Electrophilic fragment-based design of reversible covalent kinase inhibitors

Article abstract of DOI:10.1021/ja401221b

Fragment-based ligand design and covalent targeting of noncatalytic cysteines have been employed to develop potent and selective kinase inhibitors. Here, we combine these approaches, starting with a panel of low-molecular- weight, heteroaryl-susbstituted cyanoacrylamides, which we have previously shown to form reversible covalent bonds with cysteine thiols. Using this strategy, we identify electrophilic fragments with sufficient ligand efficiency and selectivity to serve as starting points for the first reported inhibitors of the MSK1 C-terminal kinase domain. Guided by X-ray co-crystal structures, indazole fragment 1 was elaborated to afford 12 (RMM-46), a reversible covalent inhibitor that exhibits high ligand efficiency and selectivity for MSK/RSK-family kinases. At nanomolar concentrations, 12 blocked activation of cellular MSK and RSK, as well as downstream phosphorylation of the critical transcription factor, CREB.

Full text of DOI:10.1021/ja401221b

Communication
pubs.acs.org/JACS
Electrophilic Fragment-Based Design of Reversible Covalent Kinase
Inhibitors
Rand M. Miller,‡ Ville O. Paavilainen,† Shyam Krishnan,† Iana M. Serafimova,‡ and Jack Taunton*,†
‡
†Department of Cellular and Molecular Pharmacology, Howard Hughes Medical Institute, and Chemistry and Chemical Biology
Graduate Program, University of California, San Francisco, California 94158, United States
S
* Supporting Information
ABSTRACT: Fragment-based ligand design and covalent
targeting of noncatalytic cysteines have been employed to
develop potent and selective kinase inhibitors. Here, we
combine these approaches, starting with a panel of low-
molecular-weight, heteroaryl-susbstituted cyanoacryla-
mides, which we have previously shown to form reversible
covalent bonds with cysteine thiols. Using this strategy, we
identify electrophilic fragments with sufficient ligand
efficiency and selectivity to serve as starting points for
Figure 1. (A) Sequence alignment of 11 human kinases containing a
the first reported inhibitors of the MSK1 C-terminal kinase
cysteine at the same position as C436 of RSK2 (yellow). The
gatekeeper position is highlighted in blue (threonine) or red (larger
hydrophobic residues). (B) Michael adducts of thiols with acrylamides
are kinetically stable, whereas cyanoacrylamides form rapidly reversible
domain. Guided by X-ray co-crystal structures, indazole
fragment 1 was elaborated to afford 12 (RMM-46), a
reversible covalent inhibitor that exhibits high ligand
efficiency and selectivity for MSK/RSK-family kinases. At
adducts with thiols.
nanomolar concentrations, 12 blocked activation of
cellular MSK and RSK, as well as downstream
functionality, which forms β-thioether adducts that eliminate
phosphorylation of the critical transcription factor, CREB.
more rapidly than acrylamide-derived adducts due to the
decreased pKa of the α-proton (Figure 1B). We sought to
exploit this reversible cysteine-targeting chemistry in the
ragment-based design is a powerful approach for
context of fragment-based ligand design.
Fdeveloping ligands that modulate protein function,
MSK1 is a close relative of RSK2, possessing two kinase
domains and a structurally homologous cysteine in its CTD.
including protein kinase inhibitors.1−3 This strategy relies on
the identification of low-molecular-weight compounds (100−
300 Da) that form specific interactions with the target protein,
often of low affinity (Kd ∼0.1−1 mM). Despite their low
binding affinity, fragments often have higher ligand efficiency
than typical high-throughput screening hits, which tend to be of
higher molecular weight (>300 Da) and therefore have a higher
probability of exhibiting steric clashes and other unfavorable
interactions with the ligand binding site.4 Elaboration of the
initial fragment hits guided by NMR or co-crystal structures
affords a ligand that is ideally more potent and selective than
the initial fragment, while retaining druglike physical properties.
