findings, Model 1 was ruled out. Simulations
showed that both Model 2 and Model 3 could,
under certain conditions, display a concave
downward Scatchard plot and a bell-shaped
homologous competition curve (fig. S5). How-
ever, fitting the kinetic SPR results with Model
2failedtosatisfactorilyaccountforthedata
(fig. S4B and text S1).
To further differentiate Models 2 and 3, we
focused on experimental validation of the de-
fining feature of Model 3: dimerization of
MSA-2 in the absence of STING.^1 H-NMR of
MSA-2 alone in aqueous buffer showed that sev-
eral nonexchangeable MSA-2 benzothiophene
ring protons underwent chemical shift per-
turbations in a concentration-dependent man-
ner (Fig. 5G), consistent with a reversible
dimerization process (KD 1 ≈18 mM; fig. S6).
The observed chemical shift perturbations were
substantially larger for protonsatocthan for
protone, mirroring the changes in local envi-
ronment experienced by the bioactive MSA-2
dimerseeninthecrystalstructure(Fig.5B;
larger shielding effects are expected for pro-
tonsatocbecause of aromatic group overlap).
This was interpreted as supporting evidence
that most MSA-2 dimers in solution are in a bio-
active configuration capable of binding STING.
As shown in Fig. 5, E and F, both the equi-
librium and kinetic properties of the MSA-2
and hSTING-WT interaction can be accom-
modated by Model 3, where monomeric and
dimeric MSA-2 are in equilibrium (KD 1 ≈
18 mM) and only MSA-2 dimers are capable of
binding hSTING-WT with single-digit nano-
molar potency (defined asKD2) and a slow
off-rate (t1/2= 1.3 hours). Using Model 3, we
were also able to determine equilibrium and
kinetic constants for theinteractionofMSA-2
with mouse STING (mSTING) and two com-
mon human STING variants ( 19 ): hSTING-
HAQ and human STING-H232 (hSTING-H232)
(tables S4 and S5 and fig. S4, D, F, and H). The
rank order of dimeric MSA-2 affinity for the
fourSTINGvariantsismSTING>hSTING-
HAQ > hSTING-WT >> hSTING-H232.
Compound 2 (Fig. 2A), an MSA-2 analog in
which the sulfur of the thiophene ring is re-
placed by nitrogen, exhibited a weaker ten-
dency to homodimerize in solution than did
MSA-2 (Fig. 5, G and H, and fig. S6). Com-
pound 2 also appeared to bind both hSTING-
WT and hSTING-HAQ more weakly than MSA-2
(fig. S7, A to D, and text S2) when assessed
with a mass spectrometry–based technique
that isolates and identifies protein-bound
compounds [automated ligand identification
system (ALIS)] ( 21 ). When the hSTING-WT (or
hSTING-HAQ) cytosolic domain was incubated
with 2 at a concentration insufficient to elicit
detectable binding (fig. S7, C and D), inclu-
sion of MSA-2 in the incubation caused con-
current binding of both MSA-2 and 2 to
STING. As shown in Fig. 5I (or fig. S7E), in-
creasing [MSA-2] caused a bell-shaped in-
crease (and decrease) in bound 2 (orange),
whereas bound MSA-2 (green) concurrently
increased in a sigmoidal fashion. These ob-
servations were interpreted as mass-action–
driven formation of MSA-2: 2 heterodimers
capable of binding STING and competition
between these heterodimers and MSA-2 homo-
dimers for binding to STING (Fig. 5I, top). To
determine whether the MSA-2: 2 heterodimer
is a functional agonist, cellular experiments
were conducted to assess the agonist potency
of MSA-2 in the presence of several fixed con-
centrations of compound 2 ,whichalonedid
not have any detectable agonist activity. Com-
pound 2 increased the apparent potency of
MSA-2asmuchas10-foldinaconcentration-
dependent manner when measuring IFN-b
secretion from THP-1 cells (Fig. 5J). These re-
sults suggest that (i) the MSA-2: 2 heterodimer
is a functional agonist, and (ii) the equilibrium
dissociation constant for heterodimer forma-
tion must be lower than that for compound 2
homodimerization. Taken together, the ob-
served interactions between MSA-2 and 2
constitutepowerfulevidenceforthebasicprem-
ise of Model 3—namely, that the bioactive
molecule is a noncovalent dimer.Design of covalent MSA-2 dimers, which are
potent STING agonists
The central tenet of Model 3 is that MSA-2
must form a noncovalent dimer in solution to
gain STING binding activity, whereas mono-
meric MSA-2 is incapable of binding STING.
