Results

Nucleotide library screen reveals selective signaling by purine-containing cyclic dinucleotides

The nucleotide library was designed to encompass the broadest possible range of candidate signaling molecules, drawing on all known cyclic nucleotides and incorporating recently discovered bacterial cyclic oligonucleotides and their structural variants.41,50,51,52 Of note, the library was not limited to known STING agonists. Rather, we extended it to include nucleotides that could potentially signal through alternative pathways, as well as molecules whose signaling capacity in human cells has not previously been established, with the goal of comprehensively profiling nucleotide-mediated signaling beyond the cGAS-STING axis. Thus, we also included linear nucleotides (LNs), given that in humans linear 2′–5′ oligoadenylates function as antiviral second messengers by activating ribonuclease L (RNase L) and inducing translational arrest.53,54,55 In addition, the linear guanosine tetraphosphate or pentaphosphate (p)ppGpp is a widespread bacterial second messenger involved in stress responses and has recently also been identified in metazoans.56,57 An additional rationale for their inclusion is that linear intermediates are generated during cyclic-nucleotide synthesis or degradation; yet it remains unknown whether these linear species possess signaling capacity in cells. The nucleotide library comprises 78 nucleotides, including 38 CDNs, 9 cyclic oligonucleotides (CONs), 13 cyclic mononucleotides (CMNs) and 18 LNs (Figure 1A; Table S1). CDNs include the natural STING ligand 2′3′-cGAMP, as well as a wide array of cyclic nucleotides that differ in nucleobase composition ranging from purine-containing and pyrimidine-containing species to purine-pyrimidine hybrids. They also encompass large structural diversity, including different combinations of 2′–5′ and 3′–5′ phosphodiester linkages, non-hydrolysable phosphorothioate analogues, and dideoxy cyclic nucleotides. These non-hydrolysable cyclic nucleotide compounds are resistant to hydrolysis by phosphodiesterases such as ENPP1 and are therefore more stable, inducing stronger and prolonged signaling.26 CONs include oligonucleotides ranging from three bases (e.g., cyclic tri-AMP) to six bases (e.g., cyclic hexa-AMP), several of which have been identified in bacteria and function in anti-phage defense systems.41,50,51,52 CONs also have the potential to signal in human cells as demonstrated by binding to and activation of mammalian reductase controlling NF-κB (RECON) by cyclic AMP-AMP-GMP (cAAG).41 CMNs include the well-known nucleotides in humans, 3′,5′-cGMP and 3′,5′-cAMP. cAMP was the first nucleotide second messenger to be discovered and regulates protein kinase A activation downstream of hormone and neurotransmitter signaling.58,59 Additional CMN diversity in the library is covered by nucleotides with 2′,3′ linkage isomers and alternative nucleobases, such as 3′,5′ xanthosine monophosphate (XMP), which has been identified in plants and mice.60,61 LNs in our library consist of adenosine or guanosine chains that range from two to five nucleobases in length. They include cleavage products of bacterial CDNs, such as linear 5′- phosphoguanylyl-(3′-5′)-guanosine (pGpG), a metabolite of c-di-GMP that may also function as an independent signaling molecule.62,63 Overall, our custom-made library comprises a diverse set of nucleotides, including molecules known to signal in human cells serving as controls as well as others whose existence or signaling capacity has not previously been established.

Figure 1: Nucleotide library screen reveals selective signaling by purine-containing cyclic dinucleotides(A) Composition of the nucleotide library with representative nucleotides shown.(B) Experimental setup. THP-1 monocytes were electroporated with 600 nM of each nucleotide in the library in duplicate or water as a control. Cells were harvested for RNA-seq analysis 6 h after electroporation. Untreated cells without electroporation were used to monitor gene expression changes due to electroporation alone.(C) Principal component analysis of transcript counts from RNA-seq analysis showing all the analyzed samples and their respective nucleotide type.(D) Ranking of nucleotides by the number of differentially expressed genes (log2 fold change ≥ [1] and adjusted p value≤0.05) when compared to control electroporation with water.CDN: cyclic dinucleotide, CMN: cyclic mononucleotide, CON: cyclic oligonucleotide, LN: linear nucleotide.See also Figure S1 and Table S1.