Covalent bond formation between an electrophilic ligand and
a poorly conserved, noncatalytic cysteine is another powerful
strategy in drug discovery that has been exploited to enhance
potency and selectivity.5−7 This is especially true for protein
kinases, which are difficult to target selectively due to the high
sequence and structural conservation of the active site. A large
fraction of the 518 human protein kinases contain a solvent-
exposed cysteine within or near the ATP binding site.8,9 We
recently reported a series of compounds that inhibit the RSK2
C-terminal kinase domain (CTD) by forming a reversible
covalent bond with a cysteine (C436) found in only 11 of the
518 human protein kinases (Figure 1A).10 This reversible
covalent interaction is made possible by an α-cyanoacrylamide
Despite this similarity, MSK1 is insensitive to our previously
developed RSK inhibitors,10 most likely because it has a large
methionine in the gatekeeper position (Figure 1A). Previous
studies have suggested that the MSK1 CTD is essential for
intramolecular phosphorylation and activation of the N-
terminal kinase domain (NTD),11 which subsequently
phosphorylates transcription factors and histone H3.12,13
Although MSK1 has been implicated in various cancers,14−16
the few known inhibitors bind the NTD and show little
discrimination among several other AGC-family kinases,
including S6K1, AKT1, PRK2, and ROCK2.17 No inhibitors
of the MSK1 CTD have been reported to date.
In this Communication, we describe an electrophilic
fragment-based approach to ligand discovery. We have used
this approach to develop the first reported inhibitor of the
MSK1 CTD. This cyanoacrylamide-based inhibitor is active
against closely related MSK/RSK-family kinases, but it is highly
selective over NEK2 and PLK1, despite the presence of a
homologous cysteine in these kinases.
Whereas our previous study started with a known RSK
inhibitor,10 there was no obvious starting point for the current
Received: February 3, 2013
© XXXX American Chemical Society
A
dx.doi.org/10.1021/ja401221b | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Communication
a
project. Moreover, it was not at all clear whether low-
molecular-weight electrophiles would be able to discriminate
among different noncatalytic cysteines. To test the feasibility of
an electrophilic fragment-based approach, we assembled a panel
of 10 aldehyde fragments (MW 96−250 Da), all with nitrogen-
containing heterocycles that are commonly found in kinase
inhibitors. Condensation with cyanoacetamide yielded the
corresponding cyanoacrylamides 1−10 (Figure 2). We
Table 1. In Vitro Kinase Assay IC50 Values (μM)
RSK2
NEK2
PLK1
1
0.36 0.01
0.23 0.01
1.4 0.2
3.4 0.7
14 0.7
0.57 0.07
2.5 0.4
2
3
38
1.4 0.1
21
1
37
25
2
2
4
96
6
5
2
>150
>300
6
6.0 0.2
4.4 0.6
0.12 0.02
8.2 0.3
6.1 0.5
1.1 0.2
2.1 0.4
26 0.2
>100
45
7
7
27 0.4
>10
8
9
<0.62
10
68
3
53
2
a
Values reported as the mean range for duplicate measurements.
Figure 2. Cyanoacrylamide-based fragments used in this study.
screened 1−10 against three human kinases, all of which
contain a cysteine at the same position: RSK2 (C436), NEK2
(C22), and PLK1 (C67). To mimic intracellular redox
conditions, glutathione (GSH, 10 mM) was included in all
kinase assays, in addition to ATP (0.1 mM) and a peptide or
protein substrate. At a molar excess of 1 million-fold over the
kinase, glutathione also increased the stringency of our screen
by acting as a competing nucleophile.
Figure 3. (A) Co-crystal structure of 1 bound to the RSK2 CTD.
2Fo − Fc map is contoured to 1σ. (B) Overlay of the structures of 1
(blue) and 8 (yellow) bound to RSK2, indicating the potential for aryl
substitution at the 3-position of the indazole scaffold.
We observed significant potency differences among the 10
cyanoacrylamides, with distinct patterns of inhibition noted for
each kinase. Isomeric indazoles 1 and 2 and azaindole 8
inhibited RSK2 at submicromolar concentrations and high
ligand efficiencies (LE = 0.54, 0.56, and 0.39, respectively; see
ref 18 for ligand efficiency calculation). By contrast, 2 and 8
were significantly less potent against NEK2 and PLK1.
Azaindole 9 was selective for PLK1 over the other kinases in
this screen, and purine 7 exhibited modest selectivity for NEK2.
