Trex1^-/- cells feature tonic DNA damage checkpoint signaling (36), increased numbers of strand breaks and spontaneous phosphorylation of p53 (35). In AGS caused by defects of RNase H2 (37), extensive genome damage results in micronucleus formation which drives the chronic IFN response of these patients (34). This response is counter-regulated by p53-dependent DNA damage responses which eliminate cells with high damage load and prevent them from prolonged IFN production (38). Micronuclear DNA or chromatin bridges were shown to trigger an IFN response in cancer cells treated with genotoxic drugs (39, 40), and TREX1 was reported to limit cGAS activation in genetically unstable cells by degradation of micronuclear DNA (41). Micronucleated cells can be quantified by flow cytometric analysis of erythrocytes, as erythrocyte precursors expel the main nucleus but frequently retain micronuclei (42). We found increased numbers of micronucleated erythrocytes in Trex1^-/- mice ( Figure 1A and Figure S1 ), reflecting spontaneous DNA damage and chromosomal aberrations. However, compared to a 30-fold increase in mice lacking RNase H2 in the hematopoietic system (manuscript in preparation), micronucleated cells were only mildly (2-fold) increased in Trex1^-/- mice. To address whether cells carrying genome damage are an important source of type I IFN in Trex1^-/- mice, we crossed Trex1^-/- to Trp53^-/- animals (43) and compared the IFN response of double knock out mice to that of Trex1^-/- littermates. Additional loss of p53 further increased fractions of micronucleated erythrocytes ( Figure 1A ), but did not increase transcript levels of type I IFN-stimulated genes (ISGs) in total blood cells ( Figure 1B ) nor enhance inflammation of the myocardium in TREX1-deficient mice ( Figure 1C ). These data strongly argue against genome damage and cytosolic chromatin as the trigger of cGAS activation in the absence of TREX1 and lead us to investigate alternative sources of TREX1-sensitive cGAS activating DNA ligands.

Primary TREX1-deficient fibroblasts from AGS patients not exposed to experimental genotoxic hazards (35, 36) showed DNA staining in the cytosol, suggesting leakage of DNA from the nucleus or mitochondria in healthy cells. Higher amounts of DNA seem to leak into the cytosol of genetically unstable cancer cells, in particular after exposure to genotoxic hazards (44–48). This DNA, likely a waste from DNA repair (44), seems to overwhelm TREX1 exonuclease activity (44, 48). Whether the same mechanisms generating large amounts of TREX1 substrates in genome-damaged cells also contribute to release of DNA from the nucleus in healthy cells is unknown. Several DNA repair pathways involve release of oligonucleotides that could, upon hybridization, form partially double-stranded structures, e.g. Y-DNA (29), capable of ligating cGAS. Nucleotide excision repair (NER) removes bulky DNA lesions by excision of short single-stranded patches containing the lesions (49). High amounts of these, 22-30 nt-long flaps may be released also in the absence of exogenous genotoxic stress, since NER is triggered already by minimal distortions of undamaged DNA resulting in low-level DNA turnover in healthy cells termed ‘gratuitous repair’ (50, 51). Moreover, we had earlier found that in contrast to cell-intrinsic activation of transcriptional responses to loss of TREX1 in several different cell types (25), this study and unpublished), inactivation of TREX1 in B cells had no detectable effect on the transcriptome of murine B cells (25). Since B cells were reported to be NER-deficient (52), our observation would be plausibly explained by NER being a source of the cGAS ligand accumulating in Trex1^-/- mice. We therefore generated mice double-deficient for TREX1 and XPA, an essential NER protein (49, 53), using CRISPR/Cas9-mediated mutagenesis in embryos ( Figure S2A ). As expected, Xpa^-/- animals ( Figure 2A ) showed a loss of repair synthesis in UV-irradiated cells ( Figure S2B ), indicating absence of NER function. We compared ISG expression in blood cells of mice lacking TREX1 or XPA and of Trex1^-/-Xpa^-/- animals ( Figure 2B ). Xpa^-/- mice proficient for TREX1 did not show enhanced ISG expression compared to control mice. IFN signaling was not reduced in the double mutants compared to Trex1^-/- mice. Moreover, the myocarditis typical of TREX1-deficient mice was not ameliorated by additional inactivation of NER ( Figure 2C ).