We monitored nucleotide-induced signaling in THP-1 monocytes, which express the STING HAQ (H72 A230 Q293) variant found in ca. 20% of human population.3,36 This variant can be activated by both the metazoan 2′3′-cGAMP and bacterial CDNs with 3′3′ linkage to induce a robust transcriptional response (Figure S1A).3,29,36 THP-1 monocytes were electroporated with each nucleotide or with water as a control and harvested 6 h later for total transcriptome analysis by RNA-seq (Figure 1B). Untreated cells without electroporation were included as a control to assess the effects of electroporation alone, which induced only minimal changes in gene expression (Figures 1C and 1D). Electroporation ensured rapid delivery of nucleotides into cells minimizing degradation by extracellular phosphodiesterases.25,26,27 THP-1 monocytes were electroporated with 600 nM of each nucleotide, a concentration at which 2′3′-cGAMP induces robust STING activation and strong expression of IFNB1 and CXCL10 (Figure S1A). Principal component analysis (PCA) of RNA-seq transcript counts revealed that all CONs, LNs, CMNs, and a subset of CDNs clustered with control and untreated cells, indicating that these nucleotides did not induce signaling (Figure 1C). The lack of signaling detected for the well-known CMNs cAMP and cGMP in humans may be explained by differences in treatment duration and stimulation conditions, which other gene-expression studies have shown to be critical for inducing a robust response.64,65,66 In contrast, a distinct group of CDNs clustered separately from control samples, consistent with induction of gene expression changes. Analysis of differentially expressed genes (DEGs), comparing each nucleotide to the control, showed that the endogenous STING ligand 2′3′-cGAMP induced the strongest transcriptional response, with a total of 1196 DEGs (Figure 1D; Table S1). The ten nucleotides inducing the greatest number of DEGs were in fact hydrolysable and non-hydrolysable isomers of cGAMP and c-di-AMP, both of which are established STING agonists. Overall, gene expression changes were induced by 27 CDNs in our library, all of which were composed of purine nucleobases guanosine, adenosine, or inosine and their hydrolysable and non-hydrolysable isomers (Figure S1B). Only 3 DEGs in total were identified for non-CDN nucleotides cyclic tetra-GMP (3′3′3′3′-c-tetra-GGGG) and ApA, which belong to the CONs and LNs class, respectively. Taken together, the results of our nucleotide library screen demonstrate that nucleotide signaling is highly specific, with many compounds failing to elicit any response, while simultaneously revealing that STING displays strong selectivity toward a narrow subset of purine-containing CDNs that activate signaling in THP-1 cells.

STING activation by distinct CDNs induces a common transcriptional signature

We next asked whether the responses induced by different CDN isomers were identical, or whether variations in nucleobase composition or phosphodiester linkage resulted in distinct transcriptional outcomes. Several structures of STING bound to different CDNs, including bacterial cGAMP, c-di-AMP, and c-di-GMP, have shown that CDNs can induce distinct STING conformations, suggesting that ligand-specific structural states may influence the magnitude or even type of downstream signaling responses.4,49,67,68 In addition, we sought to determine whether individual CDNs elicit responses beyond canonical STING signaling, as exemplified by c-di-AMP in mice, which can also engage the receptor RECON.48 Thus, we further analyzed the gene expression signatures induced by the signaling CDNs and compared them to those elicited by 2′3′-cGAMP, which induces exclusively STING-dependent transcriptional changes in THP-1 cells.69 Clustering of 2561 unique DEGs, obtained from pooling all DEGs induced by 27 signaling CDNs, revealed two major clusters (Figure 2A). The “upregulated” cluster comprised genes whose expression increased in response to CDN treatment relative to control cells, whereas the “downregulated” cluster contained genes whose expression decreased. All signaling CDNs induced a similar transcriptional signature across both clusters, with variation primarily in magnitude: 3′3′-c-di-GMP induced the weakest changes, while 2′3′-cGAMP triggered the largest number of DEGs and some of the strongest transcriptional responses. This indicates that the CDNs induce responses of varying strengths that closely resemble those triggered by 2′3′-cGAMP, consistent with STING-dependent signaling. Indeed, when analyzing each CDN individually, approximately 90% of upregulated DEGs and 65% of downregulated DEGs were shared with those induced by the endogenous STING agonist 2′3′ -cGAMP (Figure 2B).

