ARS853 | ERK Signal

The reactivity-driven biochemical mechanism of covalent KRASG12C inhibitors

Rasmus Hansen1,3, Ulf Peters 1,3, Anjali Babbar1, Yuching Chen1, Jun Feng1, Matthew R. Janes1, Lian-Sheng Li1, Pingda Ren1,2, Yi Liu1,2 and Patrick P. Zarrinkar 1*

Activating mutations in KRAS are among the most common tumor driver mutations. Until recently, KRAS had been considered undruggable with small molecules; the discovery of the covalent KRASG12C inhibitors ARS-853 and ARS-1620 has demonstrated that it is feasible to inhibit KRAS with high potency in cells and animals. Although the biological activity of these inhibitors has been described, the biochemical mechanism of how the compounds achieve potent inhibition remained incompletely under- stood. We now show that the activity of ARS-853 and ARS-1620 is primarily driven by KRAS-mediated catalysis of the chemical reaction with Cys12 in human KRASG12C, while the reversible binding affinity is weak, in the hundreds of micromolar or higher range. The mechanism resolves how an induced, shallow and dynamic pocket not expected to support high-affinity binding of small molecules can nevertheless be targeted with potent inhibitors and may be applicable to other targets conventionally considered undruggable.
ctivating mutations in KRAS are among the most common mutations found in cancer. The KRASG12C mutation in par- ticular is observed in approximately 15% of non-small cell
lung adenocarcinoma, 3% of colorectal adenocarcinoma and 1% of pancreatic adenocarcinoma (based on TCGA provisional data- sets as captured by the cBioPortal for Cancer Genomics (http://
www.cbioportal.org/), March 2017)1,2. KRAS mutations, including G12C, activate KRAS by interfering with GTPase-activating protein (GAP)-mediated GTP hydrolysis, shifting the equilibrium between the signaling-competent KRAS-GTP and signaling incompetent KRAS-GDP complexes in favor of KRAS-GTP3,4. KRAS is at the apex of multiple signaling pathways central to tumor cell prolifera- tion, and activated KRAS has long been recognized as a prominent tumor driver5–7.
Attempts to identify small-molecule inhibitors of KRAS for many years were unsuccessful, leading to the conclusion that KRAS was undruggable5,6,8,9. This notion was consistent with the lack of clearly defined, deep pockets in the structure of RAS outside of the nucleotide-binding site that might support high-affinity binding of small molecules and the challenge of targeting the nucleotide- binding site due to the extraordinarily high affinity for GTP and GDP5,9–11. Several years ago, however, multiple new strategies for inhibition of KRAS began to emerge4,12, including covalent target- ing of the mutated cysteine in KRASG12C (ref. 13). Crystal structures of these covalent KRASG12C inhibitors revealed binding to a previ- ously inferred, highly dynamic pocket near the switch II region that is not evident in other structures13,14. Optimization of the originally described inhibitors by structure-based compound design led to the identification of ARS-853 (refs 15,16), which effectively inhibits KRASG12C in tumor cell lines and, subsequently, the structurally distinct inhibitor ARS-1620, with enhanced potency in cells and improved pharmaceutical properties17. Treatment with ARS-1620 by daily oral administration induces tumor regressions in KRASG12C
ARS-1620 specifically target the GDP-bound form of KRASG12C, trapping the inactive KRAS-GDP complex, blocking nucleotide exchange, and thereby inhibiting downstream signaling15–17.
ARS-853 and ARS-1620 inhibit KRASG12C signaling in cells with nanomolar to low-micromolar potency15–17, presenting an appar- ent paradox of how a target without conventional binding pockets to accommodate small molecules can nevertheless be inhibited with high potency. To address this question, we investigated the detailed biochemical mechanism by which these inhibitors achieve potent inhibition of this challenging target and the relative contribution of reversible, noncovalent affinity (Ki) and chemical reactivity (kinact). We now demonstrate that both inhibitors exhibit only weak revers- ible binding affinity for KRASG12C, with Ki well above 64 µM, but that the chemical reaction between the inhibitors and the targeted cyste- ine is specifically and substantially accelerated. The rapid covalent bond formation is not due to elevated intrinsic reactivity of Cys12 or to high intrinsic electrophilic reactivity of ARS-853 or ARS-1620. Together, the data demonstrate that KRASG12C acts as an enzyme to specifically catalyze the covalent bond formation between groups of otherwise low inherent reactivity and that KRAS-mediated accelera- tion of the chemical reaction is the primary driver of potency for this class of KRASG12C inhibitors. The biochemical mechanism of ARS- 853 and ARS-1620 explains how an induced, shallow, dynamic pocket that is probably not capable of supporting high-affinity small-mole- cule binding can nevertheless be potently and specifically drugged and provides insight that may be relevant to other, conventionally undruggable targets, as well as to covalent inhibitor design in general.

Results
Determining kinact/Ki. The interaction of a covalent inhibitor with its target can be described in two steps, an initial reversible binding event followed by formation of the covalent bond:

tumor cell line and patient-derived mouse xenograft models, pro- viding the first in vivo validation of direct covalent targeting of KRASG12C as a promising therapeutic strategy17. Both ARS-853 and
Ki
KRAS + I ⇄

KRASI
kinact
⎯⎯⎯⎯→

KRAS-I
1Wellspring Biosciences, Inc., San Diego, CA, USA. 2Kura Oncology, Inc., San Diego, CA, USA. 3These authors contributed equally: Rasmus Hansen, Ulf Peters. *e-mail: [email protected]
The reversible binding equilibrium is determined by Ki, the first-order rate constant of the chemical step is kinact, and the over- all potency is described by the second-order rate constant kinact/Ki. If the reversible binding event is in rapid equilibrium, and the inhib- itor concentration is in excess of KRAS such that the conditions are pseudo first order, the relationship between the observed rate con- stant of formation of the covalent KRAS-I complex and inhibitor concentration can be described as:

0.08

0.06

0.04

0.02

ARS-1620 ARS-853
kobs = kinact

ti tititi ti

[I]
[I] +
Ki

ti titititi

0.00
0
20
40 (Inhibitor) (µM)
60
When the inhibitor concentration is below Ki ([I] « Ki), the rela- tionship between kobs and [I] is expected to be linear, with a slope equal to kinact/Ki. When [I] > Ki, kobs is expected to saturate until kobs = kinact (ref. 18).
To quantitatively determine each of these parameters, we mea- sured the rate of formation of the covalent KRASG12C–inhibitor complex at a range of inhibitor concentrations for both ARS-853 and ARS-1620. We used MS to quantify the unmodified protein
b
0.04

0.03

0.02

0.01

ARS-853

and the covalent protein–inhibitor adduct to directly determine the percent target engagement at each time point (Supplementary
0.00

160

240

320

Fig. 1a,b). Each time course was fit to a single exponential to obtain kobs, and kobs values were plotted as a function of inhibitor con- centration (Fig. 1a). To avoid conditions under which compound solubility may become limiting, the highest inhibitor concentration used was 64 µM. There was a linear relationship between inhibi- tor concentration and kobs through the entire concentration range tested for both compounds, with no evidence of saturation at the highest concentrations. kinact/Ki values, derived from the slopes of the plots from three independent experiments for each inhibitor, are 290 ± 60 M-1 s-1 for ARS-853 and 1,100 ± 300 M-1 s-1 for ARS-1620 (Table 1). As expected from previous studies16,17, kinact/Ki for ARS- 1620 was approximately four-fold greater than that for ARS-853. The lack of saturation at high inhibitor concentrations precluded quantitatively determining Ki and kinact values individually; how- ever, it is possible to derive limits from the data that are informa- tive. The relationship between kobs and [I] is only expected to be approximately linear when [I] < Ki. Therefore, we infer that the Ki values for both compounds must be higher than 64 µM, the highest concentration tested. By the same reasoning, kinact must be higher than 0.019 s-1 for ARS-853 and 0.066 s-1 for ARS-1620, the aver- age kobs values at the highest concentration tested (Supplementary Fig. 1a,b). These limits do not indicate that kinact for ARS-1620 is greater than for ARS-853; they simply provide lower boundaries.
To access higher concentrations, we took advantage of the greater solubility of KRASG12C protein compared to those of the small-mol- ecule inhibitors. The design of the experiment was adjusted such that inhibitor was held at a low concentration relative to protein, and the rate of covalent complex formation was determined at con- centrations of KRASG12C up to 320 µM (Supplementary Fig. 1c). This design maintains pseudo-first-order conditions and is described by the same model outlined above, except protein concentration is substituted for inhibitor concentration. To follow the reaction, the decrease over time of free inhibitor not covalently bound to KRASG12C was monitored by MS. We focused on ARS-853, as the rate of reaction of ARS-1620 is too fast at protein concentrations above 64 µM to measure without a quench-flow instrument. The kobs for ARS-853 began to saturate at high KRASG12C concentrations (Fig. 1b), yielding a Ki of 200 ± 90 µM, a kinact of 0.050 ± 0.023 s-1 and a calculated kinact/Ki value of 250 ± 20 M-1 s-1 (Table 1), consistent with the Ki and kinact limits and the kinact/Ki value derived from the experiments described above in which inhibitor concentration was varied. Collectively, the results suggest that the reversible affinity of both inhibitors for KRASG12C is low, with dissociation constants in
(KRASG12C) (µM)

