cGAS-IFN-I responses by extracting nuclear DNA from dying cells via nucleocytosis

Abstract

Self-DNA triggers cGAS-STING-mediated type I interferon (IFN-I) to induce both protective and pathogenic immune responses; however, how self-DNA activates the cytosolic cGAS-STING pathway remains unclear. Here we show that the cGAS/STING/IFN-I axis is activated by self-DNA via a process termed ‘nucleocytosis’, in which nuclear DNA is extracted from dying cells by macrophages. Mechanistically, lysosomal malfunction, via both proton loss and palmitoyl-protein thioesterase 1 (PPT1) inhibition, triggers cell death and calreticulin accumulation in the nuclei. Live-cell imaging of secretion activity reveals that macrophages access the calreticulin-enriched nuclei of dying cells and extract DNA for cGAS-STING activation. Consistent with these findings, PPT1-targeting cationic amphiphilic drugs induce a cGAS-STING-dependent IFN-I response in vitro and in vivo. Our findings thus identify nucleocytosis as a macrophage function for nuclear DNA extraction and induction of the cGAS/IFN-I axis, and suggest that nucleocytosis-inducing cell death could be a druggable target for treating self-DNA-related inflammatory diseases.

Self-DNA has been implicated in the activation of cGAS/STING/IFN-I responses in autoimmunity and inflammatory diseases. Here the authors show that macrophage uses a process termed ‘nucleocytosis’ to extract nuclear DNA from lysosome-impaired, dying target cells, thereby activating downstream cGAS-STING signaling and IFN-I production.

Introduction

Cyclic GMP-AMP (cGAMP) synthase (cGAS) is a B-form DNA (B-DNA)-sensing pattern recognition receptor (PRR) that plays a critical role in host defence^1–5. It induces a robust type I IFN (IFN-I) response upon recognising pathogen-derived cytosolic DNA and activates a protective innate immune response^1–6. For activation of the cGAS-IFN-I axis, B-DNA must be delivered to the cytosol by specific delivery mechanisms such as viruses, cationic lipids, or exosomes^1–7. Alternatively, the cGAS-IFN-I axis is involved in a wide range of diseases, including autoimmune diseases, inflammatory diseases, and cancers^1–5,8. In these cases, the cGAS-IFN-I axis is activated by endogenous cytosolic self-B-DNA^1–4. Despite its critical role in various diseases, the magnitude of cGAS-IFN-I response by endogenous B-DNA is not as clear as that observed with the response to delivered B-DNA. Most reports have focused on the IFN-I signature rather than the production of IFN-I itself ^1,2; however, a few studies have indicated that low levels of IFN-I production occur via the self-DNA-cGAS axis⁹. Thus, it remains unclear whether a robust cGAS-IFN-I response can be activated by self-DNA.

As an endogenous ligand of cGAS, considerable attention has been focused on cytosolic self-DNA released from the cytosolic organelles within IFN-inducing cells in response to various cellular stresses, injury, or DNase defects^1–4,10. However, the above cases may be subject to quantitative limitation, and large amounts of self-DNA are presumed to be mainly derived from dead cells^11,12. It is well known that dead cells often expose ‘eat-me’ signals on their surface, leading to the uptake of whole cells or cellular fragments via phagocytosis or endocytosis, a process also implicated in the activation of immune responses^13,14. Recent reports indicate that lysosome-dependent cell death induced by anticancer drugs can also expose eat-me signals on the cell surface, which contribute to immune activation¹³. In these cases, it is likely that certain immunostimulatory components from engulfed dead cells or cellular debris contribute to the activation of immune responses; however, a mechanism to selectively extract only these immune-activating components, such as nucleic acids, has not been identified. Moreover, a recent report showed that extracellular self-DNA incorporated via endocytosis is recognised by Toll-like receptors (TLRs) but not by cGAS in autoimmune diseases¹⁵. Therefore, it remains unclear whether and how external self-double stranded DNA (dsDNA) directly activates cytosolic cGAS.

During the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) pandemic, considerable efforts were devoted to antiviral drug development^16–18. However, in most of the cases, the underlying molecular mechanisms remain unclear. One such candidate is hydroxychloroquine (HCQ)^19–21, a well-known drug for treating autoimmune diseases and malaria that is being investigated in clinical trials for cancer^{16–18,22,23}. In addition to HCQ, the lysosome-tropic antivirals, cationic amphiphilic drugs (CADs), exhibit broad antiviral effects, which are not attributed to the inhibition of viral components^17,24. Although their mechanisms of viral inhibition remain unclear, the most widely presumed mechanism is the inhibition of viral entry^22,25. Relatedly, one study reported possible concerns in the screening of antiviral drugs, demonstrating that certain CADs exhibited anti-SARS-CoV-2 activity in vitro, but this activity was not reflected in antiviral activity in vivo²⁶. However, they focused on the effect of CADs on infected cells rather than on other non-infected cells. It is worth noting that some CADs, including HCQ, CQ, and amodiaquine (ADQ), exhibit in vivo activity against specific viruses^27–29. These reports indicate that some of the valuable antiviral CADs have virus species-specific in vivo antiviral effects mediated by unknown mechanisms other than direct inhibition in infected cells; however, such indirect inhibition by CADs has not been studied in detail.

In this study, we provide evidence on how antiviral CADs exert an indirect antiviral response. Our results suggest that palmitoyl-protein thioesterase 1 (PPT1)-targeting antiviral CADs induce lysosomal dysfunction by both increasing pH and inhibiting PPT1, leading to cell death with accumulation of nuclear calreticulin. Consequently, macrophages access the nucleus and extract DNA via nucleocytosis to activate the cGAS-IFN I axis. Nucleocytosis is clearly demonstrated by live-cell imaging (LCI), including LCI of secretion activity (LCI-S), which shows IFN-β secretion following DNA extraction. The concept of nucleocytosis offers a new perspective on how self-nucleic acids activate cytosolic nucleic acid-sensing pathways, with implications not only for the mechanism of antiviral CADs but also for a range of self-DNA-related diseases, including autoimmune diseases, cancer, and infectious diseases.

Results

IFN-I production by antivirals

As a possible indirect mechanism of antiviral CADs, we focused on the IFN-I response, a crucial response against viruses. During the study, we first discovered the potential of HCQ to induce IFN-β production from granulocyte-macrophage colony-stimulating factor (GM-CSF)-cultured bone marrow cells (GM-BMCs). Notably, HCQ induced Ifnb1 gene expression with a slower time course, reaching a level similar to that induced by cGAMP, one of the most potent IFN-I inducers and ligand of stimulator of IFN genes (STING) (Fig. 1a, b and Supplementary Fig. 1a). Consistently, HCQ induced the expression of IFN-inducible genes and various genes related to PRR pathways and viruses (Fig. 1a, b and Supplementary Fig. 1a–c). Of note, HCQ-specific genes (category 4) were induced early in the time course (as shown in Supplementary Fig. 1a), potentially indicating or contributing to HCQ-specific cellular stress or programmed cell death processes (see below). These results suggest that HCQ has the potential to induce an antiviral response similar to that of cGAMP. Given that HCQ is used as an inhibitor of the IFN-I response to treat autoimmune diseases²², these results are surprising and suggest the possible existence of similar IFN-I inducers among antiviral CADs. Therefore, we performed further screening of IFN-β inducers among antiviral CADs and their related agents (Supplementary Fig. 1d). We identified HCQ, CQ, ADQ, tilorone, quinacrine (QC), and AQ-13 as IFN-β inducers (Fig. 1c and Supplementary Fig. 1e). Notably, their potential to induce IFN-β production was comparable to that of pathogen-associated molecular patterns such as cGAMP and lipopolysaccharide (LPS). As HCQ is a well-known inhibitor of endosomal TLR responses²², we further investigated the underlying mechanism by focusing on cytosolic PRRs and cell surface TLR pathways. HCQ induced the phosphorylation of IFN regulatory factor 3 a critical transcription factor for Ifnb1 gene expression. Consistently, TANK-binding kinase 1 (TBK1) and mitogen-activated protein kinases (MAPKs) were also phosphorylated by HCQ (Fig. 1d and Supplementary Fig. 1f), indicating the potential involvement of the TBK1-dependent pathway. Subsequently, we investigated the involvement of PRR pathways related to TBK1 and found that HCQ-mediated IFN-β production was completely abolished in Tbk1, Sting1, and Cgas (Mb21d1)-deficient GM-BMCs (Fig. 1e–g). The contribution of STING was further supported by inhibition with H-151 (Supplementary Fig. 1h). Conversely, HCQ-mediated IFN-β production was normal in TIR domain-containing adaptor-inducing IFN-β (Trif)-deficient GM-BMCs and partially decreased in IFN-β promoter stimulator 1 (Ips1)-deficient GM-BMCs (Supplementary Fig. 1i), indicating the critical role of the cGAS-STING axis. Comprehensive analysis of HCQ-inducible genes using RNA sequencing (RNA-seq) analysis showed that HCQ induced not only Ifnb1 but also various genes in a STING-dependent manner (Fig. 1h and Supplementary Fig. 1a). Importantly, IFN-β production by all IFN-β-inducing CADs was also diminished in Cgas-deficient GM-BMCs (Fig. 1i and Supplementary Fig. 1j), suggesting a common mechanism of IFN-β production. To further clarify their commonalities, we performed in silico analysis to identify conserved structural features. As shown in Supplementary Fig. 1k, l, two-dimensional (2D) structural analysis involving 2,048 parameters revealed conserved 2D structures in five (HCQ, CQ, ADQ, QC, and AQ-13), but not all, IFN-β inducers. Conversely, 3D pharmacophore analysis identified five conserved pharmacophore features unique to all IFN-β inducers and not found in any of the other 38 non-inducers (Fig. 1j, k and Supplementary Fig. 1m), indicating the importance of these five conserved pharmacophore features in IFN-β induction. Notably, we examined several antimalarial drugs, including piperaquine and artemisinin, but found no activity in IFN-β production from any of them (Supplementary Fig. 1e). These results suggest that antimalarial activity is not related to IFN-β production. Furthermore, typical lysosome-tropic agents, including Bafilomycin A (Baf A), showed no potential (Supplementary Fig. 1n), indicating that the inhibition of lysosomes by CADs is not sufficient for IFN-β production. Therefore, our results suggest a previously unrecognised commonality among IFN-β inducers regarding their 3D structure and their mechanism of IFN-β induction, suggesting an unknown conserved target.