Figure 2: STING activation by distinct CDNs induces a common transcriptional signature(A) Heatmap showing clustering of all DEGs pooled from signaling 27 CDNs, which identified two clusters. Treatments are sorted by the number of DEGs that the nucleotides induced. Normalized transcript counts from variance stabilizing transformation (vst) were used, counts are row scaled.(B) Boxplot showing percentage of upregulated and downregulated DEGs for each signaling CDN (dots) that overlap and were identified as DEGs also for 2′3′-cGAMP. Boxes encompass the twenty-fifth to seventy-fifth percentile changes. Whiskers extend to the tenth and ninetieth percentiles. The central horizontal line indicates the median.(C) Pathway enrichment analysis for the “upregulated” and “downregulated” genes from (A).(D) Transcription factor motif enrichment analysis for the genes from “upregulated” cluster in (A).See also Figure S2.

To further exclude any STING-independent signaling or transcriptional responses, we analyzed the DEGs induced by all signaling CDNs but not by 2′3′-cGAMP by comparing the overlaps of upregulated and downregulated genes. Pooling the responses of all signaling CDNs uncovered 536 upregulated and 829 downregulated DEGs that were not significantly regulated by 2′3′-cGAMP. (Figure S2A). However, clustering analysis showed that these genes still followed a pattern of regulation similar to that induced by 2′3′-cGAMP, despite not passing the cut-off of log2 fold change of [1] and adjusted p value of 0.05 thresholds used to define DEGs (Figure S2B). Thus, signaling CDNs converge on a common STING-dependent transcriptional response. The induced genes were strongly enriched for innate immune pathways related to anti-viral defense, including interferon-stimulated genes (ISGs) and NF-κB-driven TNF signaling (Figure 2C). Induction of interferon signaling and the NF-κB pathways is a hallmark of STING activation, which is also reflected in the strong enrichment of ISRE and IRF transcription factor (TF) motifs, as well as NF-κB motifs, among the upregulated DEGs (Figure 2D). Furthermore, both upregulated and downregulated genes were enriched for pathways associated with transcriptional regulation, comprising numerous transcription factors from diverse families (Figure 2C; Figure S2C). The upregulated genes included basic leucine zipper domain superfamily members such as JUNB, FOSL1, FOS, ATF3, CEBPB, and STAT proteins, all of which are well-established regulators of inflammatory responses.70,71 In contrast, the downregulated genes were significantly enriched for Zinc finger (ZNF) transcription factors containing a Krüppel associated box (KRAB) repressor domain (Figure S2C).72 Collectively, our analysis identified a set of 27 CDNs that induce a spectrum of STING-dependent gene expression changes, modulating inflammatory and transcriptional pathways.