Mutation of the key cysteine in RSK2 (C436V) conferred
>100-fold resistance to all of the cyanoacrylamides with
submicromolar potency, suggesting that covalent bond
formation is required for potent inhibition (Table S1). Unlike
cyanoacrylamide 1, indazole itself failed to inhibit RSK2 at
concentrations below 500 μM, and indazole-5-carboxaldehyde
was weakly active against both wild-type and C436V RSK2
(IC50 ∼350 μM, Figure S1). Collectively, the structure−activity
relationships in Table 1 suggest that the free energy of covalent
bond formation is highly sensitive to structural differences in
the fragments and the cysteine-containing kinase.
To gain structural insight into the extraordinary ligand
efficiency of indazole 1, we determined its co-crystal structure
bound to RSK2 (PDB code 4JG6). The structure reveals the
same kinase hinge-binding mode observed with other indazole
inhibitors,19,20 in which the N1 proton donates a hydrogen
bond to the carbonyl of L495 while N2 accepts a hydrogen
bond from the amide of M496 (Figure 3A). Contiguous
electron density was observed between C436 and the β-carbon
of the cyanoacrylamide, consistent with covalent bond
formation. Notably, indazole 1 does not extend into the
hydrophobic pocket behind the gatekeeper T493 (Figure 3A).
This observation suggested that, unlike our previously
developed RSK2 inhibitors,10 1 would be sterically compatible
with a bulkier side chain in the gatekeeper position. Consistent
with this hypothesis, indazole 1 inhibited the T493M
gatekeeper mutant of RSK2 with an IC50 of 230 nM.
A co-crystal structure of azaindole 8 (PDB code 4JG7)
revealed that it binds RSK2 in a completely different manner
from 1 (Figure 3B). The electron density for 8 was clear and
unambiguous, allowing its accurate placement within the RSK2
active site (Figure S2). Unlike structurally related kinase
inhibitors, the azaindole moiety of 8 does not engage the hinge.
Rather, the ketone accepts a hydrogen bond from hinge residue
M496, while the azaindole projects toward the entrance of the
ATP pocket, sandwiched between the side chains of I428,
M496, and L546. This interaction is accompanied by a
conformational change in the glycine-rich loop, which contains
I428 (Figure S3). In both co-crystal structures, the nitrile and
amide groups of the covalently bound inhibitors are solvent
exposed and poorly resolved, precluding assignment of the
newly formed α-stereocenter. Together, the structures suggest
that potent RSK2 inhibition by cyanoacrylamide fragments 1
and 8 requires the simultaneous satisfaction of two geometric
restraints: at least one hydrogen bond to the hinge and an
unstrained covalent bond to C436.
An overlay of the two co-crystal structures suggested that
aromatic substituents appended to the 3-position of indazole 1
would fill the hydrophobic cleft occupied by the azaindole
moiety of 8 (Figure 3B). We therefore designed and
synthesized the trimethoxyphenyl-substituted indazole 11
(Table 2). This provided a 20-fold increase in potency toward
RSK2, but its selectivity over NEK2 and PLK1 was poor.
Addition of a 1,1-dimethyl-2-hydroxyethyl substituent to the
primary amide of 11 afforded our optimized inhibitor 12, which
retained potency against RSK2 and showed increased selectivity
over NEK2 and PLK1 (Table 2, Figure 4A). The IC50 of 12
B
dx.doi.org/10.1021/ja401221b | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Communication
RSK2 were >40-fold more sensitive than NEK2 and PLK1
when tested under identical conditions in our laboratory (Table
2 and Figure 4A, 0.1 mM ATP, 10 mM GSH). Kinases showing
negligible inhibition by 12 include, among others, MEK1,
ERK1, and p38 MAPK, all of which are upstream regulators of
RSK/MSK-family kinases (Table S2).
