The catalytic mechanism of cyclic GMP-AMP synthase (cGAS) and implications for innate immunity and inhibition

Article abstract of DOI:10.1002/pro.3304

Cyclic GMP-AMP synthase (cGAS) is activated by ds-DNA binding to produce the secondary messenger 2′,3′-cGAMP. cGAS is an important control point in the innate immune response; dysregulation of the cGAS pathway is linked to autoimmune diseases while targeted stimulation may be of benefit in immunoncology. We report here the structure of cGAS with dinucleotides and small molecule inhibitors, and kinetic studies of the cGAS mechanism. Our structural work supports the understanding of how ds-DNA activates cGAS, suggesting a site for small molecule binders that may cause cGAS activation at physiological ATP concentrations, and an apparent hotspot for inhibitor binding. Mechanistic studies of cGAS provide the first kinetic constants for 2′,3′-cGAMP formation, and interestingly, describe a catalytic mechanism where 2′,3′-cGAMP may be a minor product of cGAS compared with linear nucleotides.

Full text of DOI:10.1002/pro.3304

Article
Protein Science
DOI 10.1002/pro.3304
THE CATALYTIC MECHANISM OF CYCLIC
GMPꢀAMP SYNTHASE (cGAS) AND
IMPLICATIONS FOR INNATE IMMUNITY AND
INHIBITION
†
†
†
†
†
AUTHOR NAMES: Justin Hall *, Erik C. Ralph , Suman Shanker , Hong Wang , Laura J. Byrnes ,
†
Reto Horst , Jimson WongΦ, Amy BraultΦ, Darren DumlaoΦ, James F. SmithΦ, Leslie A. Dakin‡,
†
†
†
†
Daniel C. Schmitt , John Trujillo , Fabien VincentΦ, Matt Griffor and Ann E. Aulabaugh .
†
AUTHOR ADDRESS: Worldwide Medicinal Chemistry, Pfizer, Eastern Point Rd, Groton, CT 06340,
‡ Worldwide Medicinal Chemistry, Pfizer, 610 Main St, Cambridge, MA 02139, Φ Hit Discovery and
Lead Profiling, Pfizer, Eastern Point Rd, Groton, CT 06340, Pfizer Centers for Therapeutic Innovation
(CTI).
CORRESPONDING AUTHOR
*Justin Hall (Justin.Hall@pfizer.com)
39 Pages of text (3 table, 4 figure, 3 SI tables and 4 SI figures in separate files)
This article has been accepted for publication and undergone full peer review but has not been
through the copyediting, typesetting, pagination and proofreading process which may lead to
differences between this version and the Version of Record. Please cite this article as
doi: 10.1002/pro.3304
© 2017 The Protein Society
Received: Aug 10, 2017; Revised: Sep 12, 2017; Accepted: Sep 18, 2017
This article is protected by copyright. All rights reserved.

Protein Science
Page 2 of 52
ABSTRACT: Cyclic GMPꢀAMP synthase (cGAS) is activated by dsꢀDNA binding to produce
the secondary messenger 2ʹ,3ʹꢀcGAMP. cGAS is an important control point in the innate immune
response; dysregulation of the cGAS pathway is linked to autoimmune diseases while targeted
stimulation may be of benefit in immunoncology. We report here the structure of cGAS with
dinucleotides and small molecule inhibitors, and kinetic studies of the cGAS mechanism. Our
structural work supports the understanding of how dsꢀDNA activates cGAS, suggesting a site for
small molecule binders that may cause cGAS activation at physiological ATP concentrations,
and an apparent hotspot for inhibitor binding. Mechanistic studies of cGAS provide the first
kinetic constants for 2ʹ,3ʹꢀcGAMP formation, and interestingly, describe a catalytic mechanism
where 2ʹ,3ʹꢀcGAMP may be a minor product of cGAS compared to linear nucleotides.
KEYWORDS: cGAS, STING, 2ʹ,3ʹꢀcGAMP, cGAMP, OAS1, innate immunity.
SIGNIFICANCE: We describe a novel SPRꢀbased enzymatic assay and demonstrate its utility on
the cGAS/STING system. We expect this assay to be of broad utility as it has the advantages of
being continuous, is compatible with native substrates, and does not require the use of coupling
reactions or fluorescently labeled reagents. Using both our SPR assay and conventional
enzymatic assays, we report the first kinetic parameters for the cGAS enzymatic mechanism
which suggests its canonical product, 2ʹ,3ʹꢀcGAMP , is actually the minor product of this
reaction. We also present several new structures of cGAS with dinucleotides or small molecule
inhibitors from a fragment screen. The results of our structural analysis support an existing
model for the structural basis of DNAꢀinduced cGAS activation that may be inducible by a small
molecule binder. In combination with our activator data, we present the results from a fragment
binding study that suggest the presence of an inhibitor binding hotspot for cGAS.
2
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 3 of 52
Protein Science
INTRODUCTION
Cyclic GMPꢀAMP synthase (cGAS)* is a sensor of cytosolic doubleꢀstranded DNA (dsꢀ
DNA). dsꢀDNA in the cytosol may occur from infection or mitochondrial damage. Metazoans
have developed the ability to sense dsꢀDNA in the cytosol as a trigger for the innate immune
response.1ꢀ3 While interferon and cytokine signaling are warranted to combat infectious dsꢀDNA
contaminants, recent studies have found that selfꢀdsꢀDNA may occur in persisting autoimmune
disorders such as systemic lupus erythematosus (SLE), suggesting inappropriate dsꢀDNA sensing
may be a contributor to autoimmune disease.4ꢀ7
cGAS is activated by dsꢀDNA binding to catalyze the cyclization of ATP and GTP to
form a cyclic dinucleotide with mixed 2ʹ,5ʹꢀ and 3ʹ,5ʹꢀphosphodiester linkage (2ʹ,3ʹꢀcGAMP),
which in turn activates stimulator of type 1 interferon genes (STING).8ꢀ10 Activated STING
causes the activation of TBK1, which phosphorylates IRF3 allowing it to translocate to the
nucleus where it triggers interferonꢀinducible gene activation and interferon production.8,11ꢀ13
Interestingly, intentional activation of STING through intratumoral injections of STING agonists
has demonstrated antitumor properties and immunological memory, suggesting that while cGAS
inhibition may benefit autoimmune patients, cGAS activation may be of therapeutic benefit in
oncology.14ꢀ16
The relationship of cGAS and STING is both old (as much as much as 500 million years
of coꢀevolution),17 and interesting in that cGAS is a lowꢀactivity enzyme while STING is a
particularly avid binder of the cGAS product (Kd ~ 4 nM).18 It has also been noticed by multiple
3
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 4 of 52
investigators that cGAS produces linear homoꢀ and heteroꢀdinucleotides, including the unusual
2ʹ,5ʹꢀphosphodiester linked products GMPꢀ2ʹꢀGTP and AMPꢀ2ʹꢀGTP.19ꢀ20 GMPꢀ2ʹꢀGTP is
thought to be a side reaction while AMPꢀ2ʹꢀGTP is presumed to be the catalytic intermediate
required for 2ʹ,3ʹꢀcGAMP production; if true this is a striking phenomenon as catalytic
intermediates are seldom abundantly produced under physiological conditions, yet that appears
to be the case for AMPꢀ2ʹꢀGTP. OAS1 is a paralog of cGAS, it produces 2ʹ,5ʹꢀoligoadenylate as
a secondary messenger during dsꢀRNAꢀinduced innate immunity;21 similarly, the 2ʹ,5ʹꢀ
phosphodiester link in GMPꢀ2ʹꢀGTP and AMPꢀ2ʹꢀGTP may distinguish them for as yet
unrecognized roles. To better understand the role of cGAS and STING we undertook a study of
the cGAS enzymatic mechanism and its production of linear and cyclic dinucleotides. We report
here a novel SPRꢀbased kinetic analysis of cGAS, which in combination with HPLC, MS, and
NMR assays suggest the majority of the cGAS enzymatic process is a futile cycle in terms of
STING activation. We also present multiple structures of cGAS bound to dinucleotides and small
molecule inhibitors that allows us to expand on the existing theory for cGAS activation and
propose targeted sites for inhibitor or activator binding.
