pockets on CXCL12 to discover novel and potent CXCL12-targeting inhibitors.

Table 1

Initial SAR by Catalog

compound	R1	R2	Kd (μM)a
1	–SCH2CO2H	–Ph	169 ± 12
2	–SCH2CO2H	–CH3	∼1200
3	–H	–Ph	∼900
4	–H	–CH3	NB

NB = nonbinder.

To determine the binding affinity of the molecules to CXCL12, we employed 2D NMR spectroscopy to monitor chemical shift perturbations of amino acid residues residing in the sY12 binding pocket upon compound binding (Figure A). Prior studies in the laboratory[23,24] indicated that upon small-molecule binding to the sY12 pocket, the residues consistently demonstrating the largest and most specific shifts are I28, V39, and A40 (Figure B). Affinity was determined by nonlinear regression of the chemical shift perturbations plotted over a titratable range of concentrations (Figure C).[25] Because the overarching goal of the study was to develop a highly soluble molecule with potent binding to the sY12 binding pocket of CXCL12, only compounds that were soluble to ≥1600 μM and induced a chemical shift perturbation of ≥0.5 ppm in at least two of the three residues (I28, V39, and A40) were labeled as “binders”.

Figure 1

Data from titration of CXCL12 with compound 1. (A) HSQC spectrum of CXCL12 upon titration with increasing concentrations of 1. (B) Depiction of the maximal chemical shift perturbation of CXCL12 residues upon binding of 1 at 1600 μM. Residues I28, V39, and A40 are highlighted in orange. (C) Titration curves for binding of 1 to CXCL12.

Initially we performed a brief SAR by catalog to investigate which features of 1 are important for interactions with CXCL12 (Table ). Replacing the phenyl group with a methyl group (2) decreased the binding affinity about 7-fold (Table ). Removing the thioglycolic acid moiety from the 2-position of the pyrimidine (3) also decreased binding to CXCL12, resulting in a Kd of about 900 μM—a 5-fold decrease from the parent molecule (Table ). Finally, replacing the phenyl group with a methyl group at the 4-position and simultaneously removing the thioglycolate substituent at the 2-position (4) abolished binding to CXCL12 (Table ). This suggests that the hydrophobic 4-position and the two carboxylic acids work synergistically to bind the sY12 site of CXCL12. With this knowledge, we further investigated this scaffold to delineate important characteristics for binding via SAR and ultimately to develop more soluble and potent small molecules with favorable characteristics for future fragment linking.

Improving the simplicity of the synthesis of the target compounds was imperative to our ability to rapidly generate a library of compounds. Therefore, we began by transforming the asymmetrical pyrimidine ring on 1 into the symmetrical 1,3,5-triazine, resulting in 5 (Figure ). The ring alteration had no effect on binding of the molecule to CXCL12 (158 ± 66 μM for 5 vs 169 ± 12 μM for 1; Figures and S1). While maintaining binding, this scaffold increased the ease of synthesis by creating a symmetric molecule and allowed inexpensive, commercially available cyanuric chloride to be used as a starting material for many of the analogues. The triazine also imparts increased polarity and hydrophilicity to the molecule, thereby improving the drug-like characteristics. These improvements led us to maintain the triazine scaffold throughout the rest of our investigation. The triazine was then divided into two modular sections for SAR studies: (1) the side arms of the northern hemisphere and (2) the characteristics of the southern hemisphere attached to the triazine (Figure ).

Figure 2

Replacing the pyrimidine ring with a triazine ring retains the binding properties.

