Design, Synthesis, and Biological Evaluation of Novel Cyclic Adenosine−Inosine Monophosphate (cAIMP) Analogs That Activate Stimulator of Interferon Genes (STING)
ABSTRACT: We describe novel STING-activating cyclic dinucleotides whose constituent nucleosides are adenosine and inosine and that vary by ribose substitution, internucleotide linkage position, and phosphate modification. In mammalian cells in vitro, some of these cAIMP analogs induce greater STING-dependent IRF and NF-κB pathway signaling than do the reference agonists for murine (DMXAA) or human (2′,3′-cGAMP) STING. In human blood ex vivo, they induce type I interferons (IFNs) and proinflammatory cytokines: for the former, 3′,3′-cAIMP (9; EC50 of 6.4 μM) and analogs 52−56 (EC50 of 0.4−4.7 μM), which contain one or two 2′-fluoro-2′-deoxyriboses and/or bis-phosphorothioate linkages, are more potent than 2′,3′-cGAMP (EC50 of 19.6 μM). Interestingly, 9 induces type I IFNs more strongly than do its linkage isomers 2′,3′-cAIMP (10), 3′,2′-cAIMP (23), and 2′,2′-cAIMP (27). Lastly, some of the cAIMP analogs are more resistant than 2′,3′-cGAMP to enzymatic cleavage in vitro. We hope to exploit our findings to develop STING-targeted immunotherapies.
INTRODUCTION
Stimulator of interferon genes (STING) is an adaptor protein that is paramount in innate immunity, especially in orchestrating the body’s responses to cytoplasmic DNA, whether pathogen DNA (e.g., viral or microbial), self DNA (e.g., mitochondrial or junk), or tumor DNA. It is widely expressed in immune cells (e.g., dendritic cells, macrophages, and B cells) and endothelial cells, among others. Activation of STING leads to production of type I interferons such as IFN-α and IFN-β, via the interferon regulatory factor (IRF) pathway, and to production of proinflammatory cytokines such as tumor necrosis factor α (TNF-α) and interleukin 1β (IL-1β), via the nuclear factor κ-light-chain-enhancer of activated B cells (NF- κB) pathway. Human STING has four known natural agonists, which correspond to the four known naturally occurring cyclic dinucleotides (CDNs, Figure 1). Three of these CDNs (c- diGMP, c-diAMP, and cGAMP) are released by microbes into host cells during infection.1 However, the fourth, 2′,3′-cGAMP, is produced by the DNA sensor cyclic GMP-AMP synthase(cGAS) in mammalian cells after it detects self or foreign DNA in the cytoplasm (see, for example, Ablasser et al.2 and Zhang et al.3). Microbial CDNs contain a (3′,5′)(3′,5′) phosphodiester linkage (denoted as “3′,3′”), whereas the mammalian CDN contains a (2′,5′)(3′,5′) linkage (denoted as “2′,3′”).Since its discovery in 2008,4 STING has been widelyimputed in infectious diseases, cancer, and autoimmunity.5 Accordingly, it is attracting increasing attention as a potential immunotherapy target. Synthetic STING agonists that exhibit favorable druglike properties are currently being developed as adjuvants, antitumor agents, and other immunotherapy agents.
