STING agonist-mediated endothelial cell activation drives NK cells and neutrophils-dependent pulmonary inflammation

Stimulator of interferon genes (STING) agonists and derivative molecules have 20 been extensively developed for tumor immunotherapy. However, systemic exposure toxicity 21 risks have constrained clinical trial progression and even threatened patient lives. Currently, 22 systematic toxicity assessments for STING agonists remain lacking, with the mode of action 23 for major organ injury yet to be elucidated. Here, we focused on STING agonist-induced lung 24 injury, revealing that systemic administration of STING agonists caused pulmonary 25 hemorrhage, inflammatory alterations, and respiratory dysfunction. Through single-cell RNA 26 sequencing and immune deletion studies, we found that lung endothelial cells could be 27 stimulated by STING agonists and then secreted chemokines and IL-15 to recruit and activate 28 NK cells. NK cells could induce endothelial cell apoptosis via IFN-γ. Tbx21+ NK 29 subpopulations, which activated by endothelial cells, could produce chemokines to recruit 30 neutrophils. Neutrophils secreted IL-1β through positive feedback pathways and form 31 neutrophil extracellular traps during lung injury. This study elucidates the critical role of the 32 endothelial cell-NK cell-neutrophil axis in mediating STING agonist-associated pneumonia, 33 offering insights for developing intervention strategies for STING agonist toxicity. 34 35

STING agonist, pulmonary inflammation, immunotoxicology 36 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint

37 The cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway 38 detects pathogen DNA or damage-associated molecular patterns (DAMPs) DNA, serving a 39 vital function in the host's defense against infections and malignancies. cGAS, as a double-40 stranded DNA sensor, catalyzes the production of 2’3’ cyclic GMP-AMP (cGAMP)-the 41 endogenous ligand for STING-upon activation [1]. Subsequently, activated STING transfers 42 from the endoplasmic reticulum to the Golgi apparatus and promotes downstream effects 43 such as interferon (IFN) signaling, inflammasome activation, and light-chain 3B (LC3B) 44 lipidation through processes including TBK1-IRF3, NFκB, and proton leakage [2]. 45 Activation of the STING signaling promotes the antitumor effects of dendritic cells (DCs), 46 CD8+ T cells, natural killer (NK) cells, etc[3]. Besides, studies have further proposed that 47 STING activation in vascular endothelial cells can remodel the tumor vasculature [4] and 48 alter the structure of the tumor immune microenvironment in “cold” tumors [5]. 49 The anti-tumor effects of STING agonists were first observed in the cGAMP [6]. Studies 50 have demonstrated that injecting cGAMP into the glioma significantly reduces tumor volume 51 and improves survival rates of tumor-bearing mice in a STING-dependent manner [7]. 52 Meanwhile, amidobenzimidazole (ABZI)-based analogs were designed to improve systemic 53 delivery, which can bind to the C-terminal domain of STING and enhance the biding affinity. 54 The representative one is diABZI from linked ABZIs, which showed a potent effect in 55 ameliorating the affinity to STING and inducing the secretion of IFN-β in human peripheral 56 blood mononuclear cells (PBMCs). Administration of diABZI in mice bearing CT26 57 colorectal tumors resulted in significant tumor inhibition and enhanced survival, with 80% of 58 mice being tumor free [8]. Considering of the superior antitumor effects, numerous STING 59 agonists have now entered clinical development phases. For example, ADU-S100 [9] and 60 E7766 have been employed for treating advanced solid tumors or lymphoma [10]. Beyond 61 this, antibody-drug conjugates (ADCs) targeting STING agonists have also emerged as a 62 focal point in antitumor drug development. Examples include Takeda Pharmaceutical 63 Company's STING agonist-linked ADC, TAK-500, which targets CCR2 [11]. 64 However, multiple studies have raised concerns regarding the safety of STING agonists. 65 The injection of small-molecule STING agonists may lead to rapid systemic distribution, 66 thereby posing risks of uncontrolled inflammation and cytokine storms, tissue toxicity, and 67 autoimmune damage [12]. Chronic STING activation may also persistently stimulate 68 cytokine production, thereby fostering an inflammatory tumor microenvironment (TME) that 69 promotes tumor progression [13]. Meanwhile, Mersana's STING agonist-conjugated 70 antibody-drug conjugate XMT-2056, targeting HER2, has experienced a Grade 5 (fatal) 71 serious adverse event (SAE) in its Phase I clinical trial. The company then has announced a 72 voluntary suspension of the trial [14]. In non-human primates preclinical studies of E7766, 73 participating animals exhibited multiple respiratory adverse reactions including dyspnea, 74 hemoptysis, hypoxia, alveolar hemorrhage, pulmonary embolism, and pulmonary oedema 75 [10]. It is evident that the toxicity issues associated with STING agonists have resulted in a 76 narrow therapeutic window during clinical trials, thereby hindering dose escalation and 77 preventing trials from achieving efficacy endpoints. 78 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint And recent studies increasingly indicate that activation of the cGAS-STING pathway 79 promotes or exacerbates the development of pneumonia. For instance, following DNA 80 recognition by macrophages, the STING pathway is activated, triggering IL-6 release which 81 subsequently activates fibroblasts and intensifies airway obstruction [15]. 82 Safety concerns regarding STING agonists, such as pulmonary toxicity, have severely 83 limited their clinical development. Therefore, in this study, we focused on systematically 84 evaluating the pulmonary toxicity risk of STING agonists, identifying the characteristics of 85 lung injury induced by STING agonists, and elucidating the underlying mechanisms. 86 87 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint

88 Systemic administration of STING agonists causes lung injury in mice 89 Analysis of human single-cell databases revealed that STING is highly expressed in lung 90 tissues compared to other organs (Fig S1A). This led us to hypothesize that the lung is the 91 most likely target organ for STING agonist toxicity. Thus, we administered the human-mouse 92 cross-reactive STING-specific agonist diABZI [8] to mice and measured inflammatory 93 cytokine expression in lung, liver, kidney, and intestinal tissues. Lung tissues exhibited the 94 most significantly upregulated transcription levels of tumor necrosis factor-alpha (TNF-α), 95 interleukin-1 beta (IL-1β), and IFN-γ (Fig S1B). 96 To holistically evaluate the in vivo toxicity responses of diABZI, a concentration gradient 97 dosing regimen was established. Following intraperitoneal diABZI administration, mice 98 exhibited significant weight loss alongside elevated lung coefficients and wet-to-dry weight 99 ratios (Fig S1C, Fig 1A and B), preliminarily confirming STING agonist-induced pulmonary 100 injury. To confirm that diABZI-induced pneumonia is indeed caused by STING activation, 101 we administered natural ligand of STING, 2',3'-cGAMP to mice in vivo. The results mirrored 102 those observed with diABZI administration: the mice exhibited weight loss and elevated 103 expression of inflammatory cytokines in lung tissue (Fig S1D). Based on the the above 104 indicators, a 2 mg/kg dose was selected for subsequent in vivo assays. Further assessment of 105 pulmonary function during treatment revealed reduced peak expiratory flow rates and 106 increased respiratory rates (Fig 1B), indicating ventilatory dysfunction and respiratory 107 impairment. Histopathological examination revealed hemorrhage, congestion, disrupted 108 alveolar architecture, and increased immune cell infiltration in the lungs of diABZI-treated 109 mice (Fig 1C). Surfactant protein A (SP-A), which typically regulates alveolar gas exchange, 110 exhibits serum level elevation associated with pulmonary tissue injury [16, 17]. Western blot 111 analysis demonstrated a significant increase in serum SP-A levels 48 hours post diABZI 112 administration (Fig 1D). 113 Concurrently, the transcriptional levels of inflammatory cytokines IL-1β, IFN-γ, and IL-18 114 were substantially upregulated, rapidly peaking within 4 hours post-administration. 115 Following a decline at 6 hours, levels gradually rebounded to reach a plateau at 24 hours, 116 maintaining elevated expression thereafter (Fig 1E-G). Similarly, in vivo administration of 117 cGAMP also resulted in a significant increase in IFN-γ and IL-1β in lung tissues (Fig S2). 118 Flow cytometry analysis revealed a rapid and significant reduction in the proportion of T/B 119 lymphocytes within the pulmonary immune microenvironment following diABZI 120 administration. Regarding myeloid cells, the STING agonist induced increased neutrophil and 121 macrophage infiltration, reaching a plateau within 24 hours, concurrent with enhanced pro-122 inflammatory polarization of pulmonary macrophages (Fig 1H and I). 123 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint In summary, STING agonists possessed significant pulmonary toxicity risks, potentially 124 leading to pathological lung injury, immune dysregulation, and impaired respiratory function. 125 Single-cell sequencing reveals alterations and interactions in pulmonary cell populations 126 post STING activation 127 To elucidate the alterations in cell population proportions, gene expression changes, and 128 cellular interactions induced by STING agonists, single-cell RNA sequencing (scRNA-seq) 129 was performed on mouse lung tissue collected at 0, 4, 6, 12, and 24 hours post intraperitoneal 130 administration of diABZI. Following t-SNE dimensionality reduction, the data were 131 identified as 12 distinct cell populations. T cells and B cells exhibited markedly reduced 132 proportions following administration, whereas neutrophil and macrophage proportions 133 significantly increased (Fig 2A), consistent with flow cytometry results (Fig 1H and I). 134 Notably, NK cells, macrophages, T cells, and B cells exhibited high STING expression at rest 135 (Fig 2B). Following STING agonist administration, the IFN-α response pathway was 136 substantially activated in endothelial cells, neutrophils, and monocytes, with increased gene 137 expression (Fig 2C). Previously, we demonstrated that diABZI induced elevated expression 138 of pulmonary inflammatory cytokines IL-1β and IFN-γ (Fig 1E). Further analysis of scRNA-139 seq data revealed that myeloid cells, particularly neutrophils, constitute the primary IL-1β-140 secreting population, peaking at 12 hours post treatment. Conversely, NK cells predominantly 141 secreted IFN-γ, peaking at 4 hours; in contrast, other cell populations exhibited negligible 142 IFN-γ secretion (Fig 2D). Combined with scRNA-seq data, flow cytometry results indicate 143 significant alterations in the proportion of pulmonary immune cell populations within 4 hours 144 post-administration. 145 To investigate the mechanisms underlying changes in the proportion of pulmonary 146 immune cell populations, we analyzed levels of CC and CXC subfamily chemokines. Within 147 4 hours post-administration, Cxcl1, Cxcl2, Cxcl9, Cxcl10, Ccl3, Ccl4, and Ccl5 were 148 significantly upregulated (Fig. S3A). Analysis of these chemokine expression levels at 0 and 149 4 hours by cell population revealed that multiple cell types upregulate chemokine expression 150 following administration. Specifically, DCs, endothelial cells, NK cells, neutrophils, and 151 monocyte-macrophages exhibited markedly elevated chemokine expression (Fig. 2E). This 152 suggests potential interactions and chemotaxis between these cell types following STING 153 agonist administration. 154 To validate this hypothesis, we employed CellChat for scRNA-seq data analysis and 155 visualization of cellular communications. Results demonstrated that both the number and 156 strength of cellular communication rapidly increased following drug administration, peaking 157 at 4 hours. Subsequently, communications gradually diminished at 4 hours yet remained 158 elevated compared to resting conditions (Fig. S3B). Scatter plots visualizing signal input and 159 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint output probabilities across cell populations at different time points revealed that NK cell 160 communication probability substantially increased within 4 hours, while neutrophil and 161 endothelial cell output probabilities significantly rose. At 12 hours, neutrophil signal 162 exchange probability peaked, with intercellular communication probabilities subsequently 163 declining post-12 hours (Fig. 2F). Further analysis of specific interacting cell populations and 164 communication intensity revealed that between 0-4 hours, neutrophils and endothelial cells 165 transmitted signals to NK cells, with NK cells exhibiting the strongest signal reception 166 intensity; between 4 and 6 hours, the strength of signals transmitted by epithelial cells 167 increased, primarily received by NK cells, neutrophils, and macrophages; at 12 hours, 168 neutrophils replaced NK cells as the cell population receiving the strongest signals, primarily 169 from NK cells, macrophages, and self-activation; compared to 12 hours, at 24 hours, signal 170 exchange from non-immune cells, including endothelial and epithelial cells, was enhanced 171 (Fig. 2G). 