MUS81 is a DNA repair factor reported to be a source of cytoplasmic DNA in cancer cells (54). This structure-specific endonuclease is essential for genome integrity as it is required for resolution of Holiday junctions (55) and prevention of chromosome bridges in anaphase (56), as well as repair of stalled and collapsed replication forks (57). MUS81 function has important impact on the activity of the type I IFN system [reviewed in (58)]. Loss of the enzyme results in pronounced spontaneous IFN production (59). As MUS81 deficiency leads to chromosomal instability, a likely scenario is that micronuclear DNA is the key stimulator of this IFN response. Due to this spontaneous IFN response of MUS81-deficient cells, the question whether MUS81 contributes to production of the immunostimulatory cGAS ligand accumulating in otherwise undamaged cells in the absence of TREX1, cannot be directly addressed by inactivation of MUS81 in Trex1^-/- cells or mice. However, in case the IFN response of cells lacking TREX1 depended on DNA waste produced by MUS81, no synergistic IFN induction by TREX1 deficiency and MUS81 deficiency would be expected. We inactivated TREX1 and/or MUS81 in THP1 cells ( Figure 3A ). The robust spontaneous ISG expression of TREX1^-/- cells was rescued by reconstitution with functional TREX1 (Fig S3A&B). As expected, also MUS81^-/- cells featured prominent type I IFN signaling ( Figure 3B ). The double deficient cells, however, showed synergistically enhanced ISG expression ( Figure 3B ). This result argues that MUS81 is not required for generation of the cGAS ligand accumulating as a result of TREX1 deficiency. Collectively, our data rule out NER as a dominant source of the endogenous immunostimulatory DNA accumulating in mouse cells lacking TREX1 and argue against a significant contribution of the structure-specific endonuclease MUS81 to generation of this pathogenic cGAS ligand.

We next turned our attention to another repair factor that had been shown to contribute to release of TREX1-sensitive DNA into the cytosol of genome-damaged cells (44), exonuclease 1 (EXO1). Homology-directed repair of double-strand breaks requires long-range resection of the 5’-end in order to produce the long single-stranded 3’-overhangs that are the substrates for strand invasion (60). EXO1 is a multifunctional enzyme with 5’-3’ exonuclease activity, but also 5’-flap structure-specific endonuclease activity, involved in end processing in preparation for homology-directed repair as well as in telomere homeostasis, mismatch repair, and Okazaki fragment maturation during lagging strand synthesis (61). Upon exposure of cells to genotoxic irradiation or drugs inducing double-strand breaks, 5’-end resection, performed by EXO1 and other endonucleases, was shown to be a source of DNA waste triggering a type I IFN response (44). TREX1-mediated degradation limited the innate immune response (44). We therefore asked, whether end resection by the EXO1 endonuclease activity might yield DNA waste also during steady-state genome repair. In addition, the EXO1 5’-flap cleavage occurring during genome replication might also release immunostimulatory DNA waste. We crossed Exo1^-/- mice (62) to TREX1-deficient mice and compared littermates of different genotypes for weight gain and spontaneous type I IFN response. The conspicuous failure to thrive typical of Trex1^-/- mice was largely, albeit not completely, rescued by the additional inactivation of EXO1 ( Figures 4A, B ). Even heterozygosity for the Exo1 knock out allele resulted in a significant rescue of Trex1^-/- weight gain. Survival of Trex1^-/- mice was significantly improved by additional loss of EXO1 ( Figure 4C ). Correspondingly, type I IFN signaling was clearly reduced in Trex1^-/- mice lacking functional Exo1, as determined by quantification of ISG transcript levels in blood cells ( Figure 4D and Figure S4A ). For most ISGs, the increase in transcript levels seen in Trex1^-/- mice was blunted to about half in Trex1^-/-Exo1^-/- animals ( Figure 4D and Figure S4A ). Type I IFN bioactivity was not elevated above the background of control animals in serum of Trex1^-/- or Trex1^-/-Exo1^-/- mice ( Figure S4B ). A control experiment demonstrated that STING responses of EXO1-deficient cells are of normal intensity ( Figure S4C ). The reduced innate response was not reflected in amelioration of myocardial inflammation in the double vs. the Trex1 single mutants as assessed by histological analysis of inflammatory infiltration on H&E- ( Figure 4E ), anti-CD3- ( Figure 4F ) and van Gieson-stained sections (not shown). Collectively, we show that additional loss of EXO1 significantly ameliorates the spontaneous innate antiviral response of Trex1^-/- animals, suggesting a role for EXO1 endonuclease activity in generation of cGAS-activating ligands in TREX1-deficient cells.