We began by investigating various side-arm substitutions. These analogues (5, 7, and 9a–m) were readily accessible from commercially available 2,6-dichloro-4-phenyl-1,3,5-triazine (6) via nucleophilic aromatic substitution by refluxing in the presence of the appropriate nucleophile (Scheme ). First, we replaced the thioglycolate side arm with glycine, resulting in 9a, which retained binding affinity to CXCL12 (Table , entry 2, and Figure S2). This S-to-N substitution allowed us to use readily available (and less noxious) amino acids in our SAR studies. We then turned to the other functional group on the side arm—the carboxylic acid moiety—to investigate its role in binding to CXCL12. Hypothesizing that the acid is necessary for creating a salt bridge, we blocked the free acid as a tert-butyl ester (7) and as a ketone (9b). These changes destroyed binding of the compound to CXCL12 (Table , entries 3 and 4). Next, we probed the effect of lengthening the side arm. We first synthesized 9c as a homologated variant of 5. The homologation had no effect on binding, as 9c had a nearly identical affinity of 158 ± 73 μM (Table , entry 5, and Figure S3). However, there was a limit on the tolerable distance between the triazine core and the acid. This was apparent because homologating 9c by one carbon afforded 9d, which did not bind CXCL12 (Table , entry 6). Compound 9e, the amino variant of 9d, also displayed no binding (Table , entry 7).

Scheme 1

Synthesis of Triazines with Various Side Arms

Table 2

SAR of Various Side Arms on the Triazine

NB = nonbinder.

9m caused CXCL12 to precipitate.

Interested in whether steric bulk at the α-carbon would be tolerated in the setting of a constant distance between the triazine and the acid, we synthesized l-norleucine-derived triazine 9f, which bound with Kd = 156 ± 76 μM (Table , entry 8), demonstrating that indeed, substitution at the α-carbon actually improved the binding ∼2-fold compared with 9a. Interestingly, 9g, the enantiomer of 9f, still bound to CXCL12 (Table , entry 9), suggesting that the n-butyl side chains of norleucine are largely oriented outside the critical interactions within the binding pocket.

Finally, we examined whether carboxylic acid isosteres with varying pKa or hydrogen-bonding ability would bind as effectively. Amide 9h, trifluoromethyl-substituted alcohol 9i, and alcohol 9j are all capable of donating a hydrogen bond. However, none of them bound CXCL12 (Table , entries 10–12). To investigate different acidities, we synthesized hydroxamic acid 9k (pKa ∼ 8–10) and sulfonic acid 9l (pKa ∼ −2), and we synthesized benzoic acid 9m as a structural variant of the necessary carboxylic acid side chain. Compounds 9k and 9l did not demonstrate binding (Table , entries 13 and 14). Compound 9m caused aggregation and precipitation of the protein, revealing no spectra upon acquisition (Table , entry 15).

These results suggest that specifically the carboxylic acid moiety is critical for CXCL12 binding affinity. For structural insight into the observed specificity, we turned to computational docking. In silico docking using Glide generated 10 000 potential low-energy binding poses for the interaction of 5 with CXCL12. These poses were filtered using experimental constraints by predicting the chemical shift perturbation for each pose and correlating the predicted chemical shift perturbations to the experimental ones.[26] These top-performing filtered poses suggest that a salt bridge is formed between K27 and the carboxylic acid side arm of 5, consistent with the carboxylate moiety being necessary for efficient binding (Figure ). Compounds not able to form a salt bridge with lysine do not show any binding to the sY12 pocket of CXCL12. One exception to this is sulfonic acid 9l, which can readily form an ion pair.

Figure 3

In silico modeling. A predicted low-energy binding pose of 5 in the sY12 pocket is shown with both carboxylate arms chelating the side-chain amino group of K27 and the nonpolar phenyl ring extending toward the sY21 binding pocket.

Closer inspection of our model of 5 bound to CXCL12 revealed that both carboxylate arms are attracted inward to the lysine, providing a bidentate interaction to hold the compound in the pocket. This presumably occurs via a hydrogen-bonding interaction of one carboxylate with the ion pair formed between the other carboxylate and lysine. The necessity of this bidentate interaction is suggested in the initial SAR by the fact that 3, containing a single side arm, demonstrated >5-fold weaker binding to CXCL12 (Table ). Pike et al. demonstrated that the carboxylate anion is one of the strongest hydrogen-bond acceptors known, whereas sulfate and organosulfonates are much weaker hydrogen-bond acceptors, regardless of the countercation.[27] If the extra binding energy from the secondary hydrogen-bonding interaction with K27 is needed for efficient binding, then it is feasible that although sulfonic acid 9l may form a salt bridge, it does not have sufficient hydrogen-bond-acceptor ability to create the bidentate interaction needed to bind in the sY12 pocket of CXCL12. Additionally, the binding is hindered since the salt bridge is likely to be weaker than a carboxylate–ammonium complex because of the noncoordinating nature of an alkylsulfonic acid, necessitating less of a need for the ion pair in solution. Indeed, experiments on docking of 9l to CXCL12 failed to reveal a binding pose demonstrating bidentate interactions in the sY12 pocket. Because of the “Goldilocks effect”[28] afforded by the acidity and hydrogen-bonding properties of 5, the 2-thioglycolic acid moiety was chosen as the side arm to further study the SAR of the triazines.