RESULTS
Interaction with Asp227 causes catalytic acid alignment. dsꢀDNA binding causes two
major changes to the apo cGAS secondary structure. The first is residues Gly207ꢀVal218 (Homo
sapiens numbering for changes seen in Mus musculus (PDB 4O6A)22 and Sus scrofa (PDB
4KB6)23) change from disordered to a regular secondary structure (βꢀstrand between Gly207ꢀ
Asn210, αꢀhelix between Gly212ꢀVal218), and the second is a ~1 Å shift of the βꢀsheets containing
4
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 5 of 52
Protein Science
the catalytic acids (Glu225, Asp227 and Asp319) towards the active site (Fig. 1). In the absence of
dsꢀDNA, human cGAS can adopt a cyclic dinucleotideꢀdependent structure similar to the second
of these structural changes, where the catalytic acid containing βꢀsheets have moved towards the
active site (see PDB 4O67 and 4O69)22 while residues Gly207ꢀVal218 remain disordered. Since
only the shift in the βꢀsheets has occurred in this dinucleotideꢀdependent structural change, we
shall distinguish this conformation from the fully active form, referring to it as “βꢀpseudoꢀactive”
for the changes in the βꢀsheets. To study the βꢀpseudoꢀactive form we obtained structures of an
Nꢀterminal truncation of cGAS starting at residue 161 (cGAS161)22 in complex with five cyclic
dinucleotides (2ʹ,2ʹꢀcGAMP, 2ʹ,3ʹꢀcGAMP, 3ʹ,3ʹꢀcGAMP, 3ʹ,3ʹꢀcdIMP and 3ʹ,3ʹꢀcdUMP), and
the linear 2′,5′ꢀGpAp dinucleotide (Fig. S1, Table S1, S2). In four of these structures, cGAS
assumes the βꢀpseudoꢀactive conformation (Table 1).
There is always an interaction between the catalytic acid Asp227 and the dinucleotide
when the βꢀpseudoꢀactive conformation occurs. In most cases Asp227 forms a hydrogen bond
with the amino group of the guanine base with an approach between 2.5 and 3.3 Å (Fig. 1). The
exception to this is 3ʹ,3ʹꢀcdIMP which does not have an amino group on its base, but does
interact with Asp227 through the 2ʹꢀOH guanine ribose (3.2 Å).
There is no interaction between Asp227 and 3ʹ,3ʹꢀcdUMP or 2ʹ,2ʹꢀcGAMP; in these
structures cGAS retains an inactive pose. For 3ʹ,3ʹꢀcdUMP there is no amino group on its base,
and the smaller pyrimidine does not penetrate as deeply into the active site as purines. 2ʹ,2ʹꢀ
cGAMP contains an amino group on its guanine base; however, 2ʹ,2ʹꢀcGAMP is considerably
5
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 6 of 52
displaced from the active site compared to 2ʹ,3ʹꢀcGAMP or 3ʹ,3ʹꢀcGAMP (Fig. 1) with its amino
group interacting with Asp319 (2.9 Å) instead of Asp227.
Most of these dinucleotides have affinities weaker than 500 ꢁM (the top concentration
tested). 2ʹ,3ʹꢀcGAMP and 3ʹ,3ʹꢀcGAMP are more tightly bound than the other dinucleotides, and
we were able to detect a modest (~2ꢀfold) increase in affinity for both between apo cGAS161 and
dsꢀDNAꢀbound cGAS (Table 1, Fig. S2). The change in affinity is consistent with the
preordering of Asp227 by dsꢀDNA.
The substrate orientation is thermodynamically preferred for 2ʹ,3ʹꢀcGAMP. Structures of
dsꢀDNA bound to cGAS shows dsꢀDNA packs against residues between Gly207ꢀVal218, inducing
their active conformation.22,23 In general, Gly207ꢀVal218 do not adopt a regular secondary structure
without dsꢀDNA, though strong electron density can be seen for these residues with some
ligands. In our 2ʹ,3ʹꢀcGAMP structure, the electron density is sufficient to model Gly207ꢀVal218.
Comparison to the existing human cGAS 2ʹ,3ʹꢀcGAMPꢀbound structure (PDB 4O67) shows good
agreement, the only exception is for the modeling of residues Ile220 and Ser221. In our structure
we observe a close contact between Lys219 and Ala222 that could be modeled as a continuation of
the main chain into a βꢀturn, which is the case for 4O67; however, electron density of side chains
in our structure make it clear there is actually a break in the main chain and not a βꢀturn. 4O67 is
the only structure of cGAS from any organism that models a βꢀturn at Ile220 and Ser221 (Fig. S3).
Our structure, 4O67, and a M. musculus structure with both dsꢀDNA and 2ʹ,3ʹꢀcGAMP
(PDB 4K9B)19 all model 2ʹ,3ʹꢀcGAMP with the adenine base above Tyr436, and the guanine base
6
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 7 of 52
Protein Science
near Leu209. This is the same position seen for the adenine and guanine base in the structure of
substrateꢀbound cGAS (PDB 4KB6), we therefore refer to this base orientation for 2ʹ,3ʹꢀcGAMP
as being in the “substrate orientation”.
A second dsꢀDNAꢀ and 2ʹ,3ʹꢀcGAMPꢀbound M. musculus structure (PDB 4LEZ)24 has
the guanine and adenine bases switched relative to the positions in our model, 4O67, and 4K9B.
The discrepancy between 4LEZ and other models may be due to how 2ʹ,3ʹꢀcGAMP was
introduced to the 4LEZ crystals. In our and the 4O67 structure, 2ʹ,3ʹꢀcGAMP was added to apo
enzyme, while in 4LEZ, 2ʹ,3ʹꢀcGAMP was generated enzymatically in the crystal from ATP and
GTP. Since the AMPꢀ2ʹꢀGTP intermediate must switch base positions during cyclization, there is
the interesting possibility that the electron density captured in the 4LEZ structure represents a
mixture of reaction intermediate and final product. Thus, it would appear that the pose of 2ʹ,3ʹꢀ
cGAMP with adenine above Tyr436 is the thermodynamically favored position occurring after
product rebinding.
Tyr436 and Arg376 form a binding site for aromatic rings. A screen of the Pfizer fragment
chemical library discovered several binders of cGAS. Fragment screens are run to find low
affinity chemical leads, which are subsequently developed into more potent drugꢀlike inhibitors.
All the fragments described here bind at the active site and are weak inhibitors, the development
of one of our fragment hits from a weak (~200 ꢁM) to a potent (~200 nM) inhibitor is described
elsewhere.25 We detail here three fragments hits for their insights into the cGAS activation
mechanism and direct interested readers to our other publication for the development of a
fragment hit to a potent inhibitor of cGAS.25
7
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 8 of 52
Each of the fragment binders shown here (F1, F2, and F3) are small (~150ꢀ250 Da) with
weak affinity (~100ꢀ300 ꢁM) (Table S3, Fig. S2). The FoꢀFc maps for F2 and F3 are not
sufficient to unambiguously define atomic positions, which seems to be related to multiple
modes of binding and internal pseudoꢀsymmetry. Indeed, compared to the dinucleotide
structures which have specific interactions that define their positions shows far less ambiguity in
unbiased FoꢀFc maps, despite these dinucleotides being generally weaker binders than the
fragments and at comparable resolutions (Fig. 1, S1).