Potent STING agonists drive cell death in pancreatic cancer cells

STING agonists are currently being developed for cancer therapy to induce protective antitumor immunity as well as cancer cell death.21,73,74,75,76 The consequences of STING activation in tumors are multifaceted and shaped by both intratumoral heterogeneity and the surrounding microenvironment.20,77 On the one hand, STING signaling can exert adverse effects by promoting T cell death, thereby generating an immunosuppressive, tumor-supportive microenvironment.78,79,80,81 On the other hand, STING activation and downstream interferon signaling can be advantageous by promoting recruitment of effector T cells and eliciting inflammatory anti-tumor responses within the tumor microenvironment and more systemically, as shown in the pancreatic ductal adenocarcinoma (PDAC) models.82,83,84,85,86,87 Anti-tumor effects of STING signaling in PDAC have been attributed primarily to responses in cancer-surrounding cells, including stromal cells, epithelial cells, and tumor-associated macrophages.83,87,88,89 PDAC tumor cells express high levels of STING.82,84,90 However, the consequences of tumor cell-intrinsic STING signaling have been less studied and have been associated with metabolic reprogramming, cell-cycle arrest, and reduced tumor cell growth.82,90 Given that intrinsic type I interferon signaling in PDAC tumor cells can cause reduced cell growth and cell death, we sought to determine how such responses may be activated downstream of STING by structurally diverse nucleotides, motivating a cell death screen in PDAC cells.82,90,91,92,93 Because STING activation does not necessarily correlate with antitumor activity, we leveraged our nucleotide library together with the distinct response patterns observed in THP-1 cells to define the spectrum of CDNs capable of inducing cell death in pancreatic cancer cells. To preserve the possibility of engaging receptors beyond STING, we included the full nucleotide library, encompassing non-STING agonists, as this represented a new cell line and functional readout. To this end, we electroporated the nucleotide library into the human pancreatic cancer cell line DANG, which carries a STING WT allele94 and responds efficiently to interferon stimulation.82 Cell death was monitored by time-lapse live-cell imaging using the cell-death dye over 70 h (Figure 3A). Consistent with our transcriptional activation findings in THP-1 cells, induction of cell death in PDAC cells was observed exclusively for purine-containing CDNs (Figure 3B; Figure S3A). The most potent inducers of cell death were the non-hydrolysable isomers of cGAMP and c-di-AMP, as well as 2′3′-c-di-AMP, 2′3′-c-di-GMP and endogenous 2′3′-cGAMP (Figure 3B; Figure S3). The non-hydrolysable CDN isomers induced the highest levels of cell death when measured 60 h after electroporation, consistent with a cellular effect driven by their increased stability and resistance to phosphodiesterase-mediated cleavage (Figures 3C and 3D). To confirm that the observed cell death was STING-dependent, we electroporated 2′3′-cGAMP and non-hydrolysable 3′3′-cGAMP into STING-KO DANG cells, where both compounds failed to induce cell death and transcription of IFNB1 and CXCL10 (Figures 3E and 3F; Figure S3B). To further correlate STING signaling capacity with cell death induction, we electroporated 2′3′-cGAMP and non-hydrolysable 3′3′-cGAMP into the PDAC cell lines HPAC and PANC1, which express low STING protein levels, alongside DANG cells, which express high STING levels as previously demonstrated (Figure 3G).90 In contrast to DANG cells, neither HPAC nor PANC1 cells showed induction of cell death or upregulation of IFNB1 and CXCL10 (Figures 3H and 3I), indicating that CDN-induced cell death in PDAC cells is dependent on STING protein levels and downstream signaling capacity. Electroporation of 2′3′-cGAMP also induced Ifnb1 expression in mouse pancreatic cancer (mPC) cells (Figure S3C), indicating the potential for beneficial STING signaling across pancreatic cancer models in different organisms. Collectively, our nucleotide-library experiments highlight the remarkable specificity of nucleotide signaling in diverse cell types and demonstrate how the signaling capacity of distinct CDNs can be translated across these cellular contexts and responses.

Figure 3: Potent STING agonists drive cell death in pancreatic cancer cells(A) Experimental setup. DANG pancreatic cancer cells were electroporated with 2 μM of each nucleotide from the library in duplicate or water as a control. Cell death was monitored by live cell imaging over 70 h in the presence of red fluorescent cell death dye YOYO-3.(B) Ranking of nucleotides by the level of induced cell death (red vs. phase signal) in DANG cells at 12 h post-treatment.(C and D) Ranking of cGAMP (C) and c-di-AMP (D) isomers from the library for induction of cell death in DANG cells at 60 h post-treatment.(E) RT-qPCR analysis of IFNB1 and CXCL10 expression after electroporation of DANG STING KO and NTC KO cells with 2 μM indicated nucleotides or water (control) for 5 h.(F) Analysis of cell death in DANG STING KO and NTC KO cells 24 h after electroporation with water (control), 2 μM 2′3′-cGAMP or 2 μM 3′3′-cGAMPSS, iso2.(G) Western blot analysis showing STING levels in PANC1, HPAC, and DANG cells with anti-RPS19 as a loading control.(H) RT-qPCR analysis of IFNB1 and CXCL10 expression after electroporation of DANG, HPAC, and PANC1 cells with 2 μM indicated nucleotides or water (control) for 5 h.(I) Analysis of cell death in DANG, HPAC, and PANC1 cells 24 h after electroporation with water (control), 2 μM 2′3′-cGAMP or 2 μM 3′3′-cGAMPSS, iso2.CDN: cyclic dinucleotide, CMN: cyclic mononucleotide, CON: cyclic oligonucleotide, LN: linear nucleotide. Bars represent means from at least two replicates and error bars represent standard deviation. In (B), (C), (D) dashed line represents levels of red vs. phase (cell death) after control electroporation with water.See also Figure S3.