172 Taken together, we hypothesized that after STING agonist administration, endothelial 173 cells may recruit and activate NK cells in the early phase, and then NK cells may secrete 174 IFN-γ and other chemokines, thereby acting upon neutrophils to enhance their infiltration and 175 IL-1β secretion. 176 Endothelial cells can be directly activated by SITNG agonists 177 Following systemic administration, pulmonary vascular endothelial cells constitute the 178 primary line of direct contact with STING agonists. Analysis of differentially expressed 179 genes and IFNα response pathway genes in pulmonary endothelial cells revealed substantial 180 pathway activation within 4 hours, concurrent with upregulation of chemokines Cxcl9, 181 Cxcl10, Ccl4, Ccl5, and classical NK cell activation factors Il15 (Fig. 3A-C) . 182 To determine whether endothelial cells could be directly activated by STING agonists, 183 we directly treated human pulmonary microvascular endothelial cell line (HPMEC) with 184 diABZI. Western Blot analysis confirmed that at working concentrations, diABZI 185 significantly increased the phosphorylation levels of STING and TBK1 (Fig. 3D). Validation 186 in mouse cell lines revealed that diABZI directly activated mouse pulmonary endothelial cells 187 (MPMVEC-SV40) and upregulated chemokines (Fig. S4A). 188 Next, to validate in vitro whether STING activation promotes endothelial cell chemotaxis 189 and inflammatory cytokine secretion, and to investigate the underlying mechanisms, we 190 analyzed pathways enriched in endothelial cell differentially expressed genes at 4 h post-191 treatment compared to 0 h in scRNA-seq data. Differentially expressed genes enriched in NF-192 κB, MAPK-CREB-TGF-β, and Wnt-related pathways (Fig. 3E). Targeting these enriched 193 pathways, TBK1-IRF3 pathway and LC3B lipidation, we assessed whether cytokine secretion 194 levels were affected by adding pathway-specific inhibitors to STING activation. Results 195 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint indicated that the secretion of nearly all factors depended on IRF3 activation. Cxcl9 196 expression relied on CREB pathway activation. Cxcl10 expression was linked to LC3B 197 lipidation and other pathways but independent of NF-κB. Ccl5 expression mechanisms 198 resembled Cxcl9, largely dependent on CREB pathway activation. Whereas Il15, similar to 199 Cxcl10, was influenced by multiple pathways including LC3B lipidation (Fig. 3F). 200 Moreover, we were also investigating whether STING's newly discovered proton pump 201 function was involved in these processes [18]. Compound 53 is a STING agonist that inhibits 202 the function of the STING proton pump [2]. HPMECs activated by diABZI significantly 203 upregulated Ccl4, Ccl5, Cxcl9, Cxcl10, and Il15 expression, whereas Compound 53 204 administration markedly attenuated the upregulation of Cxcl10 and Il15 (Fig. 3G), suggesting 205 that this function depended on the STING proton pump. 206 In vivo, to confirm this process was indeed independent of other pulmonary immune 207 cells, we modelled NCG immunodeficient mice with equivalent diABZI doses and assessed 208 representative pulmonary chemokine levels. Results showed Cxcl10 and Ccl5 remained 209 significantly upregulated (Fig. 3H). To exclude interference from other non-immune cells, we 210 administered identical doses of diABZI to HPMEC, human lung epithelial cells (BEAS-2B), 211 and human lung fibroblasts (HLF-1) in vitro. Endothelial cells exhibited significantly higher 212 cytokine upregulation compared to epithelial and fibroblast cells (Fig. S3B). 213 Endothelial cells activated by STING agonists can recruit NK cells 214 Previously, we observed that within 4 hours of administration, the probability of endothelial 215 cell signal transmission increased substantially, with signals being communicated to NK cells 216 (Fig. 2, F and G). Given that STING agonists promoted endothelial cell expression of 217 chemokines, pulmonary endothelial cells we hypothesized that activated pulmonary 218 endothelial cells may recruit NK cells. Thus, we conducted transwell assays using both the 219 NK-92 cell line and primary NK cells isolated from human PBMCs. Results demonstrated 220 that, compared to the vehicle group, endothelial cells activated by diABZI attracted more NK 221 cells into the lower chamber (Fig. 3, I and J). 222 NK cells activated by endothelial cells produce IFN-γ to damage endothelial cells 223 Previously, we observed that activated endothelial cells could expression NK cell activator 224 IL-15. In subsequent experiments, we investigated the changes occurring in NK cells 225 following signal reception. Therefore, we continued to analyze whether the recruited NK 226 cells would be further activated. 227 scRNA-seq data revealed that STING expression in NK cells within the lungs rapidly 228 increased within 0–4 hours post-administration, with genes in the IFN-α response pathway 229 significantly upregulated (Fig. 4A) and signal reception intensity substantially elevated (Fig. 230 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint 2, F and G). To investigate the effects of factors secreted by endothelial cells on NK cells, 231 HPMECs were cultured in medium containing diABZI for 4 hours. NK-92 cells were 232 cultured in this medium for 4 hours, after which they were harvested for gene expression 233 analysis. Results revealed that upon receiving endothelial activation signals, NK-92 cells 234 exhibited markedly elevated expression of Ccl4, the classical receptor Ccr5 for CCL5, the 235 receptor Il15ra for IL-15, and Ifng. (Fig. 2E) Moreover, direct activation of NK-92 cells with 236 diABZI did not induce this upregulation (Fig. 4B), suggesting that the above changes 237 depended on endothelial cell-secreted factors. 238 Further analysis of NK cells in scRNA-seq data revealed four distinct subpopulations 239 after dimensionality reduction. Heatmaps indicated that clusters 0 and 1 predominantly 240 comprised Ifng-expression populations (Fig. 4C). Literature indicates that IFN-γ promotes 241 endothelial cell apoptosis, compromising the integrity of the endothelial barrier [19, 20]. We 242 hypothesized that activated NK cells could secrete elevated levels of IFN-γ, thereby 243 promoting endothelial cell apoptosis. scRNA-seq data revealed significantly increased 244 expression of endothelial apoptosis pathway genes within 0–4 hours post intraperitoneal 245 administration (Fig. 4D). Following 24-hour co-culture of NK-92 and HPMEC cells at 0:1, 246 3:1 and 5:1 ratio in vitro, NK-92 cells were removed, and then we add CCK-8 to detect the 247 viability of HPMECs. We observed a significant decrease in relative cell viability of 248 endothelial cells post-co-culture with NK-92 cells, indicating heightened endothelial cell 249 death (Fig. 4E). Then, in vivo, the neutralizing antibodies against CXCL10/CCL5/IL-15/IFN-250 γ/IL-1β were administered separately with diABZI. Through flow cytometry, results showed 251 that STING agonist reduced the proportion of pulmonary endothelial cells, indicating the 252 impairment of the endothelial barrier. Neutralizing IFN-γ mitigated this reduction, indicating 253 that NK cell activation and the IFN-γ they secrete exert damaging effects on the endothelium 254 (Fig. 4F), demonstrating the important role of IFN-γ, primarily expressed by NK cells, in this 255 process. 