In fact, a CDN STING agonist has very recently entered phase I clinical trials for monotherapy of diverse solid tumors.6Herein we report the design, synthesis, and biological screening of novel CDNs that activate STING signaling. In the context of immunotherapy drug development, we sought to develop CDNs that could offer certain advantages over natural CDNs in terms of STING-dependent cytokine induction potency and selectivity and druggability. Unlike natural CDNs, whose constituent nucleosides are guanosine and/or adenosine, the CDNs we describe here contain one adenosine nucleoside and one inosine nucleoside. Beginning with the parentmolecule 3′,3′-cAIMP (9), we designed, synthesized, and biologically tested 11 cAIMP analogs that vary by sugar (ribose, 2′-deoxyribose, or 2′-fluoro-2′-deoxyribose) and by internu- cleotide linkage position (2′,2′; 2′,3′; 3′,3′; or 3′,2′) and phosphate modification (bis-phosphodiester or bis-phosphor- othioate). We screened all 11 analogs for STING-dependent activities in vitro in two mammalian cell lines: RAW 264.7 murine macrophages (for induction of the IRF pathway) and THP-1 human leukemic monocytes (for induction of the IRF and NF-κB pathways). We also screened these analogs for ex vivo production of type I IFNs and proinflammatory cytokines in human blood from healthy donors. Lastly, we explored the resistance of the cAIMP analogs to cleavage in vitro by two enzymes: nuclease P1 (NP1) and snake-venom phosphodies- terase (SVPD). We summarize our findings and discuss our preliminary structure−activity relationship (SAR) observations on these cAIMP analogs, highlighting those features that appear to enhance STING-pathway activity and/or resistance to enzymatic cleavage relative to 2′,3′-cGAMP.
RESULTS
Design and Synthesis of cAIMP Analogs. We began our study by reflecting on the fact that all four known natural CDNs are based on the purine bases guanosine and/or adenosine. We decided to explore how using another purine nucleoside, inosine, in CDNs might influence STING-pathway agonism. Thus, we designed, synthesized, and screened 11 cAIMP analogs, whose structures are summarized in Figure 2 and Table 1. Unless otherwise specified, the cAIMP analogs of the present study were synthesized using methods described, or adapted from those described, in the following references: Hyodo et al.,7 Gaffney et al.,8 Rammler et al.,9 Shanahan et al.,10 Zhou et al.,11 and Ora et al.12We started by preparing 9, its linkage isomer 10, and its bis- phosphorothioate analog 3′,3′-cAIM(PS)2 (13) (Scheme 1). Synthesis began with the commercially available compounds N6-Bz-2′-O-TBDMS-5′-O-DMTr adenosine phosphoramidite(1) or N6-Bz-3′-O-TBDMS-5′-O-DMTr adenosine phosphor-amidite (3). Phosphoramidites 1 and 3 were converted to their corresponding H-phosphonates and subsequently detritylated with dichloroacetic acid to give the 3′-H-phosphonate 2 and the 2′-H-phosphonate 4, respectively. Then, 3′-O-TBDMS-5′-O-DMTr inosine phosphoramidite was coupled with the 5′-O-free alcohol of 2 or 4, using an ACN solution of Activator 42 (0.1 M) as promoter in the presence of 3 Å molecular sieves (MS 3 Å). The resulting phosphite products were oxidized with either5.5 M tert-butyl hydroperoxide (tBuOOH) in decane, or sulfurized phenylacetyl disulfide (PADS). The 5′-O-DMTr group in each compound was removed by treatment with a 20% solution of dichloroacetic acid in dichloromethane to give the linear dimers 5 and 11 (from 2) and 6 (from 4). The dimers were then cyclized in a high-dilution mixture of 5,5- dimethyl-2-oxo-2-chloro-1,3,2-dioxaphosphinane (DMOCP) in dry pyridine and then treated with either iodine and water (for bis-phosphodiester) or elemental sulfur (for bis-phosphoro- thioate diester) to provide the corresponding fully protected CDNs 7, 8, and 12. Finally, these CDNs were deprotected by treatment with 33% methylamine in EtOH at 50 °C to remove the benzoyl (Bz) and cyanoethyl protecting groups and then treated with Et3N·3HF in pyridine and triethylamine, toremove the tert-butyldimethylsilyl (TBDMS) protecting groups, to give the target compounds 9, 10, and 13.
