Allosteric coupling between Mn2+ and dsDNA controls the catalytic efficiency and fidelity of cGAS

Abstract Cyclic-G/AMP (cGAMP) synthase (cGAS) triggers host innate immune responses against cytosolic double-stranded (ds)DNA arising from genotoxic stress and pathogen invasion. The canonical activation mechanism of cGAS entails dsDNA-binding and dimerization. Here, we report an unexpected activation mechanism of cGAS in which Mn2+ activates monomeric cGAS without dsDNA. Importantly, the Mn2+-mediated activation positively couples with dsDNA-dependent activation in a concerted manner. Moreover, the positive coupling between Mn2+ and dsDNA length-dependent activation requires the cognate ATP/GTP substrate pair, while negative-cooperativity suppresses Mn2+ utilization by either ATP or GTP alone. Additionally, while Mn2+ accelerates the overall catalytic activity, dsDNA length-dependent dimerization specifically accelerates the cyclization of cGAMP. Together, we demonstrate how the intrinsic allostery of cGAS efficiently yet precisely tunes its activity.

It is increasingly appreciated that cGAS dynamically tunes the intensity and duration of the IFN-I activity according to the severity of stress, consequently eliciting context-dependent innate immune responses (e.g. autophagy, cellular senescence, and apoptosis) (1,2,(11)(12)(13)(14). It was recently identified that the length of cytosolic dsDNA is a key measure for cGAS to gauge stress levels (15,16): long stretches of naked dsDNA (≥ 50 base-pairs (bps)) signify major maladies such as the presence of viral genomes or mislocalized mitochondrial dsDNA, while shorter fragments indicate minor DNA damage or degraded viral genomes (15,16). How does dsDNA length regulate cGAS activity? Resting cGAS is an inactive monomer and dimerization is required for activation in addition to dsDNA binding (17,18). dsDNA length dictates the distribution of dsDNA-bound cGAS monomers and dimers, thereby grading the signaling activity (19). Importantly, such an allosteric mechanism allows dsDNA length-dependence to be conditional: substrate binding and elevated cGAS concentrations also induce dimerization, thereby reducing the dependence on dsDNA length (19).
Still, major questions remain regarding how cGAS is regulated. First, it was recently reported that cytosolic Mn 2+ originating from damaged organelles potentiates the ds-DNA sensing activity of cGAS (20). In fact, Mn 2+ was crucial for cGAS-mediated host innate immune responses against DNA viruses (20). Nevertheless, how Mn 2+ regulates GAS activity remains unknown. Second, little is known about how cGAS retains substrate specificity and signal fidelity (3,21,22). The catalytic core of cGAS contains a polymerase (pol)-␤-type nucleotidyl-transferase (NTase) domain, which is inherently promiscuous (23)(24)(25). Although ATP/GTP (1ATP:1GTP) is the cognate substrate pair, cGAS can bind and perform chemistry on non-cognate substrate pairs (e.g. ATP/ATP). Such 'off-pathway' interactions can be deleterious by not only competing against cognate substrates, but also generating dinucleotides that do not activate STING (3,21,22,26). Moreover, this issue can also be exacerbated by multiple factors: (a) ATP outnumbers GTP by at least 4-fold in vivo (27) and both NTPs, especially ATP, are present at saturating concentrations for cGAS ([ATP] and [GTP] > K M for cGAS) (8,19,22,28). Thus, noncognate ATP/ATP is likely the most abundant substrate pair for cGAS. How cGAS overcomes this issue and generates cGAMP in high fidelity remains unknown. (b) Although Mn 2+ sometimes enhances the catalytic activity of other cGAS-like NTases by substituting for Mg 2+ , its primary effect is increased substrate promiscuity (23)(24)(25). In fact, Mn 2+ is considered mutagenic for pol-␤-type NTases as it preferentially enhances the affinity for noncognate NTPs (23)(24)(25). Currently, how cGAS could compensate for the potentially deleterious effect of Mn 2+ remains unknown. (c) Human cGAS is a poor cyclase (3,8,22,28). cGAS first generates 2 -GTP to 5 -ATPlinked linear pppGpA from ATP/GTP, which dissociates and rebinds to be cyclized into 2 -5 3 -5 linked cGAMP (3,22) (simply 'cGAMP' hereafter). Because of low cyclization efficiency, the pppGpA intermediate accumulates much more rapidly than cGAMP (3,22). This could impinge on promptly mounting host responses, as the intermediate cannot activate STING or even be degraded rapidly. Again, how cGAS might address this issue remains unknown. Here, we unravel an unexpected role of allostery in resolving all these compounding issues.
We find that Mn 2+ activates human cGAS without ds-DNA and also positively couples with dsDNA-dependent activation, allowing Mn 2+ to dampen dsDNA lengthdependence in activating cGAS both in vitro and in macrophages. cGAS preferentially utilizes Mn 2+ over Mg 2+ for its NTase activity, and Mg 2+ facilitates Mn 2+ utilization instead of competing against it. Moreover, physiological amounts of Mn 2+ can activate cGAS mutants defective in dsDNA/dimerization-dependent activation both in vitro and in cells, indicating that Mn 2+ activates monomeric cGAS. The NTase activity with cognate ATP/GTP vs. noncognate ATP/ATP or GTP/GTP is similar when cGAS is bound to short dsDNA; however, longer dsDNA increases the activity only toward ATP/GTP. Likewise, although Mn 2+ accelerates the NTase activity of cGAS toward noncognate substrates, ATP/GTP benefits most. Furthermore, ATP/GTP is required for the synergy between Mn 2+ and dsDNA binding, while dsDNA binding invokes negative-cooperativity between Mn 2+ and noncognate substrates. Finally, Mn 2+ accelerates the overall catalytic activity of cGAS; however, dsDNA length-dependent dimerization specifically accelerates the rate-determining cyclization, minimizing the formation of off-pathway products even with excess ATP. Together, we demonstrate how allostery linking Mn 2+ -and dsDNA-binding maximizes the efficiency and activity of cGAS.