To elucidate the contribution of EXO1 to generation of TREX1 substrates, we first established an in vitro model of induced loss of TREX1 expression. We differentiated macrophages from bone marrow (BMDMs, Figure S5A ) of Trex1^FL/FL R26^CreERT2 mice (25) and induced Cre activity in Trex1^FL/FL R26^CreERT2 BMDMs by 4-hydroxytamoxifen (4-OHT, Figure 5A ). This resulted in efficient deletion of the loxP-flanked Trex1 allele one day after the end of 4-OHT exposure, as determined by PCR ( Figure S5B ). Western blot analysis of cells harvested at daily intervals after 4-OHT induction ( Figure 5A ) revealed rapid decline of TREX1 protein which was almost completely lost 3 days after the end of induction and was undetectable on day 4 ( Figure 5B and Figure S5C ). Increased type I IFN bioactivity was detectable (not significant) in the culture supernatant already on day 2 after end of induction, but was very prominently increased on days 3 and 4 ( Figure 5C ). Transcript levels of the ISG Ifi44 were significantly increased already on day 2 after 4-OHT and were 10-fold up-regulated on day 3 ( Figure 5D ). RNA sequencing of 4-OHT-induced BMDMs harvested on days 1 to 4 revealed robust up-regulation of ISGs ( Figure 5E and Figures S5D, E ). Enrichment of the gene set ‘IFNα response’ was not significant on day 1 after end of 4-OHT, but very much so on days 2 to 4 ( Figure 5E and Figures S5D, E ). In addition to ISG induction, loss of TREX1 was reflected in up-regulation of NF-κB and TNF-induced genes on day 2, as expected for STING responses ( Figure 5E ). Gene signatures associated with cell cycle progression, including e.g. expression of proliferation marker MKI67 were significantly downregulated, most prominently on day 4, while expression of CDKN1A encoding cell cycle inhibitor p21 was upregulated (not shown), most likely reflecting IFN-induced cell cycle arrest ( Figure 5E ). Similar results were obtained in Trex1^FL/FL R26^CreERT2 embryonic fibroblasts (MEFs) ( Figure S5F ).

Collectively, our finding that the IFN response is detectable already at day 2 after 4-OHT induction, a time point at which large amounts of TREX1 were still present, indicates that even slight reductions of TREX1 expression levels result in accumulation of the cGAS ligand and activation of cGAS/STING responses. Thus, steady-state TREX1 expression levels of healthy cells are just sufficient to degrade its DNA substrate. This finding is compatible with ongoing large-scale production of this substrate by at least a fraction of the cells in the culture at any time point.

Large amounts of oligonucleotide waste might originate from genome replication. Millions of Okazaki fragments are precisely ligated during lagging strand synthesis. Okazaki fragment synthesis by Polδ continues beyond the 5’-end of the previous Okazaki fragment, displacing a 5’-flap that requires endonucleolytic trimming (63). Nick translation by iterative rounds of strand displacement synthesis and flap cleavage not only remove the RNA primer but also the 5’-portion of primer initiator DNA of the previous Okazaki fragment synthesized by POLα. In addition to FEN1, the major flap endonuclease involved in Okazaki fragment maturation, also DNA2 and EXO1 contribute. A recent study suggested that primarily FEN1 and EXO1 compete for Okazaki fragment flap cleavage, with DNA2 playing only a minor role (64). EXO1 overexpression in yeast can partially compensate loss of FEN1 (65).