Next, we investigated important features of the southern hemisphere. We divided these compounds into three groups: triazines containing (1) a 4-aryl or 4-heteroaryl substitution, (2) a 4-alkyl or 4-alkenyl substitution, or (3) a southern hemisphere linked to the triazine via a heteroatom bond.

Synthesis of the aryl and heteroaryl analogues started with cheap, commercially available cyanuric chloride (10) (Scheme ). By careful control of the temperature, nucleophilic aromatic substitution using tert-butyl thioglycolate cleanly afforded disubstituted triazine 11 in 82% yield on a gram scale. Under microwave conditions, Pd(PPh3)4-catalyzed coupling of 11 with commercially available boronic acids afforded biaryl products 12a–q. After flash purification, trifluoroacetic acid-mediated deprotection of the tert-butyl esters afforded the desired bisacids 13a–q in quantitative yield. We carried the esters through the synthesis instead of the free acids to facilitate purification of the compounds away from impurities after the Suzuki coupling. While ethyl thioglycolate is less expensive than tert-butyl thioglycolate and is typically used for similar intermediates, we found difficulties in reproducibility of ester hydrolysis and in isolation of clean bisacids. The tert-butyl ester circumvented this problem, and the final step became a simple procedure where the bisesters were treated with trifluoroacetic acid, and then removal of solvent and excess acid in vacuo provided clean product.

Scheme 2

Synthesis of Triazines Containing Aryl Southern Hemispheres

Synthesis of triazines bearing alkyl southern hemispheres was accomplished via three different methods (Scheme ). 15 was synthesized in three steps from cyanuric chloride. First, nucleophilic aromatic substitution of cyanuric chloride with benzylmagnesium bromide afforded 2,6-dichloro-4-benzyl-1,3,5-triazine (14). Subsequent microwave-mediated nucleophilic aromatic substitution with thioglycolic acid afforded 15. Despite few reports of adamantyl-group cross-couplings to triazines in the literature, 4-adamantyl-substituted triazine 16 was synthesized in good yield from 2-adamantylzinc bromide and 11 using a variation on a previously published Negishi coupling.[29] As before, simple ester deprotection afforded 17. We attempted to synthesize hexyl-substituted triazine 18 in the same fashion as 14 using hexylmagnesium bromide. However, we were unable to control the substitution to efficiently isolate the desired product. Unsurprisingly, attempting to install the hexyl group via SNAr on 11 resulted in addition to the ester. Using elegant conditions developed by Fürstner and co-workers,[30] we successfully synthesized 18 in excellent yield. Simple deprotection of the esters afforded 19. Alkenyl compound 20 was synthesized via sp2–sp2 Suzuki coupling in good yield using the pinacol ester of cyclohexenylboronic acid followed by deprotection.

Scheme 3

Synthesis of Triazines Containing Alkyl Southern Hemispheres

Scheme depicts our strategy to synthesize compounds 24a–l, which have the southern hemisphere linked to the triazine via a heteroatom. Two different routes were used. For 24a–f and 24i–k, cyanuric chloride underwent nucleophilic aromatic substitution to afford the 2,6-dichloro-4-substituted triazines 22a–f and 22i–k. These were then treated with tert-butyl thioglycolate to afford bisesters 23a–f and 23i–k. As before, acid-mediated deprotection afforded the final bisacids. For 24g, 24h, and 24l, the synthesis began with the common intermediate 11. SNAr of 11 with the appropriate nucleophiles afforded the fully substituted triazines, which then were treated with acid to afford the final products.