Protein expression and purification
Human cGAS (encoding 2-522, cGAS FL or 157-522, cGAS cat ) was cloned into a pET28b vector (Novagen) inframe with an N-terminal 6xHis-MBP-tag and a TEV protease cleavage site. Plasmids encoding wild-type and mutant cGAS constructs were transformed in to E. coli BL21-Rosetta 2 DE3 cells and expressed via IPTG as described in ref. (19). Recombinant cGAS was then purified as described in (19), concentrated, and stored in −80 • C.

Reagents
dsDNAs longer than 90-bp were PCR amplified from the human DDX41 gene and agarose gel purified. dsDNAs shorter than 90-bp were purchased from Integrated DNA Technology. ATP and GTP were hydrated separately in 20 mM HEPES buffer and adjusted to pH 7.4 with NaOH.

Biochemical assays
All experiments were performed at least three times as described in before (19,29). Shown Figures are either representative of these independent experiments or averages with standard deviations (SD). The fits to data (the Hill form of the Michaelis-Menten equation or the binding isotherm) were generated using Kaleidagraph (Synergy Software). All reactions were performed under 25 mM Tris acetate pH 7.4, 125 mM potassium acetate pH 7.4, 0.5 mM TCEP, 5% glycerol and indicated concentrations of MgCl 2 and/or MnCl 2 at 25 ± 2 • C. cGAS concentrations are noted in Figure Legends. To avoid titrating Mn 2+ against free NTPs and to mimic the in vivo environment where Mg 2+ is present in large excess (20,30), each NTP was pre-mixed with 1 molar equivalent of MgCl 2 , with the exception of those experiments that were conducted in the complete absence of MgCl 2 , in which case the NTPs were pre-mixed with equimolar MnCl 2 instead. We used as low as 6 M MnCl 2 in our experiments to reflect the physiological amount (≤ 53 M; (20,31,32)).

Analytical fractionation of cGAS reactions by HPLC
40 l reactions containing cGAS, NTPs, dsDNA (where indicated), MgCl 2 and/or MnCl 2 were incubated in reaction buffer for the indicated time. Reactions were quenched with 25 mM EDTA, diluted to 80 l with water and filtered with 3 or 10 KDCO spin columns (Amicon). An aliquot of the flow-through was mixed 1:1 with HPLC buffer A (see below) and fractionated. HPLC fractionation was performed on the Agilent Technologies 120 Infinity II system using a Poroshell EC-C18 column (2.7m; 4.6 × 100mm). Reaction products were fractionated by the following scheme: 0% B from 0-1min, linear increase to 50% B from 1-27 min, linear increase to 100% B from 27-28 min. Buffer A: 100 mM potassium phosphate monobasic, 5 mM TBA, final pH 6.0; Buffer B: 100 mM potassium phosphate monobasic, 5 mM TBA, final pH 6.0, 30% acetonitrile. Buffers were filtered and degassed. Integrated peak intensities were calculated by the Agilent Technologies ChemStation software. Synthetic and purified compounds were used as standards for peak identification.