256 Taken together, we discovered that STING agonists could directly activate endothelial 257 cells, promoting their recruitment and activation of NK cells, and then the activated NK cells 258 could induce endothelial cell apoptosis via IFN-γ. 259 Tbx21+ NK cells activated by endothelial cells produce chemokines to recruit 260 neutrophils 261 Further analysis of NK cell alterations following administration of STING agonists, cell 262 interaction data indicated that, the strength of NK cell-to-neutrophil communication 263 progressively increased (Fig. 2G). Concurrently, neutrophil infiltration within lung tissue 264 substantially increased (Fig. 1I and 2A). Previous studies have demonstrated that during acute 265 lung injury, NK cells upregulate T-bet expression and secrete CXCL1/2 to chemotactically 266 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint recruit CXCR2+ neutrophils, thereby exacerbating disease progression [21]. Therefore, we 267 subsequently analyzed the interactions between NK cells and neutrophils. 268 Analysis of NK cell T-bet (Tbx21) expression at different time points post administration, 269 we found that STING agonists significantly upregulated Tbx21 expression in cluster0/1 NK 270 cells (Fig. 5A). NicheNet analysis of the Tbx21+ NK population predicted downstream target 271 genes potentially upregulated following NK cell activation by endothelial cell-derived 272 ligands. Circos plot results indicated that cluster0/1 NK cells, upon receiving Ccl4, Ccl5 and 273 Il15 signals from endothelial cells, leading to increased expression of chemokines Ccl3, Ccl4, 274 Cxcl10, Ccr5, and Ifng (Fig. 5B). Moreover, within 0-4 hours, differentially expressed genes 275 in cluster 0 NK cells enriched for neutrophil chemotaxis related pathways (Fig. 5C). Thus, 276 single-cell data analysis preliminarily validated our hypothesis. 277 Subsequently, we assessed the secretion levels of various chemokines by NK-92 cells 278 activated by endothelial cells. Specifically, direct diABZI treatment of NK-92 cells markedly 279 increased Ccl3/Ccl4/Cxcl10 expression. In contrast, Cxcl1/Cxcl2/Cxcl10 expression was 280 significantly amplified by endothelial cell activation signals (Fig. 5D). 281 Neutrophils secrete IL-1β through positive feedback pathways and form NETs during 282 lung injury. 283 The aforementioned experiments have demonstrated that within a short period following 284 STING agonist administration, NK cells activated by endothelial cells recruit a substantial 285 infiltration of neutrophils into the pulmonary immune microenvironment. Furthermore, 286 neutrophil signaling activity peaks at 12 hours (Fig. 2F and G), with IL-1β secretion reaching 287 its highest level at this time point (Fig. 2D). To investigate how recruited neutrophils mediate 288 lung injury, differential gene analysis and pathway enrichment of 12-hour neutrophils from 289 scRNA-seq data were performed (Fig. 6A). Differentially expressed genes were enriched not 290 only in chemotaxis but also in neutrophil extracellular trap (NET) formation and related 291 pathways (Fig. 6B). Previous studies have demonstrated that in acute lung injury in mice, 292 NETs exacerbate disease progression by mediating parenchymal cell death [22, 23]. 293 Both multiplex immunofluorescence assays and western blot confirmed increased NETs 294 during STING agonist-mediated pulmonary injury (Fig. 6B and 6C). To validate the 295 contributions of potential immune cells and cytokines, we employed neutralizing antibodies 296 to eliminate NK cells, neutrophils, and interstitial macrophages, and neutralized IL-1β in the 297 diABZI treatment models. Pulmonary tissues expression of the NET marker MPO was 298 assessed by Western blot. Results demonstrated that STING agonists increased pulmonary 299 MPO expression, whilst depletion of NK cells, neutrophils, and interstitial macrophages, and 300 neutralization of IL-1β, all reduced pulmonary MPO levels (Fig. 6C). 301 Take together, recruited neutrophils may cause lung injury through NETs. 302 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Immune cells depletion and cytokine neutralization in vivo validates the mechanism of 303 lung injury caused by STING agonist. 304 To validate the contribution of the aforementioned immune cells and cytokines to the STING-305 mediated toxicity mechanism, we performed immune cell depletion and cytokine 306 neutralization in mice. For neutrophils, which exhibited the most pronounced infiltration 307 increase post-treatment, NK cell depletion reduced their proportion among total lung cells, 308 confirming the chemotactic effect of NK cells on neutrophils during lung injury. Neutralizing 309 IL-1β reduced the increased NK cell infiltration, whereas neutrophil depletion further 310 exacerbated NK cell infiltration. Regarding the markedly reduced pulmonary endothelial 311 cells post treatment, the proportion rebounded following depletion of NK cells, neutrophils, 312 or macrophages, suggesting the role of these three cell types in STING agonist induced 313 pulmonary injury. Neutrophil depletion or IL-1β neutralization partially mitigated the 314 increased interstitial macrophages infiltration, suggesting neutrophils and their secreted IL-1β 315 may exert chemotactic effects on interstitial macrophages (Fig. 7A). 316 RT-qPCR analysis of inflammatory cytokines and chemokines in lung tissues confirmed 317 that IL-1β was primarily secreted by neutrophils. Depletion of NK cells and macrophages, 318 alongside IL-1β neutralization, partially alleviated this upregulation. These findings indicated 319 that NK cell depletion indeed reduced neutrophil infiltration, thereby lowering IL-1β 320 secretion levels, with IL-1β secretion potentially exhibiting a positive feedback response. 321 Regarding IFN-γ, primarily secreted by NK cells, deletion of neutrophils reduced its secretion 322 levels, further validating the interaction between NK cells and neutrophils in the toxic 323 mechanism. Overall, deletion of macrophages, neutrophils, NK cells, or neutralization of IL-324 1β reduced chemokines production to varying degrees. However, STING agonist-induced 325 Ccl5 upregulation was largely unaffected by these immune cell deletions, suggesting that 326 Ccl5 elevation depends on other non-immune cells, including endothelial cells (Fig. 7B). 327 This directly demonstrated that these immune cells and inflammatory mediators 328 contribute to STING agonist-induced lung injury. 329