Scheme 4

Synthesis of Triazines Containing 4-Heteroatom-Linked Southern Hemispheres

Aniline, 1-bicyclo[1.1.1]pentylamine, cyclohexylamine, dicyclohexylamine, benzylamine, pyrrolidine, or 8-oxa-3-azabicyclo[3.2.1]octane.

2-Picolylamine, 3-picolylamine, or N-methylmorpholine.

With the library of southern hemisphere variants in hand, we investigated their binding affinity toward CXCL12, beginning with the biaryl systems. We soon learned that adding bulk to the original phenyl ring is not tolerated, as 1-naphthyl-, 2-naphthyl-, anthracenyl-, and mesityl-substituted triazines (13a–d) did not demonstrate significant binding in the sY12 pocket (Table , entries 2–5). Biphenyltriazine 13e did demonstrate minimal binding in the sY12 pocket, but unsurprisingly, the compound was insoluble, and a full titration was not feasible (Table , entry 6).

Table 3

SAR of Various Aryl Southern Hemispheres

entry	R	compd	Kd (μM)a,b
1	C6H5	5	158 ± 66
2	1-naphthyl	13a	NB
3	2-naphthyl	13b	NB
4	9-anthracenyl	13c	NB
5	mesityl	13d	NB
6	4-Ph-C6H4	13e	insoluble
7	C6F5	13f	NB
8	3,5-(CF3)2-C6H3	13g	646 ± 23
9	4-MeO-C6H4	13h	357 ± 101
10	4-morpholino-C6H4	13i	306 ± 128
11	3-OH-C6H4	13j	1798 ± 218
12	3-CO2H-C6H4	13k	133 ± 12
13	4-isoquinolyl	13l	insoluble
14	4-PhO-C6H4	13m	insoluble
15	3-furyl	13n	insoluble
16	2-furyl	13o	NB
17	3-thiophene	13p	1344 ± 376
18	2-thiophene	13q	1640 ± 190

NB = nonbinder.

13e has low solubility. 13l, 13m, and 13n caused CXCL12 to precipitate.

Electron-deficient southern hemispheres showed decreased binding affinity, as pentafluorophenyl-substituted triazine 13f did not demonstrate binding to CXCL12 (Table , entry 7). 3,5-Bis(trifluoromethyl)phenyltriazine 13g bound to CXCL12 but with weaker affinity (Table , entry 8). Conversely, electron-rich southern hemispheres were well-tolerated, as demonstrated by anisole-substituted triazine 13h with Kd = 357 ± 101 μM (Table , entry 9, and Figure S4). To increase the hydrophilicity of the molecule, we synthesized 13i containing a morpholine ring at the 4-position of the phenyl ring. This compound bound comparably to the parent triazine with Kd = 306 ± 128 μM and demonstrated qualitatively greater solubility when samples were prepared for titration, again demonstrating that electron-rich southern hemispheres are tolerated (Table , entry 10, and Figure S5). Triazine 13j containing a phenol southern hemisphere demonstrated an order of magnitude loss of binding (Table , entry 11). Notably, introducing a carboxylic acid in the southern hemisphere (13k) afforded a triazine that bound with a Kd value of 133 ± 12 μM (Table , entry 12, and Figure S6). This functional group increases the polarity and hydrophilicity of the southern hemisphere and provides a handle for future molecule development.