Each compound binds to the same site formed between the side chains of Tyr436 and
Arg376 (Fig. 1). This site is occupied by the adenine base in structures of ATP, 2ʹ,2ʹꢀcGAMP,
2ʹ,3ʹꢀcGAMP, or 3ʹ,3ʹꢀcGAMP, and is composed primarily of London dispersion interactions
between the ring system of these binders with the phenyl ring of Try436. These interactions
suggest this site could accommodate most 2ꢀ or 3ꢀring aromatic systems, which is consistent with
this site needing to bind both the adenine and guanine bases during catalysis. Additionally, we
observe that in all our cyclic dinucleotides structures, this site had better defined electron density
than the other nucleobase site, even for identical ring systems like 3ʹ,3ʹꢀcdIMP and 3ʹ,3ʹꢀcdUMP.
The binding site formed by Tyr436 and Arg376 is therefore a site of binding with broad specificity
for aromatic rings, be they nucleobases or small molecule fragments.
cGAS can form an αꢀhelix at Gly212ꢀVal218 in the absence of dsꢀDNA. Compounds F1ꢀF3
bind distant (~10 Å) from Asp227 and do not elicit the βꢀpseudoꢀactive conformation. For
compounds F1 and F2, one chain of the asymmetric unit (chain A) has strong electron density for
Gly207ꢀVal218, which adopts a short βꢀstrand at Gly207ꢀAsn210, and an αꢀhelix at Gly212ꢀVal218
8
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 9 of 52
Protein Science
while the other chain in the asymmetric unit does not. Compound F3 has the same conformation
but weaker density for these residues. Interestingly, while 3ʹ,3ʹꢀcdUMP does not cause a βꢀ
pseudoꢀactive conformation, one molecule of the asymmetric unit (chain B) adopts the same
conformation for Gly207ꢀVal218 seen for compounds F1ꢀF3. Thr211 of the short βꢀstrand is
important for coordinating the nucleobase and determining the specificity of the 2ʹ or 3ʹꢀOH
bond formation (20), while Ser213 of the αꢀhelix may bind phosphates of the linear intermediate
(see PDB 4K99 and 4K9A).19 These structural changes are therefore critical for cGAS activity.
The formation of the short βꢀstrand and αꢀhelix occur in the dsꢀDNAꢀbound structures of
cGAS, yet there are no molecular contacts to explain their presence in these structures. Thus,
these data seem to describe a naturally occurring propensity for this conformation, which is
enriched in one of the two chains in the asymmetric unit of these crystals.
These structures demonstrate the ability of cGAS to form an activeꢀlike conformation for
Gly207ꢀVal218 while the catalytic acid containing βꢀsheets remain in the inactive pose. This is the
opposite effect seen for the βꢀpseudoꢀactive conformation. We call this second pseudoꢀactive
conformation “αꢀpseudoꢀactive” in reference to the αꢀhelix at residues Gly212ꢀVal218. That
distinct αꢀ and βꢀpseudoꢀactive states exist suggests these two conformations are either
independent or mutually exclusive, requiring DNAꢀbinding to coordinate both conformation
transitions (Fig. 2).
An SPRꢀbased enzymatic assay to determine kinetic constants. The Biacore T200 SPR
microfluidics were designed with a serpentine flow; samples injected must pass along all four
9
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 10 of 52
flow cells during data collection. We reasoned that if we immobilized apo cGAS, dsꢀDNA bound
cGAS, and STING155ꢀ341 in series we could inject ATP and GTP, detect binding to apo cGAS,
form 2ʹ,3ʹꢀcGAMP with dsꢀDNA bound cGAS, and finally detect 2ʹ,3ʹꢀcGAMP using STING155ꢀ
341 as a sensor [Fig. 3(A)].
KM values reflect the steadyꢀstate equilibrium between free enzyme and all other enzyme
species. In theory, KM values for DNAꢀbound cGAS161 could be determined by directly
monitoring the SPR response of this channel as a function of ATP and GTP. However, the dsꢀ
DNA bound cGAS161 channel had a nearꢀzero response (RU) signal for all concentrations of
ATP and GTP which made a direct KM calculation from the cGAS channel impossible.
Monitoring the STING155ꢀ341 response showed 2ʹ,3ʹꢀcGAMP was being produced by dsꢀDNA
bound cGAS; thus we used the STING response to determine KM in lack of a direct cGAS
response. Though STING was necessary for monitoring substrate turn over in this system, we
suspect direct analysis of the change in RU of the enzyme channel should work for other
systems.
While binding is generally seen as a positive increase in SPR response, it has been
observed that for some proteins a negative signal can occur which is attributed to conformation
changes associated with binding.26 The dsꢀDNA bound cGAS response appears to report the
summation of positive (mass accumulation) and negative (conformation changes) responses
occurring during catalysis. Since mass accumulation and conformational changes should be
distinct with distinct enzymes, we suspect this zeroꢀsum phenomenon will not occur for other
enzymes.
10
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 11 of 52
Protein Science
We have determined there are criteria that must be met for SPR to be used for activity
assays (see Discussion). One of the most important is that the sensor protein must bind and give
a measurable response for the desired analyte while being insensitive to other chemical matter
present in the assay samples (see below). STING binds 2ʹ,3ʹꢀcGAMP with nM affinity (4 nM
reported Kd),18 and gave a clear positive response when titrated with commercial 2ʹ,3ʹꢀcGAMP
standards. No response was observed when ATP or GTP was injected in the absence of the other
nucleotide indicating STING155ꢀ341 does not bind ATP or GTP at the concentrations used here.
AMPꢀ2ʹꢀGTP has been previously observed in cGAS reactions and was an important clue
to seminal studies on the cGAS enzymatic mechanism [Fig. 4(F)]19 where it was described as a
pathway intermediate. To determine if STING155ꢀ341 binds the intermediate, we used full length
cGAS to prepare four reaction mixtures with variable concentrations of 2ʹ,3ʹꢀcGAMP and AMPꢀ
2ʹꢀGTP. Samples were tested in SPR and the 2ʹ,3ʹꢀcGAMP concentration was determined from
the STING response compared to 2ʹ,3ʹꢀcGAMP standards. The total SPR response on the STING
channel was in close agreement to the expected result for total 2ʹ,3ʹꢀcGAMP as measured by
NMR, with no additional response observed from AMPꢀ2ʹꢀGTP (Table 2), even when
intermediate was present at 10ꢀfold excess over 2ʹ,3ʹꢀcGAMP. The simplest explanation for these
data is AMPꢀ2ʹꢀGTP does not bind to STING155ꢀ341 at the top concentrations tested here (~50
ꢁM). We therefore concluded that STING can be used as a selective sensor for 2ʹ,3ʹꢀcGAMP
without interference from ATP, GTP or AMPꢀ2ʹꢀGTP.
11
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 12 of 52
KM values for 2ʹ,3ʹꢀcGAMP production were determined using SPR as well as HPLC,
MS, and NMR assays. Values compiled in Table 3 show good agreement between techniques,
and between both full length cGAS and cGAS161.