330 In this study, we have demonstrated the crucial role of the “endothelial-NK-neutrophil” 331 axis in the pulmonary toxicity of STING agonists. 332 To investigate the effects of STING agonists on the pulmonary immune 333 microenvironment and gene expression, we performed dynamic monitoring of mouse lung 334 tissue at different time points post administration using single-cell RNA-seq and flow 335 cytometry. We observed that STING agonists induced dramatic alterations in the immune 336 microenvironment within a short timeframe. During the early phase, the proportion of T and 337 B lymphocytes plummeted. In fact, this phenomenon has been observed in multiple studies 338 involving gain-of-function STING mutation mice, as well as in experiments involving the 339 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint administration of STING agonists such as cGAMP [24-26]. Mechanistically, existing studies 340 suggest this is unrelated to interferon signaling or the canonical IRF3 pathway, and may 341 instead be associated with endoplasmic reticulum stress [27-30]. Studies have also indicated 342 that high-dose intratumoral administration of STING agonists can inhibit the expansion of 343 anti-tumor CD8+ T cells [31]. This immunotoxicity effect of STING agonists may 344 compromise their sustained antitumor activity. 345 Cell communication analysis revealed that during the early phase, pulmonary vascular 346 endothelial cells served as the primary signal emitters, with NK cells acting as the core signal 347 recipients. This suggests that endothelial cells, being the first to encounter circulating STING 348 agonists, may be primed for activation and subsequently initiate immune cell recruitment. 349 The role of endothelial cells in lung injury is gradually being recognized. For example, 350 researchers found that overexpression of cGAS in pulmonary endothelial cells promotes its 351 expression of CCL5 to recruit T cells [32]. These T cells then secrete IFN-γ, causing 352 endothelial damage that mediates the development of pneumonia. Furthermore, multiple 353 studies using STING N153S gain-of-function mice have revealed that sustained STING 354 activation leads to severe pneumonia in these animals [26, 33]. Chimeric mouse experiments 355 indicate that STING activation in endothelial cells initiates pneumonia, but its progression 356 still requires STING activation in immune cells [34, 35]. 357 Both NicheNet predictions from scRNA-seq and in vitro transwell assays confirmed that 358 NK cells activated by endothelial cells exhibit a phenotype characterized by high expression 359 of the transcription factor T-bet (Tbx21), secreting factors such as CXCL1, CXCL2, and 360 CXCL10. These factors strongly recruit neutrophils to lung tissue by acting on receptors 361 including CXCR2 on neutrophil surfaces [21, 36, 37]. 362 Of course, our study has many limitations, and further research is needed. For instances, 363 we have identified the role of NK cells and neutrophils in the pulmonary toxicity of STING 364 agonists, but have not evaluated their function in the antitumor efficacy of STING agonists. 365 This limitation constrains research into the efficacy-toxicity balance of STING agonists. On 366 the other hand, this study did not propose specific STING toxicity intervention strategies. In 367 future studies, we plan to modify STING agonists to avoid their action on endothelial cells, or 368 to combine them with drugs that ameliorate endothelial injury. 369 In summary, this study systematically characterized the pulmonary toxicity phenotype of 370 STING agonists and elucidated the pathogenesis dependent on the endothelial cell-NK cell-371 neutrophil axis. Following administration of the STING agonists, pulmonary endothelial cells 372 were initially targeted. Upon activation, they recruited NK cells via CCL5 and further 373 activated NK cells through IL-15. Activated NK cells promoted endothelial cell apoptosis via 374 IFN-γ and recruited neutrophils via the CXCL2-CXCR2 axis. Neutrophils then caused lung 375 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint injury through NETs, with IL-1β playing a pivotal role in this positive feedback loop. Our 376 research offers insights for developing safer STING agonists in the future, thereby enhancing 377 their potential for clinical application. 378

379 Mice. Male six- to eight-week-old C57BL/6 mice were purchased from Vital River. NCG 380 mice were purchased from GemPharmatech Co., Ltd. All mice were maintained under 381 specific pathogen-free (SPF) conditions in the animal facility of the Shanghai Institute of 382 Materia Medica, Chinese Academy of Sciences. Animal care and experiments were 383 performed in accordance with the Shanghai Institute of Materia Medica, Chinese Academy of 384 Sciences, using protocols approved by the Institutional Laboratory Animal Care and Use 385 Committee (IACUC). 386 387 Establishment of a STING agonist-mediated pulmonary inflammation model. diABZI 388 STING agonist-1 (TargetMol, T11035) was dissolved in DMSO to prepare a 40 mg/mL stock 389 solution. The administration concentration of diABZI is 2 mg/kg (200 μL, dissolved in 40% 390 PEG 300 + 8% Tween-80, then diluted to volume with saline). The pneumonia model was 391 established via intraperitoneal injection on consecutive 3 days (D0-D2). On day D3, blood 392 was collected from the posterior orbital sinus under anesthesia. Following blood collection, 393 mice were euthanized. Lung tissues were rapidly frozen in dry ice for RNA extraction or 394 fixed in formalin solution for paraffin sectioning. 395 396 Cells. NK-92 cell lines were purchased from SUNNCELL and HPMEC cell lines were 397 purchased from IMMOCELL. HPMEC and NK-92 cells were cultured in their respective 398 dedicated media. Human PBMCs were purchased from Milestone. All cells were cultured at 399 37° C in a 5% CO2 humidified atmosphere. 400 401 Immune deletion and cytokine neutralization. In the STING-associated pneumonia model, 402 anti-mouse immune cell deletion antibody (200 μg/animal; Starter) was administered 403 intraperitoneally on days D-1 and D1. On day -1, administer cytokine-neutralizing antibody 404 (200 μg/animal; BioXcell) via intraperitoneal injection. On days 0, 1, and 2, administer 405 cytokine-neutralizing antibody (100 μg/animal; BioXcell) via intraperitoneal injection to 406 maintain neutralizing effects. 