Fig. 1 | kinact/Ki determination for ARS-853 and ARS-1620. a,b, The rate of covalent engagement was measured at multiple concentrations for each compound (a) or at a fixed inhibitor concentration and multiple concentrations of KRASG12C (b), and rate constants were plotted versus
inhibitor or protein concentration. The mean of kobs from three independent experiments is shown for each point in a and from four independent experiments in b. The error bars show s.e.m. Error bars for ARS-853 are present in a, but too small to be discerned at this scale. The data in a were fit to a straight line with intercept fixed at the origin. kinact/Ki corresponds
to the slope of the line. The data in b were fit to a one-site binding function, as described in the main text. Individual reaction time courses and the corresponding kobs values at each concentration are shown in Supplementary Fig. 1.

the hundred micromolar range at least, but that the chemical reac- tion of covalent bond formation is fast, at least approximately 20- to 30-fold faster than, for example, kinact values reported for covalent EGFR inhibitors19.

Independent confirmation of poor reversible binding affinity. The apparent low reversible affinity was initially surprising, as most targeted inhibitors, both covalent and noncovalent, typically require at least modestly high-affinity reversible binding to their target to achieve potency. To independently confirm that reversible binding was indeed weak, we used three distinct experimental approaches.
First, we asked whether the rate at which inhibition of nucleotide exchange was achieved was similar to the rate of covalent engage- ment. Noncovalent binding is expected to be rapid compared to the rate of covalent bond formation. ARS-853 and ARS-1620 bind to the KRASG12C-GDP complex and inhibit the SOS1-catalyzed exchange of GDP16,17. In a standard nucleotide exchange assay, binding of a fluorescently labeled nucleotide analog to KRASG12C preloaded with GDP is monitored by observing the increase in fluorescence inten- sity upon binding of the labeled analog to KRAS as the unlabeled nucleotide dissociates3,16. To measure the rate of gain of nucleotide exchange inhibition, the exchange reaction was initiated by the addi- tion of SOS1 and BODIPY-GDP after different incubation times of 32 µM ARS-853 or ARS-1620 with the KRASG12C-GDP complex, and the fluorescence signal was determined 30 s after initiating the exchange reaction. Inhibition of nucleotide exchange results in
Table 1 | Kinetic analysis of covalent inhibitors
inhibitor kinact/Ki (M-1s-1) Ki (µM) kinact (s-1) kintrinsic (M-1s-1) [RFAACAA] kintrinsic (M-1s-1) [VGACGVGKS]
ARS-853 250 ± 20a 200 ± 90a 0.050 ± 0.023a < 0.03f < 0.03f
ARS-1620 1,100 ± 300b > 64c > 0.066 ± 0.019d < 0.03f < 0.03f
ARS-107 8.5 ± 1.6b > 320c > 0.0028 ± 0.0005d < 0.03f < 0.03f
ARS-917 29 ± 5b > 64c > 0.0019 ± 0.0004d < 0.03f < 0.03f
Compound 12 0.33 ± 0.06b > 64c > 2.4 ± 0.6 × 10-5d < 0.03f < 0.03f
Afatinib 1.5 × 107e 0.00016e 0.0024d 0.14 ± 0.03b 0.20 ± 0.02b
The values shown represent the average from three independent experiments, where the errors encompass the range of values observed. aKi, kinact and kinact/Ki values determined from a fit of data shown in Fig. 1b and Supplementary Fig. 1c. bkinact/Ki and kintrinsic values represent the slope of a linear fit of kobs vs. inhibitor or protein concentration such as those shown in Fig. 1a and Supplementary Fig. 4. cKi limits were set as the highest concentration tested, as there was no evidence of saturation at this concentration. dkinact limits represent the mean kobs at the highest concentration tested. eValues for afatinib as reported in Schwartz et al.19. fThe extent of reaction after 24 h at 64 µM inhibitor was below the limit of quantitation of 15%; the limit for kintrinsic was therefore set based on the limit of quantitation.

a lowered rate of fluorescence signal gain. The extent of covalent engagement was monitored by MS in a parallel reaction. The rate of achieving maximum nucleotide exchange inhibition mirrored the rate of covalent engagement, and there was no evidence for rapid onset of inhibition due to noncovalent binding before covalent engagement at the tested concentration (Fig. 2a).
Second, we tested the ability of ARS-853 and ARS-1620 to inhibit nucleotide exchange of wild-type KRAS (KRASwt) and KRASG12D. As the targeted Cys12 is not present in these variants of KRAS, the inhibitors cannot form covalent adducts; however, any nonco- valent binding with affinity even in the micromolar range should be detected as inhibition of nucleotide exchange. We tested both KRASwt and KRASG12D, as the G12D mutation is another common activating mutation that impairs GTP hydrolysis, and the struc- tural dynamics of the switch II pocket in KRASG12D may resemble those in KRASG12C more closely than the dynamics of KRASwt (refs 3,6). If either ARS-853 or ARS-1620 did bind with higher revers- ible affinity than suggested by the Ki estimate above, such binding may also translate, albeit potentially with reduced potency, to other variants of KRAS. There was, however, no evidence of nucleotide exchange inhibition by either ARS-853 or ARS-1620 at 32 µM with either KRASwt or KRASG12D when the inhibitors were prein- cubated for 30 min with the GDP complex of the respective KRAS variants before initiating the exchange reaction. In contrast, there was complete inhibition of SOS1-catalyzed nucleotide exchange with KRASG12C, where quantitative covalent target engagement is expected to occur during the preincubation, with exchange reduced to the non-catalyzed background rate (Fig. 2b).
Third, we tested a series of analogs of ARS-1620 and ARS-853 that lack part or all of the electrophilic warhead, and thus are unable to form a covalent bond with cysteine (Supplementary Fig. 2), for their abilities to inhibit nucleotide exchange of KRASG12C (Fig. 2b), as well as for their abilities to compete with ARS-1620 or ARS-853 covalent engagement (Fig. 2c). None of the analogs at a concentra- tion of 32 µM were able to either inhibit nucleotide exchange or inhibit covalent engagement.
Collectively, these results are consistent with weak noncovalent affinity of both ARS-853 and ARS-1620 for KRASG12C, with disso- ciation constants well above the highest concentrations tested. The potency of the inhibitors, therefore, is not driven by high-affinity reversible binding, but rather appears to be predominantly driven by the rapid covalent bond formation with Cys12.