Using the SPR method, the apparent ATP KM value was determined using a range of
GTP concentrations. Using an extra sumꢀofꢀsquares Fꢀtest, a GTP concentrationꢀindependent
KM.ATP value was supported over a GTP dependent KM value (p = 0.18), suggesting a lack of
cooperativity between ATP and GTP binding. Similarly, ATP binding to apo cGAS161 showed
no dependence on the concentration of GTP. Similar Kd and KM values for ATP (235 ± 97 and
190 ± 20 ꢁM) are consistent with a subsequent step occurring at a rate slower than the rate of
substrate association/dissociation (eg slow conformational changes or chemistry). We observed
no RU signal for GTP binding to apo cGAS161 (up to 1 mM GTP), including conditions where
ATP was preꢀbound to cGAS161 (up to 2 mM ATP). The simplest explanation of these data is
that apo cGAS161 does not bind GTP tightly; however, the absolute RU change for ATP was
muted compared to nonꢀsubstrate analytes [Fig. S2(E)]. We therefore cannot rule out that GTP
binding results in a conformation change and a netꢀzero RU signal.
Substrate inhibition occurs from competitive side reactions. ATP titrations show a clear
substrate inhibition pattern for both cGAS161 in SPR, and full length cGAS in HPLC and MS
assays. Using SPR, the apparent KIS value increased with increasing GTP concentrations.
However, no substrate inhibition was seen at up to 2 mM ATP when the ratio of ATP and GTP
was constant, suggesting substrate competition occurs between ATP and GTP.
12
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 13 of 52
Protein Science
The apparent substrate inhibition is most simply explained by the two nucleotides
competing with each other to form AMPꢀ3ʹꢀATP or GMPꢀ2ʹꢀGTP instead of 2ʹ,3ʹꢀcGAMP. In
agreement with this, cGAS has been observed to produce both AMPꢀ3ʹꢀATP and GMPꢀ2ʹꢀGTP
(Fig. S4).19,20 We found GTP behaved as a competitive inhibitor of AMPꢀ3ʹꢀATP formation
using cGAS161 or full length cGAS in HPLC titrations. This substrate inhibition is the result of a
randomꢀordered, biꢀsubstrate reaction, where one reactant can compete with the second for
binding. This also explains the GTPꢀconcentration dependence of the ATP KIS values. As
mentioned above, when ATP and GTP are kept at a constant molar ratio, no substrate inhibition
was observed because the ratio of 2ʹ,3ʹꢀcGAMP to homoꢀdinucleotides would also remain
constant. The apparent KM for AMPꢀ3ʹꢀATP formation in the absence of GTP was 3700 ± 1200
ꢁM.
Linear homoꢀ and heteroꢀdinucleotides are the major initial products of cGAS. Though
we first identified the formation of AMPꢀ3ʹꢀATP and GMPꢀ2ʹꢀGTP in HPLC and MS by running
control reactions with single nucleotide substrates, we could also observe both products in MS
experiments under physiologically relevant concentrations of ATP and GTP (2 mM and 0.5
mM)27 with cGAS161 or full length cGAS. Exact dinucleotide concentrations were not possible to
determine lacking MS ionization controls, but peak intensities suggests the homoꢀlinear products
are ~10% of the total reaction (Fig. S4). Similarly, the AMPꢀ3ʹꢀATP product could be detected at
all nonꢀzero GTP concentrations tested using HPLC [Fig. 4(D)], and at physiological substrate
concentrations,27 comprised ~20% of the product compared to 2ʹ,3ʹꢀcGAMP for full length
cGAS. The GMPꢀ2ʹꢀGTP peak was only observed in HPLC in the absence of ATP, suggesting
this product is not as readily formed compared to AMPꢀ3ʹꢀATP.
13
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 14 of 52
In addition to the homoꢀlinear products, we observed the production of linear AMPꢀ2ʹꢀ
GTP using continuous reaction monitoring of full length cGAS in NMR and reactionꢀarrested
MS and HPLC. In our studies, AMPꢀ2ʹꢀGTP is produced in significant abundance initially, while
2ʹ,3ʹꢀcGAMP is not [Fig. 4(AꢀC)]; this result is independent of the cGAS form used in the study
(ie cGAS161 or full length). These data strongly suggest a mechanism where the majority of
substrate flows through AMPꢀ2ʹꢀGTP, which is released to solution and then must rebind in
competition with ATP and GTP to form 2ʹ,3ʹꢀcGAMP. Consistent with this mechanism, there is a
lag in the initial rate of 2ʹ,3ʹꢀcGAMP formation, but no lag in the initial rate of AMPꢀ2ʹꢀGTP
formation at saturating ATP and GTP concentrations (Fig. 4).
cGAS can produce 2ʹ,3ʹꢀcGAMP without releasing AMPꢀ2ʹꢀGTP to solution. In a closed
system (such as an NMR tube), AMPꢀ2ʹꢀGTP released from the enzyme will accumulate until it
reaches a concentration sufficient to compete with ATP and GTP for rebinding to free enzyme.
Thus, end point activity assays may miss the difference in AMPꢀ2ʹꢀGTP and 2ʹ,3ʹꢀcGAMP
concentrations at early time points while continuous methods like NMR will not. In contrast to a
closed tube, the continuous flow through the SPR system limits the concentration of AMPꢀ2ʹꢀ
GTP that can accumulate according to the rate of production and the rate of flow through the
cell. Thus, as ATP and GTP concentrations increase past their KM in SPR they will competitively
block AMPꢀ2ʹꢀGTP rebinding to cGAS, decreasing 2ʹ,3ʹꢀcGAMP production. In effect this would
look like substrate inhibition. To separate substrate inhibition of 2ʹ,3ʹꢀcGAMP production due to
ATP and GTP competing with AMPꢀ2ʹꢀGTP, from inhibition by side reactions (eg AMPꢀ3ʹꢀATP
and GMPꢀ2ʹꢀGTP formation), we used a fixed ratio of ATP to GTP with final concentrations up
to 25ꢀfold their KM and did not observe inhibition of 2ʹ,3ʹꢀcGAMP production [Fig. 4(E)]. These
14
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 15 of 52
Protein Science
data are consistent with a catalytic mechanism where AMPꢀ2ʹꢀGTP is not subject to ATP or GTP
competition; this is most easily explained by a mechanism where AMPꢀ2ʹꢀGTP is not released
but instead reorients on the enzyme.
These SPR data appear to be in conflict with the NMR data which demonstrate a large
accumulation of AMPꢀ2ʹꢀGTP in solution. Reconciliation of the NMR and SPR data results in a
mixed mechanism, where the majority of AMPꢀ2ʹꢀGTP dissociates and can rebind after enough
has accumulated to compete, but where a minor portion reorients directly onꢀenzyme to form
2ʹ,3ʹꢀcGAMP in a process that is not competitive with ATP or GTP [Fig. 4(F), red arrows].
NMR experiments using full length cGAS and physiological concentrations of ATP and
GTP (2 mM and 0.5 mM, 10ꢀfold their KM)27 show ~250 ꢁM AMPꢀ2ʹꢀGTP accumulates at
steadyꢀstate [Fig. 4(A)]. These data suggest either slow chemistry occurs for AMPꢀ2ʹꢀGTP loss
relative to formation, or the KM of AMPꢀ2ʹꢀGTP is around 25 ꢁM. Consistent with a weak KM,
we did not observe binding for AMPꢀ2ʹꢀGTP at a top concentration of 50 ꢁM using apo cGAS161
in SPR.