407 408 Immunophenotype analysis. After obtaining the whole mouse lung, mince it and place it in 409 digestion buffer (3 mL phenol red-free RPMI 1640 + 2.1 mg collagenase I). Shake at 37° C 410 and 220 rpm for 30 minutes. Subsequently, filter the digested solution through a 70 μm filter 411 membrane. Centrifuge, then lyse with erythrocyte lysis buffer at room temperature for 8 412 minutes. Filter again after lysis, then centrifuged, washed, and resuspended to obtain a single-413 cell suspension. Then the cells were blocked with 4% FBS and anti-CD16/32 (BD 414 Biosciences), incubated with surface marker antibodies for 20 minutes at 4℃. Flow 415 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint cytometry analysis was performed using ACEA NovoCyte and data processing was done 416 through NovoExpress software. Antibody staining was performed following the 417 manufacturer’s recommendations. 418 419 Immunofluorescence. NETs visualization was performed using immunofluorescence 420 confocal microscopy. Formalin-fixed and paraffin-embedded lung specimens from mice were 421 stained with anti-citrullinated histone-3 (citH3, 1:200; Abcam) and anti-myeloperoxidase 422 (diluted 1:200; Abcam 134132), a polyclonal goat anti-mouse Alexa Fluorite 647 antibody 423 (Thermo Fisher) and anti-rabbit Alexa Fluorite 488 antibody (Thermo Fisher) as secondary 424 Abs. The DNA was stained using DAPI (Sigma-Aldrich). All the samples were observed 425 under laser scanning confocal microscopy. 426 427 Cell preparation single cell for RNA-seq. After harvested, lung tissues were washed in ice-428 cold PBS (Hyclone SH30256.01) and dissociated using SeekMate Tissue Dissociation 429 Reagent Kit A Pro (SeekGene K01801-30) from SeekGene as instructions. DNase Ⅰ (Sigma 430 9003-98-9) treatment was optional according to the viscosity of the homogenate. Cell count 431 and viability was estimated using SeekMate Tinitan Fluorescence Cell Counter (SeekGene 432 M002C) with AO/PI reagent after removal erythrocytes (Solarbio R1010) and then debris and 433 dead cells removal was decided to be performed or not (Miltenyi 130-109-398/130-090-101). 434 Finally fresh cells were washed twice in the RPMI1640 (Gibco 11875119) and then 435 resuspended at 1×106 cells per ml in RPMI1640 and 2% FBS (Gibco 10100147C). 436 437 Single cell RNA-seq library construction and sequencing. Single-cell RNA-Seq libraries 438 were prepared using SeekOne® DD Single Cell 3’ library preparation kit (SeekGene Catalog 439 No.K00202). Briefly, appropriate number of cells were mixed with reverse transcription 440 reagent and then added to the sample well in SeekOne® chip S3. Subsequently Barcoded 441 Hydrogel Beads (BHBs) and partitioning oil were dispensed into corresponding wells 442 separately in chip S3. After emulsion droplet generation reverse transcription were performed 443 at 42℃for 90 minutes and inactivated at 85℃ for 5 minutes. Next, cDNA was purified from 444 broken droplet and amplified in PCR reaction. The amplified cDNA product was then 445 cleaned, fragmented, end repaired, A-tailed and ligated to sequencing adaptor. Finally, the 446 indexed PCR were performed to amplified the DNA representing 3’ polyA part of expressing 447 genes which also contained Cell Barcode and Unique Molecular Index. The indexed 448 sequencing libraries were cleanup with V AHTS DNA Clean Beads (Vazyme N411-01), 449 analyzed by Qubit (Thermo Fisher Scientific Q33226) and Bio-Fragment Analyzer (Bioptic 450 Qsep400). The libraries were then sequenced on illumina NovaSeq X Plus with PE150 read 451 length. 452 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint 453 Statistical analysis. The in vivo experiments were randomized but the researchers were not 454 blinded to allocation during experiments and results analysis. Statistical analysis was 455 performed using GraphPad Prism 8 Software. A Student's t test was used for comparison 456 between the two groups. Multiple comparisons were performed using one-way ANOVA 457 followed by Tukey’s multiple comparisons test or two-way ANOVA followed by Tukey’s 458 multiple comparisons test. Detailed statistical methods and sample sizes in the experiments 459 are described in each figure legend. All statistical tests were two-sided and P-values < 0.05 460 were considered to be significant. ns not significant; *p < 0.05; **p < 0.01; ***p < 0.001. 461 462 Data availability 463 All data are available in the main text or the supplementary materials. Correspondence and 464 requests for materials should be addressed to Y.L. 465 466

467 1. Zhang, J., et al., Emerging mechanisms and implications of cGAS-STING signaling in 468 cancer immunotherapy strategies. Cancer Biology & Medicine, 2024: p. 1-20. 469 2. Gao, Z., B. Wang, and L. Zhang, Activated STING: an ion channel to trigger non-470 interferon-related functions. Signal Transduction and Targeted Therapy, 2023. 8(1). 471 3. Jian Wang, et al., STING licensing of type I dendritic cells potentiates antitumor 472 immunity. Sci Immunol, 2024. 9(92). 473 4. Yang, H., et al., STING activation reprograms tumor vasculatures and synergizes with 474 VEGFR2 blockade. Journal of Clinical Investigation, 2019. 129(10): p. 4350-4364. 475 5. Luo, K., et al., Activation of Stimulation of Interferon Genes (STING) Signal and Cancer 476 Immunotherapy. Molecules, 2022. 27(14). 477 6. Ross, P., et al., Regulation of cellulose synthesis in Acetobacter xylinum by cyclic 478 diguanylic acid. Nature, 1987. 325(6101): p. 279-81. 479 7. Ohkuri, T., et al., STING contributes to antiglioma immunity via triggering type I IFN 480 signals in the tumor microenvironment. Cancer Immunol Res, 2014. 2(12): p. 1199-208. 481 8. Ramanjulu, J.M., et al., Design of amidobenzimidazole STING receptor agonists with 482 systemic activity. Nature, 2018. 564(7736): p. 439-443. 483 9. Meric-Bernstam, F., et al., Phase I Dose-Escalation Trial of MIW815 (ADU-S100), an 484 Intratumoral STING Agonist, in Patients with Advanced/Metastatic Solid Tumors or 485 Lymphomas. Clinical Cancer Research, 2022. 28(4): p. 677-688. 486 10. Study of Intratumorally Administered Stimulator of Interferon Genes (STING) Agonist 487 E7766 in Participants With Advanced Solid Tumors or Lymphomas - INSTAL-101. 2024 488 [cited 2026 02-08]; Available from: https://clinicaltrials.gov/study/NCT04144140. 489 11. A Study of TAK-500 With or Without Pembrolizumab in Adults With Select Locally 490 Advanced or Metastatic Solid Tumors. 