Whether DNA waste originating from Okazaki fragment maturation can trigger cGAS/STING responses is unknown. To approach this question, we asked whether the IFN response of cells lacking TREX1 requires DNA replication. This is not easily addressed by experiments aiming to arrest the cell cycle of constitutively TREX1-deficient cells, because immunostimulatory cGAS ligands, formed before halting proliferation, may be stable in the absence of TREX1 and maintain the IFN response. We therefore induced inactivation of Trex1 in cells in vitro and simultaneously arrested the cell cycle. Treatment of BMDM cultures with the CDK4/6 blocker Palbociclib, arresting cells in G1 (66), for 3 days, reduced the fraction of cells in S-phase 10-fold ( Figures 6A, B ). Palbociclib administration in parallel to 4-OHT-induced knock out of Trex1 ( Figure S6A ) abrogated the IFN response triggered by loss of TREX1, as determined by quantification of type I IFN bioactivity in culture supernatant ( Figure 6C ) and ISG transcript levels ( Figure 6D ). A control experiment demonstrated that BMDMs arrested by Palbociclib are capable of mounting cGAS/STING responses ( Figure S6B ). To independently confirm that the IFN response to induced loss of TREX1 depends on proliferation, we repeated the experiment in a different cell type, MEFs derived from Trex1^FL/FL R26^CreERT2 mice, using a different cell cycle inhibitor, Mimosine, that is chemically unrelated to Palbociclib and arrests cells in late G1 by activating checkpoint signaling independent of DNA damage (67). 4-OHT-induced inactivation of Trex1 resulted in an IFN response that was blunted by Mimosine-mediated G1 arrest ( Figure 6E and Figure S6C ). A control experiment demonstrated unimpaired IFN responses of Mimosine-exposed cells to STING activation by DMXAA ( Figure S6C ). Collectively, the IFN response of TREX1-deficient cells requires ongoing proliferation and functional EXO1, compatible with oligonucleotides released by flap endonucleases during lagging strand synthesis as the pathogenic ligand for cGAS.

For reasons of clarity, we collectively refer to both, the Trex1 knock out line published by Morita et al. (23) and the Trex1^Δ478/Δ478 line generated in this study, as ‘Trex1^-/-’. Trex1^Δ478/Δ478 mice, which phenocopy the published line, were used in experiments of Figure 2 , and Figure S2 . Trex1^Δ478/Δ478Xpa^Δ11/Δ11 mice were generated by microinjection of a mixture of four individual Cas9 RNPs into zygotes of C57BL/6 mice allowing simultaneous targeting of Trex1 (GTGACTTCGGGCCGAGACGA and CGGTGCTTGCCAGTACAGGC) and Xpa (GTGTTTATCATCAGCATCTC and GAACCCACGCCATTCACAGTG). The injection mix contained 1.73 µM of each crRNA (IDT), 6.9 µM tracrRNA (IDT) and 6.9 µM Cas9 protein (ToolGen). Trex1^FL/FL Rosa26-CreERT2^WT/KI mice were described earlier (25). Mice were housed under specific pathogen-free conditions at the Experimental Center, Medical Faculty Carl Gustav Carus, TU Dresden. All procedures were in accordance with institutional guidelines on animal welfare and were approved by the Landesdirektion Dresden (permit number TVV-88/2017).