STING activation is skewed toward cyclic dinucleotides with 2′3′ phosphodiester linkage

Since no additional signaling was observed from nucleotide compounds other than the CDNs that activate STING and trigger the canonical 2′3′-cGAMP response, albeit with different magnitudes, we next sought to use these findings to identify the shared features that determine STING’s nucleotide activation preferences. Having profiled the complete nucleotide library in both THP-1 cells, which carry the STING HAQ variant,3,68 and DANG cells, which carry the STING WT allele,94 our dataset enables a systematic comparison of CDN structural requirements across two naturally occurring human STING variants. Multiple STING SNPs exist in the human population, influencing responsiveness to different CDNs. Both human STING WT and HAQ variants respond to CDNs with a 2′3′ linkage as well as bacterial CDNs with a 3′3′ linkage, whereas the R232H variant is unresponsive to CDNs containing a 3′3′ linkage.3,29,36 Altogether, our analysis in THP-1 and DANG cells demonstrated STING-dependent signaling for 27 out of 38 CDNs included in our nucleotide library. The overall pattern of active and inactive CDNs was broadly consistent across both cell lines, although a larger number of CDNs induced signaling in THP-1 cells than in DANG cells. Among CDNs with 3′3′ phosphodiester linkages, STING activation was observed exclusively for molecules composed of purine nucleobases in both STING alleles, with pyrimidine-containing CDNs failing to induce signaling (Figures 4A and 4B). We further examined STING's preference for specific phosphodiester linkage types using purine-containing cGAMP, c-di-AMP and c-di-GMP isomers. All 2′2′, 2′3′, 3′3′, or 3′2′ (cGAMP only) linkage isomers of c-di-AMP and cGAMP induced high levels of STING activation (Figures 4C and 4D) with a strong and consistent preference for 2′3′ linkages evident across both STING alleles. Both 2′3′-c-di-GMP and 2′3′-c-di-AMP induced robust signaling comparable to 2′3′-cGAMP (Figures 4C and 4D), whereas 3′3′-c-di-GMP acted as a weak agonist in THP-1 cells but not in DANG cells, and 2′2′-c-di-GMP showed no activity in either. Consistent with this, treatment with 2′3′-c-di-GMP, but not its 3′3′ isomer, induced IFNB1 and CXCL10 transcription in DANG cells, further corroborating the preference for 2′3′ phosphodiester linkage (Figure 4L). Notably, in THP-1 cells we identified 2′2′-c-di-AMP as a previously unrecognized STING agonist, thereby expanding the repertoire of known activating CDNs (Figure 4C). The weak activity of 3′3′-c-di-GMP is consistent with earlier reports in mouse and human STING1,4,29,67,68 and is thought to result from its lower affinity, which stabilizes a more open STING conformation and promotes cooperative activation rather than the closed-state polymerization induced by 2′3′-linked ligands and 3′3′-c-di-AMP.4,67,68