Low intrinsic reactivity of Cys12 in KRASG12C. One possible explanation for rapid covalent bond formation could be that Cys12 is intrinsically activated and will react rapidly with any sterically accessible electrophile. To test this possibility, we measured the rate of reaction of the small model electrophile iodoacetamide
with Cys12 in KRASG12C, as well as with cysteine embedded in a short, unstructured model peptide of sequence RFAACAA and a peptide corresponding to the sequence surrounding Cys12 in KRASG12C (VGACGVGKS). On the basis of the crystal structure of GDP-bound KRASG12C, Cys12 should be readily accessible to iodoacetamide20. To directly assess the effect of tertiary protein structure, we also measured the rate of reaction of iodoacetamide with KRASG12C Cys12 under denaturing conditions (6 M urea). As a control, we assessed the reactivity with Cys80, which is in the interior of the protein and should be inaccessible to iodoacetamide under native conditions, but accessible under denaturing condi- tions. The second-order rate constants of iodoacetamide reacting with the cysteines in the model and KRASG12C-derived peptides and with Cys12 under native and denaturing conditions reveal that the intrinsic reactivity is similar for all substrates tested, with a two- to three-fold higher rate for Cys12 under denaturing con- ditions only (Table 2). Cys80 was unreactive under native condi- tions and similarly reactive to Cys12 under denaturing conditions (Table 2). Cys12 in native KRASG12C is therefore not inherently highly reactive.
The reactivity of cysteine at physiological pH will in part depend on the pKa of the sulfhydryl group, and it has been suggested that the pKa of Cys12 may in fact be lowered relative to the pKa for free cysteine21. To experimentally determine the pKa of Cys12 and cys- teines in the peptides, we measured the reactivity of iodoacetamide as a function of pH for each of the substrates (Supplementary Fig. 3a and Supplementary Table 1). The pKa of Cys12 under native conditions was within error of the pKa under denaturing condi- tions, as well as of the pKa of cysteine in the model peptides and Cys80 under denaturing conditions (Table 2), suggesting that the KRAS structural environment does not induce a shift in pKa for Cys12. To confirm the integrity of the inhibitor-binding site across the pH range used, we compared the engagement rate of ARS-1620 with that of KRASG12C at pH 7.5 and 10 (Supplementary Fig. 3b). As the engagement rate of ARS-1620 is highly dependent on the structural context of KRAS, any denaturation of the binding site would be expected to dramatically lower the observed rate. The kobs for engagement of 8 µM ARS-1620 did not decrease at the higher pH, but rather was enhanced, as expected for a sulfhydryl reaction when raising the pH, thus demonstrating that the bind- ing site remains intact. To confirm that the observed pKa values are independent of the electrophile used, we measured the pKa for KRASG12C Cys12 and for cysteine in the RFAACAA peptide by determining the rate of reaction of N-benzylacrylamide with the respective cysteines, which yielded similar values as those deter- mined with iodoacetamide (Supplementary Fig. 3c). Together, these results show that Cys12 in KRASG12C is not inherently acti- vated for nucleophilic attack.
a ARS-1620 ARS-853

100

Covalent engagement Nucleotide exchange
100

Covalent engagement Nucleotide exchange

0
50 100 150 200 250
Time (s)
0
50 100 150 200 250
Time (s)
b KRASwt KRASG12D

9,000
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0
9,000
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0

0
1,000 2,000 3,000 4,000
Time (s)
0
1,000 2,000 3,000 4,000
Time (s)
9,000
8,000
7,000
6,000
5,000
4,000
3,000
2,000
1,000
0
G12C
KRAS

DMSO
No SOS1 ARS-853 ARS-1620

ARS-3006 ARS-1372 ARS-1408 ARS-1440 ARS-1448

0 1,000 2,000 3,000 4,000
Time (s)

100

ARS-1620
DMSO ARS-1372 ARS-1408 ARS-1440 ARS-1448

100

ARS-853
DMSO ARS-3006

0 100 200 300 400 500 0 2,000 4,000 6,000
Time (s) Time (s)

Fig. 2 | independent confirmation of low reversible affinity of ARS-853 and ARS-1620. a, Covalent engagement and nucleotide exchange inhibition are achieved at the same rate. Covalent engagement was monitored by MS and nucleotide exchange was monitored by measuring the initial rate of exchange of BODIPY-GDP for GDP in parallel reactions, both with 32 µM inhibitor. Percent engagement values and fluorescence readings from the nucleotide exchange reaction were normalized such that maximum inhibition was set to a value of 100%. This experiment was performed once. b, ARS-853 and
ARS-1620 (32 µM) inhibit nucleotide exchange of KRASG12C, but not KRASwt or KRASG12D. Analogs of ARS-1620 (ARS-1372, ARS-1408, ARS-1440, ARS- 1448) and ARS-853 (ARS-3006) that do not bind covalently do not inhibit nucleotide exchange of KRASG12C at 32 µM final concentration. The chemical structures of the analogs are shown in Supplementary Fig. 2. This experiment was independently repeated one additional time, with similar results.
c, Noncovalent analogs of ARS-1620 and ARS-853 do not compete with covalent engagement of KRASG12C by 8 µM ARS-1620 or ARS-853. Covalent engagement by ARS-1620 in the presence of DMSO or 32 µM ARS-1372, ARS-1408, ARS-1440 or ARS-1448 is shown at left, and covalent engagement by ARS-853 in the presence of DMSO or 32 µM ARS-3006 is shown at right. This experiment was performed once as shown.
Low intrinsic reactivity of ARS-853 and ARS-1620. Another possible explanation for the rapid bond formation between the inhibitors and Cys12 is that the compounds possess high intrinsic
reactivity. Previous global cellular cysteine profiling demonstrated that both ARS-853 and ARS-1620 highly selectively modify Cys12 in KRASG12C, with few cellular off-targets, suggesting that they do
not possess high intrinsic reactivity16,17. To quantify their inherent biochemical reactivity and selectivity for Cys12, we assessed the reactivity of ARS-853 and ARS-1620 with cysteine in the RFAACAA and VGACGVGKS peptides and compared their reactivity to that of the covalent EGFR inhibitor afatinib19. The extent of covalent modi- fication by 64 µM inhibitor of both peptides was below the limit of quantitation of approximately 15% after 24 h for both ARS-853 and ARS-1620. Their intrinsic reactivity is therefore low, and at least four- to seven-fold lower compared to that of afatinib (Table 1). The second-order rate constants for reaction with the VGACGVGKS peptide are at least 3.7 × 104-fold lower than the kinact/Ki for reaction with KRASG12C for ARS-1620 and at least 8.3 × 103-fold lower for ARS-853. These data show that ARS-853 and ARS-1620 do not pos- sess inherent high electrophilic reactivity. Taken together, the results indicate that the chemical reaction of ARS-853 and ARS-1620 with Cys12 is highly specifically activated in the context of KRASG12C.
For comparison to the optimized inhibitors ARS-853 and ARS- 1620, we also determined the reactivity of the originally described covalent KRASG12C inhibitor compound 12 (ref. 13), as well as of ARS-107, an early lead compound16 structurally closely related to ARS-853, and ARS-917, an earlier compound in the ARS-1620 quin- azoline chemical series22,23 (Fig. 3 and Supplementary Fig. 2), against both KRASG12C and the two peptides (Table 1 and Supplementary Fig. 4a). The intrinsic reactivity of compound 12, ARS-107 and ARS-917 against the peptides was detectable, but below the limit of quantitation in our assay, as for ARS-853 and ARS-1620. The kinact/Ki for reaction with KRASG12C was approximately 30-fold lower for ARS-107 compared to ARS-853, 40-fold lower for ARS-917 compared to ARS-1620, and approximately 3 × 103-fold lower for compound 12 compared to ARS-1620 (Table 1). While we cannot accurately quantify the overall rate enhancements because of the inability to quantitatively measure the reactivity with the peptides, the warhead and proximal structure is identical for ARS-107 and ARS-853, as well as for ARS-917 and ARS-1620, suggesting that the intrinsic reactivities of each compound pair are likely very close. With this assumption, the KRASG12C-mediated overall rate enhance- ment for ARS-107 and ARS-917 afforded by the aggregate of revers- ible binding and the chemical reaction is probably substantially less than that observed for ARS-853 and ARS-1620, respectively. The lower potency of ARS-107 compared to that of ARS-853 is at least in part due to a loss of noncovalent binding affinity, as the Ki for ARS-107 was well above 320 µM when covalent complex formation rates were measured at high protein concentrations (Supplementary Fig. 4b), compared to 200 µM for ARS-853 (Table 1).