Measuring cGAS activity in THP-1 cells
THP1-Dual and THP1-Dual cGAS-KO reporter monocytes were differentiated into macrophages with 50 nM PMA for 16 hours at 37 • C and 5% CO 2 . The PMA containing media was replaced by fresh media and cells were incubated an additional 48 hours at 37 • C 5% CO 2. The media was again replaced with fresh media or media plus MgCl 2 or MnCl 2 and incubated for 2 hrs at 37 • C and 5% CO 2 . Cells were then transfected with various lengths of ds-DNA (1.67 g/ml final) using Lipofectamine LTX containing OptiMEM media. Cells were incubated an additional 24 hours at 37 • C and 5% CO 2 before supernatants were collected and Lucia production was determined using the Luciferase Reporter Assay System (Promega).

Measuring cGAS activity in HEK293T cells using dualluciferase reporter
HEK293T cells were pre-incubated for 2 hours at 37˚C and 5% CO 2 with growth media, or media supplemented with 50 M of MnCl 2 or MgCl 2 . 10 or 50 ng of pCMV plasmids encoding empty vector or various cGAS FL variants were then transfected using Lipofectamine LTX (Invitrogen), along with 5 ng of Renilla Luciferase plasmid, 10 ng of plasmid encoding human STING, and 10 ng of plasmid encoding Firefly Luciferase under an IFN-I␤ promoter. Transfected cells were incubated overnight at 37˚C and 5% CO 2, washed once with PBS and lysed. Lysates were analyzed for Firefly Luciferase and Renilla Luciferase activities independently using the Dual-Luciferase Reporter Assay System (Promega). Data shown is normalized as the ratio of Firefly output divided by log 10 Renilla output of individual wells.
While investigating the underlying mechanism, we found that Mn 2+ alone does not enhance the dsDNA binding activity of human full-length cGAS (cGAS FL ; Supplementary Figure S1B-D and Supplementary Table S1), which prompted us to test whether Mn 2+ regulates the catalytic activity. Using our PP i ase-coupled NTase assay (19), we found that Mn 2+ drastically increased the maximal dsDNAinduced NTase activity of cGAS FL (k max ) independent of duplex length ( Figure 1B), revealing that Mn 2+ diminishes dsDNA length-dependence by directly enhancing the catalytic activity. Next, we tested whether Mn 2+ enhances the K M or the maximal turnover (k cat ) in the presence of saturating amounts of dsDNA with various lengths ( Figure  1C). For all dsDNA lengths, K M s improved moderately, but k cat s were consistently and significantly higher ( Figure  1C, Supplementary  Table S2), suggesting positive cooperativity between Mn 2+ and dsDNA-binding in activating cGAS; see (33,34) for reference. To explore this unexpected mechanism, we determined whether Mn 2+ enhances the apparent affinity of dsDNA (K act ) in the presence of ATP/GTP. Here, we used the catalytic domain (cGAS cat ) for accurate measurements as it binds dsDNA more weakly than cGAS FL (19). Indeed, cGAS cat showed enhanced k max s and K act s for both 19-and 72-bp dsDNA with Mn 2+ (Figure 1D, Supplementary Table S3). Together, we concluded that Mn 2+ not only increases the catalytic activity of cGAS, but also allosterically enhances its dsDNA binding activity in conjunction with ATP/GTP.

cGAS preferentially couples Mn 2+ to its allosteric activation
We noted that the effect of Mn 2+ was evident despite excess Mg 2+ in our assays ( Figure Tables S4 and S5). Moreover, in contrast to the Hill constant of 1.2 for Mg 2+ , the Hill constants were ∼2 for Mn 2+ ( Figure 1E versus Figure S1E; see also Supplementary Tables S4 and S5), indicating Mn 2+ can substitute for Mg 2+ in a positively cooperative manner. In the presence of 5 mM Mg 2+ ( Figure 1F), the EC 50 of Mn 2+ improved 2to 3-fold and the Hill constants also decreased to ∼1, albeit the k max decreased ∼ 2-fold (Supplementary Table S5). Considering that cGAS utilizes two divalent metals at the active site (3), our observations suggest that occupying one metal binding site by Mg 2+ increases the affinity of Mn 2+ for the second site. The decreased k max is also consistent with this idea, as the catalytic activity is supported by both Mg 2+ and Mn 2+ . Collectively, our observations suggest that free Mg 2+ in the cytoplasm (∼1 mM (35)) would facilitate Mn 2+ (≤50 M (20,31,32)) utilization rather than compete against it (e.g. at 10 M Mn 2+ , the activity of cGAS with 72-bp ds-DNA would be ∼10-fold higher in the presence of Mg 2+ ).  (19), saturating [dsDNA] = 10-fold higher than their apparent affinity (K act ) as determined in (19).