Bone marrow cells from Trex1^FL/FLR26^CreERT2 or Trex1^WT/FL R26^CreERT2 control mice were cultured overnight in RPMI 1640 medium supplemented with 10% heat-inactivated FCS, 1% Pen/Strep, 2 mM L-Alanyl-L-glutamine and 1 mM sodium pyruvate (all Biochrom). Non-adherent cells were transferred to 10 cm petri dishes and differentiated for six days in RPMI medium (supplemented as described) containing 30 % L929 supernatant. Fresh differentiation medium was added after three days. BMDMs were harvested by trypsinization and scraping. For experiments addressing the kinetics of loss of TREX1, 2.5x10⁵ cells were seeded onto 6-well plates and treated with 2 µg/ml (Z)-4-Hydroxytamoxifen (Sigma, Cat#H7904) for 22 h. To address effects of cell cycle arrest on the IFN response after induced knock out of Trex1, 2x10⁶ cells were seeded onto 10 cm petri dishes and induced with 4-OHT. One day after washing away 4-OHT, cells from 10 cm petri dishes were split to 6 well plates at a density of 2.5x10⁵ and maintained in the presence of 5 µM Palbociclib (Selleckchem, Cat#S1116) for 72 h. Efficiency of Palbociclib-induced cell cycle arrest was verified by pulsing with 10 µM EdU for 4h at 37°C. Upon harvest, incorporated EdU was quantified flow cytometrically using the Click-iT™ EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit (Invitrogen, Cat#C10419) following the manufacturer’s instructions.

MEFs were generated by standard procedures. Briefly, after caesarean section, E14.5 mouse embryos were decapitated and internal organs were removed. Tissue was cut into small pieces, digested with 1x trypsin (0.25%, Invitrogen) for 30 min at 37°C and disaggregated by pipetting. The cell suspension was seeded into 10 cm petri dishes with DMEM (Gibco) supplemented with 10% heat inactivated FCS, 1% Pen/Strep, 1x nonessential amino acids and 100 µM β-mercaptoethanol. Induction of Cre activity was done as described above. To address effects of cell cycle inhibition on the IFN response after induced knock out of Trex1, 1.1x10⁵ cells were seeded onto 6 cm petri dishes and induced with 4-OHT. On the next day, 4-OHT-containing medium was removed and fresh supplemented DMEM containing 200µM L-Mimosine (Sigma-Aldrich, Cat# M0253) was added to the cells for four days.

Total RNA was isolated using NucleoSpin RNA Kit (Macherey-Nagel) and reverse transcribed into cDNA using PrimeScript RT Reagent Kit (Takara). qRT-PCR was performed using Luna^® Universal qPCR Master Mix (New England BioLabs) on a CFX384 Touch Real-Time PCR Detection System (Bio-Rad). Primer Sequences are listed in Supplementary Table S1 . Target gene expression was normalized to the housekeeping gene Tbp1.

Blood was sampled using retrobulbar puncture and mixed with heparin/PBS (250 units/ml, Biochrom). 20 µl of the blood/heparin/PBS mixture was added to 2 ml ice-cold Methanol, inverted and stored at -80°C until further use. For quantification of micronucleated erythrocytes, the protocol published by Balmus et al. was followed (42). Briefly, fixed blood cells were washed in bicarbonate buffer and stained against CD71 and Ter119 in the presence of RNase A. After washing and addition of PI, cells were analysed by flow cytometry and gated for single Ter119⁺ CD71^- PI⁺ micronucleated normochromatic erythrocytes (MN-NCE). 5x10⁵ cells in P1 SSC/FSC gate were acquired.

To detect ‘unscheduled’ DNA synthesis occurring in non-S phase cells repairing UV-induced genome damage (‘repair synthesis’), 5x10⁶ splenocytes from Xpa^WT/WT and Xpa^KO/KO mice were seeded onto 10 cm petri dishes in 3 ml PBS and irradiated with 20 J/m² UVC (xy min) using an OSRAM HNS 15W OFR 3224 lamp. 6 ml of serum-free RPMI containing 1% Pen/Strep, 1x NEA and 10 µM EdU was added for 3h. Cells were harvested and incorporated EdU was quantified flow cytometrically using the Click-iT™ EdU Alexa Fluor™ 647 Flow Cytometry Assay Kit (Invitrogen, Cat#C10419) following the manufacturer’s instructions.