Fig. 1. cGAS/STING-mediated robust IFN-I production by antiviral CADs.

a RNA-seq analysis of mRNA expression in GM-BMCs treated with HCQ (100 µM) or cGAMP (5 µg/mL). The top 50 genes induced by HCQ in 12 h among the genes are shown in the heatmap. The line graph of Ifnb1 and IFN-inducible genes is shown in (b). c IFN-β concentration in the culture supernatant of GM-BMCs treated with HCQ (80, 100, 120, and 140 µM), CQ (20, 40, 60, and 80 µM), tilorone (20, 40, 60, and 80 µM), ADQ (20, 40, 60, and 80 µM), QC (10, 12.5, 15, and 17.5 µM), cGAMP (5 µg/mL), or LPS (100 ng/mL) for 20 h. The chemical structural formulas of IFN-β inducers are shown above each column. d Immunoblot analysis for indicated proteins in GM-BMCs treated with the indicated concentrations of HCQ or cGAMP (5 µg/mL) for 6 h. e–g IFN-β concentration in the culture supernatant of control or gene-deficient GM-BMCs (e; Tnf^-/-/Tbk1^-/-, f; Sting1^-/-, g; Cgas^-/-) treated with the indicated concentration of HCQ, cGAMP (5 µg/mL), or poly(dA:dT) (5 µg/mL) for 16 h. h RNA-seq analysis of mRNA expression in wild-type (WT) or Sting1^-/- GM-BMCs treated with HCQ (100 µM) for 6 h. i. IFN-β concentration in the culture supernatant of WT or Cgas^-/- GM-BMCs treated with IFN-β inducers as in c for 16 h. j, k Pharmacophore model constructed from IFN-β inducers. Features without parentheses indicate features all compounds have in common, while features in parentheses indicate features some compounds have in common (j). Superimposition of IFN-β inducers and pharmacophore are shown in k. Data in b are expressed as the mean ± SD of biological triplicates. Data in (c, e–g), and i are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (each experiment was repeated three times). Data in d are representative of one experiment (experiment was repeated three times). See also Supplementary Fig. 1.

Robust IFN-β production by limited cells

To further examine the mechanism of antiviral CAD-mediated IFN-β production, we established a time-lapse monitoring system for IFN-β secretion using LCI-S^30,31. LCI-S revealed significant variation in the pattern of IFN-β secretion, including differences in the time course, dose, and shape of secreted IFN-β, even among cGAMP-treated cells (Fig. 2a–d, Supplementary Fig. 2a and Supplementary Movies 1, 2). Only a subset of cells secreted IFN-β, and each of these IFN-β-producing cells exhibited one-time secretion despite surviving the secretion process. Conversely, HCQ induced IFN-β secretion more slowly and less frequently than cGAMP (Fig. 2a–d, Supplementary Fig. 2a and Supplementary Movies 1, 2), although the total IFN-β production by HCQ was comparable to that by cGAMP (Fig. 1c). These results suggest that compared with cGAMP, HCQ induces robust IFN-β production from a limited number of cells. Of note, the perforated shape of secreted IFN-β indicated strong cell attachment on culture plates induced by HCQ (Fig. 2c and Supplementary Movies 3, 4). Consistent with these results, single-cell RNA-seq (scRNA-seq) analysis revealed a divergence of the gene expression pattern in IFN-β-expressing cells between HCQ and cGAMP treatments (Fig. 2e, f and Supplementary Fig. 2b). Interestingly, Ifnb1 expression was induced by HCQ only in macrophage cell clusters, whereas cGAMP induced Ifnb1 expression in both macrophage and dendritic cell (DC) clusters (Fig. 2e, f), indicating that a macrophage-specific function is required for HCQ-mediated IFN-β production. scRNA-seq analysis also confirmed a lower number of IFN-β-expressing cells but a higher expression level of strong inducers with HCQ treatment (Fig. 2f, g), confirming robust IFN-β production in a limited number of cells. Notably, there was no remarkable difference in IFN-inducible gene expression between HCQ and cGAMP (Supplementary Fig. 2c), consistent with their similar total production of IFN-β (Fig. 1a–c). These results demonstrate the potential of HCQ to induce IFN-β production from a limited number of cells and a significant difference in cell behaviour during IFN-β secretion between HCQ and cGAMP.

Fig. 2. HCQ-induced cell death and robust IFN-β production from living cells.

a, b LCI-S of GM-BMCs treated with cGAMP (a; 5 µg/mL) or HCQ (b; 100 µM). IFN-β secretion (integral value at indicated time point from the previous time point), DNA staining, and bright field are shown in the indicated sets of colours. c High-resolution pictures of cGAMP (upper; 5 µg/mL) or HCQ (lower; 100 µM)-treated GM-BMCs. IFN-β secretion (integral value, middle panel; total value, upper panel) and bright field are shown in the indicated sets of colours. d Quantitative analysis of IFN-β-secreting cells and dying cells. Right: number of IFN-β spots at each time point (HCQ; n = 197, cGAMP; n = 396). A line graph of DNA intensity over time in 169 fields of view is shown (left). The coloured line graph shows the average. e–h scRNA-seq analysis of GM-BMCs treated with HCQ (100 µM) or cGAMP (5 µg/mL) for 6 h. e UMAP plots showing cell types (left three panels) annotated by the expression of key marker genes (Z-score) (right panel). f UMAP plots showing the expression level (left) and density (right) of Ifnb1-expressing cells. Number of Ifnb1-expressing cells (unique molecular identifier > 1) and all cells: HCQ; n = 197 and 3856, cGAMP; n = 1,352 and 6,158. g Violin plots showing the expression of Ifnb1 (HCQ; n = 197, cGAMP; n = 1352) and mitochondria genes (HCQ; n = 1,112, cGAMP; n = 894). UMAP plots showing the percentage (left) and density (right) of mitochondrial genes. Number of injured cells (percentage of mitochondrial genes > 10%): HCQ; 1,112, cGAMP; 894. i Lactate dehydrogenase (LDH) activity and IFN-β concentration in the culture supernatant of GM-BMCs treated with indicated concentrations of HCQ, cGAMP (5 µg/mL), or LPS (100 ng/mL) for 16 h. Data in a–c are representative pictures from one experiment, which was repeated three times. Data in g, the box plot within the violin plot shows the median (centre line), and the interquartile range (IQR) (the box boundaries, spanning the 25th to 75th percentiles). The whiskers extend to data points within 1.5 times the IQR, defining the non-outlier range. Points outside the whiskers are outliers. Data in i are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (experiment was repeated three times). Statistical differences in g were examined using the two-sided Wilcoxon rank sum test. *p < 0.05, **p < 0.01, ***p < 0.001, n.s., not significant. See also Supplementary Fig. 2.

Time-lapse analysis revealed the induction of cell death by HCQ but not cGAMP (Fig. 2a, b, d, Supplementary Fig. 2a and Supplementary Movies 1, 2). scRNA-seq analysis also confirmed a higher frequency of injured or dying cells with HCQ treatment compared to cGAMP (Fig. 2g, h)³². Consistently, HCQ induced cell death in a dose-dependent manner, and the dose-response curve of IFN-β exhibited a bell-shaped pattern (Fig. 2i), indicating that high HCQ concentrations induce cell death in IFN-β-producing cells. Notably, cell death was observed with all of the other IFN-β inducers (Supplementary Fig. 2d). These results indicate the involvement of cell death-related events in IFN-β production by antiviral CADs.

Direct access to the dead cell nucleus and DNA extraction

As HCQ-mediated IFN-β production is dependent on cGAS (Fig. 1g), the above results indicate the possible involvement of dead cells as a source of B-DNA for cGAS-mediated IFN-β production. We further examined interactions among dying cells within HCQ-treated cell populations using an imaging flow cytometer. Consistently, HCQ treatment increased the SYTOX green-positive population, indicating the induction of cell death (Fig. 3a). This increase was significant, affecting not only singlets but also multiplets that included both living and dying cells (Fig. 3a, band Supplementary Fig. 3a). Furthermore, the importance of dsDNA transfer, but not cGAMP, was confirmed by the diminished production of IFN-β from mixed cells of Cgas^-/- and sting1^-/- cells (Fig. 3c). These results suggest the possibility that dsDNA transfer may occur from dying cells to living cells during HCQ-induced IFN-β production.