To synthesize the next two analogs (Scheme 2), the cAIMP linkage isomers (3′,2′)cAIMP (23) and (2′,2′)cAIMP (27), we began by simultaneously protecting the hydroxy groups in the 3′ and 5′ positions of commercially available inosine via treatment with 1,3-dichloro-1,1,3,3-tetraisopropyldisiloxane (TIPSCl).13,14 The resulting protected nucleoside 14 was treated with 3,4-dihydro-2H-pyran and pyridinium p-toluene- sulfonate (PPTS) in dry CH2Cl2 to give the corresponding fully protected inosine intermediate 15, which was subsequently treated with tetrabutylammonium fluoride (TBAF) on silica gel in tetrahydrofuran (THF) at room temperature to afford 2′-O-THP-inosine (16). After sequential protection of the 5′position with DMTr and the 3′ position with TBDMS, the 5′ position of the resulting intermediate was deprotected in acidicmedia to afford compound 19. Coupling of 19 with phosphoramidites 1 or 3 gave the dimer 21 or dimer 25, respectively. These dimers were then converted into their corresponding 5-H-phosphonates and deprotected in acid media to remove the THP groups. Finally, the resulting intermediates were cyclized and deprotected as described for preparation of 9 to give 23 and 27.Finally, we synthesized (Scheme 3) c-(dAMP-dIMP) (51), c- (2′FdAMP-2′FdIMP) (52), c-[2′FdAMP(S)-2′FdIMP(S)](53), c-(dAMP-2′FdIMP) (54), c-(2′FdAMP-dIMP) (55),and c-[2′FdAM(PS)-dIM(PS)] (56), all of which, relative to cAIMP, differ by the 2′-substitution at either or both sugars and by the internucleotide linkage position and phosphatesubstitution. Thus, inosine phosphoramidites 28 and 29 were each condensed with the 5′-O-free nucleoside allyl phospho- triesters 30−32, using the standard procedure, to afford the corresponding linear dimers 33−38. Subsequently, the allyl group in each dimer was removed by treatment of sodiumiodide in acetone (for compound 33) or tetrakis(triphenyl- phosphine)palladium(0) and N-methylaniline in THF (for compounds 34−38).
The resulting products were cyclized with DMOCP in dry pyridine. Finally, the cyclic compounds 45−50 were fully deprotected by successive treatment with 33% methylamine in ethanol to afford the desired analogs 51−56. Due to their high polarity, these compounds had to be purifiedusing reverse-phase HPLC.Biological Evaluation. In Vitro Screening of cAIMP Analogs in a Murine Macrophage Cell Line. We first screened the cAIMP analogs for IRF pathway induction in a murine immune cell line that expresses STING and then compared their activities to that of the reference agonists for murine (DMXAA15,16) and human (2′,3′-cGAMP) STING. We used the murine macrophage reporter cell line RAW-Lucia ISG, which reports on IRF pathway activation as an indirectmeasure of type I IFN induction. Each cAIMP analog induced IRF activation to distinct levels (Table 2). Interestingly, compared to DMXAA (EC50 = 35.2 μM) and 2′,3′-cGAMP (EC50 = 18.8 μM), most of the cAIMP analogs gave a superior IRF response, with 53 (EC50 = 0.6 μM) and 56 (EC50 = 2.2 μM) exhibiting strikingly higher activity. We next screened the analogs in the corresponding STING knockout cell line, RAW- Lucia ISG-KO-STING. As expected, none of them exhibited any IRF activity in the knockout cells (data not shown). Together, these findings indicate that in a murine immune cell line, the cAIMP analogs trigger STING-dependent IRFpathway induction and that their activities differ from that of DMXAA and 2′,3′-cGAMP.In Vitro Screening of cAIMP Analogs in a HumanMonocyte Cell Line. We next evaluated the ability of the cAIMP analogs to induce both the IRF and the NF-κB pathways in a human immune cell line that expresses STING and then compared their activities to that of 2′,3′-cGAMP (in this case we did not test DMXAA, as it does not bind to human STING16). We employed the THP-1 human monocyte reporter cell line THP1-Dual, which uses a dual-reporter system to report on IRF activation as an indirect measure of type I IFN induction, and NF-κB activation as an indirect measure of proinflammatory cytokine induction. Each cAIMP analog activated both pathways to varying degrees (Table 2). Interms of IRF activity relative to that of 2′,3′-cGAMP (EC50 = 7.2 μM), six of the cAIMP analogs (13 and 52−56) exhibited greater than 3-fold activity, whereas the remaining five showed roughly similar activity.