Mn 2+ activates cGAS without dsDNA and also synergizes with dsDNA-dependent activation
The NTase activity of cGAS FL without dsDNA was notably higher with Mn 2+ (Figure 1B, dark blue), raising the possibility that Mn 2+ alone activates the enzyme. Indeed, cGAS FL showed robust dsDNA-free NTase activity with increasing Mn 2+ (Figure 2A). The maximal activation without dsDNA was comparable to the dsDNA-induced activity ( Figure 2A,   that Mn 2+ plus ATP/GTP enhances the dsDNA binding affinity of cGAS ( Figure 1D), our results indicated that dsDNA-binding couples with Mn 2+ -utilization in a concerted manner. Collectively, we reveal an unexpected Mn 2+dependent, dsDNA-independent activation mechanism of cGAS. Moreover, our results indicate that dsDNA binding is crucial not only for allowing Mn 2+ to be utilized at physiological concentrations, but also for synergizing with Mg 2+ .

Mn 2+ activates monomeric cGAS
We found previously that ATP/GTP binds cGAS even without dsDNA and modestly induces dimerization (19). Considering the positive relationship between dsDNA and Mn 2+ , we asked whether Mn 2+ also promote substratemediated dimerization more efficiently than Mg 2+ . We thus monitored the changes in Förster resonance energy transfer (FRET) ratios between donor-and acceptorlabeled cGAS FL populations with increasing Ap c pp/Gp c pp (non-hydrolyzable analogues that prevent turnover (19)). Ap c pp/Gp c pp produced higher FRET signals with Mn 2+ , suggesting that substrate-mediated dimerization is enhanced ( Figure 2D). Nevertheless, FRET ratios did not change as robustly as those induced by 72-bp dsDNA (Figure 2D versus Supplementary Figure S2A), indicating that the dimeric fraction was still marginal. Moreover, unlike the NTase activity assays, FRET ratios did not change with increasing Mn 2+ when Ap c pp/Gp c pp was fixed at a low concentration ( Figure 2E; 200 M, equivalent to [ATP/GTP] used in NTase assays in Figure 2A and C). These observations suggest that dimerization at this substrate concentration is minimal even with Mn 2+ . Adding Mn 2+ and/or Ap c pp/Gp c pp did not alter dsDNA-induced FRET ratios, again indicating that dimerization is dictated by dsDNA (Supplementary Figure S2A). Nevertheless, the Mn 2+ -induced NTase activity of cGAS without dsDNA was as high as those induced by long dsDNA (Figure 2A-C), suggesting that Mn 2+ activates monomeric cGAS (each monomer has an intact active-site (3,17,18,36,37)).
To test whether Mn 2+ activates monomeric cGAS, we monitored the NTase activity of a mutant that cannot dimerize, K394E ( Figure 2F, Figure S2B) (19). Both K394E-cGAS FL and K394E-cGAS cat displayed robust NTase activities with Mn 2+ without dsDNA ( Figure 2G, Supplementary Figure S2C-F), supporting the mechanism by which Mn 2+ activates monomeric cGAS. Of note, cGAS has a Zn 2+ -finger adjacent to the dimer interface ( Figure  2F), and mutating two coordinating Cys residues abrogates the signaling activity of cGAS in cells (17,36,37). Thus, to disrupt the dimerization in a different manner, we removed Zn 2+ via EDTA or by mutation (C396A/C397A (36), Figure 2F). Unlike wild-type (WT) cGAS FL that behaved a mixture of monomer and dimer in size-exclusion chromatography (SEC), both variants eluted as monomers (Supplementary Figure S2B), corroborating that the Zn 2+finger is important for dimerization. Both EDTA-treated and C396A/C397A-cGAS FL variants bound dsDNA (Figure S2G); however, dsDNA failed to activate either variant with Mg 2+ ( Figure 2G). Nevertheless, both mutants showed robust NTase activities with Mn 2+ in the absence of dsDNA or Mg 2+ ( Figure 2G). Our observations not only support the mechanism by which Mn 2+ alone activates monomeric cGAS, but also indicate that Mn 2+ does not substitute for Zn 2+ . Next, to explore biological relevance, we monitored the signaling activity of cGAS FL variants with or without Mn 2+ in HEK293T cells. A physiological amount of Mn 2+ substantially increased the WT cGAS FL -STING induced IFN-␤ activity ( Figure 2I). Mn 2+ also induced IFN-␤ activity from K394E-and C396A/C397A-cGAS when they were sufficiently overexpressed ( Figure 2I versus J); the dead variant showed no IFN-␤ activity in all conditions (Figure 2I-J). Together, these results suggest that Mn 2+ activates monomeric cGAS.