Table 2. In Vitro Kinase Assay IC50 Values (nM) for 11 and
12
a
A co-crystal structure of 12 bound to T493M RSK2 (PDB
code 4JG8) confirmed the binding mode anticipated by our
structural analysis of 1 and 8 (Figure 4B,C). Overlaying the co-
crystal structures of 12 and 1 revealed that the indazole cores of
both inhibitors were perfectly aligned, with the indazole of 12
packed against the methionine gatekeeper (M493). Similar to
the azaindole of 8, the trimethoxyphenyl substituent of 12 was
nestled between the side chains of I428, M496, and L546. The
1,1-dimethyl-2-hydroxyethyl amide group of 12 extended
beneath the glycine-rich loop (Figure 4B,C). Given the low
sequence conservation and conformational heterogeneity of
this loop,21,22 this interaction may account for the improved
selectivity of 12 compared to 11 (Table 2). Finally, a covalent
bond was observed between C436 and the electrophilic β-
carbon of the cyanoacrylamide. Unfolding the 12/RSK2
complex with guanidinium·HCl resulted in the quantitative
release of 12, confirming that the covalent bond is reversible
and that its stability requires the intact folded structure of the
kinase domain (Figure S4).
RSK2
RSK2 T493M
NEK2
PLK1
37
2200 200
b
11
12
15
12
2
3
3
1
c
62
8
3
<2.5
530 16
a
Values reported as the mean SD (n ≥ 3) unless otherwise noted.
All assays contain 10 mM GSH and 100 μM ATP. Value is the mean
range for duplicate measurements. Calculated IC50 is less than half
the concentration of kinase present (5 nM).
b
c
against T493M RSK2 was below 2.5 nM, the lower limit of our
assay (LE >0.35).
The ability of 12 to inhibit T493M RSK2 with >1000-fold
higher potency than our previous cyanoacrylamide inhibitors10
prompted us to test the related kinase, MSK1, which also has a
methionine in the gatekeeper position (Figure 1A). To date, it
has not been possible to perform kinase assays with the isolated
MSK1 CTD. We therefore asked whether 12 could prevent
CTD-mediated autophosphorylation of full-length MSK1 in
stimulated cells. Cells expressing HA-tagged MSK1 were
treated with increasing concentrations of 12, followed by
stimulation with phorbol myristate acetate (PMA). As revealed
by a phospho-specific antibody that recognizes pS376, 12
inhibited MSK1 CTD activity in the context of the full-length
protein, with an IC50 of ∼100 nM (Figure 4D). In the same
cell-based assay format, 12 was equipotent against the closely
related kinases, MSK2, RSK2, and RSK3 (Figure S5).
MSK1/2 are among several kinases that have been shown to
regulate the cAMP-response element binding transcription
factor (CREB) via phosphorylation of S133. Studies with
knockout mice have demonstrated a role for MSK1/2 in CREB
S133 phosphorylation in response to mitogens (e.g., PMA) and
cellular stress (e.g, UV-C).13 However, until now, it has not
been possible to test whether the CTD of endogenous MSK1/2
is required for CREB phosphorylation. Importantly, 12 was
found to be inactive toward several kinases previously shown to
phosphorylate CREB, including AKT1, PKA, CAMK2, and
PKCθ (Table S2). Treatment of HeLa cells with 12 reduced
CREB phosphorylation (IC50 ∼300 nM) under both PMA and
UV-C stimulation conditions (Figure 4E). By contrast, a potent
RSK inhibitor,10 which is inactive against MSK1/2, failed to
block CREB phosphorylation under these conditions (Figure
S6). These results indicate that the MSK1/2 CTD plays a
critical role in promoting CREB phosphorylation by the NTD.
In this study, we have developed a new approach to
fragment-based ligand design that exploits the reversible
conjugate addition of thiols to cyanoacrylamides. This
chemistry, when deployed against a poorly conserved non-
catalytic cysteine, provides enhanced selectivity and potency,
while avoiding the formation of irreversible covalent adducts.
Figure 4. (A) Kinase assay dose−response curves for 12 (mean
SD). (B,C) Co-crystal structure of 12 bound to T493M RSK2. (D)
Inhibition of MSK1 autophosphorylation in stimulated COS7 cells.
(E) Inhibition of CREB phosphorylation in HeLa cells after
stimulation with PMA or UV-C.
The selectivity of 12 was further defined by screening a
commercial panel of 26 kinases (Table S2), 12 of which have a
cysteine at various positions throughout the ATP-binding site
(e.g., BTK, EGFR, JAK3, KDR). At a screening concentration
of 1 μM, NEK2 and PLK1 were the only kinases that were
significantly inhibited (77% and 84% inhibition, respectively;
[ATP] = 200 and 15 μM). We note that wild-type and T493M
C
dx.doi.org/10.1021/ja401221b | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX

Journal of the American Chemical Society
Communication
(5) Potashman, M. H.; Duggan, M. E. J. Med. Chem. 2009, 52, 1231−
1246.