Mechanism of catalysis. To gain additional insight into the struc- tural basis for the variation in overall potency and KRAS-mediated catalysis of the reaction between the inhibitors and Cys12, we solved the crystal structures of ARS-107 and ARS-917 bound to KRASG12C (Table 3) and compared the inhibitor binding modes to those of ARS-853 and ARS-1620 (refs 16,17) (Fig. 3). The overall binding mode of all four compounds and the conformations of the binding pockets are similar. Alignment of the cocrystal structure of ARS-107 with that of ARS-853 reveals that the bound compounds are almost superimposed (Fig. 3a), as are bound ARS-917 and ARS-1620 (Fig. 3b). The apparent loss of reversible affinity of ARS-107, com- pared to that of ARS-853, suggests that the interaction between the methylcyclopropane on ARS-853, the only site divergent between the two compounds, and the small hydrophobic binding pocket it occupies at least in part drives noncovalent binding. In a recently described independent structure of ARS-917 bound to KRASG12C the inhibitor is in a similar orientation as that observed here; how- ever, there are some conformational differences in the protein, including a considerable upward shift of helix α222, suggesting that the switch II pocket may be somewhat dynamic even when inhibi- tors are bound. The static view of the crystal structures, however, does not capture any differences in the dynamics of the interactions between the compounds and the protein before and during chemi- cal bond formation, which may play an important role in governing the potency.
To more directly investigate the mechanism by which KRASG12C accelerates the chemical reaction, we took advantage of the ability to independently measure effects on Ki (binding) and kinact (cataly- sis) for ARS-853. To test the role of specific conformational align- ment and protein side chains proximal to the inhibitor warhead, we changed the location of the cysteine in KRAS from residue 12 to residue 92 by introducing a D92C mutation in the context of KRASwt at residue 12 (ref. 24). The D92C mutation does not impair RAS function25. On the basis of inspection of the crystal structure, this location should be sterically accessible to the acrylamide in ARS- 853, but reaction at this site would require substantial movement of the acrylamide relative to the position in the KRASG12C complex (Fig. 3a). Remarkably, the Ki for ARS-853 was unchanged, whereas kinact was reduced approximately ten-fold for KRASD92C relative to that of KRASG12C (Supplementary Fig. 5a and Table 1) and was in the same range as the kinact values measured for covalent EGFR inhibi- tors19. The kinact/Ki, however, was still at least approximately 1,000- fold higher than the reactivity with the unstructured peptides. These data show that there are two mechanisms that contribute to catalysis of the chemical reaction between inhibitor and Cys12. One is the increase of the local concentration of the electrophile achieved through the weak noncovalent binding interaction, which occurs in

Table 2 | Context-independent reactivity of cysteines with iodoacetamide
both KRASG12C and KRASD92C. In addition to this general mecha- nism, a second, more specific mechanism of catalysis is the precise
Substrate kintrinsic (M-1s-1)a Cysteine pKaa
RFAACAA peptide 0.75 ± 0.16 9.2 ± 0.1
VGACGVGKS peptide 1.0 ± 0.20 8.7 ± 0.1
KRASG12C Cys12 (native) 0.67 ± 0.14 9.0 ± 0.2
KRASG12C Cys12 (denatured) 2.6 ± 0.3 8.8 ± 0.1
KRASG12C Cys80 (native) < 0.02b Not determined
KRASG12C Cys80 (denatured) 2.6 ± 0.2 8.8 ± 0.1
Intrinsic reactivity of cysteines in model peptides and KRAS protein with iodoacetamide. aThe values shown represent the average from three independent experiments, except for the kintrinsic for the RFAACAA peptide, which is from four independent experiments, where the errors encompass the range of values observed. kintrinsic values represent the slope of a linear fit of kobs vs. inhibitor concentration. pKa values were measured by determining kobs of 1 mM iodoacetamide reacting with the respective substrate across a pH range. The data were fit as described in Methods to determine pKa values. kintrinsic and the pKa for Cys12 and Cys80 in denatured KRAS were measured with 1 mM iodoacetamide in the presence of 6 M urea. bNo modification was detected. The value is based on the limit of quantitation of the MS engagement assay.
orientation of the electrophile for attack of Cys12 in the context of KRASG12C, which is disrupted in the context of KRASD92C. The simi- lar Ki for ARS-853 observed with KRASG12C and KRASD92C also dem- onstrates that the majority of the binding energy for noncovalent binding, as reflected in the Ki, is probably contributed by the sub- stituted aromatic ring region on the lefthand side of the inhibitor, with little contribution from the region proximal to the acrylamide, consistent with the higher Ki observed for ARS-107.
In the crystal structures the carbonyl oxygen of the acrylamide warhead forms a hydrogen bond with Lys16 of KRASG12C that may contribute to orienting the electrophile (Fig. 3). The positive charge on this lysine may also stabilize the transition state by neu- tralizing the transient negative charge on the acrylamide carbonyl (Fig. 4a). Such a mechanism would be dependent on the proton- ated state of Lys16, such that the reaction rate would be negatively impacted at higher pH values approaching or exceeding the pKa of Lys16. The pKa of free lysine is approximately 10.5, but it can

a b

ARS-853
ARS-917

ARS-107

K16

K16 ARS-1620

D92
D92

ARS-853 ARS-107 ARS-1620 ARS-917

Fig. 3 | Cocrystal structures of inhibitors covalently bound to KRASG12C-GDP. a, ARS-853 and ARS-107. ARS-853 is shown in purple, ARS-107 in green. b, ARS-1620 and ARS-917. ARS-1620 is shown in purple, and ARS-917 is shown in green. Lys16 and Asp92 are marked in both panels. The structures for ARS-853 and ARS-1620 have been previously published16,17.

Table 3 | Data collection and refinement statistics
KRASG12C bound to ARS-107 (PDB 6B0V) KRASG12C bound to ARS-917 (PDB 6B0Y)
Data collection
Space group P1 P1
Cell dimensions
a, b, c (Å) 33.1, 39.9, 62.4 33.4, 39.7, 61.9
α, β, γ (°) 77.0, 81.2, 77.6 77.5, 81.7, 77.1
Resolution (Å) 1.29–38.17 (1.29–1.36)a 1.43–37.94 (1.43–1.51)a
Rmerge 0.085 (0.341) 0.076 (0.281)
I/σ(I) 6.1 (2.1) 5.5 (2.6)
CC1/2 0.991 (0.935) 0.986 (0.781)
Completeness (%) 90.4 (87.5) 92.7 (89.9)
Redundancy 2.8 (3.0) 1.6 (1.6)
Refinement
Resolution (Å) 1.29–38.17 (1.29–1.32) 1.43–37.94 (1.43–1.47)
No. reflections 65,427 (4,699) 49,052 (3,531)
Rwork / Rfree 0.182/0.218 (0.272/0.270) 0.174/0.207 (0.235/0.230)
No. atoms
Protein 2,711 2,746
Ligand/ion 114 (Ca2+ ions, GDP, ARS-107) 134 (Ca2+ ions, GDP, glycerol, ARS-917)
Water 435 350
B factors
Protein 15.7 13.4
Ligand/ion 14.0 12.1
Water 24.8 22.0
R.m.s. deviations
Bond lengths (Å) 0.023 0.021
Bond angles (°) 2.48 2.21
One crystal used per dataset. 3,371 (4.9%) and 2,501 (4.9%) reflections were used in the respective test sets. aValues in parentheses are for highest-resolution shell.
a b
+NH2
H
Lys16
+NH2
H
Lys16
+NH2
H
Lys16
K16

R2N O R2N O–
+H+
R2N O
S–

Cys12

γ-phos.