Fig. 3. HCQ-induced interaction between the dead cell nucleus and living cells.

a Image cytometry with GM-BMCs treated with HCQ (100 µM) for 6 h. SYTOX green-positive population (circles; upper panel) and “multiples” population (indicated region; lower panel) are shown. Statistical analyses of both populations is shown in the upper and lower right panels (n = 3). b SYTOX green-positive population within the “multiplets” population is shown. Upper right panel: statistical analysis of SYTOX green higher population (n = 3). Lower panels: images of multiplets in SYTOX green higher and lower populations. c (left) Schematic of the experiment. (right) IFN-β concentration in the culture supernatant of WT or cell mixture (WT/ Sting1^-/-, WT/ Cgas^-/-, Sting1^-/-/ Cgas^-/-) of GM-BMCs treated with HCQ (100 µM), cGAMP (5 µg/mL), or poly(dA:dT) (5 µg/mL) for 16 h. As this experiment was done with the data in Fig. 1g, data with WT cells are shared. d–h. Microscopic analysis of HCQ-treated GM-BMCs. d Snapshot analysis of GM-BMCs treated with HCQ (80 µM) for indicated time periods. Bright field with DNA staining is shown in both maximum intensity projection (MIP) and Z-slice images. e Time-lapse analysis of GM-BMCs treated with HCQ (80 µM). Bright field with DNA staining is shown in the left five panels. DNA intensity is summarised in the middle. The Z-slice picture at the indicated time point is shown on the right. f Holotomographic analysis of GM-BMCs treated with HCQ (80 µM). Snapshot images in indicated time points are shown in both MIP and Z-slice pictures. g, h Holotomographic time-lapse analysis of GM-BMCs treated with HCQ (100 µM) for the indicated time periods. DNA staining is shown in red in MIP images. Z-slice images at the indicated time points are shown in (h). i Concentration of IFN-β in the culture supernatant of GM-BMCs treated with HCQ (100 µM), or cGAMP (5 µg/mL) in the presence of the indicated concentration of inhibitor. Both panels were performed together, sharing unstimulated controls. Data in c and i are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (each experiment was repeated three times). Yellow arrows indicate the protrusion-like structure. Data in a, b are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (experiment was repeated two times). See also Supplementary Fig. 3.

Nuclear components, such as dsDNA and histone H3, were not released into the culture supernatant following HCQ treatment (Supplementary Fig. 4a). In this case, IFN-β-producing cells must directly access the nucleus to obtain B-DNA for cGAS-mediated IFN-β production. Based on the abovementioned results (Fig. 3a–c), we hypothesised that, through contact between dead and live cells, DNA is transferred to IFN-β-producing cells, thereby activating the STING-cGAS response. According to the results of LCI-S, within 6 h of HCQ treatment, at which time IFN-β secretion was observed, the nuclear structures of SYTOX-positive dead cells could be detected (Fig. 2b, d). Given the fact that HCQ induced robust IFN-β production, enough DNA was efficiently taken up from dead cells. Therefore, we next used microscopic analysis to determine whether live cells make contact with dead cell nuclei and subsequently take up DNA. The results showed that despite variations in the distance between dead and live cells and the degree of dead cell disintegration, in all cases both maximum intensity projection and slice images revealed protrusion-like structures and direct access to the nucleus (Fig. 3d). Remarkably, time-lapse analysis further demonstrated a decrease in DNA intensity in dead cell nuclei accompanied by an increase in DNA intensity in live cells (Fig. 3e, Supplementary Fig. 3b, c). Likewise, holographic microscopy confirmed the presence of protrusion-like structures, and importantly, DNA signals within the protrusion were also detected (Fig. 3f, g). Cross-sectional image analysis revealed that this protrusion extended into the nucleus, and that low-density regions were observed inside the nuclei at sites of contact (Fig. 3g, h). Consistent with these findings, inhibitors of actin polymerisation specifically suppressed HCQ-induced IFN-β production (Fig. 3i). Collectively, these results suggest that live cells directly access the nuclei of dead cells and extract DNA.

Nucleocytosis before IFN-β secretion

To further clarify the detailed mechanism, we performed high-resolution LCI-S and monitored the interaction between dying cells and IFN-β-secreting cells. Consistently, we observed interactions between dead cells and IFN-β-secreting cells before IFN-β production (Fig. 4a, Supplementary Fig. 4b, c, Supplementary Movies 5–7). Notably, IFN-β-secreting cells were observed to attach to dead cells near the nucleus, with dsDNA transfer into the cytosol occurring before IFN-β secretion (Fig. 4a, Supplementary Fig. 4b, c, Supplementary Movies 5–7). These results indicate that HCQ-induced IFN-β secretion is triggered by nuclear invasion to extract dsDNA. Consistently, imaging flow cytometry revealed that living cells interacted with dying cells within the dsDNA-positive population (Fig. 4b, c).

Fig. 4. Nucleocytosis and nuclear translocation of calreticulin.

a LCI-S of GM-BMCs treated with HCQ (100 µM). IFN-β secretion (blue) and DNA staining (magenta) are shown. DNA intensity is summarised in the right. b Representative images of SYTOX Green-positive duplets analysed in Fig. 3b. Six representative images from each population are shown. (-)P2: P2 population without HCQ treatment (Fig. 3b). HCQ P1: P1 population with HCQ treatment (Fig. 3b). c Quantitative analysis of DNA accumulation in the boundary region. Left: schematic of calculation. Right: DNA ratios within the cells in the unstimulated P2 and HCQ-stimulated P1 population in Fig. 3b. d–g Confocal microscopy of GM-BMCs treated with HCQ (100 µM). d Calreticulin (green) and DNA (blue) staining. e Intensity at the site of white dashed lines. The X-axis indicates the distance from the left edge of the white dashed lines. f 3D pictures based on sectional 2D pictures. g Statistical analyses of nuclear calreticulin intensity in each cell. Upper panel: schematic of the calculation. Data are presented as the mean ± SEM (0 h; n = 33, 6 h; n = 42) h–k Image cytometry of GM-BMCs treated with HCQ (100 µM) for 6 h. h Singlet cells are shown in dot plots. Calreticulin high and low populations are gated within live and dead cell populations. Right: proportions of the P1 and P2 populations. i Mean fluorescence intensity of calreticulin in dead or live cells. Gating strategy for both h and i are shown in Supplementary Fig. 7. j Representative images of each population. Merged regions are shown in yellow. Based on the corrected images, correlation values between calreticulin and the nucleus were calculated and are presented in panel (k). l LCI-S of GM-BMCs treated with HCQ (80 µM). IFN-β secretion (blue) and DNA staining (magenta) are shown. m Concentration of IFN-β (left) and LDH activity (right) in the culture supernatant of GM-BMCs treated with HCQ (100 µM) in the presence of the indicated concentrations of inhibitors. n LCI-S of GM-BMCs treated with HCQ (100 µM) in the presence or absence of Fasudil (100 µM). Representative image of endpoint (12 h) is shown in left. Quantitative analysis of three independent experiments are shown on the right with the number of IFN-β spots detected. Data in a and l are representative pictures from one experiment (experiment was repeated three times). Data in (h, i, k, and m) are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (each experiment was repeated two to three times). Statistical analysis was performed using two-sided Wilcoxon test (c), two-sided Welch’s t test (g). See Supplementary Fig. 3.

The above observations represent unique phenomena for obtaining nuclear components by IFN-I-inducing cells, which we named nucleocytosis. We further examined the molecules that can trigger the induction of nucleocytosis. As a candidate, we focused on calreticulin, which is known as an eat-me signal in the immunogenic cell death (ICD) of tumour cells^33,34. Notably, HCQ induces ICD in certain types of tumour cells with calreticulin translocation to the cell surface^33,34. Another reason for focusing on calreticulin is that it triggers dendrite extension via Rho/Rho-associated protein kinase (ROCK) signalling^35–38. Therefore, we focused on calreticulin and examined its localisation in HCQ-treated cells. Notably, calreticulin was detected in DNA-positive regions after HCQ treatment, indicating its translocation from the cytosol to the nucleus (Fig. 4d–g, Supplementary Fig. 4d, e and Supplementary Movie 8). To exclude the possibility that this observation was due to HCQ-induced morphological changes, we performed nuclear western blotting and confirmed calreticulin nuclear translocation (Supplementary Fig. 4f). These results indicate the role of calreticulin in the nucleus³⁴. To further examine the nuclear translocation of calreticulin, we performed imaging cytometry. Interestingly, although nuclear western blotting (Supplementary Fig. 4f) showed no marked changes in calreticulin protein levels after HCQ treatment, intracellular staining revealed increased signals, particularly in dead cells (Fig. 4h, i). Calreticulin gene expression was also examined, but no increase was observed with HCQ (Supplementary Fig. 4g). These findings indicate that HCQ induces structural or localisation changes in calreticulin that affect its antibody reactivity or accessibility. We further assessed the relationship between calreticulin and the nucleus, and found that the correlation was significantly stronger in cells with high calreticulin signals than in those with low levels (Fig. 4j, k). Of note, Baf A treatment also enhanced calreticulin signals, particularly in live cells (Fig. 4h, i). Taken together, these results suggest that lysosomal inhibition induces some changes in the structure or localisation of calreticulin, and that a fraction of the protein translocates into the nucleus. As extracellular calreticulin is important for the extension of dendrites^35–38, we next examined dendrite extension toward the nucleus using LCI-S. To examine dendrite extension, we focused on IFN-β-producing cells near, but not attached to, dying cells. Interestingly, we observed dendrite extension to dying cells and further interaction with the nucleus during IFN-β secretion (Fig. 4l and Supplementary Movie 9), which is consistent with microscopic analysis, as shown in Fig. 3d–h. Notably, the dendrite was still connected to the nucleus following IFN-β secretion, indicating a robust connection, which was supported by the nuclear accumulation of calreticulin in HCQ-treated cells (Fig. 4d–k). To further explore the importance of calreticulin in nucleocytosis, we investigated the involvement of Rho kinase, which plays a critical role in calreticulin-induced dendrite extension^35–38. The inhibition of Rho kinase resulted in the inhibition of IFN-β production, while cell death remained unaffected (Fig. 4m). This inhibition was further confirmed by LCI-S (Fig. 4n). Interestingly, however, the Rho kinase inhibitor did not block calreticulin nuclear translocation by HCQ or IFN-β production induced by cGAMP (Supplementary Fig. 4h–k). These results suggest that Rho kinase functions after calreticulin nuclear translocation but before cGAS-STING activation during nucleocytosis. Taken together with previous reports^35–38, our findings support a possible involvement of Rho kinase in dendrite extension toward calreticulin during nucleocytosis.