Despite the apparent weak affinity of AMPꢀ2ʹꢀGTP, and its competition with ATP and
GTP, this may not have a significant biological effect since STING is a tight binder of 2ʹ,3ʹꢀ
cGAMP (4 nM Kd)18 requiring only a small amount to cause signaling. The SPR results
demonstrate a mechanism exists where a fraction of AMPꢀ2ʹꢀGTP is not subject to ATP or GTP
competition, such as through reorientation on the enzyme. Based on the full length NMR steadyꢀ
state concentration of AMPꢀ2ʹꢀGTP, a rough approximation suggests that if the fraction of AMPꢀ
15
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 16 of 52
2ʹꢀGTP that stays onꢀenzyme is greater than 0.1%, the onꢀenzyme path should produce the
requisite levels of 2ʹ,3ʹꢀcGAMP needed for STING sensing before enough AMPꢀ2ʹꢀGTP
accumulates to compete with ATP and GTP for enzyme rebinding. Thus AMPꢀ2ʹꢀGTP appears to
be both the major initial product of cGAS and its release is a futile cycle in terms of 2ʹ,3ʹꢀ
cGAMP production and STING activation.
DISCUSSION
During catalysis cGAS must accommodate a swap of adenine and guanine nucleobases in
its active site. This is enabled by Tyr436 and Arg376, which create a binding site for disparate
aromatic rings. Since this site is essential for binding of both substrates and intermediate, small
molecules binding this site cause enzyme inhibition (Fig. 1).25
When analyzing the inactiveꢀtoꢀactive transition of cGAS, we see there are αꢀ and βꢀ
pseudoꢀactive conformations that mimic the true dsꢀDNAꢀdependent active state. The αꢀpseudoꢀ
active state occurs without clear ligand provocation, suggesting it may be a regularly occurring
state in the absence of dsꢀDNA, whereas the βꢀpseudoꢀactive state is observed when Asp227 is
engaged.
In the absence of dsꢀDNA, we only observe αꢀ and βꢀpseudoꢀactive states separately,
suggesting these states are either independent or mutually exclusive. Our structural analysis
supports a model where αꢀ and βꢀpseudoꢀactive states are mutually exclusive, which is consistent
with a model for activation proposed by Civril et al.23 According to this model, when dsꢀDNA
16
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 17 of 52
Protein Science
binds, it breaks the long Nꢀterminal helix of cGAS into two daughter helices (α1 and α2) at
Ser175 (Fig. 2). This break is observed in dsꢀDNAꢀbound structures where the positive dipole of
the α2 helix interacts with the DNA backbone,19,23 but also occurs in the cyclic dinucleotideꢀ
induced βꢀpseudoꢀactive states. The potential importance of the long Nꢀterminal helix formed the
basis for a Leu174Asn mutant by Civril et al. (in recognition of its importance they refer to this
helix as the “spine” of cGAS) where they hypothesized the helical break allows Leu174 to
stabilize the active form through the formation of an αꢀhelix at Gly212ꢀVal218. Strikingly, they
showed Leu174Asn is able to bind DNA but no longer produces 2ʹ,3ʹꢀcGAMP, these results are
consist with the idea that αꢀ and βꢀpseudoꢀactive confirmations are mutually exclusive. Similarly,
others have found amino acid substitutions along this helix (eg Lys173Ala/Arg176Ala in H.
sapiens or Arg158Ala in M. musculus) greatly reduce catalytic activity.23,24
In the αꢀpseudoꢀactive structures, the long Nꢀterminal helix is intact and the singleꢀturn αꢀ
helix is formed at Gly212ꢀVal218. When dsꢀDNA breaks the long Nꢀterminal helix, it also
positions the Cꢀterminal end of the daughter helix α2 towards dsꢀDNA, aligning Gly207ꢀAsn210 to
form a short βꢀstrand with Val228ꢀLys231. The importance of the short βꢀstrand for cGAS activity
has been described by others, who liken it to the “activation loop” of kinases.22 Though dsꢀDNA
binding may also help form the αꢀhelix at Gly212ꢀVal218 through packing, we consistently observe
that the short βꢀstrand at Gly207ꢀAsn210 occurs with the formation of the αꢀhelix at Gly212ꢀVal218,
suggesting these structures are linked. In the βꢀpseudoꢀactive structures, daughter helix α2 is
pulled away from the dsꢀDNA binding site, suggesting the dsꢀDNA interaction is needed to
position the daughter helix α2 and facilitate the formation of the βꢀstrand at Gly207ꢀAsn210. Thus,
αꢀ and βꢀpseudoꢀactive states seem mutually exclusive, with the long Nꢀterminal helix acting like
17
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 18 of 52
a spring to pull the αꢀ and βꢀpseudoꢀactive states apart, an effect that is removed when dsꢀDNA
binding breaks this helix (Fig. 2).
The αꢀ and βꢀpseudoꢀactive states observed here are therefore consistent with existing
data and support the Civril et al. model for dsꢀDNA activation. Interestingly, the 4KB6 structure
shows Asp227 (Asp202 in S. scrofa numbering) interacting with Mg2+ and the αꢀphosphate
oxygens of ATP. Since the physiological concentration of ATP is ~10ꢀfold its Kd or KM, these
data suggest cGAS may be in the βꢀpseudoꢀactive state in cells with the long Nꢀterminal helix
broken. If so, the element missing for cGAS activation would be the induction and alignment of
the short βꢀstrand at Gly207ꢀAsn210 and the αꢀhelix at Gly212ꢀVal218. We therefore propose efforts
to stimulate the innate immune response through cGAS should focus on the discovery of binders
that facilitate the formation of the short βꢀstrand at Gly207ꢀAsn210 and the αꢀhelix at Gly212ꢀVal218.
Since we have always observed both the short βꢀstrand at Gly207ꢀAsn210 and the αꢀhelix at Gly212ꢀ
Val218 to occur simultaneously in our αꢀꢀpseudoꢀactive states, it may be possible to induce both
these structures through stabilizing the short βꢀstrand at Gly207ꢀAsn210 in the presence of high
concentrations of ATP. We therefore envision a small molecule binder that could mimic the
phosphate backbone interactions of dsꢀDNA in aligning the end of the daughter helix α2 after the
Nꢀterminal helix is broken, this should induce a short βꢀstrand at of Gly207ꢀAsn210, thus
stimulating cGAS activation at high ATP concentrations.
In addition to structural studies, we have engaged in an analysis of the catalytic
mechanism of cGAS. These studies include a novel SPRꢀbased enzymatic assay that should be
applicable to other systems (Fig. 3). In these experiments, dsꢀDNA bound cGAS did not show a
18
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 19 of 52
Protein Science
direct RU effect during catalysis, necessitating the use of STING155ꢀ341 as a sensor protein for
2ʹ,3ʹꢀcGAMP. While it is possible to use other binding sensors, such as antibodies tailored to
detect specific analytes, a prior SPRꢀbased catalytic assay showed a nonꢀzero RU effect for an
enzyme undergoing catalysis;28 it is therefore reasonable to speculate a direct use of SPR for
enzymology studies should be possible for other systems.
Compared to the NMR and HPLC assays used here, SPR has the advantage of being a
continuous and relatively quick data collection method. Its disadvantages are that it cannot
separate side reactions from the main reaction without associated sensor proteins (eg STING155ꢀ
341 here), and it is hard to quantitate turnover rates. Furthermore, since SPR systems have a
continuous flow, their use for enzymology must meet certain constraints: 1) as KM is a steadyꢀ
state measurement, the analyte produced must reach steadyꢀstate response (RU plateau) within
the experiment; 2) the same flow rate must be used for all injections; and 3) the reaction cannot
proceed exclusively through intermediates with weak affinities that will be lost to the flow.
cGAS produces AMPꢀ2ʹꢀGTP as an intermediate to 2ʹ,3ʹꢀcGAMP. Full length cGAS
experiments using NMR and HPLC clearly demonstrate that AMPꢀ2ʹꢀGTP is a significant initial
product of cGAS in a closed reaction vessel. These data are consistent with prior studies which
also show linear dinucleotides are produced by cGAS;19,20 however, these data are distinct in that
they are from continuous assays capable of probing early time points to distinguish the relative
rates of formation of linear v cyclic dinucleotide. Furthermore, using SPR we demonstrate 2ʹ,3ʹꢀ
cGAMP can be produced through a process that is not competitive with ATP and GTP [Fig.