This model therefore predicts that a stable
compound dimer would be a good ligand. We
thus used this compound as the starting point
for development of a more potent class of
STING agonists. Although various substitutions
of the heterocycle and oxobutanoic acid regions
of the molecule did not improve potency, anal-
ysis of the x-ray cocrystal structure with STING
suggested the possibility of synthetically linking
thetwocloselypackedMSA-2unitstoproducea
single molecule, a covalent dimer, that would
bind with reduced entropic penalty.
To predict the optimal linkers for this design,
we developed a computational method in which
thousands of tethered benzothiophene cores
were generated in silico (enumerated). Each
of their conformations was scored for the
estimated free energy required for selection
out of the conformational ensemble versus the
quality of overlay on the crystal structure of
MSA-2 bound to STING (Fig. 6A). The results
highlight linking between the 5-positions—
especially replacement of both 5-methoxy
groups with a propane linker (Fig. 6A, teal
circle labeled 3 )—as particularly promising.
We thus synthesized covalent dimer 3 (Fig.
6C) and found that it is a highly potent STING
agonist. Confirmation that the binding mode
of 3 was similar to the MSA-2 noncovalentpose was provided by an x-ray cocrystal struc-
ture with human STING (Fig. 6B), which il-
lustrated that the key interactions of both
the ketone and carboxylic acid moieties with
STING lid residues replicate those of MSA-2.
Notably, the loss of the 5-methoxy groups
and their interactions with the side chains of
Ser^162 did not abrogate cellular potency.
Having demonstrated the viability of a co-
valent linking strategy with a three-atom, all-
carbon propyl linker in 3 , we investigated a
diverse set of modifications and found that
both homologation to four- and five-atom link-
ers and incorporation of oxygen at the linker
attachment points were generally well toler-
ated, particularly when one or both benzo-
thiophene cores were fluorinated alpha to the
5-position linker attachment point (Fig. 6C
and fig. S8A). This effect was particularly evi-
dent for the 1,2-dioxoethane linker, for which
fluorination of the core was required for po-
tency (e.g., 6 and 7 versus 5 ). Within the
computational analysis, linkers with oxygen
attachment points were predicted to have in-
creased strain due to the required out-of-plane
geometry. Fluorination at the neighboring po-
sition helps eliminate this strain through pre-
organization, consistent with observation.
In addition to linker composition, we also
observed a surprisingly broad accommodation
of different linker attachment points on the
benzothiophene cores (Fig. 6C and fig. S8),
with both 5,6- and 6,6-linked analogs of 9 pro-
viding potent agonists. With these data validat-
ing additional permutations beyond 5,5-tethers,
we revisited the x-ray structure of MSA-2 bound
to STING and noted the proximity of the 6-
methoxy group of one MSA-2 unit to the 4-
position of the benzothiophene core of its
partner. On the basis of modeling that suggested
that a shorter tether would be preferred, we pre-
pared the 4,6-propyl linked analog 12 ,which
again demonstrated potent STING agonism.
X-ray crystallographic data obtained for a
number of these covalent dimers bound to
STING revealed that the oxobutanoic acid
moiety is the dominant feature in determining
the binding pose for these molecules (Fig. 6D, fig.
S8B, and table S6). Notably, all analogs preserve
the same configuration and interactions of the
ketone and carboxylic acid portions of the oxo-
butanoic acid moieties, regardless of linker at-
tachment points. To preserve these interactions,
tethered molecules adopted benzothiophene
conformations to maintain symmetric or pseu-
dosymmetricp-stacking arrangements that
conserve the oxobutanoic acid interaction
with STING.
Taken together, the results described for
MSA-2 and its covalent dimer analogs estab-
lish that presentation of two oxobutanoic acid
substructures in a specific conformation—
either through noncovalent interactions or
linked by various methods and modified byPanet al.,Science 369 , eaba6098 (2020) 21 August 2020 5of10
RESEARCH | RESEARCH ARTICLE