Figure 4: STING activation is skewed towards cyclic dinucleotides with 2′3′ phosphodiester linkage(A and B) Bar plot showing number of DEGs in THP-1 cells (A) and red vs. phase (cell death) in DANG cells (B) for all nucleotides in the library with 3′3′ phosphodiester linkage grouped by purine and/or pyrimidine base composition.(C and D) Bar plot showing number of DEGs in THP-1 cells (C) and red vs. phase (cell death) in DANG cells (D) for all c-di-GMP, c-di-AMP, and cGAMP isomers in the library grouped by the phosphodiester linkage type.(E and F) Bar plot showing number of DEGs in THP-1 cells (E) and red vs. phase (cell death) in DANG cells (F) for all hydrolysable and non-hydrolysable cGAMP isomers in the library.(G) Depiction of different cGAMP isomers containing thiophosphate modifications at different positions.(H and I) Bar plots showing number of DEGs in THP-1 cells (H) and red vs. phase (cell death) in DANG cells (I) for all hydrolysable and non-hydrolysable c-di-AMP isomers in the library.(J) Depiction of 2′3′-cUAMP and 3′2′-cUAMP.(K) RT-qPCR analysis of IFNB1 and CXCL10 expression after electroporation of THP-1 monocytes with 600 nM 2′3′-cGAMP, 2′3′-cUAMP, 3′2′-cUAMP, or water (control) for 6 h.(L) RT-qPCR analysis of IFNB1 and CXCL10 expression after electroporation of DANG cells with 2 μM indicated nucleotides or water (control) for 5 h. In (B), (D), (F), (I), (K) and (L) bars represent means from at least two replicates and error bars represent standard deviation. In (B), (D), (F) and (I) dashed line represents levels of red vs. phase (cell death) after control electroporation with water.See also Figure S4.

Our nucleotide library includes several phosphorothioate-containing isomers of c-GAMP, c-di-GMP, and c-di-AMP, in which a non-bridging oxygen on one or both phosphates is substituted with a sulfur atom (Figure 4G). This modification confers resistance to phosphodiesterase-mediated cleavage, making these analogues more attractive candidates for immunomodulatory applications.26,29 Among the cGAMP isomers, all non-hydrolysable linkage isomers produced robust STING activation, except for the 3′2′-cGAMPSS isomer 1 and 3′3′-cGAMPSS isomer 1 in DANG cells (Figures 4E and 4F). For the 2′3′-cGAMP analogues, signaling strength was similar whether sulfur was incorporated at both phosphates or restricted to the 2′–5′ or 3′–5′ linkage (Figures 4E, 4F, and 4G), although these phosphorothioate forms generally induced slightly lower activation than unmodified 2′3′-cGAMP in THP-1 cells. In contrast, the behavior of other phosphorothioate-modified CDNs varied by phosphodiester linkage type. In THP-1 cells, while natural 2′2′-c-di-AMP activated STING, introducing sulfur atoms at both phosphates eliminated its activity (Figure 4H). Non-hydrolysable 3′3′-c-di-AMP analogues retained STING activation capacity, albeit at reduced levels compared with unmodified 3′3′-c-di-AMP (Figures 4H and 4I). Conversely, thiophosphate modification of 3′3′-c-di-GMP completely abolished signaling in THP-1 cells (Figure S4A and S4B). The ability of 2′3′-c-di-AMP to induce STING signaling was highly dependent on the stereochemistry of the thiophosphate groups and STING allele. The Rp, Rp dithio-substituted diastereomers triggered strong STING activation, while the Sp, Sp form failed to elicit a response in both THP-1 and DANG cells (Figures 4H, 4I, and L). The mixed Sp, Rp diastereomer displayed intermediate and cell type-dependent activity, inducing strong signaling in THP-1 cells and weaker signaling in DANG cells (Figures 4H, 4I, and 4L). To determine whether these differences in cellular activity reflect differential binding to STING, we performed thermal shift assays with purified recombinant STING ligand-binding domain (LBD). The Sp, Rp form stabilized the STING LBD, whereas the Sp, Sp form did not induce measurable stabilization (Figures S4F and S4G), demonstrating that the capacity for cellular signaling correlates with stabilization of STING’s active conformation. Together, these data indicate that phosphorothioate stereochemistry imposes strict constraints on CDN recognition, with the Rp configuration at one or both positions being required for productive STING engagement. In fact, the Rp, Rp-2′3′-c-di-AMP analogue, also known as ADU-S100 and MIW815, has progressed into clinical trials for advanced and metastatic solid tumors as well as lymphomas, albeit with limited clinical efficacy.33,34 In summary, our analysis of STING responses to 38 distinct CDNs revealed a strong preference for 2′3′ phosphodiester linkage, which outweighed the influence of base composition. Even weaker agonists, such as 3′3′-c-di-GMP, exhibited markedly enhanced signaling when configured with a 2′3′ linkage. In addition, we identified several non-hydrolysable c-di-AMP and cGAMP isomers that induced robust STING activation, highlighting candidates with improved stability and prolonged signaling potential for therapeutic applications.