Lysosomal malfunction to IFN-β production

Our results showed that the inhibition of lysosomes was not sufficient for robust IFN-β production (Supplementary Fig. 1d, e, n), although Baf A induced a similar change in the calreticulin signal in living cells (Fig. 4h). These results indicate the existence of an unknown common target of IFN-inducing CADs that triggers cell death after calreticulin accumulation in the nucleus. As HCQ is a direct inhibitor of palmitoyl-protein thioesterase 1 (PPT1)³⁹, a lysosomal depalmitoylating enzyme, and CQ is a direct agonist of Mas-related G protein-coupled receptor A3 (MRGPA3)⁴⁰, a sensory neuron-specific receptor implicated in itch sensation, we examined the involvement of PPT1 and MRGPA3 in IFN-β production. The PPT1 inhibitor DC661, but not MRGPA3 agonists, induced IFN-β production, although the level was considerably lower than that induced by cGAMP (Fig. 5a and Supplementary Fig. 5a). Conversely, compensation of esterase activity of PPT1 by N-(tert-butyl) hydroxylamine³⁹ resulted in the inhibition of HCQ-induced IFN-β production (Fig. 5b). Similar inhibitory effects of N-(tert-butyl) hydroxylamine were observed for all IFN-β inducers (Supplementary Fig. 5b), suggesting a common involvement of PPT1 inhibition in their IFN-β production. To examine the relationship between PPT1 and IFN-β inducers, we performed in silico analysis. HCQ bound to the central pocket of PPT1 (Fig. 5c and Supplementary Fig. 5c), masking the binding region of the substrate⁴¹. Based on pharmacophore analysis (Fig. 1j, k), we examined the importance of the five conserved pharmacophore features in HCQ and identified the predicted interacting amino acids near all five conserved pharmacophore features (Fig. 5d). Consistently, the binding affinity of modified HCQ, which lacks each conserved pharmacophore feature but not the non-conserved pharmacophore features, decreased to a level similar to that of the substrate (Fig. 5e), indicating the importance of conserved structures in binding to PPT1. Consistently, all IFN-β inducers with conserved pharmacophore features bound to the central pocket of PPT1 and masked the binding region of the substrate, similar to the PPT1 inhibitors DC661 and Lys05 (Fig. 5f). Moreover, their binding affinity was stronger than that of the substrate (Fig. 5g). Conversely, non-IFN-β inducers did not mask the binding region of PPT1 to the substrate, and the affinity was lower than that of the substrate (Fig. 5f, g). Furthermore, we observed a strong correlation between binding affinity to PPT1 and the potential to induce IFN-β (Fig. 5h). These results suggest that PPT1 is the target of IFN-β inducers.

Fig. 5. Target protein of IFN-β-inducing CADs.

a IFN-β concentration in the culture supernatant of GM-BMCs treated with the indicated concentrations of DC661, cGAMP (5 µg/mL), or LPS (100 ng/mL) for 16 h. b IFN-β concentration in the culture supernatant of GM-BMCs treated with HCQ (100 µM) in the presence of indicated concentrations of N-(tert-Butyl) hydroxylamine (10 mM) for 16 h. c, d In silico analysis of the interaction between human PPT1 and compounds. Yellow circle: binding pocket of palmitate. An enlarged picture of the binding pocket with detailed information is shown in (c). e In silico analysis of the interaction between human PPT1 and HCQ lacking each sub-structure. Left: schematic of HCQ with conserved pharmacophore features. Red, conserved pharmacophore features. Right: binding energy of HCQ lacking each pharmacophore feature of human PPT1. f, g In silico analysis of the interaction between human PPT1 and compounds. Graphical images are shown in e. The binding energy of the indicated compound to human PPT1 is shown in f. Binding energy is calculated under near-neutral conditions (pH 7.4). h Correlation analysis between IFN-β-inducing activity and binding energy to human PPT1. i, j In silico analysis of the interaction between human PPT1 and compounds under acidic conditions. The binding energy of the indicated compound to is shown in h. Graphical images of the interaction are shown in i. Interactions between each compound and H₃O⁺ were also examined and shown in each right panel. k IFN-β concentration in the culture supernatant of WT or Cgas^-/- GM-BMCs treated with the indicated concentration of DC661 in combination with Baf A (15 nM) for 16 h. l LCI-S of GM-BMCs treated with DC661 (3 µM) in combination with Baf A (15 nM). IFN-β secretion (blue) and DNA staining (magenta) are shown. DNA intensity is summarised in the lower right. Data in (a, b and k) are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (each experiment was repeated three times). See also Supplementary Fig. 5.

We also examined the binding affinity of IFN-β inducers to PPT1 under acidic conditions similar to those in the lysosome. Notably, the binding affinity of IFN inducers, with the exception of DC661, decreased in acidic conditions (Fig. 5i). We further found that hydronium (H₃O⁺) bound strongly to the conserved HY3 structure of HCQ under acidic conditions (Fig. 5j). Similar results were obtained with CQ and ADQ, but not with DC661 (Fig. 5j). These results suggest that IFN-β inducers cannot inhibit lysosomal PPT1 by binding H₃O⁺ before the increase of lysosomal pH by their accumulation and inhibition of protons via their reactivity as CADs. Conversely, DC661 was expected to inhibit PPT1 before proton inhibition but could not induce robust IFN-β production (Fig. 5a), indicating the possible importance of proton inhibition in robust IFN-β production. We further examined the contribution of lysosomal proton inhibition to IFN-β production. Interestingly, DC661 treatment under proton inhibition by Baf A enhanced IFN-β production in a cGAS-dependent manner (Fig. 5k and Supplementary Fig. 5d). LCI-S observed interactions involving dying cells, a DNA signal in IFN-β-producing cells, and IFN-β secretion following DC661 and Baf A treatment (Fig. 5l and Supplementary Fig. 5e). These results suggest that PPT1 inhibition can trigger IFN-β production, but proton inhibition is required for a robust response via nucleocytosis (Fig. 5a, k, l). As shown in Fig. 4h, Baf A treatment led to relocalisation of calreticulin mainly in living cells but induced only weak cell death and failed to trigger IFN-β production (Supplementary Fig. 1n). By contrast, DC661 induced cell death but elicited only minimal IFN-β production (Fig. 5a). Taken together, these findings indicate that proton inhibition-induced calreticulin relocalization and PPT1 inhibition-induced cell death both contribute to robust IFN-β induction.

In the case of ICD in tumours, PPT1 inhibition induces multiple types of cell death accompanied by surface calreticulin expression¹³. We further examined the cell death pathways involved in HCQ-mediated IFN-β production. HCQ-induced IFN-β production was observed even in cells deficient in Z-DNA binding protein 1 (Zbp1), mixed lineage kinase domain-like protein (Mlkl), or gasdermin D (Gsdmd) (Supplementary Fig. 5f–h), although the peak concentration was affected in Gsdmd-deficient cells, indicating a slight but not essential contribution of these molecules. We subsequently investigated the effect of various cell death inhibitors on HCQ-induced IFN-β production; these inhibitors did not inhibit IFN-β production (Supplementary Fig. 5i, j), suggesting that HCQ-mediated cell death leading to IFN-β production involves unknown pathways or a combination of multiple pathways.

Nucleocytosis in tumour cells

The above results suggest a previously unrecognised function of macrophages in inducing an IFN-I response via nucleocytosis. Consistently, we could not detect IFN-β production using various cell types, including splenocytes, peritoneal macrophages, and cell lines, indicating the unique mechanism of IFN-β production by specific cell subsets in GM-BMCs (Supplementary Fig. 6a). Conversely, GM-BMCs produced IFN-β after mixing with HCQ-treated dying tumour cells (Fig. 6a, b). Consistently, HCQ-induced nuclear translocation of calreticulin was observed in these cells (Fig. 6c and Supplementary Fig. 6b, c), indicating that nucleocytosis-inducing cell death is induced in these two types of tumour cells.