2025 [cited 2026 03 -06]; Available from: 491 https://clinicaltrials.gov/study/NCT05070247. 492 12. Barber, G.N., STING: infection, inflammation and cancer. Nat Rev Immunol, 2015. 15(12): 493 p. 760-70. 494 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint 13. Chen, Q., et al., Carcinoma-astrocyte gap junctions promote brain metastasis by cGAMP 495 transfer. Nature, 2016. 533(7604): p. 493-498. 496 14. A Study of XMT-2056 in Advanced/Recurrent Solid Tumors That Express HER2. 2023 497 [cited 2026 03-06]; Available from: https://clinicaltrials.gov/study/NCT05514717. 498 15. Chen, Y., et al., Macrophage STING signaling promotes fibrosis in benign airway stenosis 499 via an IL6-STAT3 pathway. Nature Communications, 2025. 16(1). 500 16. Ohnishi, H., et al., Comparative study of KL-6, surfactant protein-A, surfactant protein-501 D, and monocyte chemoattractant protein-1 as serum markers for interstitial lung 502 diseases. American Journal of Respiratory and Critical Care Medicine, 2001. 165(3): p. 503 378-381. 504 17. Wright, J., Immunomodulatory functions of surfactant. Physiological Reviews, 1997. 505 77(4): p. 931-962. 506 18. Bingxu Liu, R.J.C., Ivan S. Pires, Matteo Gentili, Ellie Feng, Quentin Hellier, Marc A. 507 Schwartz, Paul C. Blainey, Darrell J. Irvine, Nir Hacohen, Human STING is a proton 508 channel. Science, 2023. 381(6657): p. 508-514. 509 19. PAN Jian, Z.S.-q., CHEN Bing, Interferon-γ induces the apoptosis of vascular 510 endothelial cells in vitro. Journal of Bengbu Medical University, 2005. 30(5): p. 388-389. 511 20. Ng, C.T., L.Y. Fong, and M.N.H. Abdullah, Interferon-gamma (IFN-γ): Reviewing its 512 mechanisms and signaling pathways on the regulation of endothelial barrier function. 513 Cytokine, 2023. 166. 514 21. Hoegl, S., et al., NK cells regulate CXCR2+ neutrophil recruitment during acute lung 515 injury. Journal of Leukocyte Biology, 2017. 101(2): p. 471-480. 516 22. Zhang, H., et al., Neutrophil extracellular traps mediate m6A modification and regulates 517 sepsis-associated acute lung injury by activating ferroptosis in alveolar epithelial cells. 518 International Journal of Biological Sciences, 2022. 18(8): p. 3337-3357. 519 23. Sha, H.-X., et al., Neutrophil extracellular traps trigger alveolar epithelial cell necroptosis 520 through the cGAS-STING pathway during acute lung injury in mice. International 521 Journal of Biological Sciences, 2024. 20(12): p. 4713-4730. 522 24. Dynamic STING repression orchestrates immune cell development and function. 2025. 523 25. Tang, C.-H.A., et al., STING regulates BCR signaling in normal and malignant B cells. 524 Cellular & Molecular Immunology, 2020. 18(4): p. 1016-1031. 525 26. Bouis, D., et al., Severe combined immunodeficiency in stimulator of interferon genes 526 (STING) V154M/wild-type mice. Journal of Allergy and Clinical Immunology, 2019. 527 143(2): p. 712-725.e5. 528 27. Warner, J.D., et al., STING-associated vasculopathy develops independently of IRF3 in 529 mice. Journal of Experimental Medicine, 2017. 214(11): p. 3279-3292. 530 28. Gulen, M.F., et al., Signalling strength determines proapoptotic functions of STING. 531 Nature Communications, 2017. 8(1). 532 29. Tang, C.H., et al., Agonist-Mediated Activation of STING Induces Apoptosis in Malignant 533 B Cells. Cancer Res, 2016. 76(8): p. 2137-52. 534 30. Wu, J., et al., STING-mediated disruption of calcium homeostasis chronically activates 535 ER stress and primes T cell death. Journal of Experimental Medicine, 2019. 216(4): p. 536 867-883. 537 31. Sivick, K.E., et al., Magnitude of Therapeutic STING Activation Determines CD8+ T Cell-538 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Mediated Anti-tumor Immunity. Cell Reports, 2018. 25(11): p. 3074-3085.e5. 539 32. Zhao, R., et al., cGAS-activated endothelial cell-T cell cross-talk initiates tertiary 540 lymphoid structure formation. Sci Immunol, 2024. 9(98): p. eadk2612. 541 33. Luksch, H., et al., STING-associated lung disease in mice relies on T cells but not type I 542 interferon. Journal of Allergy and Clinical Immunology, 2019. 144(1): p. 254-266.e8. 543 34. Gao, K.M., et al., Endothelial cell expression of a STING gain-of-function mutation 544 initiates pulmonary lymphocytic infiltration. Cell Reports, 2024. 43(4). 545 35. Kevin MingJie Gao, M.M., Thomas Tedder, Ann Marshak-Rothstein, Katherine A 546 Fitzgerald, Radioresistant cells initiate lymphocyte-dependent lung inflammation and 547 IFNγ-dependent mortality in STING gain-of-function mice. Proceedings of the National 548 Academy of Sciences, 2022. 549 36. Benigni, G., et al., CXCR3/CXCL10 Axis Regulates Neutrophil–NK Cell Cross-Talk 550 Determining the Severity of Experimental Osteoarthritis. The Journal of Immunology, 551 2017. 198(5): p. 2115-2124. 552 37. Jaeger, B.N., et al., Neutrophil depletion impairs natural killer cell maturation, function, 553 and homeostasis. Journal of Experimental Medicine, 2012. 209(3): p. 565-580. 554 555 Acknowledgments 556 This work was supported by the Shanghai Rising-Star/Sailing Program (24YF2756000). All 557 the schematics are were created with BioRender.com. 558 559 Author contributions 560 Conceptualization, L.G. and Y.L.; Methodology, Y.L. and C.C.; Formal Analysis, Y.L.; 561 Investigation, C.C., Y.Z., F.D., R.L., X.Z., S.W., Y.W., F.Q., L.C., R.C., and F.L.; Resources, 562 L.G. and Y.L.; Writing – Original Draft, C.C. and Y.L.; Writing – Review & Editing, L.G. 563 and Y.L.; Supervision, L.G.; Funding Acquisition, L.G. and Y.L. 564 565 Competing interests 566 The authors declare that they have no competing interests. 567 568 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 1 569 570 Fig. 1. Systemic administration of STING agonists causes lung injury in mice. Female 571 C57BL/6 mice received intraperitoneal injections of 2 mg/kg diABZI every day for three 572 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint days. (A) Body weight change percent of mice was monitored after diABZI treatment from 573 day1 to day4 and lung coefficient was detected on day4 (n=6). (B) Lung wet/dry weight ratio 574 was detected on day4 and mice peak expiratory flow and respiratory rate was monitored 575 during day1 to day4 (n=6). (C) Representative H&E staining images of lung tissues from 576 mice. Scale bar, 200 μm. (D) Western blot analysis of surfactant protein A proteins in mouse 577 serum. (E-G) Transcription levels of multiple proinflammatory factor in mouse lung tissues 578 were detected (n=6). (H) Flow cytometric analysis of neutrophil macrophage M1-like and 579 M2-like macrophages in the lung tissues (n=6). (H) Flow cytometric analysis of lymphocytes 580 and myeloid cells in the lung tissues (n=6). The data are presented as the mean ± SEM. * p < 581 