Next, we investigated heteroaromatic southern hemispheres. Isoquinoline 13l (Table , entry 13) and phenyl ether 13m (Table , entry 14) precipitated the protein from solution. There was a noticeable difference between a furan or thiophene ring as the southern hemisphere. 3-Furyl-substituted triazine 13n (Table , entry 15) precipitated the protein from solution, and 2-furyl-substituted triazine 13o did not demonstrate binding to CXCL12 (Table , entry 16). In contrast, 3- and 2-substituted thienyltriazines 13p and 13q were competent ligands for CXCL12, albeit with weakened affinity (Table , entries 17 and 18). While the exact reason for the discrepancy between the binding of the furyl- and thienyl-substituted triazines is not readily apparent, it may be attributable to hydrogen-bond-acceptor ability. Both furan and thiophene are electron-rich heterocycles, but sulfur does not appreciably hydrogen-bond. The oxygen within the furan ring can accept a hydrogen bond. Therefore, the furan ring may engage in a disfavorable interaction within the pocket that disrupts its ability to bind, whereas thiophene does not have this issue. Nevertheless, further studies are needed to definitively answer this question.

These results suggest that there is a defined pocket in which the southern hemisphere sits when bound in the sY12 pocket. The fact that increasing the bulk of the southern hemisphere via multiple structures abolished ligand binding to CXCL12 supports this, as does our docking model (Figure ). The data also suggest that prominent cation−π interactions are present in the binding pocket. Many proteins participate in cation−π interactions as part of their secondary and tertiary structure,[31,32] and small-molecule inhibitors have been designed to take advantage of these interactions for binding to their target.[33] Arginine[34,35] and lysine[31,36] residues are typically most important to these interactions, and multiple positively charged residues are in proximity to the sY12 binding pocket. Further supporting this notion is the observation that electron-deficient southern hemispheres, which have decreased ability to participate in cation−π interactions, bind either weakly or not at all.

Triazines with alkyl southern hemispheres did not yield molecules that bind to CXCL12 (Table ). This lack of binding is most likely due to the inability of these compounds to participate in cation–π interactions within the sY12 pocket. Benzyltriazine 15 shows specific binding to the sY12 pocket (Table , entry 1). However, the chemical shift perturbation did not cross our threshold for binding as described above. Adamantyl- and hexyl-substituted triazines (17 and 19, respectively) did not bind to CXCL12 (Table , entries 2 and 3). Alkenyltriazine 20 caused the protein to precipitate from solution (Table , entry 4). Removing the southern hemisphere altogether in compound 21 also afforded a nonbinding molecule (Table , entry 5).

Table 4

SAR of Alkyl Southern Hemispheres

NB = nonbinder.

20 caused CXCL12 to precipitate

The final group of southern hemispheres we investigated were those attached to the triazine via a heteroatom. Creating a thioether by replacing the phenyl group of 5 with a thiophenyl ring afforded 24a, which demonstrated excellent binding to CXCL12 with Kd = 80 ± 18 μM (Table , entry 1, and Figure S7). Swapping the phenyl group for an aniline (24b) maintained the binding affinity while increasing the polarity (Table , entry 2, and Figure S8). Interestingly, bicyclopentylamine 24c bound with a Kd that was an order of magnitude less than that of the parent molecule, even though bicyclopentane is a common isostere of a phenyl ring[37] (Table , entry 3, and Figure S9). As above, steric bulk plays a role in fitting within the discrete pocket, and thus, (cyclohexylamino)triazine 24d (Table , entry 4, and Figure S10) binds with Kd = 170 ± 44 μM, but (dicyclohexylamino)triazine 24e (Table , entry 5) does not bind to CXCL12.

Table 5

SAR of 4-Heteroatom-Linked Southern Hemispheres

NB = nonbinder.

Creating a positively charged molecule by using N-methylmorpholine as opposed to morpholine as the southern amine resulted in a molecule with very weak binding. Our hope was to create a molecule that would be extremely soluble to circumvent solubility issues. Unfortunately, adding the methyl group to 24j, affording 24l, increased the Kd value from 234 ± 4 μM (Table , entry 10) to 3024 ± 626 μM (Table , entry 12). We attribute this to a combination of positively charged electrostatic repulsion and increased steric bulk.