Fig. 6. IFN-β production by HCQ-treated tumour cells and in vivo effects of HCQ.

a Schematic of the experiment. b IFN-β concentration in the culture supernatant of GM-BMCs mixed with the indicated cell lines. Tumour cell lines were treated with the indicated concentrations of HCQ before being mixed with GM-BMCs. c Confocal microscopy of B16 cells treated with HCQ (250 µM) for the indicated periods. Left: intracellular calreticulin (green) and DNA staining (blue). Intensities of calreticulin and DNA staining were calculated at the site of the white dashed lines and shown in the upper left panel. The X-axis indicates the distance from the left edge of the white dashed lines. Schematic of the calculation is shown in the lower right. Data are presented as the mean ± SEM (0 h; n = 39, 6 h; n = 36). d Schematic of the in vivo experiment. e–i scRNA-seq analysis of lung cells from mice treated with HCQ. e UMAP plots showing cluster numbers and cell types (left two panels) annotated by the expression of key marker genes (Z-score) (right panel). f Violin plots of the sting1 gene in each cell clusters. g Dot plots showing enriched gene sets for each cell type in WT-HCQ compared with that in WT-water (left) or Sting^−/−-HCQ (right) in GSEA. NES, normalised enrichment score. The top 10 gene sets with high NES and the bottom 10 with low NES are shown. h, i Violin plots of indicated genes (h) and ISG signature (i) in each cell type. Data in b are expressed as the mean ± SD of n = 3 replicates of each in vitro culture condition in one representative experiment (each experiment was repeated three times). Data in (h and i), the box plot within the violin plot shows the median (centre line), and the interquartile range (IQR) (the box boundaries, spanning the 25th to 75th percentiles). The whiskers extend to data points within 1.5 times the IQR, defining the non-outlier range. Points outside the whiskers are outliers. Statistical differences in (c) was examined using Welch’s t test. Statistical differences in (h, i) were examined using the two-sided Wilcoxon rank sum test. *p < 0.05, **p < 0.01, ***p < 0.001, n.s., not significant. See also Supplementary Fig. 6.

HCQ-induced local IFN-I response in vivo

Finally, we examined the physiological effect of severe lysosomal inhibition by HCQ treatment. As previous reports have shown the in vivo effects of HCQ against viruses, including respiratory viruses^27,28, we examined the HCQ-induced IFN-I response in the lung by scRNA-seq analysis (Fig. 6d). First, we classified the cell types detected in the scRNA-seq analysis and found that the Sting1 gene was expressed in macrophage subsets, including monocyte-derived macrophages (MoMFs) and alveolar macrophages (Fig. 6e, f). Interestingly, when we analysed the gene expression patterns across the detected cell types, the IFN-I response emerged as one of the most prominent subsets among both the total gene set and the STING-dependent genes (Fig. 6g). Importantly, IFN-I-related genes were expressed in a STING-dependent manner in multiple cell types, particularly MoMFs, epithelial cells, and neutrophils, and IFN-stimulated gene signatures were also induced in these cells (Fig. 6h, i). These results indicate that HCQ triggers a STING-dependent IFN-I response in the lung. Based on the pattern of Sting1 expression, these findings suggest that MoMFs initiate STING-dependent IFN-I responses, which in turn activate the expression of secondary IFN-stimulated genes in other cell types (Fig. 6e–i and Supplementary Fig. 6e, f). This is consistent with our in vitro observations of nucleocytosis by macrophages within GM-BMCs (Fig. 2e, f), indicating that MoMFs are candidate inducers of nucleocytosis in vivo. Notably, systemic production of IFN-β was not observed (Supplementary Fig. 6d), indicating a local effect of HCQ on the lung. Considering the different magnitudes of the STING-dependent IFN-I responses across cell types (Fig. 6h, i and Supplementary Fig. 6e, f), such cell type-dependent responses may underlie the localised IFN-I response in the lung. In conclusion, these results demonstrate the physiological relevance of HCQ in inducing a STING-dependent local IFN-I response in vivo.

Discussion

Cytosolic cGAS, one of the most potent B-DNA sensors to activate the IFN-I response, is involved in various diseases^1–4; however, it is unclear whether and how cytosolic cGAS is directly activated by external self-DNA. DNA cannot pass through the cell membrane and needs to be delivered into the cytosol by certain mechanisms as in the case of a virus or exosome^1–7. As the mechanism underlying the direct delivery of external self-DNA to the cytoplasmic cGAS is unknown, previous studies have focused mainly on self-DNA originating from inside the cells, such as from the mitochondria and nucleus, under cellular stress or DNase dysfunction^1–4. Moreover, many studies have shown an IFN-I signature that is mediated by self-DNA-cGAS-mediated signalling, yet the actual production of IFN-I itself is not evident for reasons that remain unclear^1,2. Therefore, our study provides insights into cGAS-mediated diseases by showing how nucleocytosis allows external DNA to enter the cytosol, inducing a local but robust cGAS-IFN-I response at the site of cell death (Fig. 7). We also found that CAD antivirals activate cGAS-IFN-I. Among the antiviral agents we investigated, HCQ stands out as one of the most well-studied drugs^16–18,22. Although various cellular effects have been observed in in vitro analyses, it is unclear how HCQ exerts its antiviral effects in vivo. Considering that HCQ and some IFN inducers are used to suppress the immune response, our findings are surprising and provide some important aspects as discussed below. Notably, this is the first study to visualise IFN-β secretion from live cells by LCI-S. This technique enables us to identify IFN-β secretion from a limited number of cells and discover unique interactions between living and dying cells.

Fig. 7. Graphical summary.

The findings of this study are summarised in a graphical representation.

An important previous discovery relevant to the discussion of the uniqueness of nucleocytosis is the developmental abnormality observed in DNase II-deficient mice^11,12. In the absence of DNase II, self-DNA derived from apoptotic cells is engulfed by macrophages but fails to undergo degradation, resulting in its abnormal accumulation within endosomes. Consequently, this leads to exposure of DNA to the cytosol and activation of the cGAS-STING pathway. These findings clearly demonstrate the critical role of extracellular DNA degradation within endosomes and provide a rational mechanism by which dead cells can be processed without triggering inflammation. In this sense, when considering how normal cells with DNases activate an immune response via the cGAS-STING, another mechanism that bypasses the endosomal route is necessary. In this context, the discovery of nucleocytosis further advances our understanding of self-nucleic acid-mediated activation of the cGAS-STING pathway, and future studies are expected to elucidate its involvement in diverse immune responses and diseases.

As shown in Figs. 3d–h, nucleocytosis is markedly different from phagocytosis both morphologically and mechanistically. In conventional phagocytosis¹⁴, cells or cellular fragments are engulfed and internalised through endocytosis, after which their contents are digested. By contrast, in nucleocytosis, protrusion-like structures were observed (Fig. 3d–h). Consistent with this, inhibitors that suppress protrusion formation or extension also reduced IFN production (Figs. 3i, 4m). These observations indicate that despite clear morphological differences between processes, the fundamental commonality is a shared requirement for cellular shape changes. The exact mechanism by which dead cells are recognised before nucleocytosis remains to be elucidated. Future studies are needed to determine whether the same eat-me signals utilised during general phagocytosis are also employed in this specific recognition process. Moreover, regarding the mechanism by which protrusion-like structures incorporate nucleic acids, a potential relationship with tunnelling nanotubes (TNTs) represents an interesting issue for future investigation, because TNTs enable the transfer of cytosolic molecules, including nucleic acids, to the cytosol of other cells, and the involvement of Rho kinase in TNT formation has been reported⁴². Further analyses of TNT between ‘the cytosol and the nucleus’ present an intriguing avenue for future research. Furthermore, how nucleic acids are extracted from the nucleus and in what state they exist remains to be elucidated. In Figs. 3d, 4a, DNA appears to diffuse into the cytosol of IFN-β-producing cells, consistent with activation of the cytoplasmic cGAS-STING pathway. Although it is currently difficult to determine the precise state of these nucleic acids, the most important point of our findings is that their translocation has been confirmed. Further high-resolution analyses, such as electron microscopy, are required to clarify this issue.

Although further analyses are required on the hitherto unknown phenomenon of nucleocytosis as a function of macrophages, our study revealed a unique mechanism of immune activation in the local site of cell death. This phenomenon can be a mechanism to maintain concentrated IFN-β in a local area without systemic adverse responses. Notably, IFN-β-producing cells also died after secretion. Consistently, the IFN-I response, but not IFN-β-positive cells, was detected in the lungs of HCQ-treated mice, indicating a transient response by HCQ. This point may also be important to explain the safety of short-term HCQ treatment and the side effects of prolonged HCQ treatment. We also speculate that a similar local robust response via nucleocytosis may be involved in various diseases with an IFN signature, where systemic IFN-I production is difficult to detect. Consistently, lysosomal dysfunction and calreticulin have been reported in relation to diseases, suggesting their deep relationship^43–45. Furthermore, the local IFN-I response provides the answer to the previously mysterious antiviral effects of HCQ in vivo. The local effects of HCQ could result in virus-type-specific inhibition in a local area, dependent on the target cells of the virus. Our speculation about the effect of HCQ on COVID-19 is that HCQ may induce an IFN-I response in a local area; however, the response in that area may not be sufficient to inhibit SARS-CoV-2. Notably, the intracellular accumulation of HCQ is known to be associated with its efficacy^19,22,23,46. Previous studies have reported that in vivo blood levels of HCQ decrease rapidly within 24 h, consistent with its intracellular accumulation^47–49. Based on these reports, we speculate that HCQ dosage control in the local area is challenging, which may be another reason why HCQ did not demonstrate a clinical therapeutic effect against SARS-CoV-2.