0.05; ** p < 0.01; *** p < 0.001; ns not significant by ns not significant by unpaired t test or 582 ANOVA followed by Tukey’s multiple comparisons test. 583 584 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 2 585 586 Fig. 2. Single-cell sequencing reveals alterations and interactions in pulmonary cell 587 populations during inflammation. (A) t-SNE plots and relative proportion of the indicated 588 cell types for scRNA-seq data of lung samples. (B) Tmem173 expression in t-SNE plots. (C) 589 Violin plot of gene expression scores for the HALLMARK INTERFERON ALPHA 590 RESPONSE pathway across cell populations. (D) Heatmap of IL-1β and IFN-γ expression 591 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint levels in different cell populations at various time points (E) Heatmap of chemokine 592 expression levels in various cell populations at 0 and 4 hours. (F) Scatter plots visualized the 593 primary senders and receivers of cellular communication. The x-axis and y-axis respectively 594 represent the total outgoing or incoming communication probabilities associated with each 595 cell population. The size of each point indicates the number of relationships (both outgoing 596 and incoming) with each cell population. (G) Heatmap of relative signal strength at different 597 time points (red indicates an increase compared to the previous time point, blue indicates a 598 decrease). 599 600 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 3 601 602 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 3. Endothelial stimulated by STING agonist secret chemokines and cytokine to 603 recruit and activate NK cells. (A) Volcano plot of gene expression differences of 604 endothelial cells at different time points from scRNA-seq data of lung samples. (B) Heatmap 605 of chemokines expression level of endothelial cells. (C) Violin plot of gene expression scores 606 for the HALLMARK INTERFERON ALPHA RESPONSE pathway of endothelial cells. (D) 607 Western blot analysis of total and phosphorylated TBK1 and STING proteins in HPMEC cell 608 lines with or without diABZI treatment (n=3). (E) KEGG pathway enrichment analysis of 609 differentially expressed genes in NK cell at 4 h compared to 0 h. (F) Transcriptional levels of 610 Ccl5, Cxcl9/10 and Il15 of HPMEC with or without diABZI or inhibitor treatment (n=3). (G) 611 Transcriptional levels of Ccl4/5, Cxcl9/10 and Il15 of HPMEC with or without diABZI or 612 Compound 53 treatment (n=6). (H) Transcriptional levels of Ccl5 and Cxcl10 of NCG mice 613 with or without diABZI treatment (n=6). (I-J) Flow cytometry counting of NK-92 or 614 hPBMC-NK cells recruited to the basement membrane by HPMECs (n=3). The data are 615 presented as the mean ± SEM. * p < 0.05; ** p < 0.01; *** p < 0.001; ns not significant by ns 616 not significant by unpaired t test or ANOVA followed by Tukey’s multiple comparisons test. 617 618 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 4 619 620 Fig. 4. STING-activated endothelial cells release IL-15 to activate NK cells, which 621 produce IFN-γ to damage endothelial cells. (A) Heatmap of Tmem173 expression level and 622 violin plot of gene expression scores for the HALLMARK INTERFERON ALPHA 623 RESPONSE pathway of NK cells at different time points from scRNA-seq data of lung 624 samples. (B) Schematic diagram of HPMEC and NK-92 conditioned medium cultivation 625 workflow and transcriptional levels of Ccr5, Il15ra and Ifng of NK-92 (n=6). (C) UMAP 626 plots and heatmap of Ifng expression level of NK cell subsets for scRNA-seq data of lung 627 samples. (D) Violin plot of gene expression scores for the HALLMARK INTERFERON 628 ALPHA RESPONSE pathway of NK cell subsets. (E) Relative cell viability after co-629 culturing NK-92 cells with endothelial cells for 24 hours was determined by measuring the 630 absorbance at 450 nm following incubation with CCK-8 for 2 hours. (F) Flow cytometric 631 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint analysis of endothelial in the lung tissues after neutralization of chemokines and cytokines 632 (n=4). The data are presented as the mean ± SEM. * p < 0.05; ** p < 0.01; *** p < 0.001; ns 633 not significant by ns not significant by unpaired t test or ANOVA followed by Tukey’s 634 multiple comparisons test. 635 636 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 5 637 638 Fig. 5. Tbx21+ NK cells activated by endothelial cells produce chemokines to recruit 639 neutrophils (A) Heatmap of Tbx21 expression level of NK cell subsets from scRNA-seq data 640 of lung samples. (B) Circos plot of NicheNet analysis of the Tbx21+ NK cell population 641 predicted downstream target genes potentially upregulated following NK cell activation by 642 endothelial cell-derived ligands. (C) GOBP and KEGG pathway enrichment analysis of 643 differentially expressed genes in cluster0 NK cell at 4 h compared to 0 h. (D) Transcriptional 644 levels of Ccl3/4 and Cxcl1/2/10 of NK-92 after conditioned medium cultivation with HPMEC 645 (n=6). The data are presented as the mean ± SEM. * p < 0.05; ** p < 0.01; *** p < 0.001; ns 646 not significant by ns not significant by unpaired t test or ANOVA followed by Tukey’s 647 multiple comparisons test. 648 649 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 6 650 651 Fig. 6. Neutrophils secrete IL-1β through positive feedback pathways and form NETs 652 during lung injury. (A) Volcano plot of differentially expressed genes in Neutrophil at 12 h 653 compared to 6 h. (B) GOBP and KEGG pathway enrichment analysis of differentially 654 expressed genes in Neutrophil at 12 h compared to 6 h. (C) Representative images of IF 655 staining of citH3 and MPO in lung tissues. (D) Western blot analysis of MPO in mice lung 656 tissues after depletion of immune cells. The data are presented as the mean ± SEM. * p < 0.05 657 by unpaired t test or ANOVA followed by Tukey’s multiple comparisons test. 658 659 660 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint Fig. 7 661 662 Fig. 7. Immune cells depletion and cytokine neutralization in vivo validates the 663 mechanism of lung injury caused by STING agonist. (A) Flow cytometric analysis of NK 664 cell, neutrophil, endothelial and IM in the lung tissues (n=4). (B) Transcriptional levels of 665 Il1b, Ifng, Ccl4/5 and Cxcl9/10 of the lung tissues after neutralization of chemokines and 666 cytokines (n=5). The data are presented as the mean ± SEM. * p < 0.05; ** p < 0.01; *** p < 667 0.001; ns not significant by ns not significant by unpaired t test or ANOVA followed by 668 Tukey’s multiple comparisons test. 669 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted March 12, 2026. ; https://doi.org/10.64898/2026.03.10.710764doi: bioRxiv preprint