Alternatively, for NF-κB pathwayactivation, 13, 53, and 56 were also more active (>3-fold) than 2′,3′-cGAMP (EC50 = 39.1 μM), with 53 showing nearly 20-fold greater activity; in contrast, the remaining analogs had similar or lower activity than 2′,3′-cGAMP (in all cases, EC50 of approximately 15.4 μM to 110.9 μM). Together, these findings indicate that in a human immune cell line, the cAIMP analogs induce the IRF and the NF-κB pathways and that some of them are more active than 2′,3′-cGAMP at activating either or both pathways.Ex Vivo Screening of cAIMP Analogs in Human Blood. Having demonstrated that the cAIMP analogs trigger STING signaling in vitro, we then assessed their cytokine-induction activities in human blood ex vivo. To this end, we tested bloodaReagents and conditions: (a) H2O, pyridinium trifluoroacetate (Pyr·TFA), dry acetonitrile (ACN), rt; (b) tert-butylamine (tBuNH2); (c) dichloroacetic acid (DCA), H2O, dichloromethane (CH2Cl2); (d) 5′-O-DMTr-2′-O-TBDMS-inosine phosphoramidite, Activator 42, ACN, 3 Å molecular sieves (MS); (e) 5.5 M tert-butyl hydroperoxide (tBuOOH) in decane for Z = O, or phenylacetyl disulfide (PADS) for Z = S; (f) DCA, H2O, CH2Cl2; (g) 5,5-dimethyl-2-oxo-2-chloro-1,3,2-dioxaphosphinane (DMOCP), pyridine (Pyr); (h) iodine (I2), H2O for Z = O or S8 for Z = S;(i) methylamine (MeNH2) 33% in ethanol (EtOH), rt or 50 °C; (j) Et3N·3HF in pyr/Et3N.samples collected from healthy donors (three experiments; five donors per experiment) by incubating them with either a cAIMP analog or 2′,3′-cGAMP at a range of CDN concentrations and then evaluating the induction of type I IFNs (IFN-α and IFN-β), type III interferons (IL-28 A/B), interleukin 1 (IL-1), and interleukin 6 (IL-6). Figure 3A shows dose−response curves for these activities for 9, as a representative example. However, the only activity for which we were able to calculate EC50 values was type I IFN induction (see Table 2 and Figure 3B), as the other activities were characterized by a lack of saturating responses, even at the highest test dose.
In terms of inducing type I IFNs, every one ofthe cAIMP analogs showed either significantly greater (up to∼45-fold; p < 0.05), or similar, activity relative to 2′,3′-cGAMP (EC50 = 19.6 μM ± 6.7 μM) (Table 2 and Figure 3B). The most active analogs were 53 (EC50 = 0.4 μM ± 0.1 μM; p = 0.003) and 52 (EC50 = 0.7 μM ± 0.1 μM; p = 0.002), whichalso gave the strongest IL-1 production (see Figure 3C). In contrast, 9, whose EC50 for type I IFN induction activity (6.4 μM ± 2.3 μM; p = 0.046) was 3 times lower than that of 2′,3′- cGAMP, was highly selective for this activity over induction of IL-1 (see Figure 3C). Together, the above findings indicate that in whole blood from healthy human donors, the cAIMP analogs induce type I IFNs and other cytokines to varying degrees andaReagents and conditions: (a) TIPSCl2, Pyr; (b) 3,4-dihydro-2H-pyran, PPTS, dry CH2Cl2; (c) TBAF on silica gel, THF; (d) DMTr-Cl, Pyr/ CH2Cl2 (4:1); (e) TBDMS-Cl, imidazole, Pyr; (f) ZnBr2 (0.5 M) i-PrOH/CH2Cl2 (1:1); (g) adenosine 3′- or 2′-O-phosphoramidite, Activator 42, ACN, 3 Å MS; (h) 5.5 M tBuOOH in decane; (i) diphenylphosphite, pyr TEAA 0.1 M; (j) DCA 10% in CH2Cl2; (k) DMOCP, pyr; (l) I2, H2O;(m) 33% MeNH2 in EtOH, rt or 50 °C; (n) Et3N·3HF in pyr/Et3N.that they exhibit distinct cytokine induction activities to that of 2′,3′-cGAMP. Importantly, some of these analogs preferentially induce type I IFNs over IL-1, whereas others give a more balanced response.Enzymatic Stability. As previously mentioned, we are endeavoring to design and synthesize STING-activating CDNs that can offer advantages over natural CDN STING agonists not only in terms of activity but also as concerns druggability. Considering that CDNs can be degraded