Ears of Xpa^WT/WT and Xpa^KO/KO animals were homogenized and lysed in RIPA buffer [50 mM Tris HCl, 150 mM NaCl, 1% NP-40, 0,1% SDS, 0,5% sodium deoxycholate, 1x Complete protease inhibitor (Roche)] for 30 min on ice. Lysates were cleared by centrifugation (16.000 xg, 10 min, 4°C) and supernatant was incubated with 6x Laemmli sample buffer at 95°C for 5 min. Cell pellets were directly lysed in 2x Laemmli sample buffer and incubated at 95°C for 5 min. Proteins were separated on 12% SDS polyacrylamide gels and transferred to a nitrocellulose membrane (Amersham Hybond-ECL, GE Healthcare). Membranes were blocked in 5% low fat dried milk/TBS-T or 1x Roti^®Block and incubated overnight with the respective primary antibody. After washing in TBS-T, membranes were incubated with secondary antibodies for 1 h at room temperature and washed again in TBS-T. For signal detection, membranes were incubated in ECL (Amersham) or AP buffer following incubation in NBT/BCIP substrate (Roche).

The following primary antibodies were used: β-Tubulin (9F3, Cell Signaling Technology, Cat#2128), β-actin (13E5, Cell Signaling Technology, Cat#4970), Xpa (abcam, Cat#ab58464, 1:500), mTrex1 (C-11, Santa Cruz, Cat#sc-133112, 1:250), hTrex1 (E-6, Santa Cruz, Cat#sc-271870, 1:250), Mus81 (MTA30 2G10/3, Santa Cruz, Cat#58382, 1:250). Secondary antibodies used were AP-coupled anti-rabbit (Dako) and anti-mouse (Cell Signaling Technology).

Hearts were fixed in 4 % formalin and embedded in paraffin. 3 µm paraffin sections were stained with hematoxylin and eosin. Inflammation was quantified by an experienced histopathologist in a blinded fashion. A histological score for H&E staining was given depending on the grade of inflammation (0 – no sign of inflammation, 1 - slight diffuse increase in inflammatory cell infiltration, 2 - few small dense focal inflammatory infiltrates, 3 - large inflammatory infiltrates extending through more than half of the thickness of the ventricular muscle plus focal destruction of myocardial structure). To judge the effect of additional loss of EXO1 on the intensity of myocarditis of Trex1^-/- mice, density of T cell infiltration on sections stained for CD3 and tissue fibrosis on Masson’s trichrome-stained sections were analyzed. CD3 immunohistochemical staining was performed on a Ventana Discovery Ultra Instrument. In brief, antigen retrieval using cell conditioning 1 solution (Ventana Medical Systmes) was performed at 95°C for 32 min followed by incubation with anti-CD3 antibody (SP7, abcam, Cat#ab16669, 1:200) at 36°C for 32 min, anti-rabbit OmniMap HRP-coupled secondary antibody (Ventana Medical Systems) for 12 min and finally Opal 650 TSA fluorophore (Akoya Biosciences, 1:100) at RT for 8 min. Primary and secondary antibodies were removed by denaturation at 100°C for 24 min in cell conditioning 2 buffer (Ventana Mediacl Systems) before counterstaining with DAPI. Sections were scanned (100x magnification) and images aquired (x200 magnification) by Vectra 3.0 Automated Imaging System (Akoya Biosciences). CD3⁺ cells were quantified using inForm (Akoya Biosciences) and R software. Masson’s trichrome staining was done according to standard protocols (Sigma-Aldrich, Procedure No.HT15).

5 ng/µl RNA were sequenced on Illumina NovaSeq 6000 and data were demultiplexed using JE demultiplexer tool. Read quality was checked by FastQC. Transcript abundance was quantified using Salmon tool (88) and tximport (89) was used to generate gene level count matrices. The murine genome GRCm39 (90) was used as the reference. All post mapping analysis and visualization was conducted in R unless specifically stated otherwise. Normalization, exploratory analysis and differential expression analysis was performed using DESeq2 (91). Genes were considered significantly differentially expressed at FDR <0.01 and absolute log fold change >1.5. GGplot2 (92) was used to visualize results. List of significant DE genes was used to conduct a further gene set enrichment analysis (GSEA) (93, 94).

Unless stated otherwise, significance was calculated by One-way ANOVA with Tukey’s post-hoc test ( Figures 1 – 4 ) or Two-way ANOVA with Sidak’s post-hoc test ( Figures 5 , 6 ).