Another important issue to be discussed is the mechanism of HCQ as a drug for autoimmune diseases. Considering that HCQ is used to inhibit the immune response via the inhibition of endosomal TLRs^22,50, our findings may appear contradictory. However, there is a notable difference between TLR inhibition and cGAS activation by HCQ; one is systemic inhibition, whereas the other is local activation without a systemic response. This site difference may explain why HCQ acts as a suppressor of IFN-I response in systemic autoimmune diseases. On the other hand, our findings can explain the side effects of HCQ and other agents. For example, HCQ retinopathy is one of the well-known side effects induced by long-term treatment with HCQ⁵¹. One possible cause is HCQ accumulation in retinal pigment epithelium (RPE) cells, resulting in RPE dysfunction⁴⁶. This accumulation may activate a local response via cGAS, which has the potential to induce several harmful effects^1,2. Additionally, in the case of cardiotoxicity, which is a known adverse effect of CQ/HCQ⁵², Rho kinase inhibition decreases CQ-related cardiotoxicity⁵³, which is consistent with our findings. Conversely, HCQ suppresses tumours and has the potential to enhance the efficacy of anti-cancer drugs⁵⁴, indicating the possible involvement of nucleocytosis in the activation of anti-tumour immunity via cGAS-IFN-I. According to our results, DC661 treatment in combination with a lysosome inhibitor, such as Baf A, may activate nucleocytosis against tumours, thereby activating anti-tumour immunity. Of note, as previously reported, lysosomal proton inhibition in IFN-β-producing cells by Baf A may also block the negative feedback regulation of STING activity through STING degradation⁵⁵, thereby contributing to the enhanced STING response observed by Baf A treatment under PPT1 inhibition. Considering all of the above issues, in silico analysis provides important information for the further development of not only antiviral drugs but also anti-cancer drugs and adjuvants. Conversely, our findings provide insights into developing drugs for autoimmune diseases without immune activation.

PPT1, the molecular target of IFN-β inducers, is an intriguing molecule that has also attracted attention as a therapeutic target in cancer⁵⁶. PPT1 is a lysosome-resident PPT that primarily contributes to protein degradation by depalmitoylating lysosomal proteins⁵⁶. Inhibition of PPT1 activity is known to trigger various forms of cell death through the accumulation of undegraded proteins and activation of stress pathways⁵⁶. In our study, the binding affinity of antiviral CADs to the active pocket of PPT1 correlated with their ability to induce IFN-I (Fig. 5h). These findings suggest that cell death caused by inhibition of palmitoyl-protein hydrolase activity plays a critical role in driving nucleocytosis. Given that depalmitoylation also regulates protein localisation, a role for PPT1 in calreticulin relocalisation warrants further investigation.

In addition to lysosomal PPT1 inhibition, our results suggest the importance of calreticulin relocalisation induced by lysosomal proton inhibition in nucleocytosis (Fig. 5f–l). As shown in Fig. 4h and Supplementary Fig. 4f, calreticulin protein levels remained unchanged, yet fluorescence-activated cell sorting analysis detected increased signals. This finding suggests the possibility of structural changes in calreticulin, which is a future interesting issue to clarify. The localisation pattern of calreticulin is also of particular interest. In cancer cells, calreticulin relocalises to the cell surface upon treatment with certain chemotherapeutic agents, showing striking redistribution from the intracellular compartment to the plasma membrane⁵⁷. By contrast, in our case, calreticulin was dispersed throughout the cytoplasm, with granule formation and partial nuclear translocation observed, indicating clearly distinct intracellular dynamics (Fig. 4d). This observation is also consistent with the fact that entire dead cells were not phagocytosed. As noted above, it remains to be elucidated why calreticulin exhibits such intracellular localisation.

It is worth discussing how HCQ-treated cells attach to dying cells. While various studies have demonstrated the importance of the eat-me signal on dying cells upon ICD, an interesting hypothesis was proposed by Herbst et al.⁵⁸, who proposed cytocytosis as a function of DCs to patrol inside the cell for detecting abnormalities. Their picture of cytocytosis may be similar to that of the GM-BMC during nucleocytosis. Related to this issue, clarifying the specific macrophage cell type capable of inducing nucleocytosis represents an intriguing future direction. As shown in Fig. 2f, Ifnb1 expression was observed in the macrophage population, whereas peritoneal macrophages and splenocytes did not produce IFN-β upon HCQ treatment (Supplementary Fig. 6a). These results suggest the possible existence of a specialised IFN-I-producing macrophage subset acting through nucleocytosis. Identification of this distinct IFN-I-producing macrophage subset would be highly intriguing and remains an important subject for future investigation.

This study had some limitations. First, we did not investigate the detailed molecular mechanisms by which nucleocytosis is regulated. Second, the key cell types that induce nucleocytosis in vivo and the significance of nucleocytosis in cGAS-mediated diseases require further clarification. Nonetheless, our study highlights the previously unknown self-DNA-cGAS-dependent IFN-I response mechanism, characterised by the newly identified process of nucleocytosis. Our findings provide important insights for drug development and understanding various diseases linked to the self-DNA-cGAS pathway, potentially clarifying the previously mysterious self-DNA-cGAS-IFN-I axis.

Method

Reagents

HCQ was purchased from Tokyo Chemical Industry (Tokyo, Japan). DNA primers were obtained from Fasmac (Kanagawa, Japan). Details of the antiviral agents and cell death inhibitors used in this study are summarised in the Supplementary Tables 1 and 2.

Mice

C57BL/6 mice were purchased from CLEA Japan (C57BL/6Jcl; Tokyo, Japan). Sting1^-/- mice were generated as previously described⁵⁹. Tnf^-/-/Tbk1^-/- and Zbp1^-/- mice were established as previously described⁶⁰. Ips1^-/- and Trif^-/- mice were kindly provided by Dr. Shizuo Akira^61,62. Mlkl^-/- mice were kindly provided by Dr. Manolis Pasparakis⁶³. Gsdmd^-/- mice were provided by RIKEN BioResource Research Center (RBRC10761; Tsukuba, Ibaraki, Japan) through the National BioResource Project of the MEXT, Japan⁶⁴. Mice (male and female, all C57BL/6 background) were used at 7–16 weeks of age, unless otherwise specified. Mice were maintained under specific pathogen-free conditions with a 12 h/12 h light/dark cycle, with a controlled room temperature of 22 ± 2 °C and relative humidity of 55 ± 5 %. CO₂ gas was used for euthanasia. All animal experiments were conducted in accordance with the guidelines of the University of Tokyo (Tokyo, Japan). Permission was granted to perform animal experiments by the Animal Experiment Committee at The University of Tokyo (Tokyo, Japan). Cgas^-/- mice were generated in the Research Institute for Microbial Diseases (Osaka University; Osaka, Japan) by following method⁶⁵. The sgRNA sequence 5′-gaaagctgcggcccgcaaag-3′, or 5′-gggggctcgatcgcggcggg-3′ targeting exon 1 of the mouse cGAS gene, was cloned into the BbsI site of the pX330_hSpCas9 vector (Addgene, Watertown, MA, USA). C57BL/6 N female mice (6 weeks old) were superovulated and mated with C57BL/6 N males; fertilised one-cell-stage embryos were collected from the oviducts and injected into the pronucleus or cytoplasm with the sgRNA-expressing pX330_hSpCas9 plasmids. Then the injected embryos were transferred into the oviducts of pseudopregnant ICR females at 0.5 days post coitum. F0 founder males carrying the mutation were crossed with C57BL/6 N females, and germline transmission was confirmed by sequencing. For sequence, genomic DNA was extracted from tail or ear tissue using the NucleoSpin Tissue kit (MACHEREY-NAGEL GmbH & Co. KG, Düren, Germany) according to the manufacturer’s instructions, and the region surrounding the sgRNA target site in the cGAS locus was amplified by PCR using forward primer 5′-AAGCTAGCCACGCGTGCTCCTGCGCCTGCTCGCGGCGG-3′ and reverse primer 5′-AAGTCGACACAACTTTATTCACCGTCTCGGCCGCCTCC-3′. PCR products were purified with the FastGene Gel/PCR Extraction Kit (Nippon Genetics Co., Ltd., Tokyo, Japan) and sequenced by FASMAC Co., Ltd. (Kanagawa, Japan). The primer for sequencing was 5′-CACGCGTGCTCCTGCGCCTGCTCGC-3′, and chromatograms from wild-type and cGAS^−/− alleles were aligned to confirm disruption of exon 1. Related information is summarised in Supplementary Fig. 9.

Cell culture

GM-BMCs were prepared according to a standard method⁶⁶. Briefly, BMCs were isolated from the legs of mice and cultured in RPMI 1640 (Nacalai Tesque) containing 10% foetal bovine serum (FBS; Sigma-Aldrich, St. Louis, MO, USA) and 10 ng/mL GM-CSF (PeproTech, Cranbury, NJ, USA). Red blood cells (RBCs) were removed with RBC lysis buffer (Thermo Fisher Scientific, Waltham, MA USA) prior to culture. After 4 days, additional GM-CSF was added (10 ng/mL) and used for the experiment on days 6–7. Splenocytes were also prepared using a standard method. Spleens were isolated from mice and mashed through a cell strainer (Corning, Corning, NY, USA) and then used for experiments after the removal of RBCs using RBC lysis buffer (Thermo Fisher Scientific). Peritoneal macrophages were isolated by peritoneal gavage with phosphate-buffered saline (PBS) (Nacalai Tesque) containing 2% FBS. B16 cells and NIH3T3 cells were obtained from RIKEN BioResource Research Center.

RNA analysis

Total RNA was extracted using TRIsure (Nippon Genetics, Tokyo, Japan) with the Direct-zol RNA Kit (Zymo Research, Irvine, CA, USA) or NucleoSpin RNA (Takara Bio, Shiga, Japan). RNA-seq was performed by ImmunoGeneTeqs, Inc. (Chiba, Japan).

Cytokine analysis

Mouse IFN-β concentrations in the culture supernatant or plasma were measured using an enzyme-linked immunosorbent assay (ELISA) (R&D Systems, Minneapolis, MN, USA).