Fig. 4 | Transition state stabilization as a proposed mechanism of KRASG12C-mediated catalysis of covalent bond formation. a, In the transition state of the reaction, there is a negative charge on the acrylamide carbonyl, which may be stabilized by positively charged Lys16. b, The position of the acrylamide carbonyl relative to Lys16 in the inhibitor-KRASG12C-GDP cocrystal structures is very close to the position of the γ-phosphate of GTP in the KRAS-GTP complex. This positioning is consistent with an analogous role of negative charge stabilization by Lys16 in each of the complexes. Shown are the aligned structures of ARS-853 (PDB 5F2E) and ARS-1620 (PDB 5V9U) bound to KRAS-GDP, with the inhibitors in green, and of the complex of KRAS with the non-hydrolysable analog GDPNP (PDB 3GFT), with GDPNP in orange.
be shifted lower in the context of protein structure26. To determine whether the protonated state of Lys16 contributes to catalysis, we measured the Ki and kinact of ARS-853 at pH 9.5 and compared the values to those obtained at pH 7.5. There was no substantial change in the Ki, and an approximately four-fold increase of the kinact at pH 9.5 relative to that at pH 7.5 (Supplementary Fig. 5b and Table 1). The four-fold increase in kinact is substantially less than the approximately 20- to 30-fold increase of the intrinsic reactivity of Cys12 at pH 9.5 relative to that at pH 7.5, based on the pH depen- dence of Cys12 reacting with model electrophiles (Supplementary Fig. 3a,c and Supplementary Table 1). These results indicate that a second deprotonation event, such as loss of a proton on Lys16, negatively impacts the reaction rate and counteracts the positive effect of Cys12 deprotonation at high pH. There are no side chains other than Lys16 in the vicinity of the inhibitors that are likely to account for the observed effect, consistent with the lack of impact on reversible binding at the higher pH. We therefore infer that protonated Lys16 contributes to KRASG12C-mediated catalysis of covalent bond formation between Cys12 and ARS-853. A compa- rably small increase in reaction rate between pH 7.5 and high pH is observed with ARS-1620 (Supplementary Fig. 3b), suggesting a similar mechanism for ARS-1620.
Together, these data demonstrate that there are two primary mechanisms by which KRASG12C accelerates the covalent reaction with ARS-853, and likely with ARS-1620. First, an increase in the local concentration of the electrophile afforded by weak nonco- valent binding substantially accelerates the reaction relative to the reaction with unstructured peptides. Second, the optimized, specific orientation of the electrophile relative to Cys12 further promotes and facilitates the reaction beyond the more general increase in local concentration. The specific orientation may be mediated by the contact with Lys16, which may also stabilize the transition state.
Discussion
KRAS inhibition has been one of the most prominent unsolved problems in oncology drug discovery for many years. The identi- fication of ARS-853 and ARS-1620, the first compounds directly targeting KRAS that have demonstrated potent activity with proof of mechanism in cellular and animal models, represents signifi- cant progress toward solving KRAS as a drug target. An unresolved question has been how the potency of ARS-853 and ARS-1620 can be reconciled with the flexible, transient nature of their bind- ing pocket on KRAS, as targeted inhibitors typically require at least modestly high affinity binding to drive potency. This is the case for both noncovalent and covalent inhibitors, including covalent kinase inhibitors19,27,28. A detailed understanding of how this class of cova- lent KRAS inhibitors achieves their potency could thus provide a
foundation for further improvement and may be relevant to other, conventionally undruggable targets.
The results presented here demonstrate that both ARS-853 and ARS-1620 exhibit only very weak reversible affinity for KRASG12C and that their potency is primarily driven by substantial KRASG12C- mediated acceleration of covalent bond formation between the inhibitors and Cys12. To provide context, it is instructive to compare ARS-853 and ARS-1620 to covalent EFGR inhibitors, which are among the best-characterized covalent kinase inhibi- tors19. The reversible affinity (Ki) of covalent EGFR inhibitors is in the low- to subnanomolar range, at least 105- to 106-fold bet- ter than that of ARS-853 and ARS-1620 (ref. 19). The kinact values for EGFR inhibitors, however, are at least approximately 20- to 50-fold lower than those of ARS-853 and ARS-1620, based on the data shown in Table 1 and the assumption that kinact for ARS-1620 is at least two-fold higher than the low boundary estimate. The cellular IC50 of ARS-1620 for covalent engagement of KRASG12C after 2 h of treatment is approximately 300 nM, and the average IC50 for inhibition of cell growth in proliferation assays across multiple KRASG12C-harboring cell lines is 150 nM (ref. 17). These values are within approximately 10- to 100-fold of the potency of clinical covalent EGFR inhibitors in cellular target occupancy and proliferation assays19,29,30, a much smaller difference than the dif- ferences in Ki. The cellular potency of covalent EGFR inhibitors correlates with and is driven by Ki (ref. 19), whereas the cellular potency of ARS-1620 and ARS-853 (refs 15,16) is substantially bet- ter than what could be expected based on affinity alone. The bio- chemical mechanism that drives cellular potency of ARS-853 and ARS-1620 therefore differs from that of covalent EGFR inhibi- tors. While it has been suggested that cellular inhibition of KRAS may be achieved with some switch II pocket targeted quinazoline compounds related to ARS-1620 without covalent ligation23, our results indicate that a substantial improvement in affinity relative to ARS-1620 would be required to achieve cellular potency with a noncovalent inhibitor.
The weak reversible affinity of ARS-853 and ARS-1620 is con- sistent with the shallow and dynamic nature of the induced switch II binding pocket, which is unlikely to support high-affinity bind- ing of small molecules. Although the reversible affinity is weak for all compounds tested, the binding event must contribute to the proper positioning of inhibitors for electrophilic attack on Cys12 and overall potency, as suggested by the increased Ki, and therefore weaker binding, of ARS-107 compared to ARS-853, as well as by the dramatic difference in potency between the active S-atropisomer of ARS-1620, characterized here, and the essen- tially inactive R-atropisomer17. In the absence of strong reversible affinity, however, the substantial KRASG12C-mediated accelera- tion of the chemical reaction of the inhibitor electrophiles with
Cys12 provides a rationale for the observed potency, including cellular potency, of both ARS-853 and ARS-1620, as supported by the high kinact values. At least two mechanisms contribute to this acceleration. One is an increase of the local concentration of inhibitor afforded by the weak noncovalent binding interaction. This mechanism provides substantial rate acceleration in the con- text of the KRASD92C mutant relative to the baseline reaction rate with unstructured peptides. The kinact for reaction of ARS-853 with this mutant is in the same range as the kinact values determined for covalent EGFR inhibitors, suggesting that EGFR inhibitors may also benefit from this general mechanism. The second is the opti- mized alignment of the inhibitor electrophile, which is lost in the KRASD92C mutant, relative to KRASG12C under standard conditions. We suggest that the EGFR inhibitors do not benefit from this addi- tional catalytic mechanism.
Lys16 makes a direct hydrogen bond with the inhibitor acryl- amide carbonyl oxygen in the crystal structures and we propose that through this contact lysine 16 contributes to electrophile ori- entation as well as stabilizes the transition state. Transition state stabilization is promoted when Lys16 is protonated (Fig. 4a) and, therefore, should contribute less to the rate acceleration at high pH, where Lys16 is at least partially unprotonated relative to stan- dard conditions. The attenuated pH dependence of the reaction of Cys12 with the inhibitors, relative to the intrinsic pH dependence of Cys12 reactivity, as determined with iodoacetamide, supports the contribution of protonated Lys16 to catalysis. Lys16 is highly conserved31,32 and stabilizes the negative phosphate charges of the γ-phosphate in the KRAS-GTP complex33,34. Indeed, in the crystal structures, the position of the inhibitor acrylamide carbonyl relative to Lys16 is very close to that of the γ-phosphate in the complex of KRAS bound to the nonhydrolyzable analog GDPNP (Fig. 4b), con- sistent with an analogous role of Lys16 in both contexts. In addition to Lys16, the backbone amide of Gly60, although not charged, may be sufficiently close to contribute to stabilization of the transition state enolate. Future work to study the dynamics of the interaction of inhibitors with KRASG12C before and during covalent bond for- mation may provide additional insight into the interaction between KRASG12C and the inhibitors.
Most inhibitor discovery campaigns focus on optimizing revers- ible contacts between inhibitors and their target to enhance binding affinity. The results described here reinforce the notion that high reversible affinity is not always required to achieve high potency for covalent inhibitors and that equal weight should be given to opti- mizing the alignment, positioning and potentially dynamics of the electrophile in the binding pocket to achieve target-specific rate acceleration of covalent bond formation35. We have shown here that an extreme version of this general strategy can be applied to target- ing a marginal binding pocket in an otherwise undruggable protein, provided a residue for covalent modification is available, and may be relevant to other conventionally undruggable targets beyond KRAS.

Methods
Methods, including statements of data availability and any asso- ciated accession codes and references, are available at https://doi. org/10.1038/s41594-018-0061-5.
Received: 14 November 2017; Accepted: 4 April 2018; Published: xx xx xxxx

References
1.Cerami, E. et al. The cBio cancer genomics portal: an open platform for exploring multidimensional cancer genomics data. Cancer Discov. 2, 401–404 (2012).
2.Gao, J. et al. Integrative analysis of complex cancer genomics and clinical profiles using the cBioPortal. Sci. Signal. 6, pl1 (2013).
3.Hunter, J. C. et al. Biochemical and structural analysis of common
cancer-associated KRAS mutations. Mol. Cancer Res. 13, 1325–1335 (2015).