The positive coupling between Mn 2+ and dsDNA occurs within the cGAS monomer
Dimerization is critical for the dsDNA-mediated activation of cGAS (15)(16)(17)19). Because our results suggest that Mn 2+ activates monomeric cGAS, we then asked whether Mn 2+ could restore dsDNA-mediated activation of cGAS mutants defective in dimerization ( Figure 2F and G). Here, we determined the NTase activity of cGAS FL mutants using a sub-optimal amount of Mn 2+ for dsDNA-free activation of cGAS FL ( Figure 2H). As expected, all cGAS FL variants, except for the catalytically dead, showed moderate Mn 2+ -induced NTase activity without dsDNA ( Figure  2H light purple). Strikingly, dsDNA further stimulated the Mn 2+ -induced NTase activities of K394E, EDTA-treated, and C396A/397A variants ( Figure 2H dark purple), which suggested that the positive coupling between Mn 2+ and ds-DNA occurs within a monomer (i.e. in cis). Next, to further test the cis coupling mechanism, we mutated two sidechains at the hinge of the 'spine helix' that changes conformation upon dsDNA binding (K173E/R176E (36), Figure  2F, Supplementary Figure S2H); the dsDNA-induced conformation change of the spine helix restructures the 'activation loop' in cis (3,18,36) (Supplementary Figure S2H). K173E/R176E-cGAS FL migrated mostly as a dimer in SEC and bound dsDNA ( Supplementary Figures S2B and G). 5 mM Mn 2+ also activated K173E/R176E-cGAS FL without dsDNA and Mg 2+ ( Figure 2G), but dsDNA failed to further activate this cGAS FL mutant even with sub-saturating Mn 2+ ( Figure 2H). Surprisingly, however, K173E/R176E-cGAS showed robust IFN-␤ activity when overexpressed in HEK293T cells ( Figure 2J). Considering that dsDNA does not activate this mutant ( Figure 2H), our observation suggests that even the marginal catalytic activity of cGAS induced by Mn 2+ alone could be sufficient for robust IFN-␤ signaling in vivo. Overall, our observations consistently indicate that the positive cooperativity between Mn 2+ and ds-DNA occurs within the cGAS monomer.

Allosteric coupling between dsDNA and Mn 2+ underpins substrate specificity
Although ATP/GTP is the cognate substrate pair, the active-site of cGAS is inherently promiscuous; cGAS can create linear dinucleotides using either ATP or GTP alone (pppApA and pppGpG, respectively) (3,21,22). This could be problematic in vivo, as ATP outnumbers GTP and neither pppApA nor pppGpG activates STING (3,21,22,26). Moreover, Mn 2+ could exacerbate this issue, as it increases the substrate promiscuity of cGAS-like NTases (23)(24)(25). Nonetheless, cGAS can sustain the fidelity and efficiency of its signaling pathway. To understand the underlying mechanism, we first investigated how efficiently cGAS utilizes either ATP/ATP or GTP/GTP with Mg 2+ and varying lengths of dsDNA using our NTase assay. With 19-bp ds-DNA, both k cat s and K M s for noncognate substrates were ∼2-fold worse than those from ATP/GTP, resulting in ∼3fold lower k cat /K M s ( Figure 3A, B and Supplementary Table S8; see Supplementary Figure S3A for GTP/GTP). Our observations suggest that cGAS produces cGAMP and offpathway dinucleotides with similar efficiency when bound to short dsDNA. Strikingly, and in contrast to ATP/GTP, longer dsDNA failed to enhance either k cat or K M against ATP/ATP or GTP/GTP, resulting in an up to 12-fold difference in k cat /K M between cognate and noncognate substrates ( Figure 3A and B, Supplementary Table S8). These results suggest an unexpected role of dsDNA length in governing the substrate specific catalytic activity of cGAS.
Mn 2+ -induced dsDNA-free NTase activity of cGAS FL toward ATP/ATP or GTP/GTP was ∼3-fold higher compared those induced by dsDNA and Mg 2+ ( Figure 3C versus Figure 3A and Supplementary Figure S3A; see also Supplementary Table S8). Nevertheless, ATP/GTP still produced the highest Mn 2+ -induced k cat (Figure 3C), which suggested that Mn 2+ preserves the substrate preference of cGAS unlike several other related NTases (23-25). We then investigated whether ATP/ATP or GTP/GTP allows Mn 2+ to synergize with dimerization/dsDNA-mediated activation. Here, we found that increasing Ap c pp or Gp c pp alone boosted the FRET ratios between two labeled cGAS FL populations ( Figure 3D), suggesting that binding of any NTP at the active-site promotes dimerization. However, Mn 2+ did not further increase the FRET ratio for either NTP alone ( Figure 3D versus 2D), suggesting that noncognate substrates do not allow Mn 2+ to promote dimerization. Next, we asked how well cGAS utilizes Mn 2+ in its NTase activity against noncognate pairs. Without dsDNA or excess Mg 2+ , the EC 50 of Mn 2+ against ATP/ATP or GTP/GTP was 2fold weaker compared to ATP/GTP, but the Hill constants were still near two (Figures 2A versus 3E, F, Supplementary Figure S3B). Adding 5 mM Mg 2+ in the absence of dsDNA moderately enhanced the EC 50 of Mn 2+ and reduced the Hill constant to near one (Supplementary Figure  S3B-C and Table S9). However, with 339-bp dsDNA, the EC 50 of Mn 2+ against noncognate NTPs could not be determined due to apparent negative-cooperativity (Hill < 1; Figure 3E, F and Supplementary Figure S3B-C and Supplementary Table S9). These results suggest that dsDNA bind- ing allosterically suppresses the utilization of Mn 2+ in catalyzing noncognate substrates. Fixing the Hill constant to one for data fitting estimates the EC 50 s around 350 M for both NTPs, which is 4-fold tighter than in the absence of ds-DNA (Supplementary Table S9). Such changes pale in comparison to the 100-fold enhancement for ATP/GTP (EC 50 : 18 M, Supplementary Table S5). Moreover, the low affinity would make the utilization of Mn 2+ against noncognate substrates unlikely in vivo (20,30). Together, we concluded that positive coupling between dsDNA and Mn 2+ requires cognate ATP/GTP.