by circulating and cytoplasmic enzymes such as nucleases and phosphodiesterases (see, for example, Shanahan et al.17), which can ultimately influence their pharmacokinetic (PK), we subjected our cAIMP analogs and 2′,3′-cGAMP to a preliminary in vitro enzyme cleavage assay. To this end, we incubated each analog for up to2 h with either nuclease P1 (NP1) or snake venom phosphodiesterase (SVPD), both of which cleave CDNs at their phosphodiester linkages.18,19 To measure the resistance of each CDN to each enzyme, we monitored the reactions by HPLC. In terms of resistance to NP1, some of the cAIMP analogs were more resilient than 2′,3′-cGAMP (83% remaining): in fact, 13, 52, and 53 did not show any signs of degradation (Figure 4, top). The remaining analogs showed a similar level of resistance to that of 2′,3′-cGAMP except for 9, 23, and 51, which had been totally, or near totally, degradedwithin 2 h, exhibiting half-lives (T1/2) of 23.1 min (9), 7.1 min (23), and 5.7 min (51). Regarding resistance to SVPD, all of the cAIMP analogs exhibited either markedly greater (13, 51, 52, 53, 55, and 56), or similar, stability relative to 2′,3′- cGAMP. Interestingly, even after a 2 h incubation, 53 and 56 had not undergone any degradation (Figure 4). These results reveal that some of the cAIMP analogs are more stable than 2′,3′-cGAMP to cleavage by NP1 and/or SVPD. DISCUSSION We provide here a preliminary SAR analysis for the cAIMPanalogs. On the basis of the ensemble of biological results, we compare these analogs to the parent compound, 9, based on each structural parameter explored. We also consider the activities of all the analogs relative to those of 2′,3′-cGAMP.Considering literature precedent describing substantialdifferences in the STING-binding activity and modes of cGAMP linkage isomers,3,20,21 we sought to evaluate the role of the internucleotide linkage position in the four linkage isomers of cAIMP (9, 10, 23, and 27). Intriguingly, whereas2′,3′-cGAMP has been described to bind to STING approximately 300 times more strongly than does 3′,3′- cGAMP,3 our screening results suggest that 10 and 9 induce STING signaling to similar levels. We did observe that 9 wasgenerally more active in vitro and gave slightly stronger induction of type I IFNs in blood. Extending this analysis to the remaining cAIMP linkage isomers 23 and 27, we noted that all four linkage isomers gave comparable results for induction of IL-1 in blood but differed in their ability to induce type I IFNs.Specifically, 9 was the strongest inducer of type I IFNs, especially relative to 27. Thus, in terms of internucleotide linkage geometry, it appears that AI CDNs do not exhibit the same SAR as the corresponding (naturally occurring) cGAMP linkage isomers. Accordingly, for the future design of CDNSTING agonists, the choice of internucleotide linkage should account for the two constituent bases and vice versa.In terms of enzymatic degradation, 10 and 2′,3′-cGAMP were similarly more resistant than 9 to NP1, but all threeCDNs were susceptible to hydrolysis by SVPD, which suggests that the linkage position influences stability in an enzyme- specific fashion. In line with these findings, Danilchanka and Mekalanos22 and Gao et al.23 have previously reflected on the possible existence, in eukaryotic cells, of linkage-specific phosphodiesterases that could distinguish between self (i.e., 2′,3′-cGAMP) and foreign (i.e., 3′,3′) CDNs.Considering the ribose modifications, we first observed that 9 and 51 generally had similar activity in vitro, although the former gave mildly higher induction of type I IFNs in blood (based on EC50). In terms of enzymatic stability, although both CDNs had been totally degraded by NP1 within 2 h, 51 did show a 4-fold shorter