Indeed, it is not clear that a fragment-based approach would be
feasible with electrophiles that form irreversible covalent bonds.
Inhibition/binding would occur under kinetic control and
would be sensitive to the intrinsic nucleophilicity of a given
cysteine.23 Although intriguing results with acrylamide frag-
ments have been reported,24 it remains to be seen whether
useful levels of kinetic discrimination can be achieved.
The structure−activity relationships observed with our
collection of 10 cyanoacrylamides suggest that potency is not
solely driven by the free energy of covalent bond formation.
Rather, specific noncovalent interactions (hinge hydrogen
bonds, van der Waals contacts, and steric complementarity)
are required in concert with the covalent bond to cooperatively
stabilize the complex. Given the geometric restraints imposed
by an unstrained thioether bond, it may be possible to
computationally predict the binding poses of fragment hits,
potentially extending the utility of this approach to cases where
experimental structures cannot be obtained. Relative to
disulfide-based fragments,25 the cyanoacrylamide electrophile
is more compatible with the thiol-rich cellular environment and
may therefore be retained during inhibitor optimization for use
in biological studies. Finally, we anticipate that cyanoacrylamide
fragments can be applied to cysteine-containing targets beyond
kinases, enabling the discovery of new chemical probes.
(6) Fry, D. W.; Bridges, A. J.; Denny, W. A.; Doherty, A.; Greis, K.
D.; Hicks, J. L.; Hook, K. E.; Keller, P. R.; Leopold, W. R.; Loo, J. A.;
McNamara, D. J.; Nelson, J. M.; Sherwood, V.; Smaill, J. B.; Trump-
Kallmeyer, S.; Dobrusin, E. M. Proc. Natl. Acad. Sci. U.S.A. 1998, 95,
12022−12027.
(7) Hagel, M.; Niu, D.; St. Martin, T.; Sheets, M. P.; Qiao, L.;
Bernard, H.; Karp, R. M.; Zhu, Z.; Labenski, M. T.; Chaturvedi, P.;
Nacht, M.; Westlin, W. F.; Petter, R. C.; Singh, J. Nat. Chem. Biol.
2011, 7, 22−24.
(8) Zhang, J.; Yang, P. L.; Gray, N. S. Nat. Rev. Cancer 2009, 9, 28−
39.
(9) Leproult, E.; Barluenga, S.; Moras, D.; Wurtz, J. M.; Winssinger,
N. J. Med. Chem. 2011, 54, 1347−1355.
(10) Serafimova, I. M.; Pufall, M. A.; Krishnan, S.; Duda, K.; Cohen,
M. S.; Maglathlin, R. L.; McFarland, J. M.; Miller, R. M.; Frodin, M.;
̈
Taunton, J. Nat. Chem. Biol. 2012, 8, 471−476.
(11) Deak, M.; Clifton, A. D.; Lucocq, J. M.; Alessi, D. R. EMBO J.
1998, 17, 4426−4441.
(12) McCoy, C. E.; Campbell, D. G.; Deak, M.; Bloomberg, G. B.;
Arthur, J. S. C. Biochem. J. 2005, 387, 507−517.
(13) Wiggin, G. R.; Soloaga, A.; Foster, J. M.; Murray-Tait, V.;
Cohen, P.; Arthur, J. S. C. Mol. Cell. Biol. 2002, 22, 2871−2881.
(14) Odgerel, T.; Kikuchi, J.; Wada, T.; Shimizu, R.; Kano, Y.;
Furukawa, Y. Leukemia 2010, 24, 1087−1090.
́
(15) Perez-Cadahía, B.; Drobic, B.; Espino, P. S.; He, S.; Mandal, S.;
Healy, S.; Davie, J. R. J. Biol. Chem. 2010, 286, 42−49.
(16) van der Heide, L. P.; van Dinther, M.; Moustakas, A.; ten Dijke,
P. J. Biol. Chem. 2011, 286, 5003−5011.
ASSOCIATED CONTENT
■
S
* Supporting Information
(17) Naqvi, S.; Macdonald, A.; McCoy, C. E.; Darragh, J.; Reith, A.