Immunoblot analysis

Whole cell lysates were prepared in M-PER Mammalian Protein Extraction Reagent (Thermo Fisher Scientific) containing a protease inhibitor cocktail (Sigma-Aldrich). Proteins were resolved on the NuPAGE 4–12% Bis-Tris Gel (Thermo Fisher Scientific), and electrotransferred to nitrocellulose membranes included in the iBlot Transfer Stack using the iBlot Dry Blotting System (Thermo Fisher Scientific). The membranes were blocked in 5% bovine serum albumin (BSA) in PBS with 0.1% Tween 20 and then incubated with primary antibodies against GAPDH (14C10; Cell Signalling Technology [CST], Danvers, MA, USA), IRF-3 (D83B9; CST), phosphorylated IRF-3 (p-IRF-3, Ser396) (4D4G; CST), TBK1/NAK (CST), p-TBK1/NAK (Ser172) (D52C2; CST), p44/42 MAPK (ERK1/2) (137F5; CST), p-p44/42 MAPK (ERK1/2, Thr202/Tyr204) (197G2; CST), stress activated protein kinase (SAPK)/c-Jun N-terminal kinase (JNK) (CST), and p-SAPK/JNK (Thr183/Tyr185) (G9; CST). After washing, the membranes were incubated with anti-rabbit horseradish peroxidase (HRP)-conjugated secondary antibody (CST) or anti-mouse IgG HRP-conjugated secondary antibody (CST), and detected by HRP chemiluminescence. Nuclear and cytoplasmic proteins were collected using NE-PER extraction reagents (Thermo Fisher Scientific) according to the manufacturer’s instructions. Protein concentrations were determined using the Bradford assay, and equal amounts of protein from each fraction were separated by SDS-PAGE and analysed by western blotting using an anti-calreticulin antibody (1:1000, ab92516; Abcam, Cambridge, UK) and an anti-histone H3 antibody (1:1000, #9715; CST).

Principal component analysis of compounds and machine learning

Principal component analysis (PCA) was used to obtain a panoramic view of the structural features of the active compounds. Extended-connectivity fingerprints with a diameter of 4 (ECFP4) was calculated and used as the feature vector for the compounds. ECFP4 has a length of 2,048 bits. RDKit (version 2023.03.2) was used to calculate ECFP4, and scikit-learn was used for the PCA.

Derivation of the pharmacophore model

The main goal of 3D ligand-based pharmacophore modelling is to construct a common 3D pharmacophore for active compounds. To construct the pharmacophore model, six active compounds were used. The “idbgen” module of LigandScout 4.4^67,68 was used for conformer generation of the active compounds. The parameters for conformer generation were set to the “icon-best” option. Pharmacophore models were constructed from six active compounds using the ligand-based pharmacophore creation mode of LigandScout 4.4 with default parameters. After the pharmacophore creation, 10 pharmacophore models were proposed, and the pharmacophore model with the best score was selected.

Pharmacophore screening

Using the pharmacophore model derived from six active compounds, pharmacophore screening was performed to assure that only active compounds were selected from a compound library consisting of 44 compounds. In the pharmacophore screening process, we utilised the “iscreen” module within LigandScout 4.4. The scoring function chosen was “Relative Pharmacophore-Fit,” the conformation match mode was set to “BEST,” and the check for exclusion volume clashes was set to “disabled.” The Relative Pharmacophore-Fit score was obtained from the number of matching pharmacophore features and the root mean square deviation of the pharmacophore alignment, normalised by [0,1]^67,68. Compounds that did not match the pharmacophore were assigned a score of zero.

Immunofluorescence staining and imaging

Cells attached to glass coverslips for overnight culture were stimulated with HCQ (100 or 250 µM) for the indicated time periods, fixed in 100% MeOH (Nacalai Tesque) for 5 min, and permeabilised with 0.1% Triton X-100 (Nacalai Tesque) for 5 min. After washing twice with PBS (Nacalai Tesque), the cells were incubated in BlockAce (Snow Brand Milk Products, Tokyo, Japan) for 1 h and then incubated with an anti-calreticulin antibody (1:500, ab92516; Abcam) in PBS containing 2% BSA (Nacalai Tesque) for 6 h at 4 °C. After washing three times with PBS, a secondary antibody (anti-rabbit IgG, 1:2000, A-21246; Thermo Fisher Scientific) was applied and incubated in the dark for 1 h at room temperature. DNA staining was performed with DAPI (0.025 μg/mL; Sigma-Aldrich) during the staining with the secondary antibody. Finally, the cells were washed three times with PBS and mounted in Fluorsave (345789; Merck, Darmstadt, Germany). Confocal images were captured using the Nikon A1Rsi laser scanning microscopy system (Nikon, Tokyo, Japan). Fluorescence intensity profile analysis and 3D image reconstruction were performed using Nikon NIS-Elements software. Fluorescence images were also obtained with the BZ-X800 all-in-one fluorescence microscope (KEYENCE, Osaka, Japan). The nuclear and cytoplasmic areas and the fluorescence intensity of calreticulin were automatically calculated using the Hybrid Cell Counting application BZ-H4C (KEYENCE) of the BZ-X Analyser software BZ-H4A (KEYENCE).

scRNA-seq

scRNA-seq was performed by KOTAI Biotechnologies (Osaka, Japan). Sequencing was performed in paired-end mode (read1: 28 bp; read2: 100 bp) using the DNBSEQ-G400 sequencer (MGI Tech, Shenzhen, China). The resulting scRNA-seq data were processed using the CellRanger pipeline (ver. 7.2.0) configured for 3’ v3 chemistry, and aligned to the refdata-gex-mm10-2020-A reference genome (10x Genomics, Pleasanton, CA, USA). Subsequent analysis was performed using the Seurat R package (ver. 5.0.2). Cells were filtered based on thresholds of 200–10,000 detected genes and a minimum unique molecular identifier count of 1000. Data were normalised by the NormalizeData function and integrated by the IntegrateData function for reference-based integration⁶⁹. Canonical correlation analysis was used as the reduction method to identify anchors using the FindIntegrationAnchors function. Cluster identification of cells was performed by a shared nearest neighbour modularity optimisation-based algorithm using the FindNeighbors and FindClusters functions. Cell types for each cluster were annotated based on differentially expressed genes identified by the FindAllMarkers function and known marker genes. Uniform Manifold Approximation and Projection (UMAP) and violin plots were generated by the DimPlot and VlnPlot functions, respectively. Gene expression and density on the UMAP plots were visualised using the FeaturePlot_scCustom and Plot_Density_Custom functions, respectively, from the scCustomize package (ver. 2.0.1). IFN-stimulated gene signature score was calculated using the AddModuleScore_UCell function from the UCell package (ver. 2.6.2) with IFN-induced protein with tetratricopeptide repeats 1 (Ifit1), Ifit3, 2’−5’-oligoadenylate synthetase 1, Irf7, and Zbp1 genes as input. Gene set enrichment analysis (GSEA) was conducted with the fgsea package (ver. 1.28.0) and mouse MSigDB hallmark gene sets. The results were visualised using the geom_point function from the ggplot2 package (ver. 3.4.4). Statistical comparisons between groups of cells were conducted using the Wilcoxon rank sum test.

Flow cytometry

For the nuclear staining of dying cells, the cells were incubated with SYTOX green (0.5 µM; Thermo Fisher Scientific) for 10 min. After washing, the cells were analysed using the BD FACSDiscover S8 Cell Sorter (Becton, Dickinson and Company [BD], Franklin Lakes, NJ, USA). To analyse cell surface calreticulin, the cells were stained with AlexaFluor^® 405 calreticulin antibody (Abcam) after blocking the Fc receptor with anti-CD16/32 antibody (BioLegend, San Diego, CA, USA). Then the cells were incubated with SYTOX green (0.5 µM; Thermo Fisher Scientific) for 10 min and measured using the BD LSRFortessa flow cytometer (BD). For intracellular staining of calreticulin, the cells were stained with LIVE/DEAD^® Fixable Near-IR stain (Thermo Fisher Scientific). After fixation and permeabilisation with Foxp3/Transcription Factor Staining Buffer (Thermo Fisher Scientific), non-specific staining was blocked with mouse serum. Then the cells were stained with the AlexaFluor488 anti-calreticulin antibody (Abcam). After washing, the cells were stained with DRAQ5 (5 µM; Thermo Fisher Scientific) for 10 min and analysed using the BD FACSDiscover S8 Cell Sorter.

Analysis of images by flow cytometry

Cell regions were cropped from grayscale images and underwent brightness correction using Contrast Limited Adaptive Histogram Equalisation. Noise was then reduced by Gaussian filtering, and the image was binarised using Otsu’s method. Holes in the resulting binary image were filled to define cellular regions. The Watershed algorithm was utilised to segment these cells. Within each segmented cell region, the number of DNA pixels was quantified. The region with the higher DNA pixel count was selected for further analyses, provided that the original cell region was a singular entity capable of being subdivided into two distinct segments. For the selected cell, the boundary delineated using the Watershed segmentation was dilated twice, effectively expanding it by two pixels. An AND operation was then performed between this expanded boundary and the mask of the analysed cell. This procedure identified the adjacent contact area of the cells. Finally, we measured the number of pixels in this adjacent contact surface area, as well as the count of DNA pixels within it. DNA ratios are calculated as the ratio of the number of DNA pixels to the number of pixels in the adjacent contact area.