4(E)]. The simplest explanation of the NMR, HPLC and SPR data is that while the majority of
19
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 20 of 52
AMPꢀ2ʹꢀGTP product is released, a portion can instead reorient in the cGAS active site to
produce 2ʹ,3ʹꢀcGAMP in a mechanism that is not competitive with ATP or GTP.
NMR and HPLC experiments using full length cGAS demonstrate AMPꢀ2ʹꢀGTP or linear
homoꢀnucleotides are the major initial products at physiological concentrations of ATP and GTP
[Fig. 4(D)]. To our knowledge no one has yet demonstrated the presence of AMPꢀ2ʹꢀGTP in
cells, though many have now demonstrated its presence in vitro. Our protein was produced
without any post translational modifications, it is possible that such modifications, or multiple
localization of cGAS upon long segments of dsꢀDNA in cells,29 could lead to a lower fraction of
linear nucleotides produced. Indeed, our SPR data could be thought of as a mimic for the dsꢀ
DNA localized condition. However, lacking cell data, we can still say these HPLC and MS data
show linear homoꢀdinucleotides are readily produced at high (> 1 mM) concentrations of a single
nucleotide and do not elicit an SPR binding response for STING155ꢀ341, suggesting STING155ꢀ341
does not bind AMPꢀ3ʹꢀATP or GMPꢀ2ʹꢀGTP. Furthermore, neither apo cGAS161 nor STING155ꢀ
341 binds AMPꢀ2ʹꢀGTP at up to 50 ꢁM. While weak AMPꢀ2ʹꢀGTP binding may be dismissed as
an artifact of cGAS truncation or the SPR system, these data are borne out by an apparent weak
affinity for dsꢀDNAꢀbound full length cGAS in solution NMR experiments. That AMPꢀ2ʹꢀGTP
has relatively weak affinity for cGAS is supported by its steadyꢀstate concentration of 250 ꢁM at
physiologically relevant (~10ꢀfold their KM) ATP and GTP concentrations [Fig. 4(A)].27 If the
catalytic efficiency (kcat/KM) of AMPꢀ2ʹꢀGTP formation and depletion are equivalent the
intermediate would have a KM around 25 ꢁM.
20
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 21 of 52
Protein Science
The apparent reconciliation of the weak affinity of AMPꢀ2ʹꢀGTP with the importance of
producing 2ʹ,3ʹꢀcGAMP for immune signaling is that STING is a strong binder of 2ʹ,3ʹꢀcGAMP,
thus only a small amount of 2ʹ,3ʹꢀcGAMP is need to activate STING. SPR shows a portion of the
cGAS161 reaction occurs without competition with ATP and GTP. If the portion of 2ʹ,3ʹꢀcGAMP
produced through a nonꢀcompetitive mechanism is greater than 0.1% of the cellular
concentration of ATP (~ 2 mM), this minor portion will produce the requisite concentration of
2ʹ,3ʹꢀcGAMP needed for STING signaling before the AMPꢀ2ʹꢀGTP released from cGAS can
accumulate to sufficiently compete with ATP and GTP for enzyme rebinding.
Given their unusual 2ʹ,5ʹꢀphosphodiester bond, which would distinguish them from
normal RNAꢀlike oligos, there is a possibility AMPꢀ2ʹꢀGTP and GMPꢀ2ʹꢀGTP have nonꢀSTING
binding partner. However, until such a partner is identified it would seem the large amount of
linear homoꢀnucleotides and AMPꢀ2ʹꢀGTP released compared to 2ʹ,3ʹꢀcGAMP is part of a futile
cycle for cGAS. In conclusion, given the low basal activity of cGAS,18,19,30 its apparent bias to
not bind GTP, and therefore not produce the unusual 2ʹ,5ʹꢀphosphodiester bond, and that much of
its activity seems to be lost in apparent futile cycles involving homoꢀ or heteroꢀdinucleotide
release, this enzyme seems to contain several levels of control to limit 2ʹ,3ʹꢀcGAMP production
in the absence of stimulatory dsꢀDNA.
MATERIALS AND METHODS
Protein expression. The genes for full length Homo sapiens cGAS, Nꢀterminal truncated
cGAS beginning at residue 161 (cGAS161), and H. sapiens STING residues 155 through 341
21
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 22 of 52
(STING155ꢀ341) were ordered from GeneWiz (South Plainfield, New Jersey). Genes were cloned
into pET28 containing an Nꢀterminal SUMOꢀHIS6 tag, a ULP1 cleavage site, a BIRA
recognition sequence, and a TEV cleavage site.
Escherichia coli BL21(DE3) cells transformed with the above constructs were grown in
LB medium (Invitrogen) at 37°C to an OD600nm of 0.8 before inducing protein expression with
0.1 mM isopropylꢀ1ꢀthioꢀβꢀdꢀgalactopyranoside at 15°C for 16ꢀ20 hrs. Harvested cells were
suspended in 20 mM HEPES pH 7.5, 1 M NaCl, 10% (v/v) glycerol, 30 mM imidazole and 1
mM TCEP, and gently killed using a Branson Ultrasonic Disintegrator (VWR Scientific
Products, Chicago, IL) with 7 rounds of 10 s 10% duty cycle sonication separated by 50 s rest
periods.
The soluble fraction was separated using centrifugation (30,000 RCF, 1 h), applied to a
HisTrap FF column (GE Healthcare), washed with 10 column volumes of buffer containing 20
mM HEPES pH 7.5, 250 mM NaCl, 10% glycerol, and 1 mM TCEP containing 30 mM
imidazole, then eluted in the same buffer but with 300 mM imidazole and 300 mM NaCl. The
protein was concentrated using a 10 kDa MWCO Amicon spin column (Millipore), buffer
exchanged into 50 mM HEPES pH 7.5, 250 mM NaCl, 10% (v/v) glycerol and 1 mM TCEP. For
SPR studies, cGAS161 or STING155ꢀ341 was treated with ULP1 and BirA ligase (Avidity) to
generate Nꢀterminal biotinꢀtagged protein; otherwise, samples were incubated for 16ꢀ20 h with
TEV protease (Life Technologies) to liberate untagged protein. Proteaseꢀtreated samples were
passed through a HisTrap FF column to remove tags and residual tagged protein. Full length
cGAS or cGAS161 was applied to a Heparin FF column in 20 mM HEPES pH 7.5, 250 mM
22
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 23 of 52
Protein Science
NaCl, 10% glycerol, and 1 mM TCEP, then eluted with a gradient of 0.25 to 1 M NaCl. All
proteins were subjected to a final purification step using a HiLoad Superdex75 column (GE
Healthcare) in 20 mM HEPES pH 7.5, 150 mM KCl and 1 mM TCEP. Protein purity was
verified by SDSꢀPAGE and ESIꢀTOF mass spectrometry.
Crystallization and structure determination. Dinucleotides were purchased (InvivoGen).