Mn 2+ does not change product identity and accelerates both first and second linkage formation of cGAMP
cGAS first generates pppGpA from ATP/GTP then cyclizes the intermediate into cGAMP (3,22). It remained unclear which catalytic step was affected by Mn 2+ and whether the same products were being produced. We thus analyzed various cGAS FL reactions using High Pressured Liquid Chromatography (HPLC). We first established the retention times for ATP, GTP, and cGAMP using synthetic materials ( Figure 4A, gray). With 5 mM Mg 2+ , we then confirmed that the peaks corresponding to cGAMP or pppGpA appeared only if dsDNA was present ( Figure 4A, black versus orange). With saturating Mn 2+ , but without dsDNA or Mg 2+ , the same two peaks emerged ( Figure 4A, blue), indicating that Mn 2+ does not change the intrinsic cGAMP synthesis activity. dsDNA/Mg 2+ -induced cGAS reactions with or without 50 M Mn 2+ also produced the same peaks ( Figure 4B). Notably, however, both cGAMP and pppGpA peaks were almost equally higher when Mn 2+ was present ( Figure 4B and C). These results suggest that either the first catalytic step, or both catalytic steps were accelerated. We thus concluded that Mn 2+ potentiates the NTase activity of cGAS without altering product identity.

The product fidelity is dictated by dsDNA length
Next, we asked whether dsDNA also accelerates both the first and second linkage formation in a length-dependent manner. Here, cGAS reaction products were identical across different dsDNA lengths ( Figure 4D). However, while the peak corresponding to pppGpA was increased by 1.4-fold from 19-to 339-bp dsDNA, the cGAMP peak was 4.1-fold higher ( Figure 4D and E), which are in stark contrast to when Mn 2+ accelerated both steps equally ( Figure  4C, b). Importantly, our results also suggest that dsDNA length allosterically dictates the rate-limiting cyclization. To further test this, we monitored reaction products resulting from different cGAS amounts, as promoting dimerization by raising protein concentrations decreases the dependence on dsDNA length in activating cGAS (19). pppGpA production increased linearly with increasing cGAS concentrations and dsDNA length-dependence was consistently marginal ( Figure 4F and Supplementary Figure S4A-C). On the other hand, while cGAMP production depended on dsDNA length at lower cGAS concentrations, it became essentially length-independent at the highest concentration ( Figure 4F, b, Supplementary Figure S4D-F), Our observations corroborate that dsDNA length-dependent dimerization is crucial for cyclization. Mn 2+ did not alter this trend, but increased the amount of pppGpA and cGAMP for all dsDNA lengths even at the highest cGAS concentration (Supplementary Figure S4). Our observations again indicate that Mn 2+ alone activates cGAS, and also synergizes with the allosteric activation. Overall, these results consistently suggest that dsDNA length-dependent dimerization regulates the cyclization of cGAMP from the intermediate.