half-life. Interestingly, Li et al. reportedthat 2′,3′-cGAMP and its mono-2′-deoxyribose analog 2′,3′- cdGAMP underwent similar levels of degradation (T1/2 ≃ 1 h for both compounds) by ecto-nucleotide pyrophosphatase/ phosphodiesterase 1 (ENPP1);24 however, the authors did not test the corresponding bis-2′-deoxyribose analog. Nevertheless,whereas 9 had also been degraded by SVPD within 2 h, 51 was partially resistant, having undergone less than 40% degradation. Thus, the bis-2′-deoxyribose substitution appears to increase or decrease the resistance of cAIMP to enzymatic cleavage on an enzyme-specific basis.We observed that the cAIMP analogs containing at least one 2′-fluoro-2′-deoxyribose sugar (52−56) were typically among the most active compounds in all assays. Specifically, in blood the bis-2′-fluoro-2′-deoxyribose analogs 52 and 53 were markedly more active for induction of type I IFNs and of IL-1 than were the mono-2′-fluoro-2′-deoxyribose compounds 54−56, which in turn were more active than all the remaining (i.e., nonfluoro) analogs for induction of type I IFNs. Likewise,52 and 53 were also among the most resistant analogs to enzymatic degradation and were more resistant than 54 and 55. Interestingly, 54 and 55, which differ only by the location of the 2′-fluoro group (2′-FdI vs 2′-FdA, respectively), exhibited similar activities in nearly all of the assays, which suggests that this moiety enhances STING-pathway signaling and enzymatic stability relative to 9 irrespective of which nucleoside is modified. Regardless, we cannot yet conclude whether the apparently superior activity of these fluoro-cAIMP analogs relative to both 9 and 2′,3′-cGAMP is due to superior enzymatic stability, stronger STING binding, greater cellular uptake (e.g., owing to the lipophilicity of the 2′-F atoms relativeto the corresponding 2′-OH groups in 9), or a combination ofthese or other factors known to be conferred by fluorine atoms.25In oligonucleotide chemistry, replacement of phosphodiester linkages with phosphorothioate linkages is a well-known strategy for lengthening the half-life of drugs in vivo, as it confers partial or total resistance to enzymatic cleavage (for areview on the PK of phosphorothioate oligonucleotides, see, for example, Srinivasan et al.26). Furthermore, phosphorothioate analogs of oligonucleotides have been reported to exhibit distinct activities relative to their parent (phosphodiester) compounds. In fact, phosphorothioate CDNs have been described to be more stable to enzymatic resistance than their corresponding phosphodiester CDNs in vitro;24 to induce more STING-dependent production of type I IFNs in vitro in monocytes,27 lung fibroblasts,24 and modified HEK293 cells;24 and to provide greater efficacy as STING agonists for immunotherapy of cancer in mouse tumor models in vivo.27 Consistent with these and other reports on the relative activity of phosphorothioate CDNs, in our study all three phosphor- othioate analogs (13, 53, and 56) were more active than their corresponding phosphodiester analogs (9, 52, and 55, respectively) for cytokine induction in vitro. The cytokine induction activity of CDNs depends partly on their stability to enzymatic cleavage. For example, McWhirter et al. reported that c-diGMP induced IFN-β in bone marrow-derived macrophages, but the product resulting from hydrolysis of c- diGMP by SVPD was inactive.28 Similarly, Li et al. attributed the superior in vitro cytokine induction capacity of 2′,3′-cGAM(PS)2 relative to 2′,3′-cGAMP exclusively to the greaterresistance of the former to phosphodiesterase digestion rather than to superior STING binding, as the two compounds exhibited similar affinities in a STING-binding assay.24 Consistent with this premise, we also observed that our