Compounds F1, F2 and F3 were discovered as the result of an NMRꢀbased fragment screen (see
Hall et al. for general description of library and methods31). Crystals of cGAS161 were grown
using conditions similar to a previous report.30 Protein was concentrated to 6 mg/ml, and then
mixed at a 2:1 ratio with PEG 3350 (18ꢀ20% v/v), 0.2 M ammonium citrate pH 7 in a sitting
drop well at 4°C. Rodꢀshaped crystals were observed within 2 days, and grew to their final size
within 5ꢀ7 days. Cryoprotectant was made using mother liquor at a final concentration of 23%
PEG 3350. Compounds were dissolved into cryoprotectant at 50 mM, and soaks were performed
at 4°C for 5ꢀ10 min. Crystals were flash frozen in liquid nitrogen, and data were collected at the
Argonne National Lab (IMCA) beamline. Data were scaled and merged using AIMLESS.32
Initial phases were obtained from MR using PDB 4LEV in PHASER.33 Refinement was
performed using BUSTERꢀTNT and Phenix Refine. Omit maps were calculated using Phenix
Refine with simulated annealing.
HPLC assays. GTP titrations experiments were performed using 500 nM full length
cGAS, 1 ꢁM interferon stimulatory dsꢀDNA (ISD) (Integrated DNA Technologies) in 10 mM
HEPES pH 7.5, 140 mM NaCl, 5 mM MgCl2 and 0.01% Tweenꢀ20 at 37°C. ATP was held at 1.1
mM, while GTP was titrated down from 1.1 mM to 80 ꢁM using a 60% dilution series. Reactions
23
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 24 of 52
were quenched with 50 mM EDTA and separated on a Zorbax SBꢀC8 column (5 ꢁm, 4.6x150
mm) using a methanolꢀphosphate gradient (buffer A: 20 mM potassium phosphate, pH 6.0;
buffer B: equal volumes methanol and buffer A) at 35°C. Peaks were identified using ATP,
GTP, and 2ʹ,3ʹꢀcGAMP chemical standards. 2ʹ,3ʹꢀcGAMP peak areas were converted to molar
concentrations using a standard curve. Formation of this product was fit to a timeꢀdependent
approach to steadyꢀstate using equation 1:
ꢃ − ꢃ
ꢄ
ꢅ ∗ (1 − ꢊꢋꢌ(−ꢆꢇꢈꢉ ∗ ꢍ)) + ꢃ ∗ ꢍ
ꢀ ꢂ
ꢁ =
ꢅ
ꢆꢇꢈꢉ
(Eq. 1)
where V1 is the initial rate and was constrained to zero, kobs is the rate constant for the approach
to the steady state rate (V2), and t is time. Steadyꢀstate values were analyzed for substrateꢀ
dependent inhibition using equation 2:
ꢀ ꢂ
ꢃꢎꢏꢐ ∗ ꢑ
ꢃꢇꢈꢉ
=
ꢀ ꢂ
ꢑ
ꢒꢕꢖ
ꢀ ꢂ
ꢒꢓ + ꢑ ꢔ1 +
ꢗ
(Eq. 2)
where Vobs is the observed reaction velocity at substrate concentration S, Vmax is the theoretical
maximal rate, and KM and KIS are the apparent Michaelis constant and the apparent inhibition
constant. While titrating GTP, a second product peak was identified with an inverse dependence
on the GTP concentration. Maximal rate of production was seen in the absence of GTP,
suggesting the peak was the linear AMPꢀ3ʹꢀATP product reported by Gao et al.19 (Fig. S4). The
GTPꢀdependence for the rate of AMPꢀ3ʹꢀATP formation was fit to a standard halfꢀmaximal
inhibitory concentration (IC50) model, described in equation 3:
ꢃ
ꢘ
ꢃꢇꢈꢉ
=
1 + ꢀGTPꢂ/ꢙꢚꢛꢘ
24
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 25 of 52
Protein Science
(Eq. 3)
where V0 is the observed rate in the absence of GTP. Additionally, cGAS was titrated with 0.3 to
3 mM ATP in the absence of GTP. The rate of AMPꢀ3ʹꢀATP formation showed a hyperbolic
concentration dependence, and was fit to the Michaelis Menten equation.
NMR assays. Samples for continuous monitoring of cGAS reactions were prepared with
0.5 ꢁM cGAS, a top concentration of 2 mM ATP with 2ꢀfold dilutions over three points, and a
fixed concentration of 500 ꢁM GTP. Reactions were performed in SPR running buffer (see
below) at 23°C. Reactions were monitored through 1D spectra collected at 8 min intervals. Peak
identities were determined from comparison to ATP, GTP and 2ʹ,3ʹꢀcGAMP standards, the
assignment of additional peaks as the linear AMPꢀ2ʹꢀGTP intermediate was made after mass
spectra analysis revealed a mass of 853 Da (predicted 853 Da) in addition to the substrates and
product. Compound concentrations were determined through peak integration using ATP as an
internal standard.
Steadyꢀstate rates of 2ʹ,3ʹꢀcGAMP formation were fit using Eq. 1. Initial rates of
intermediate formation were determined by fitting the intermediate concentration to Eq. 1 with
an unconstrained V1 value and a negative V2 value.
KM values were determined using fixed ATP (690 ꢁM) and variable GTP concentrations
(90, 180, 360, or 550 ꢁM) or fixed GTP (770 ꢁM) and variable ATP concentrations (70, 125,
300, or 500 ꢁM) at 37°C. Data were fit to the Michaelis Menten equation.
25
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 26 of 52
MS assays. cGAS (100 nM) was incubated for 30 min at 37°C with 100 nM ISD in 10
mM HEPES pH 7.5, 140 mM NaCl, 5 mM MgCl2 and 0.01% Tweenꢀ20, and varied substrate
concentrations. The substrate dependence was assessed by titrating ATP or GTP over a range of
5 ꢁM to 1.3 mM, keeping the invariant substrate at 0.3 mM (GTP) or 1 mM (ATP). Additional
samples were prepared using 2 mM ATP or GTP in the absence of the second nucleotide, or 0.5
mM GTP, 2 mM ATP. All samples were quenched with 50 mM EDTA prior to analysis.
Quenched samples were diluted in H2O, vortexed, and centrifuged at 3000 RCF. Soluble analytes
were separated in a hypercarb column (5 ꢁm, 2x30 mm) using an acetateꢀacetonitrile/acetone
gradient (Buffer A: 20 mM ammonium acetate, pH 10; Buffer B: 45% acetonitrile, 45% acetone,
10 mM ammonium acetate, pH 10, 0.1% formic acid) at 60°C. Analytes were identified by mass
spec analysis using a 1290 Agilent UHPLC in conjunction with a Sciex 5500 tripleꢀquadrupole
mass spectrometer in MRM mode. Data were processed using Sciex's Multiquant 3.0. Reaction
products were identified by mass and MRM transitions for each nucleotide: AMPꢀ3ʹꢀATP at 837
m/z (predicted and observed), GMPꢀ2ʹꢀGTP at 869 m/z (predicted and observed), AMPꢀ2ʹꢀGTP
at 853 m/z (predicted and observed) and 2’3’ꢀcGAMP at 675 m/z (predicted and observed). ATP,
GTP and 2’3’ꢀcGAMP quantities were calculated from standard curves of peak intensity. Data
were analyzed using equation 2.
SPR binding assays. SPR experiments were performed using CM5 sensor chips in a
BIACORE T200 (GE Healthcare). CM5 channels were functionalized with neutravidin (Pierce)
through amine coupling (GE Healthcare) at 25°C to a density of 25,000ꢀ30,000 resonance units
(RUs). Nꢀterminal biotinylated cGAS161, STING155ꢀ341, or 3ʹꢀbiotinylated ISD were
immobilization at 5 ꢁL/min, 4°C. No protein was added to channel 1, channel 2 held apo
26
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 27 of 52
Protein Science
cGAS161 (final density ~7000 RU), channel 3 held a 1:1 molar equivalence of ISD:cGAS161
(final density ~6000 RU), and channel 4 held STING155ꢀ341 (final density ~10,000 RU).