Allostery ensures the product fidelity even with excess ATP
The in vivo concentration of ATP is well above the K M for cGAS and also present in ∼5-fold excess to GTP (27). Because Mn 2+ boosts the NTase activity of cGAS against ATP/ATP (Figure 3), we envisioned that cytosolic Mn 2+ might instigate futile catalytic cycles, accumulating an off-pathway product (pppApA). To test this possibility, we thus analyzed cGAS reaction products using physiologically relevant amounts of ATP:GTP (5ATP/1GTP). With 5 mM Mn 2+ , and without dsDNA or any Mg 2+ , we noted an additional peak indicating the production of pppApA along with cGAMP and pppGpA ( Figure 5A, blue, see also Supplementary Figure S5A-C for ATP/ATP and GTP/GTP reactions). Adding 339-bp dsDNA did not change the pppGpA peak, but increased the cGAMP peak ( Figure 5A, B, Supplementary Figure S5D). Remarkably, dsDNA also decreased pppApA production ( Figure 5A, B, Supplementary Figure S5D), suggesting that the acceleration of cyclization by long dsDNA in turn suppresses off-pathway product formation. We also analyzed dsDNA/Mg 2+ -stimulated cGAS reaction with 5ATP/1GTP in the absence or presence of 50 M Mn 2+ ( Figure 5C). Here, Mn 2+ specifically increased the production of pppGpA and cGAMP without altering pppApA formation ( Figure 5C, D, Supplementary Figure S5E), indicating that Mn 2+ and longer dsDNA synergistically ensure the efficiency and fidelity of cGAS.
Finally, to further test the role of dsDNA length in suppressing off-pathway product formation, we monitored the kinetics of cGAS FL reaction with 5ATP:1GTP in the presence of 19-or 339-bp dsDNA. In the presence of 19-bp ds-DNA, both pppApA and pppGpA accumulated linearly, with pppGpA accumulating 5-fold faster than pppApA ( Figure 5E, Supplementary Table S10). cGAMP production occurred with a prominent lag then reached a steadysteady (linear) rate ( Figure 5E), corroborating the stepwise synthesis model (3,22). With 339-bp dsDNA, both pppGpA and pppApA accumulated about 1.3-fold faster than 19-bp dsDNA-induced production ( Figure 5E versus F, Supplementary Table S10). However, the steady-state production of cGAMP was reached 4.3-fold faster than that induced by 19-bp dsDNA, quickly outpacing pppApA production ( Figure 5E versus 5F, Supplementary Table S10). The negligible boost in pppApA formation by longer dsDNA also agrees with the dsDNA length-independent NTase activity of cGAS against ATP/ATP ( Figure 3A). These results further support the mechanism by which dsDNA lengthdependent allostery controls the signal fidelity of cGAS.