4.Ostrem, J. M. & Shokat, K. M. Direct small-molecule inhibitors of KRAS: from structural insights to mechanism-based design. Nat. Rev. Drug Discov. 15, 771–785 (2016).
5.Downward, J. Targeting RAS signalling pathways in cancer therapy. Nat. Rev. Cancer 3, 11–22 (2003).
6.Downward, J. RAS’s cloak of invincibility slips at last? Cancer Cell 25, 5–6 (2014).
7.van Hattum, H. & Waldmann, H. Chemical biology tools for
regulating RAS signaling complexity in space and time. Chem. Biol. 21, 1185–1195 (2014).
8.Papke, B. & Der, C. J. Drugging RAS: know the enemy. Science 355, 1158–1163 (2017).
9.Wang, W., Fang, G. & Rudolph, J. Ras inhibition via direct Ras binding—is there a path forward? Bioorg. Med. Chem. Lett. 22, 5766–5776 (2012).
10.Mao, Y., Yao, H., Wang, H., Cheng, P. & Long, D. Microsecond timescale dynamics of GDP-bound Ras underlies the formation of novel inhibitor- binding pockets. Angew. Chem. Int. Edn Engl. 55, 15629–15632 (2016).
11.Müller, M. P., Jeganathan, S., Heidrich, A., Campos, J. & Goody, R. S. Nucleotide based covalent inhibitors of KRas can only be efficient
in vivo if they bind reversibly with GTP-like affinity. Sci. Rep. 7, 3687 (2017).
12.Welsch, M. E. et al. Multivalent small-molecule pan-RAS inhibitors. Cell 168, 878–889.e29 (2017).
13.Ostrem, J. M., Peters, U., Sos, M. L., Wells, J. A. & Shokat, K. M.
K-Ras(G12C) inhibitors allosterically control GTP affinity and effector interactions. Nature 503, 548–551 (2013).
14.Taveras, A. G. et al. Ras oncoprotein inhibitors: the discovery of potent, ras nucleotide exchange inhibitors and the structural determination of a
drug-protein complex. Bioorg. Med. Chem. 5, 125–133 (1997).
15.Lito, P., Solomon, M., Li, L. S., Hansen, R. & Rosen, N. Allele-specific inhibitors inactivate mutant KRAS G12C by a trapping mechanism. Science 351, 604–608 (2016).
16.Patricelli, M. P. et al. Selective inhibition of oncogenic KRAS output with small molecules targeting the inactive state. Cancer Discov. 6, 316–329 (2016).
17.Janes, M. R. et al. Targeting KRAS mutant cancers with a covalent G12C- specific inhibitor. Cell 172, 578–589.e17 (2018).
18.Copeland, R.A. Evaluation of Enzyme Inhibitors in Drug Discovery: A Guide for Medicinal Chemists and Pharmacologists (Wiley, Hoboken, NJ, 2013).
19.Schwartz, P. A. et al. Covalent EGFR inhibitor analysis reveals importance of reversible interactions to potency and mechanisms of drug resistance. Proc. Natl Acad. Sci. USA 111, 173–178 (2014).
20.Hunter, J. C. et al. In situ selectivity profiling and crystal structure of SML-8-73-1, an active site inhibitor of oncogenic K-Ras G12C. Proc. Natl Acad. Sci. USA 111, 8895–8900 (2014).
21.Huynh, M. V. & Campbell, S. L. Getting a handle on RAS-targeted therapies: cysteine directed inhibitors. Mini Rev. Med. Chem. 16, 383–390 (2016).
22.Lu, J. et al. KRAS G12C drug development: discrimination between switch II pocket configurations using hydrogen/deuterium-exchange mass spectrometry. Structure 25, 1442–1448.e3 (2017).
23.Zeng, M. et al. Potent and selective covalent quinazoline inhibitors of KRAS G12C. Cell Chem Biol. 24, 1005–1016.e3 (2017).
24.Patricelli, M. P., Peters, U., Li, L., Ren, P. & Liu, Y. Method for screening inhibitors of Ras. US Patent 9, 810690 (2017).
25.Bandaru, P. et al. Deconstruction of the Ras switching cycle through saturation mutagenesis. eLife 6, e27810 (2017).
26.Isom, D. G., Castañeda, C. A., Cannon, B. R. & García-Moreno, B. Large shifts in pKa values of lysine residues buried inside a protein. Proc. Natl Acad. Sci. USA 108, 5260–5265 (2011).
27.Hagel, M. et al. First selective small molecule inhibitor of FGFR4 for the treatment of hepatocellular carcinomas with an activated FGFR4 signaling pathway. Cancer Discov. 5, 424–437 (2015).
28.Harling, J. D. et al. Discovery of novel irreversible inhibitors of interleukin (IL)-2-inducible tyrosine kinase (Itk) by targeting cysteine 442 in the ATP pocket. J. Biol. Chem. 288, 28195–28206 (2013).
29.Cross, D. A. et al. AZD9291, an irreversible EGFR TKI, overcomes
T790M-mediated resistance to EGFR inhibitors in lung cancer. Cancer Discov. 4, 1046–1061 (2014).
30.Solca, F. et al. Target binding properties and cellular activity of afatinib (BIBW 2992), an irreversible ErbB family blocker. J. Pharmacol. Exp. Ther. 343, 342–350 (2012).
31.Gay, N. J. & Walker, J. E. Homology between human bladder carcinoma oncogene product and mitochondrial ATP-synthase. Nature 301, 262–264 (1983).
32.Walker, J. E., Saraste, M., Runswick, M. J. & Gay, N. J. Distantly related sequences in the alpha- and beta-subunits of ATP synthase, myosin, kinases and other ATP-requiring enzymes and a common nucleotide binding fold. EMBO J. 1, 945–951 (1982).
33.Pai, E. F. et al. Refined crystal structure of the triphosphate conformation of H-ras p21 at 1.35 A resolution: implications for the mechanism of GTP hydrolysis. EMBO J. 9, 2351–2359 (1990).
34.Privé, G. G. et al. X-ray crystal structures of transforming p21 ras mutants suggest a transition-state stabilization mechanism for GTP hydrolysis. Proc. Natl Acad. Sci. USA 89, 3649–3653 (1992).
35.Zhang, T. et al. Discovery of potent and selective covalent inhibitors of JNK. Chem. Biol. 19, 140–154 (2012).

Author contributions
R.H. and U.P. contributed equally as co-first authors, and A.B. and Y.C. contributed equally as co-second authors. R.H. directed biochemical experiments, performed the MS and analyzed MS data. U.P. performed and coordinated the crystallography,
solved the crystal structures, developed and performed the high protein concentration kinetic assays and expressed and purified recombinant KRAS proteins. A.B., Y.C. and U.P. performed the biochemical kinetics experiments and analyzed experimental data. J.F. synthesized compounds. L.-S. L. and P.R. designed compounds and directed the

chemistry. M.R.J. directed experiments and edited the manuscript. Y.L. suggested

Acknowledgements
We thank J. Edwards, K. Shokat and D. Dhanak for critical reading of the manuscript and helpful discussions, A. Borum for technical assistance with chemistry and compound management and Shanghai Langtze Biomedical Technology for chemistry support on compound synthesis. Use of the IMCA-CAT beamline 17-ID at the Advanced Photon Source was supported by the companies of the Industrial Macromolecular Crystallography Association through a contract
with Hauptman-Woodward Medical Research Institute. This research used resources of the Advanced Photon Source, a US Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory
under Contract No. DE-AC02-06CH11357 and of the Advanced Light Source (beamline 8.2.2), which is a DOE Office of Science User Facility under contract no. DE-AC02- 05CH11231. We would like to thank R. Alexander and J. C. Spurlino for acquiring the crystallographic dataset for ARS-917. The KRASG12C project is financially supported by Janssen Biotech, Inc. under a collaboration between Wellspring Biosciences, Inc. and Janssen Biotech, Inc.
the study, supervised the work and edited the manuscript. P.P.Z. conceptualized, designed and directed the study and analyzed data. R.H., U.P., L.-S. L. and P.P.Z. wrote the manuscript.

Competing interests
All authors are employees of Wellspring Biosciences, Inc. and shareholders of Araxes Pharma LLC, which holds the rights to the inhibitors characterized in the paper.

Additional information
Supplementary information is available for this paper at https://doi.org/10.1038/
s41594-018-0061-5.
Reprints and permissions information is available at www.nature.com/reprints. Correspondence and requests for materials should be addressed to P.P.Z.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Methods
Proteins, peptides and inhibitors. Hexahistidine-tagged KRASG12C, KRASwt and KRASG12D proteins (isoform 2B, residues 1–169, based on the construct used for PDB 3GFT) were expressed in Escherichia coli, loaded with GDP and purified similarly to what has been described previously13 (details are provided in the Supplementary Note). Attempts to test the contribution of Lys16 experimentally more directly by mutation were inconclusive, as any mutation of Lys16 results in major structural disruption and inactivation
of KRAS, making it difficult to attribute loss of inhibitor potency specifically to loss of Lys16-mediated electrophile activation25. In the case of the KRASG12C protein used for X-ray crystallography and the KRASD92C protein, the corresponding cysteine-light constructs (each harboring C51S, C80L,
and C118S mutations) were used. The cysteine-light mutations have previously been shown not to materially impact KRAS structure13 or RAS function25. Flash- frozen aliquots were stored in 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl2 and 1 mM DTT at –80 °C. The catalytic domain of SOS1 for nucleotide exchange assays was expressed and purified as described13. Peptides were purchased from Life Technologies (Carlsbad, CA). ARS-853, ARS-107, and ARS-1620 were synthesized as described16,17, and ARS-917, ARS-3006, ARS-1372, ARS-1448,
ARS-1440 and ARS-1408 were synthesized as described in the Supplementary Note. Compound 12 and afatinib were purchased from Selleckchem (Houston, TX). Inhibitors were dissolved in DMSO or N-methyl-2-pyrrolidone (ARS-853 for some experiments, ARS-107, Compound 12) to produce 10 mM stock solutions. Iodoacetamide was purchased from Millipore Sigma (St. Louis, MO) and dissolved in water to produce a 500 mM stock solution. N-benzylacrylamide was purchased from Cambridge Chemicals (Woburn, MA) and dissolved in DMSO to produce a
1 M stock solution.