Concurrent detection of IFN-β secretion and nuclear staining using LCI-S

Imaging of IFN-β secretion using LCI-S was performed as previously described with some modifications^30,31. Briefly, the time-resolved measurement was performed with a completely automated inverted microscope ECLIPSE Ti2-E (Nikon) equipped with a high numerical aperture (NA) objective lens (CFI Apo TIRF 60× Oil, NA  =  1.49; Nikon), the INUBG2TF-WSKM stage-top incubator (Tokai Hit, Shizuoka, Japan), and the ORCA-flash4.0 V3 digital CMOS camera (Hamamatsu Photonics K.K., Shizuoka, Japan). A light-emitting diode (D-LEDI; Nikon) and laser diode (640 nm, LDI-7; 89 North, Williston, VT, USA) were used as light sources. The following sets of excitation and emission filters and a dichroic mirror were used: for epi-illumination of SYTOX Blue, a fluorescence filter cube for large field of view (FOV) imaging (CFP C-FLL-C) from Nikon was used; for total internal reflection fluorescence (TIRF) illumination of LCI-S of IFN-β, the ZET405/470/555/640x excitation filter (Chroma Technology, Bellows Falls, VT, USA), ZT561dcrb-UF2 dichroic beamsplitter (Chroma Technology), and ZT405/470/555/640rpc-UF1 multiband dichroic beamsplitter (Chroma Technology) were used. Cells were suspended in medium at a density of 2.5 × 10⁶ cells/mL and added to each chamber (5 mm diameter) of a 4-condition LCI-S chip (LCI-SPQ002; Live Cell Diagnosis, Saitama, Japan) at 50 μL, where the anti-IFN-β antibody (capture antibody from the Mouse IFN-beta DuoSet ELISA, DY8234-05; R&D Systems) was immobilised. Immediately before observation, the culture supernatant was replaced with freshly prepared culture medium containing 0.1 µM SYTOX Blue and 30 nM biotinylated antibody against IFN-β (detection antibody from the Mouse IFN-beta DuoSet ELISA, DY8234-05; R&D Systems) coupled with CF660R-labelled streptavidin (29040; Biotium, Fremont, CA, USA). Stimuli (80 or 100 µM HCQ and 5 µg/mL cGAMP) were replaced immediately before observation. Mineral oil (M5310; Merck, Darmstadt, Germany) was layered on top of the medium to prevent evaporation. A total of 169 fields of view were scanned in each chamber repetitively to continuously detect extracellular release of IFN-β using LCI-S, and dead cell nuclei were stained with SYTOX Blue.

LCI-S imaging analysis

Microscopic data obtained underwent image analysis using NIS-Elements 5.4 (Nikon). To visualise the IFN-β secretion activity at a moment in each frame, we determined the incremental fluorescence intensity at every time point using a specific image analysis procedure. First, we corrected for the small pixel drift occurring during each scanning cycle. Subsequently, noise reduction was applied, followed by a rolling minimisation over six frames to eliminate fluorescence spots from intracellular vesicles formed by pinocytosis of the fluorescent detection antibody. To obtain consecutive images, the fluorescence intensity from the previous time point was subtracted to derive the fluorescence intensity increment for each time point. To track changes over time in the number of IFN-β-secreting cells and damaged cells, we developed an artificial intelligence model for cell recognition. For this purpose, images of total and damaged cells in 10% and 1% of the total field, respectively, were manually identified in brightfield and SYTOX Blue-stained images, and the model was trained using NIS.ai on NIS-Element. The time-dependent increase in cell damage was assessed as the proportion of the number of damaged cells at each time point to the total number of cells at the start of observation in each FOV. Moreover, the frequency and intensity of IFN-β secretion were evaluated by the time of appearance and maximum brightness of each spot in the IFN-β secretion activity signal (incremental images) stitched over the entire area of observation. Each spot was identified by spot detection and tracking methods using NIS-Elements. To evaluate the effects of inhibitors, imaging was performed using the same inverted microscope setup equipped with a wide-field LCI-S module (wfLCI-S illuminator; Live Cell Diagnosis, Saitama, Japan) and an LCI-S chip (model LCI-SWFP6002; Live Cell Diagnosis) in combination with a 10 × objective lens (CFI Plan Apochromat Lambda D 10×; Nikon). Differential interference contrast optics were used for transmitted light observations.

In silico analysis

In silico analysis was conducted based on the quantum molecular dynamics method^70,71 using DMol3^72,73 in Materials Studio made by Dassault Systemes and parallel computers with 100 TFLOPS. The simulations were performed based on the density functional theory⁷², and the functional used in this study is a hybrid functional called the B3LYP⁷⁴ functional, which stands for “Becke, 3-parameter, Lee–Yang–Parr.” Instead of using pseudopotentials as core treatment, all-electron calculations were conducted, and double numerical plus d-functions⁷⁵ were used as the atomic basis sets. The time step of the dynamics simulations was set at 0.1 fs. By putting a compound molecule at an initial position on the PPT1 surface, the molecule was observed to move to an energetically stable position as a result of molecular dynamics simulation. By randomly changing the initial position on the PPT1 surface 100 times, the stable position was confirmed to always be the same and independent of the initial position. Thus, the stable state of the system where the compound molecule adheres to PPT1 was obtained, and the energy of this state, E_cP, was calculated. The binding energy of the compound to human PPT1, E_bin, was defined as E_bin = |E_c + E_P – E_cP | , where E_c and E_P are the energies of the isolated compound and isolated PPT1, respectively^74,76,77. The binding energy is often called the adhesion energy.

Microscopic analysis

Cells were suspended in RPMI 1640 without phenol red (Nacalai Tesque) containing 10% FBS (Sigma-Aldrich) and seeded into each well (2.25 or 2.0 × 10⁵ cells/well) of a 96-well glass-bottom plate (Matsunami Glass Ind., Ltd., Osaka, Japan). Before observations, HCQ and SYTOX Green were added to a final concentration of 80 or 100 µM, and 0.1 µM, respectively. After 1 h, microscopic analysis was performed with the BZ-X1000 fluorescence microscope (KEYENCE). Images were obtained every 600 s for 12 h using a 60 x objective lens (Plan Apochromat 60x Oil / N.A.1.40 / WD 0.13). Statistical analyses were carried out with the BZ-X1000 Analyser application (KEYENCE)

Holographic analysis

Cells were suspended in RPMI 1640 without phenol red (Nacalai Tesque) containing 10% FBS (Sigma-Aldrich) and seeded into each well (2.0 × 10⁵ cells/well) of a 96-well plate that was coated with polylysine (Sigma-Aldrich). Before observations, HCQ and SYTOX Red (Thermo Fisher Scientific) were added to a final concentration of 80 or 100 µM, and 0.167 µM, respectively. Holotomographic imaging was performed using the 3D Cell Explorer microscope (Nanolive SA, Tolochenaz, Switzerland) according to the standard protocol provided by the manufacturer. After 1 h, images were obtained every 10 min for 12 h with an acquisition area of XY = 217 µm × 217 µm with a Z-stack range of 30 µm for each position. Holotomographic images were obtained using EVE Version 2.2 operating software. Sytox images were also obtained every 10 min using the inbuilt fluorescence unit.

3D holotomography

Cells were suspended in RPMI 1640 without phenol red (Nacalai Tesque) containing 10% FBS (Sigma-Aldrich) and seeded into each well (4.0 or 6.0 × 10⁶ cells/well) of a 6-well plate (IWAKI, Tokyo, Japan). Before observations, HCQ and SYTOX Green were added to a final concentration of 80 or 100 µM, and 0.01 µM, respectively. After 30 min, holotomography imaging was conducted with the HT-X1 microscope (Tomocube Inc., Daejeon, South Korea). Images were obtained every 100 s for 12 h with an acquisition area of XY = 164.7 µm × 164.7 µm with a Z-stack range of 60 µm for each position. 3D HT images were obtained using the operating software (TomoStudio X, Tomocube Inc.). Sytox images were obtained every 4 h.

Statistical analyses

Data are expressed as the mean ± standard deviation or as described in each figure legend. Statistical analyses were performed using GraphPad Prism 8 (GraphPad Software, Boston, MA, USA). The results are shown in each figure or described in each figure legend.

Materials availability

The mouse lines generated in this study will be deposited to the RIKEN BioResource Research Center after acceptance for publication. Any additional data will be provided by the corresponding author upon request.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

Conceptualisation, H.N., Y.W., Y.S., and K.I.; Methodology, H.N., Y.W., Y.S., T.H., Y.K., T.Iw., M.K., T.S., M.Y., A.Y., C.K., and K.K.; Investigation, H.N., Y.W., Y.S., T.H., Y.K., T.Iw., T.S., T.B., D.M., Y.S., T.K., Y.M., S.L.I., B.T., A.Y., C.K., T.T., K.K., T.I., N.T-S., M.T., C.C., and K.I.; Visualisation, H.N., Y.W., Y.S., T.H., Y.K., and K.D.; Writing – Original Draft, H.N.; Writing – Review & Editing, H.N., T.H., N.T-S., C.C., and K.I.; Resources, K.D., M.O., A.I., Y.N., and E.K.; Funding Acquisition, H.N., K.K., and K.I.; Supervision, H.N., Y.S., TT., T.I., N.T.-S., M.T., C.C., and K.I.; Project Administration, H.N., and K.I.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The RNA sequencing data have been deposited in the NCBI Gene Expression Omnibus (GEO) under accession numbers GSE311546, GSE311755, and GSE312134. All data are included in the Supplementary Information or available from the authors, as are unique reagents used in this article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided in this paper.

Code availability

All original codes will also be available after publication.

Competing interests

Tomio Iwasaki is an employee of Hitachi, Ltd. Mai Yamagishi is a founder of Live Cell Diagnosis, Ltd. Atsushi Yoshimori is the CEO of Institute for Theoretical Medicine, Inc. Chisato Kanai is an employee of INTAGE Healthcare. The other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-68839-w.