DISCUSSION
cGAS is central to the host defense against various maladies that give rise to cytosolic dsDNA (1). Despite extensive cellular and structural studies (3,15,18,19,28,36,37), how human cGAS operates remain poorly understood at the molecular level. For instance, how cytosolic Mn 2+ released from damaged organelles potentiates cGAS activity remains unknown (20). Moreover, it is unclear how cGAS manages the inherent substrate promiscuity at its NTase active-site in vivo when its NTP substrates exist well above their respective K M s ( Figure 3) with ATP present in large excess over GTP (27). Here, we identify an unexpected activation mechanism in which Mn 2+ activates monomeric cGAS without dsDNA. More importantly, this noncanonical activation synergizes with the canonical dsDNA-dependent activation. Furthermore, we show that dsDNA lengthdependent dimerization regulates the substrate specificity and cyclase activity of cGAS, with Mn 2+ further polarizing these activities toward cognate ATP/GTP. Together, we propose that this uniquely synergetic coupling mechanism underpins the signaling efficiency, substrate specificity, and product fidelity of cGAS ( Figure 6).
Cytosolic Mn 2+ is crucial for the dsDNA sensing activity of cGAS (20). Unlike abundant Mg 2+ , Mn 2+ is present in trace amounts and also sequestered in mitochondria and the Golgi (31,32,35). Pathogen invasion or exposure to toxic reagents damages these vital organelles, releasing Mn 2+ to the cytosolic space (20). That is, the co-presence of Mn 2+ and long dsDNA in the cytosol would signal one of the most critical conditions such as viral replication and/or massive mitochondrion deaths. Our work demonstrates the mechanisms by which cGAS senses and integrates multiple danger-associated molecular patterns within the environment that gives rise to cytosolic dsDNA, thus dialing its signaling outputs accordingly. The concentration of Mn 2+ for dsDNA-free activation (EC 50 ∼ 1 mM; Figure 2) is much higher than physiological levels (e.g. <10 M in THP1 cells (20) and ∼50 M in brain cells (30)); however, the affinity Figure 6. A scheme describing how allostery tunes the biochemical activities of cGAS With increasing stress levels (colored dots within the ramp), all signaling activities (substrate specificity, cyclization efficiency, Mn 2+ utilization and cooperativity) are maximized, while the opposite occurs with decreasing stress levels. Such a dynamic mechanism would then lead to changes in product distribution, consequently tuning the signaling activity. Text sizes for dinucleotides indicate their relative abundance.
for Mn 2+ drastically improves by dsDNA ( Figure 1E and F). Such a highly cooperative mechanism would then ensure the host to properly initiate IFN-I responses.
The active-sites of cGAS-like NTases show varying degrees of promiscuity (23)(24)(25)28), and cGAS accepts different NTP pairs and generate off-pathway products (3,18,21,22,28). By contrast, human STING is hyperselective towards cGAMP (3,21,22). cGAS is an unusually slow enzyme, as one round of cGAMP synthesis takes ∼20 s even under the best-case scenario ( Figure 1C). Thus, if cGAS excessively generates off-pathway dinucleotides or fails to cyclize the on-pathway intermediate in a timely fashion would impinge on promptly mounting stress-responses. Further compounding issues include the imbalance between ATP and GTP (27), rate-limiting cyclization efficiency (3,8,22,28) and Mn 2+ that often found to be mutagenic in other NTases (23)(24)(25). We find that the intrinsic allostery of cGAS solves these potentially graving problems. For instance, all biochemical activities required for efficient signaling are concertedly and selectively maximized with the correct signal, cognate substrates, and a specific metal cofactor ( Figure 6). On the flip side, the same allosteric mechanism also permits greater promiscuity under normal, non-stressful conditions to minimize spurious activation ( Figure 6). For example, unlike other NTases where Mn 2+ favors noncognate NTPs (23)(24)(25), Mn 2+ preferentially accelerates the k cat of ATP/GTP for cGAS ( Figure 3). Second, ATP/GTP is essential for the synergy between Mn 2+ and dsDNA length-dependent activation ( Figure 3). Third, ds-DNA binding invokes anti-cooperativity in utilizing Mn 2+ against either ATP or GTP alone, thereby further suppressing misincorporation of Mn 2+ with noncognate substrates. Fourth, dsDNA length is irrelevant for the catalysis against ATP/ATP or GTP/GTP (Figure 3). Finally, longer dsDNA specifically accelerates the rate-limiting cyclization of cGAMP (Figures 3-5), which also suppresses off-pathway product formation even with Mn 2+ and excess ATP ( Figure 5). Considering that longer dsDNA promotes oligomerization (19) and subsequent liquid-phase condensation (LPC) of cGAS (38), it is tempting to speculate that phase-separated cGAS•dsDNA complexes trap the pppGpA intermediate to facilitate cyclization. Together, all these factors would synergistically maximize host innate immune responses against major intracellular assaults. On the other hand, short (or no) dsDNA decreases the affinity for Mn 2+ (Figures 1 and 2), reduces the cyclization efficiency (Figures 4 and 5), and allows off-pathway reaction with noncognate NTPs to be almost as favorable as the onpathway reaction with ATP/GTP ( Figure 3B). All these actions would then synergistically minimize the signaling activity of cGAS (i.e. viral dsDNA has been degraded by host nucleases). Combined, such a tightly coupled allosteric mechanism would control the host innate immune system in a precise yet efficient manner.
Mg 2+ is thought to be the canonical cofactor for cGASlike NTases (23)(24)(25)39). However, a recent study revealed that Mn 2+ could activate the catalytic efficiency of several bacterial cGAS-like NTases (39); the ligand requirements for these NTases remain largely unknown (39). Considering the highly concerted relationship between the dsDNA ligand and the Mn 2+ cofactor we observe here, future mechanistic studies will reveal whether other NTases with signaling functions might entail similar ligand-cofactor selection mechanisms as cGAS. On the other hand, it is noteworthy that 'short' dsDNA (<40 bp) is relatively abundant in the cytoplasm (1,40,41). We found that a physiologically relevant amount of Mn 2+ (50 M) activated cGAS with minimal (or no) dsDNA when both wild-type and various mutant enzymes were overexpressed in HEK293T cells ( Figure  2J). For instance, K173E/R176E-cGAS FL did not respond to dsDNA in vitro, but showed robust activity in HEK293T cells with Mn 2+ (Figure 2G versus J). Also importantly, the catalytic activity of these cGAS mutants were nonetheless marginal compared to wild-type ( Figure 2H), indicating that cGAS does not have to operate at the maximal catalytic capacity to induce robust IFN-I signaling. Considering that IFN-I drives the overexpression of cGAS via a positive feedback loop (42), it is tempting to speculate that the Mn 2+ -induced activity of cGAS might underpin many type-I interferonopathies that do not involve pathogen infection (i.e. sterile inflammation), which include manganism (32), Parkinson's disease (43,44), and a wide range of autoimmune disorders such as Aicardi-Goutieres syndrome and systemic lupus erythematosus (7)(8)(9)(10).