Covalent engagement assays. For experiments at low protein/peptide and high compound concentration, KRASG12C protein (2 µM in engagement reactions with inhibitors, 10 µM in reactions with iodoacetamide or
N-benzylacrylamide) loaded with GDP was incubated with the specified test compounds at indicated concentrations for the indicated times in 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM DTT and 1 mM MgCl2 at room temperature. Urea
(6 M final concentration) was added to reactions probing the denatured state. Reactions with the peptide substrates (5 µM each of the substrate and unreactive normalization peptide, see below) were performed similarly, except DTT was omitted. To determine the pKa of cysteines in KRAS and the RFAACAA and VGACGVGKS peptides, the pH was varied using buffers as described below.
All engagement reactions were quenched with 0.1 volume of 2% formic acid and analyzed by LC-MS as described below. To assess covalent adduct formation specifically at C12 and C80 in KRASG12C, following the reaction, the protein was

150 °C. Sheath, auxiliary and sweep gas flow rates were 40, 5 and 1, respectively. The capillary temperature was 320 °C and S-lens RF level was set to 55.
The MS raw files were processed using Protein Deconvolution software version 4.0 (Thermo Scientific; Waltham, MA) to deconvolute the charge envelope into protein masses. The output file containing protein masses and their associated peak intensities was used to estimate percent modification as the fraction of compound-modified KRAS protein over the sum of unmodified and modified KRAS protein in the assay. Any points where reaction progress was less than 15%, which was considered the limit of quantitation of the readout, were omitted from the subsequent analysis.

KRAS sample processing and mass spectrometry analysis for experiments
with model electrophiles. To assess covalent adduct formation by iodoacetamide or N-benzylacrylamide specifically at C12 and C80 of KRASG12C, following
formic acid quenching, iodoacetamide or N-benzylacrylamide were removed by gel filtration using Zeba Spin Desalting plates with a 7K molecular weight cutoff (ThermoFisher; Carlsbad, CA). Approximately 25% of each sample in a buffer containing 100 mM ammonium bicarbonate, 6 M urea, 2 mM DTT was diluted in three volumes 100 mM ammonium bicarbonate and then digested with 1 µg trypsin (sequencing grade; Promega; Madison, WI) at 37 °C for 1 h. Trypsin digests were quenched with 0.1 volume 2% formic acid. The tryptic digest samples were analyzed using LC-MS, as described above with the following modifications. An injection of 20 µl peptide sample was separated
on a Kinetex 5 µm EVO C18 100 Å, LC Column (150 × 2.1 mm; Phenomenex; Torrance, CA) at a flow rate of 600 µl per minute. LC solvent A was 0.1% formic acid in 99.9% H2O, and solvent B was 0.1% formic acid in 99.9% acetonitrile. The column particles were equilibrated at 1% solvent B for 60 s. Following equilibration, a 90 s gradient from 1% to 35% solvent B and a 30 s gradient
from 35% to 80% solvent B was applied to separate protein, before re-equilibration before the next sample injection. The MS acquisition was set to acquire mass spectra at 70,000 resolution over a mass range from 350 to 1,850 Da, with an
AGC target of 3 × 106 ions, and the maximum allowed ion accumulation time was 100 ms. The LC-MS raw data were processed using Skyline software version 3.6 (University of Washington; Seattle, WA). The peak area for each peptide was the sum of the total integrated area for the three most abundant isotopes within set chromatographic peak boundaries. The signal for each isotope was extracted based on the predicted m/z and a mass window around the ion, set to 2 × 60,000 resolution (FWHM) at 400 m/z. The peptide sequences and associated masses are shown in the table below. The percentage cysteine modification was calculated as depletion of the normalized unmodified peptide according to the following formula:

digested with trypsin and analyzed by LC-MS, as described below. Reactions with high protein and low inhibitor concentrations were performed and analyzed as described below.
pKa determination. For pKa determinations the following buffers were used: HEPES (pH 7.0, 7.5), Trizma (pH 8.0, 8.5), CHES (pH 9.0 to 10.5) and CABS (pH 11.0, 11.4). To obtain pKa values, kobs and pH values were fit to the following form
% cysteine modification = 100
ti ti sample
Cysteine peptideDMSO,average
×

Normalization peptideDMSO,average
Normalization peptidesample

ti tititititi

ti titititititi

of the Henderson–Hasselbalch equation36:
kobs = kS-∕ (1 + 10 (pKa-pH) )

where kS– is the pH-independent rate constant at high pH, corresponding to reaction of the fully deprotonated thiolate form. To produce the plots shown in Supplementary Fig. 3a,c, kobs values were normalized as described in the figure legend.

Mass spectrometry analysis of intact protein samples. To analyze intact protein samples from KRASG12C engagement assays with inhibitors, the samples were analyzed after formic acid quenching on a Dionex RSLCnano instrument that was coupled to a Q Exative Plus mass spectrometer (both from Thermo Scientific; Waltham, MA), as a sample volume of 20 µL equaling approximately
40 pmol KRAS protein was injected for LC-MS mass analysis. Briefly, the injected protein was desalted and separated on an Aeris 3.6 µm WIDEPORE C4 200 Å,
LC Column (50 × 2.1 mm; Phenomenex; Torrance, CA) kept at 40 °C at a flow

rate of 600 µl per minute, using the loading pump on the Dionex RSLCnano instrument. LC solvent A was 0.1% formic acid in 99.9% H2O, and solvent B was 0.1% formic acid in 99.9% acetonitrile. The column particles were equilibrated in 20% solvent B for 60 s. Following equilibration, a 90 s gradient from 20% to 60% solvent B and a 30 s gradient from 60% to 90% solvent B was
applied to separate protein species, before re-equilibration before the next sample injection. The MS acquisition was set to acquire mass spectra at 70,000 resolution over a mass range from 800 to 1,850 Da, with an AGC target of 3 × 106 ions
and maximum allowed ion accumulation time was 250 ms. In-source CID was set to 5 eV and the “intact protein mode” gas setting was used to optimize ion transmission. Heated electrospray was applied for ionization using an IonMax source installed with a HESI-II probe (Thermo Scientific; Waltham, MA).
The spray voltage was set to 4 kV with an auxiliary gas heater temperature set to

Sample processing and mass spectrometry analysis for experiments with synthetic peptides. To enable normalization of the MS signal, a normalization peptide (see table below) was included in the reaction at the same concentration
as the corresponding substrate peptide. To analyze samples from synthetic peptide modification experiments, reactions were quenched with formic acid. The peptide samples were analyzed using LC-MS, as described above for the tryptic peptide samples. The synthetic peptide masses used for the analysis are shown in the
table below. The peptides purchased from Life Technologies were synthetized N-terminal acetylated with a purity above 90%. The percentage cysteine
modification was calculated as depletion of the normalized unmodified peptide according to the formula above.
Synthetic peptide substrates Peptide sequence
[ + 42 is MW of N-terminal acetyl] Precursor peptide ions (m/z) extracted ions
RFAACAA model cysteine peptide R[ + 42]
FAACAA 751.3556 (z = 1) Monoisotope(M) M + 1, M + 2
RFAAAAA model normalization peptide R[ + 42]
FAAAAA 719.3835 (z = 1) Monoisotope(M) M + 1, M + 2
KRASG12C peptide V[ + 42]
GACGVGKS 819.4029 (z = 1) Monoisotope(M) M + 1, M + 2
KRAS normalization peptide V[ + 42]
GASGVGKS 803.4258 (z = 1) Monoisotope(M) M + 1, M + 2
measurement. Quantitation of the unreacted compound amount in each sample and determination of kobs was carried out analogously to the samples at normal pH, as described above. Reaction progress was assessed by setting the 100 ms time point to 0% progress and assuming a proportional compound signal. For time points longer than 30 s, the described reactions were carried out analogously, but by manual pipetting.