phosphorothioate analogs were among the most stable to enzymatic degradation. However, in the present work we did not directly measure the STING-binding affinity of the cAIMP analogs and thus cannot speculate on this parameter. Regardless of whether the superior in vitro activity of the phosphorothioates relative to the phosphodiesters was indeed due to the greater enzymatic stability of the former, we cannot yet explain why this advantage did not consistently translate to better cytokine induction activity in blood. For example, concerning type I IFN induction, although 53 was slightly more active than 52, 13 and 56 were actually slightly less active than 9 and 55, respectively.A crucial factor to consider when assessing the activity ofphosphorothioate CDNs is stereochemical purity: since each of the two internucleotide linkages in a CDN can exist as either of two diastereomers (Rp or Sp), each CDN can theoretically exist as four possible stereoisomers ([Rp, Rp]; [Rp, Sp]; [Sp, Sp]; or [Sp, Rp]). Interestingly, Corrales et al. reported that for induction of type I IFNs in THP1 cells by bis-phosphorothioate CDNs such as 2′,3′-cdiAM(PS)2, each Rp,Rp stereoisomer was more active than its corresponding Rp,Sp stereoisomer.27 In our study, we screened 13, 53, and 56 as mixtures of their constituent stereoisomers.The results for cytokine induction in blood demonstrated that the cAIMP analogs vary in their relative abilities to drive STING-dependent induction of type I interferons and of IL-1. For drug development, such differences might ultimately represent a therapeutic advantage according to the desired activity: for example, for selective induction of type I IFNs (as with 9 and 10) or for a more balanced response between type I IFNs and IL-1 (as with 52−54).Not only did we observe that type I IFN induction activity inblood differed with each CDN, but we also observed that it varied widely by donor. Interestingly, this variability was the smallest for the two most active CDNs (52 and 53). There is some precedent for such heterogeneity in assays of STING-activating compounds. Fu et al. stimulated PBMCs from human donors with different CDN STING agonists and found that production of IFN-β and TNF-α varied not only by CDN but also by each donor’s STING (Tmem173) genotype.29 Among the four genotypes in their study, they observed the weakest cytokine induction for the heterozygous variant R232H/R71H- G230A-R293Q.30 Given their findings, and the variability that we observed in the present study, one can reason that the therapeutic utility of a STING-modulating drug might ultimately have to be ascertained on a patient-by-patient basis, for example, by using a blood-based companion diagnostic test. CONCLUSIONS We have reported the design, synthesis, and biologicalscreening of a set of novel STING-activating cAIMP analogs. Compared to the reference agonists for murine (DMXAA) and(ESI) and controlled by Chemstation software. The LC system was equipped with an Aquity CSH C18 column (50 mm × 2.1 mm, 1.7 μm) using gradients of 10 mM aq triethylammonium sodium bicarbonate and acetonitrile at a flow rate of 300 μL/min. The UV detection wavelength was 254 nm. The mass spectrometer was run in positive and negative ESI modes. Purity for all final compounds was confirmed to be greater than 95% by LC−MS. Preparative HPLC was done on a Waters preparative 150Q HPLC system with monitoring at 254 nm on a SunFire Prep C18 column (5 μm OBD; 30 mm × 150 mm) using gradients of 10 mM aq triethylammonium bicarbonate and acetonitrile at a flow rate of 100 mL/min. 1H NMR and 31P NMR spectra were acquired on a Bruker Fourier 300 at room temperature and are reported in ppm. Molecular sieves (3 Å; Aldrich) were employed after drying the commercially