Introduction

In metazoans, recognition of cytosolic dsDNA by cyclic GMP-AMP synthase (cGAS) triggers synthesis of the second messenger 2′3′-cyclic GMP-AMP (2′3ʹ-cGAMP), a CDN carrying an unusual mixed 2′5′–3′5′ phosphodiester linkage and the most potent endogenous STING ligand.1,2,3,4 Upon ligand binding, STING undergoes major conformational changes and traffics from the endoplasmic reticulum to the Golgi, eliciting several downstream signaling outcomes: the canonical IRF3-driven type I interferon response,5,6,7,8 NF-κB-driven production of pro-inflammatory cytokines,9,10,11,12 LC3-associated autophagy and lysosomal remodeling, as well as promotion of lysosomal biogenesis.13,14,15,16,17

The potency of STING-dependent immune activation has made CDNs attractive candidates for pharmacological development, particularly as cancer immunotherapy and vaccine adjuvants.18,19,20,21 From a therapeutic standpoint, understanding STING’s cyclic nucleotide preferences is essential for agonist design, yet the clinical path has proven unexpectedly difficult. Natural CDN agonists face significant pharmacological obstacles: 2′3′-cGAMP is negatively charged rendering it non-cell permeable and dependent on specific transporters for uptake, which vary across cell types.22,23,24 2′3′-cGAMP is also subjected to cleavage by extracellular enzymes.25,26,27 Ectonucleotide pyrophosphate phosphodiesterase 1 (ENPP1) is even exploited by cancer cells to degrade the immunotransmitter 2′3′-cGAMP and to generate immunosuppressive adenosine, thereby dampening anticancer immune responses within the tumor microenvironment.28 These liabilities motivated development of stabilized, non-hydrolysable CDN analogues, such as ADU-S100 (the Rp, Rp diastereomer of 2′3′-c-di-AMPSS) and synthetic STING agonists including diABZI and E7766.26,29,30,31 Among CDN-class agonists, ADU-S100 and MK-1454 demonstrated limited clinical efficacy in Phase 1/2 trials, and no STING agonist has yet advanced to Phase 3.32,33,34,35 These outcomes have nonetheless informed a clearer mechanistic understanding of the requirements for effective STING-targeted therapy, and multiple programs continue to pursue improved strategies. Several interconnected biological challenges underlie this translational gap between preclinical promise and clinical efficacy (reviewed elsewhere in the study by Temizoz and Ishii18), including STING’s nucleotide selectivity, which is modulated by naturally occurring human polymorphisms. Multiple single nucleotide polymorphisms (SNPs) in human STING create functionally distinct population variants: both the wild-type (WT) and HAQ (H72/A230/Q293) variant respond to CDNs with 2′3′ and 3′3′ linkages, while the R232H variant is unresponsive to 3′3′-linked CDNs.3,29,36 Rational optimization of next-generation STING agonists and delivery platforms therefore requires a deeper mechanistic and STING allele-specific understanding of which structural determinants, such as nucleobase identity, phosphodiester linkage, and chemical modifications determine agonist activity.

The chemical space of CDN second messengers has expanded considerably in recent years. Bacterial cGAS-like enzymes synthesize a chemically diverse repertoire of CDNs that differ in phosphodiester linkage geometry and nucleobase composition, including mixed purine-pyrimidine molecules, and activate STING homologs in bacteria, Drosophila, and cnidarians as part of antiviral defense.37,38,39,40,41 Notably, STING was first characterized as a receptor for bacterial CDNs (c-di-GMP, c-di-AMP, and 3′3′-cGAMP) before its endogenous metazoan ligand 2′3′-cGAMP was identified.1,2,3,42,43,44,45 This evolutionary precedent raises the question of whether additional, yet untested, CDNs can engage human STING and shape its downstream signaling.

Bacterial CDNs also operate in a wide range of pathways beyond antiviral defense, mediating diverse signaling functions outside the cGAS-STING axis, including roles in biofilm formation and metabolic regulation in bacteria, as well as differentiation in amoeba.46,47 Additional evidence for alternative CDN-mediated signaling pathways comes from studies in mice, where c-di-AMP not only activates STING but also signals through the cytosolic oxidoreductase RECON.48 This raises the question of whether humans harbor additional receptors or pathways dedicated to detecting cyclic oligo-nucleotides beyond STING-dependent 2′3′-cGAMP signaling.

STING exhibits a binding and activation pattern that is highly selective for nucleotides with a specific nucleobase composition and phosphodiester linkage types.4,49 Despite extensive structural and biochemical work, no comprehensive cellular analysis has evaluated the full spectrum of CDN classes, including recently discovered bacterial and invertebrate CDNs containing pyrimidine bases and non-canonical linkages,37,38,39,40,41 for their ability to activate endogenous STING in a single, consistent cellular system. Here, we report systematic functional profiling of STING (HAQ allele)-dependent transcriptional responses in human THP-1 cells and STING (WT allele)-dependent cell death in pancreatic cancer cell line (DANG) by a library containing 78 nucleotide compounds. We identify a subset of CDNs that robustly activate STING and define the common structural features underlying STING’s nucleobase and linkage preferences. This approach enabled unbiased structure-activity profiling across a broad chemical space of CDNs, establishing the structural requirements for STING activation. Notably, we also show that many bacterial CDNs fail to elicit any signaling in the selected cell lines, further underscoring the high specificity of CDN recognition and the strong selectivity of human STING for its endogenous ligand, 2′3′-cGAMP. In addition, we demonstrate how CDN modifications and their isomers, such as phosphorothioate-containing CDNs, that confer resistance to hydrolysis, modulate STING activation, providing insights with direct relevance for the development of therapeutic STING agonists. Mechanistically, we dissect the contribution of nucleobase identity at each position, an especially underexplored determinant of CDN recognition. We demonstrate that 2′3′-cUAMP and 3′2′-cUAMP exhibit dramatically different STING activation profiles despite differing only in nucleobase positioning. Molecular dynamics (MD) simulations indicate reduced 3′2′-cUAMP interaction with STING caused by subtle conformational changes, however, no clearly defined altered interaction pattern can be observed.