D.; Arthur, J. S. C. Biochem. J. 2012, 441, 347−357.
(18) Ligand efficiency is calculated according to ref 2, with IC50 as a
surrogate for Kd at 293 K, using ΔG = −RT ln(IC50) and LE = ΔG/
(no. of heavy atoms).
Detailed experimental procedures, synthesis and spectral
characterization of compounds, crystallographic statistics, and
collection parameters. This material is available free of charge
via the Internet at http://pubs.acs.org.
(19) Zhu, G.; Gandhi, V. B.; Gong, J.; Thomas, S.; Woods, K. W.;
Song, X.; Li, T.; Diebold, R. B.; Luo, Y.; Liu, X.; Guan, R.; Klinghofer,
V.; Johnson, E. F.; Bouska, J.; Olson, A.; Marsh, K.; Stoll, V. S.; Mamo,
M.; Polakowski, J.; Campbell, T. J.; Martin, R. L.; Gintant, G. A.;
Penning, T. D.; Li, Q.; Rosenberg, S. H.; Giranda, V. L. J. Med. Chem.
2007, 50, 2990−3003.
AUTHOR INFORMATION
■
Corresponding Author
jack.taunton@ucsf.edu
Notes
(20) McTigue, M.; Murray, B. W.; Chen, J. H.; Deng, -L; Solowiej, J.;
Kania, R. S. Proc. Natl. Acad. Sci. U.S.A. 2012, 109, 18281−18269.
(21) Patel, R. Y.; Doerksen, R. J. J. Proteome Res. 2010, 9, 4433−
4442.
The authors declare the following competing financial
interest(s): The authors have filed a patent application on
cyanoacrylamide kinase inhibitors (licensed to Principia
Biopharma, of which J.T. is a co-founder).
(22) Guimaraes, C. R. W.; Rai, B. K.; Munchof, M. J.; Liu, S.; Wang,
̃
J.; Bhattacharya, S. K.; Buckbinder, L. J. Chem. Inf. Model. 2011, 51,
ACKNOWLEDGMENTS
1199−1204.
■
(23) Weerapana, E.; Wang, C.; Simon, G. M.; Richter, F.; Khare, S.;
Dillon, M. B. D.; Bachovchin, D. A.; Mowen, K.; Baker, D.; Cravatt, B.
F. Nature 2010, 468, 790−795.
This work was supported by the NIH (GM071434 to J.T.), the
Academy of Finland and the Sigrid Juselius Foundation (to
V.P.), and the California Tobacco Related Disease Research
Program (19FT-0091 to S.K.). We thank Christopher
Waddling at the UCSF Crystallography Facility for assistance
with instrumentation and software.
(24) Nonoo, R. H.; Armstrong, A.; Mann, D. J. ChemMedChem.
2012, 7, 2082−2086.
(25) Erlanson, D. A.; Braisted, A. C.; Raphael, D. R.; Randal, M.;
Stroud, R. M.; Gordon, E. M.; Wells, J. A. Proc. Natl. Acad. Sci. U.S.A.
2000, 97, 9367−9372.
REFERENCES
■
(1) Erlanson, D. A.; McDowell, R. S.; O’Brien, T. J. Med. Chem. 2004,
47, 3363−3482.
(2) Howard, S.; Berdini, V.; Boulstridge, J. A.; Carr, M. G.; Cross, D.
M.; Curry, J.; Devine, L. A.; Early, T. R.; Fazal, L.; Gill, A. L.;
Heathcote, M.; Maman, S.; Matthews, J. E.; McMenamin, R. L.;
Navarro, E. F.; O’Brien, M. A.; O’Reilly, M.; Rees, D. C.; Reule, M.;
Tisi, D.; Williams, G.; Vinkovic,
52, 379−388.
́
M.; Wyatt, P. G. J. Med. Chem. 2009,
(3) Bamborough, P.; Brown, M. J.; Christopher, J. A.; Chung, C.;
Mellor, G. W. J. Med. Chem. 2011, 54, 5131−5143.
(4) Hann, M. M.; Leach, A. R.; Harper, G. J. Chem. Inf. Comput. Sci.
2001, 41, 856−864.
D
dx.doi.org/10.1021/ja401221b | J. Am. Chem. Soc. XXXX, XXX, XXX−XXX