Twoꢀfold dilution series of compounds F1ꢀF3 (300 to 0.3 ꢁM) or cyclic dinucleotides (500
to 0.5 ꢁM) were prepared in 20 mM HEPES pH 7.5, 150 mM KCl, 20 mM MgCl2 and 1 mM
TCEP, 1% DMSO. Samples were injected for 30ꢀ60 s, and dissociation was measured for 30ꢀ60
s at 60 ꢁL/min, 4°C.
Sampleꢀdependent responses on channels 2 through 4 were subtracted from channel 1 to
account for interactions with neutravidin, and data were further corrected by subtracting a zero
concentration blank from all compounds to account for mismatches in the sample buffer and
running buffer. Dissociation equilibrium constants (Kd) were determined using a binary
association model (T200 BIA Eval software):
ꢚ ∗ ꢝꢎꢏꢐ
ꢒꢜ =
− ꢚ
ꢝꢞ − ꢝꢘ
(Eq. 4)
where RC is the response at compound concentration C, Rmax is the maximum response of the fit,
and R0 is a global data offset from zero (Table 1).
SPR activity assays. The ATPꢀconcentration dependence of cGAS activity was assessed
by injecting ATP (2.0 mM to 2.0 ꢁM, 2ꢀfold dilutions) in the presence of 1.0 mM to 1.0 ꢁM GTP
(2ꢀfold dilutions). ATP and GTP titrations were also performed in the absence of the other
nucleotide. 2ʹ,3ʹꢀcGAMP (20 ꢁM to 20 nM, 2ꢀfold dilutions) was injected before and after ATP
and GTP samples to establish the 2ʹ,3ʹꢀcGAMPꢀdependent change on the STING channel.
27
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 28 of 52
SPR instrumentation and sensor chip setup was as described above. Samples were
injected for 900 s, dissociation was measured for 300 s at 5 ꢁL/min, 4°C. The STING response
(channel 4) values showed an initial negative deflection, especially at higher nucleotide
concentrations; therefore, blank subtracted values were normalized to have zero absorbance at
the signal minimum (~100 seconds). Normalized data were fit to a single exponential
association from 600 to 900 s, and the extrapolated plateau RU values were then replotted as a
function of the ATP concentration. Data sets corresponding to different GTP concentrations were
simultaneously fit to equation 2 using GraphPad Prism with a shared KM.ATP value. In turn, the
resulting Vmax values were fit as a function of the GTP concentration using the Michaelis Menten
equation (Eq. 2 when [S]/KIS <<1).
To further probe the cGAS substrate inhibition observed above, SPR injections were
made using a 2:1 molar ratio of ATP to GTP over an ATP range of 5 mM to 90 ꢁM (11 points,
1.5ꢀfold dilutions). Data collection and analysis were as described above.
To ensure that STING155ꢀ341 response was specific to 2ʹ,3ʹꢀcGAMP and not influence by
any of the linear products present in the enzymatic samples (eg the cGAS intermediate),
reactions were prepared in the buffer used for SPR with 2 ꢁM cGAS, 3 ꢁM ISD, and a 2:1 molar
ratio of GTP to ATP at 0.25, 0.5, 1.0, and 4 mM ATP. Reactions were quenched with 85 mM
EDTA after 3 h at 23°C and analyzed by NMR to determine the concentration of 2ʹ,3ʹꢀcGAMP
and linear 2ʹ,5ʹꢀGMPꢀ3ʹꢀAMP dinucleotide. The samples were then diluted 4ꢀfold into SPR
running buffer to minimize noise associated with running buffer and sample buffer mismatch and
injected for 60 s, and dissociation was measured for 60 s at 60 ꢁL/min, 4°C.
28
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Page 29 of 52
Protein Science
ACKNOWLEDGMENT
The authors thank Karen L. Leach for project support critical to these studies; Bruce Lefker and
Dave Hepworth for helpful discussions; Andrea Hall and Felix Vajdos for manuscript evaluation.
ABBREVIATIONS
AMPꢀ3ʹꢀATP, a linear homonucleotide with a 3ʹ,5ʹꢀphosphodiester bond; AMPꢀ2ʹꢀGTP, a linear
heteronucleotide with a 2ʹ,5ʹꢀphosphodiester bond; GMPꢀ2ʹꢀGTP, a linear homonucleotide with a
2ʹ,5ʹꢀphosphodiester bond; 2ʹ,2ʹꢀcGAMP, the cyclic guanine adenine dinucleotide with 2ʹ,5ʹꢀ
phosphodiester bonds; 2ʹ,3ʹꢀcGAMP, the cyclic guanine adenine dinucleotide with mixed 2ʹ,5ʹꢀ
and 3ʹ,5ʹꢀphosphodiester bonds, the enzymatic product of cGAS and a secondary messenger for
STING activation; 2′,5′ꢀGpAp, a linearized 2ʹ,3ʹꢀcGAMP resulting in break in the 3ʹ,5ʹꢀbond and
the phosphate retained on the 3ʹꢀhydroxyl of the adenine ribose; 3ʹ,3ʹꢀcGAMP, the cyclic guanine
adenine dinucleotide with 3ʹ,5ʹꢀphosphodiester bonds; 3ʹ,3ʹꢀcdIMP, the cyclic inosine
dinucleotide with 3ʹ,5ʹꢀphosphodiester bonds; 3ʹ,3ʹꢀcdUMP, the cyclic uracil dinuceleotide with
3ʹ,5ʹꢀphosphodiester bonds; cGAS, the cyclic GMPꢀAMP synthase; cGAS161, a truncated
construct of cGAS corresponding to residues 161ꢀ522 (numbering relative to H. sapiens cGAS);
EC50, halfꢀmaximal effective concentration; IC50, halfꢀmaximal inhibitory concentration; ISD,
interferon stimulatory dsꢀDNA, a 45ꢀbp oligomer (5ʹꢀ
TACAGATCTACTAGTGATCTATGACTGATCTGTACATGATCTACAꢀ3ʹ) used for cGAS
stimulation; Kd, dissociation constant; KM, Michaelis constant; RU, resonance unit equivalent to
one picogram of protein per square millimeter of the SPR surface; SLE, systemic lupus
erythematosus; STING155ꢀ341, stimulator of type 1 interferon genes corresponding to residues
29
John Wiley & Sons
This article is protected by copyright. All rights reserved.

Protein Science
Page 30 of 52
155ꢀ341 (numbering relative to H. sapiens STING) with the H232 SNP. PBD accession codes
are 5VDQ (cGAS161●2′,5′ꢀGpAp), 5VDO (cGAS161●2′,2′ꢀcGAMP), 5VDP (cGAS161●2′,3′ꢀ
cGAMP), 5VDT (cGAS161●3′,3′ꢀcGAMP), 5VDR (cGAS161●3′,3′ꢀcdIMP), 5VDS
(cGAS161●3′,3′ꢀcdUMP), 5VDW (cGAS161●F1), 5VDU (cGAS161●F2), and 5VDV
(cGAS161●F3).
AUTHOR CONTRIBUTIONS
All authors contributed to the intellectual development of these works. JH and ER wrote the
manuscript. JH, LB and ER created illustrations. SS and JH produced protein. JH and LB
performed structural studies. JH and ER developed the SPR assay, JH performed SPR assays.
HW developed the NMR assays, HW and RH performed NMR assays. ER performed HPLC
assays and kinetic analysis. AB and DD performed MS assays.
FUNDING SOURCES
We would also like to acknowledge financial support from Pfizer.
NOTES
The authors declare no competing financial interests.
30
John Wiley & Sons
This article is protected by copyright. All rights reserved.