Recent studies have expanded the diversity of known CDNs as well as cGAS-like enzymes and STING homologs in bacteria, Drosophila and metazoans.37,38,39,40,41 Of note, STING homologs in coral S. pistillata can be activated not only by 2′3′-cGAMP but also by purine-pyrimidine CDN 2′3′-cUAMP, which contains a 2′3′ phosphodiester linkage40 (Figure 4J). Structural studies further demonstrated that 2′3′-cUAMP binds both to S. pistillata and human STING and induces a closed conformation similar to that triggered by 2′3′-cGAMP.4,40,67 The strong preference of human STING for 2′3′-linked CDNs (Figures 4C and 4D), together with the ability of 2′3′-cUAMP to induce a closed, active STING conformation,40 suggests that human STING can also be activated by pyrimidine-containing 2′3′-linked CDNs. Indeed, electroporation of THP-1 monocytes with 2′3′-cUAMP resulted in robust induction of IFNB1 and CXCL10 expression, with levels comparable to those elicited by 2′3′-cGAMP (Figure 4K). In DANG cells 2′3′-cUAMP-induced signaling was lower when compared to 2′3′-cGAMP (Figure 4L). In contrast, 3′2′-cUAMP and 3′3′-cUAMP failed to induce signaling, indicating a requirement for both the specific pyrimidine base position and the 2′3′ phosphodiester linkage (Figures 4A, B and 4J-L). To confirm the dependence on STING, we generated STING-KO THP-1 cells and verified that the induction of IFNB1 and CXCL10 by 2′3′-cUAMP, as well as by the other signaling CDNs in our screen (Figure 2), was abolished in the absence of STING (Figures S4C and S4D). Dose-response analysis of 2′3′-UAMP and 2′3′-cGAMP further showed that 2′3′-cGAMP activated STING more efficiently at lower concentrations, although both CDNs achieved similar activation at higher concentrations (Figure S4E). Taken together, our data reveal that STING can tolerate pyrimidine bases at a specific position for activation and signaling when presented within its preferred 2′3′ phosphodiester linkage.

Molecular dynamics simulations of STING in complex with 2′3′-cUAMP and 3′2′-cUAMP

Previous structural and biochemical studies have investigated why CDNs containing purine bases and mixed 2′3′ phosphodiester linkages are more potent activators of human STING than those with canonical 3′3′ or 2′2′ linkages, establishing the mixed linkage topology as a key determinant of receptor activation.4,49,67,68,95 Our results extend this picture by showing that even within the preferred 2′3′ linkage context, nucleobase positioning is a critical and previously underappreciated variable: substitution of a purine with a pyrimidine at specific positions can reduce STING activation or lead to its complete abolishment.

Our cellular data demonstrate that 2′3′-cUAMP activates STING in THP-1 cells carrying the HAQ variant at levels comparable to 2′3′-cGAMP, and similarly activates STING WT, while the isomer 3′2′-cUAMP fails to induce signaling in either STING allele (Figures 4K and 4L; Figures S4D and S4E). To determine whether this functional difference reflects differences in ligand binding and STING dimer stabilization, we performed thermal stability assays with recombinantly purified STING ligand-binding domain (LBD). 2′3′-cUAMP stabilized STING LBD by 1.4°C, whereas 3′2′-cUAMP produced no measurable thermal stabilization, suggesting an inability to stabilize the closed STING dimer conformation required for signaling, consistent with the cellular results (Figure 5A; Figure S5A).