Overall, linking the southern hemisphere via an amine appears to be well-tolerated (Table ), and in doing so hydrophilicity and polarity are increased in the molecule. While there is not yet a straightforward rationale for the ability of alkylamine southern hemispheres to bind well in the sY12 pocket of CXCL12, NMR data clearly demonstrate binding in the same pocket as the triazines possessing the aryl southern hemispheres reported above (Figure ). Although these molecules cannot participate in a cation−π interaction as the aryl southern hemispheres can, there is the possibility of a hydrophobic interaction between the alkyl carbons on the substituent and the residues in the pocket. There is also the possibility that there is an electrostatic interaction between the nitrogen at the 4-position of the triazine and residues within the pocket; however, more structural studies are needed. Regardless, the plethora of commercially available amines allows for a nearly endless number of southern hemispheres to be synthesized in future SAR studies.

Figure 4

4-Morpholinotriazine 24j maintains binding to the sY12 pocket. (A) HSQC spectrum of CXCL12 upon titration with increasing concentrations of compound 24j. (B) Depiction of the maximal chemical shift perturbation of CXCL12 residues upon binding of compound 24j at 1600 μM. Residues I28, V39, and A40 are highlighted. (C) Titration curves for binding of compound 24j to CXCL12.

On the basis of the complete set of results of our SAR studies, the modularity of the triazine scaffold notably allowed us to increase the lipophilic ligand efficiency (LLE)[38] without sacrificing binding affinity to CXCL12 (Table ).[39] As stated earlier, increasing the ease of library synthesis by changing our scaffold from pyrimidine 1 to triazine 5 resulted in essentially unchanged binding affinity while simultaneously increasing the solubility and LLE (Table , entries 1 and 3). While compound 24a demonstrated excellent affinity compared with the rest of our series, this compound is quite lipophilic and qualitatively was not as soluble as some of our other analogues (Table , entry 2). It also does not have a chemical handle for future transformations. Because it lacks desirable characteristics in both categories, we decided against carrying 24a forward for development. Adding a carboxylate arm to the southern hemisphere (13k) improved the LLE to 2.06 (Table , entry 4). Finally, we realized maximal optimization of the LLE in this study using 24j (Table , entry 5); qualitatively, this molecule was notably more soluble than its analogues as well. Ultimately, we have chosen to carry forward 13k and 24j in future studies. The carboxylic acid in the southern hemisphere of 13k provides a synthetic handle for future derivatization, and there are commercially available analogues of morpholine 24j that can be explored for further derivatizations. Compound 24k from this study is one example.

Table 6

Demonstration of Improved Lipophilic Ligand Efficiency

entry	compd	Kd (μM)	LEa	cLogP	LLEb
1	1	169 ± 12	0.23	2.82	0.95
2	24a	80 ± 18	0.24	3.05	1.05
3	5	158 ± 66	0.24	2.41	1.39
4	13k	133 ± 12	0.21	1.82	2.06
5	24j	234 ± 4	0.23	1.56	2.07

LE = ligand efficiency.

LLE = lipophilic ligand efficiency.

In conclusion, we synthesized 50 trisubstituted 1,3,5-triazines as the first reported ligands of the sY12 pocket of CXCL12. Using a combination of NMR chemical shift perturbation measurements and computational modeling, we probed the structure–activity relationships of the triazine that are necessary for binding to the sY12 pocket on the chemokine. In this family of molecules, carboxylic acid substituents at the 2- and 6-positions of the triazine are necessary to maintain binding affinity. Modeling suggests that these carboxylic acids interact with K27 to promote binding in the CXCL12 pocket. Provided that the molecule is soluble and not bulky, the southern hemisphere modification shows tolerance to aryl and amine modifications. Cation−π interactions may play a role in the interaction of aryl groups with CXCL12; more structural studies are necessary to determine the interactions that the amine southern hemispheres have with CXCL12.

The necessity of increasing the hydrophilicity during future compound optimization will be important, as many of the compounds screened were either insoluble or precipitated the protein when added. Thus, the molecules that we decided were optimal to carry forward for further development had increased hydrophilicity compared with the starting material along with adequate affinity for the sY12 pocket. They were designed to increase the lipophilic ligand efficiency and to introduce functional handles that can be used for growth of the molecule in the future. Further structural studies are underway to increase our understanding of the binding of these triazines to CXCL12. The results of these X-ray crystallography, NMR spectroscopy, and docking experiments will be reported in due course.