supplied product at 250 °C for 12 h under vacuum. Exact mass characterization was done on a Waters ESI-QTOF operating in negative-ion acquisition mode. The compounds were dissolved in a 1:1 (v/v) mixture of water (0.1% formic acid) and acetonitrile. Flow injection analysis (FIA) was performed at a rate of 0.15 mL/min.human (2′,3′-cGAMP) STING, some of these cAIMP analogs trigger greater induction of the IRF and/or NF-κB pathways in vitro in human and murine immune cell lines. Similarly, these cAIMP analogs induce cytokine production ex vivo in human blood from healthy donors, with some of them, such as 9,showing selective induction of type I IFNs over IL-1. In contrast to literature reports that 2′,3′-cGAMP is a dramatically stronger STING agonist than is its linkage isomer 3′,3′- cGAMP, we did not observe any marked difference in activity between 10 and 9 in most of our assays, although in humanblood, 9 did demonstrate slightly stronger induction of type I IFNs relative to all three of its linkage isomers. Provisional SAR analysis suggests that the combination of adenosine and inosine as constituent nucleosides, the 3′,3′ internucleotide linkagegeometry, the presence of one or two 2′-fluoro-2′-deoxyribosesugars and, in some cases, a bis-phosphorothioate linkage each seems to enhance STING activation potency relative to 2′,3′- cGAMP. However, the advantages of each of these structural features are not necessarily additive. Moreover, the 2′- deoxyribose and 2′-fluoro-2′-deoxyribose sugars and the bis- phosphorothioate linkage also seem to confer resistance to cleavage by certain phosphodiesterases. It would be interestingto ascertain how these same features influence the PK/PD of such compounds in vivo. Accordingly, we are currently evaluating these cAIMP analogs in healthy animals and in animal models of disease, with the aim of identifying viable preclinical candidates for STING-based immunotherapy of cancer, infectious diseases, and other disorders.Anhydrous solvents were purchased from Sigma-Aldrich, and all reagents for nucleoside and nucleotide synthesis were purchased from Sigma-Aldrich, TCI, Carbosynth, or ChemGenes. Unless noted otherwise, all commercially obtained solvents and reagents were used directly. Moisture- or air-sensitive reactions were run under argon atmosphere in oven-dried (120 °C) glassware and in the presence of activated 3 Å molecular sieves. Amidite coupling reactions and cyclizations were done in anhydrous acetonitrile or pyridine under dry argon or nitrogen. The starting materials for all reactions in dry pyridine were dried by concentration (three times) from pyridine. Preparative silica gel flash chromatography was done on Fluka 60 Å high-purity grade or Merck grade 9385 silica, using gradients of methanol in dichloromethane. LC/ES-MS was performed on an Agilent 1290 Infinity UHPLC system coupled to an Agilent 1260 Infinity diode array detector (DAD) and an Agilent 6130 quadrupole mass spectrometer equipped with an electrospray ionization source2′-O-(TBDMS)-3′-O-(H-phosphonate)-N -(Bz)adenosine (2).The adenosine phosphoramidite 1 (11.7 g, 14.73 mmol) was dissolved in a solution of ACN (50 mL). To the solution were added water (0.53 mL, 29.46 mmol) and pyridine·TFA (3.41 g, 17.67 mmol), and the resulting mixture was stirred for 15 min at rt. Then, tert-butylamine was added, the solution was stirred for 15 min, and the solvents were removed in vacuo. The residue was treated with a 3% solution of DCA in CH2Cl2 (100 mL) and water (10 equiv) for 15 min. The reaction was quenched with MeOH and pyridine. The solvents were removed in vacuo, and the residue was purified by DMXAA silica gel column chromatography, using CH2Cl2/EtOH/MeOH (80/10/10) as eluent,to give 2 (6.8 g; 84% yield).