Figure 5: Molecular dynamics simulations of STING in complex with 2′3′-cUAMP and 3′2'-cUAMP(A) Melting temperature analysis of recombinantly purified STING ligand-binding domain (LBD) with water (control), 2′3′-cGAMP, 2′3′-cUAMP, 3′2′-cUAMP by thermal shift assays. Dots: independent replicates, horizontal lines: mean. Paired t test ∗∗ p ≤ 0.01, ∗∗∗∗ p ≤ 0.0001.(B) Top: structure of STING ligand binding domain (LBD) dimer bound to a cyclic dinucleotide. Bottom: close-up view of MD simulation models of STING bound to 2′3′-cUAMP and 3′2′-cUAMP.(C) Calculated mean interaction energies to STING-ligand binding domain (LBD) for the indicated nucleotides using MMGBSA (molecular mechanics generalized born surface area) method. Error bars represent standard errors of the mean.See also Figure S5.

Because available STING crystal and cryo-EM structures4,67,95,96 are inherently symmetric and do not fully resolve the discrimination of asymmetric ligands, we performed all-atom molecular dynamics (MD) simulations to characterize how nucleobase positioning differences between 2′3′-cUAMP and 3′2′-cUAMP influence conformational dynamics within the STING ligand-binding pocket. STING dimer complex with 2′3′-cUAMP remained stable throughout several independent simulations with low overall root-mean-square deviation (RMSD) values of the protein (Figure S5B) and of the bound 2′3′-cUAMP ligand (∼1–1.7 Å), with slightly greater conformational fluctuations observed in the 2′3′-cAUMP-bound state (Figure S5B). The simulations identify R232, R238, Y163, Y167, Y240, and E260 as the key coordinating residues in the CDN-binding and lid region, consistent with contacts previously described for 2′3′-cGAMP4,68 (Figure 5B; Figure S5C). Cluster analysis of sampled conformational states indicates that the coordination is not static but transient changes in the interaction pattern of R238 and R232 with the 2′3′-cUAMP as well as 2′3′-cAUMP ligand are possible (illustrated as conformational clusters in Figure S5C). Critically, we observed that E260 occasionally forms direct interactions with R238 (Figure S5C). Binding of negatively charged CDN ligands potentially promotes protonation of E260, disrupting the transient R238–E260 interaction and repositioning R238 to coordinate the phosphate of the incoming CDN. Consistent with a functional role for E260 in ligand coordination, disruption of the R238–E260 hydrogen bond has been independently observed in MD simulations of STING bound to DMXAA, a non-nucleotide murine STING agonist, and mutation of E260I has been shown to improve DMXAA-dependent signaling of human STING,97 further supporting the mechanistic importance of this interaction. To further probe the role of E260, we generated E260Q and E260D mutants and observed reduced STING stability during purification of the recombinant protein, suggesting that the E260–R238 interaction plays a stabilizing role in the apo STING dimer in the absence of ligand (Figures S5D and S5E). In 2′3′-cGAMP, the endogenous vertebrate STING ligand, guanine occupies the 2′ position and directly stabilizes E260, resulting in the highest binding affinity.4 Notably, in 3′2′-cGAMP, the cGAS-like receptor (cGLR) product in Drosophila,38,39 guanine occupies the 3′ position instead, and human STING binds this isomer with lower affinity,4 consistent with our cellular data (Figures 4C and 4D). Nonetheless, guanine stabilizes E260 more effectively than uracil regardless of its position in the CDN scaffold. In 2′3′-cUAMP, uracil occupies the 2′ position in place of guanine and, being less bulky, provides additional space for R238 to retain interaction with E260, resulting in partial dissociation of R238 from the CDN phosphate. This reduces binding stability relative to 2′3′-cGAMP but remains compatible with productive STING activation. In 3′2′-cUAMP, uracil instead occupies the 3′ position in place of adenine. At this position the nucleobase sits directly on the sugar ring and is more geometrically constrained than at the 2′ position, which is slightly displaced (Figure 5B). Calculations of the mean STING interaction with 2′3′-cGAMP, 3′2′-cGAMP, 2′3′-cUAMP, and 3′2′-cUAMP are in line with these findings, showing that the calculated binding energies correspond to the relative STING signaling capacity of each CDN observed in cells (Figures 5C and 4K, and 4L).