Nucleotide exchange assays. GDP-loaded KRAS protein (0.5 µM) in 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM MgCl2 and 0.01% Triton X-100 was incubated with the indicated concentration of compound for 30 min before the exchange reaction was initiated by addition of BODIPY-GDP (0.36 µM final concentration) and SOS1 (0.5 µM final concentration) where indicated.
Fluorescence was monitored in a Tecan Infinite M1000 Pro plate reader (Tecan Group; Maennedorf, Switzerland), with readings taken every 30 s.
For BODIPY-GDP the excitation wavelength was 484 nm, and emission wavelength was 509 nm. Raw fluorescence values were normalized by setting the initial reading in each well after initiation of the reaction to a value of 0. All subsequent time points in each well were normalized by subtracting the initial fluorescence reading from the value at time t.
To measure the rate of gain of nucleotide exchange inhibition, inhibitor at the
Reaction of the KRASG12C inhibitors with the RFAACAA and KRAS peptides was very slow, with less than ~15% adduct formation (limit of quantitation)
after 24 h at 64 µM inhibitor for ARS-1620, ARS-853, ARS-917, ARS-107 and Compound 12. The upper limit for the rate constant was estimated based on 15% extent of reaction at 24 h and assuming linear reaction progression from t = 0. The reaction of afatinib with the peptides was sufficiently fast to determine kobs values by single exponential fit of complete reaction time courses.
Covalent engagement assays with high protein concentration. To measure covalent engagement at low inhibitor concentration and increasing protein concentration, KRASG12C or KRASD92C protein (~0.4 to 1 mM, GDP-loaded, previously purified and frozen) was first incubated overnight at 37 °C in 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM DTT and 1 mM MgCl2 to guarantee the reduced state of cysteines. The protein was then repurified by Gel Filtration (Superdex 75, as described above), concentrated (Amicon Ultra 0.5, 10,000 Da molecular weight cut-off; Millipore Sigma; Burlington, MA), and the concentration was determined (BCA assay; ThermoFisher; Carlsbad, CA). The protein concentration was adjusted if necessary, and the stock prepared in reaction
buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM DTT, 1 mM MgCl2, 0.01% Triton X-100).
The experiment was carried out by adding 53.9 µL protein (at the indicated concentrations) to 1.1 µL DMSO containing the compound of interest and an internal standard compound (2% DMSO, 8 µM compound, 2 µM standard
(a noncovalent analog of ARS-853), all final concentrations), followed by mixing through pipetting up and down. At the indicated time points (adjusted to match the different reaction rates of the compounds) 10 µL of the mix were removed and added to 30 µL of quench mix (reaction buffer containing 0.3% formic acid). The quenched reactions were then frozen until measurement. Reactions were carried out in 96-well plates (PCR-plate, ThermoFisher; Carlsbad, CA) with each protein concentration run in duplicate. For each compound/plate, a standard curve was prepared by two-fold serial dilution of compound in reaction buffer starting at
8 µM followed by immediate quenching, with the internal standard compound held constant at 2 µM. To determine the reaction progress, the compound and internal standard signal was determined for each well by MS (Agilent RapidFire QQQ instrument; MS was performed at PureHoney Technologies; Billerica, MA). The compound signal was normalized by dividing it by the internal standard signal and then converted to concentration using the prepared standard curve (quadratic fit due to slight saturation) for each plate. The concentrations from duplicate wells were averaged, and the reaction progress for each time point was calculated from comparison with the reaction containing no protein. The obtained kinetic data were fit to a single exponential (assuming pseudo-first-order kinetics; any points where reaction progress was less than 15%, which was considered the limit of quantitation of the readout, were omitted) for each protein concentration to obtain kobs (Supplementary Fig. 1c).
For experiments at pH 9.5, the reaction buffer was prepared with CHES at pH 9.5, instead of HEPES, pH 7.5. Owing to the faster reaction rate at high pH, the experiment setup was further adjusted as follows. Protein stock containing 2 × of the final concentration in reaction buffer, compound stock at 2 × concentration (buffer without DTT) and quench solution (containing 0.6% formic acid) were prepared separately and loaded into a QFM400 quench flow instrument (Bio- Logic USA, Knoxville, TN). For each time point (not performed in duplicate for these experiments), 24 µL of protein and 24 µL of compound were combined and allowed to react for the desired amount of time before quenching with 48 µL of quench solution, all in the instrument. 10 µL of each quenched sample was then added to 40 µL acetonitrile containing 0.3% formic acid and vigorously mixed,
and the resulting mix was centrifuged (~3,500 g, 10 min). 10 µL of the acetonitrile supernatant was then added to 30 µL of 0.3% formic acid and frozen until
indicated concentration was preincubated with 2 µM KRASG12C for the indicated times in 4.5 µL 20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM DTT, 1 mM MgCl2 and 0.01% Triton-X at room temperature, then added to 10.5 µL BODIPY-GDP (1.44 µM final concentration) and SOS1 (2 µM final concentration) to initiate
the nucleotide exchange reaction. Separate reactions were set up for each preincubation time point and initiated in a staggered fashion, such that addition of the inhibitor/protein mixture to SOS1 and labeled nucleotide occurred at the same time for each reaction. After initiating the nucleotide exchange reaction, reaction plates were inserted into a Tecan Infinite M1000 Pro plate reader (Tecan Group; Maennedorf, Switzerland), and fluorescence was read 7.5 s after the plate had been inserted. The total time elapsed from initiating the reaction to the fluorescence reading was approximately 28–30 s. For each well, the fluorescence reading was normalized by setting the reading corresponding to maximum inhibition (based on a single exponential fit of the raw 7.5 s fluorescence data) to 1, and the reading from a reaction with no preincubation time to 0. The percent engagement from an engagement reaction run under identical conditions was similarly normalized.
Crystallization, data collection and refinement. For X-ray crystallography,
the KRAS 1–169 cysteine-light construct (harboring the G12C along with C51S, C80L and C118S mutations, see above) was used, and the protein covalently was bound to compound and prepared as described below and previously13. MgCl2
(1 mM final concentration) and GDP (40 µM final concentration) were added to the freshly purified protein. After high-speed centrifugation, hanging-drop crystallization conditions were set up by mixing 1 µL protein and 1 µL reservoir solution (27% (ARS-917)/ 29% (ARS-107) PEG4000, 0.2 M CaCl2, 0.1 M
Tris pH 8.5) in 15-well EasyXtal plates (Qiagen; Hilden, Germany). After several days at 20 °C, plate-shaped crystals were observed. The crystals were cryoprotected in the crystallization solution supplemented with 10% glycerol, flash frozen and stored in liquid nitrogen before obtaining diffraction data at beamline 8.2.2 (100 K nitrogen stream, wavelength = 0.999995 Å) at the Berkeley Lab Advanced Light Source for ARS-107 and at IMCA-CAT beamline 17-ID (100 K nitrogen stream, wavelength = 1.00000 Å) at the Advanced
Photon Source for ARS-917.
Data were initially processed with iMosflm, solved by molecular replacement using Phaser and refined to the indicated statistics using Refmac37. The refined model for ARS-107 showed no Ramachandran outliers, with 98.2% of the residues in the favored region, and the refined model for ARS-917 shows two Ramachandran outliers and 98.2% of the residues in the favored region38.
Reporting Summary. Further information on experimental design is available in the Nature Research Reporting Summary linked to this article.
Data availability. Coordinates and structure factors have been deposited in the Protein Data Bank under accession codes PDB 6B0V (KRASG12C bound to ARS- 107) and PDB 6B0Y (KRASG12C bound to ARS-917). Source data for Figs. 1 and 2 and Supplementary Figs. 1, 3, 4 and 5 are available with the paper online. All other data are available from the authors upon reasonable request.

References
36.Bulaj, G., Kortemme, T. & Goldenberg, D. P. Ionization-reactivity relationships for cysteine thiols in polypeptides. Biochemistry 37, 8965–8972 (1998).
37.Winn, M. D. et al. Overview of the CCP4 suite and current developments. Acta Crystallogr. D Biol. Crystallogr. 67, 235–242 (2011).
38.Chen, V. B. et al. MolProbity: all-atom structure validation for macromolecular crystallography. Acta Crystallogr. D Biol. Crystallogr. 66, 12–21 (2010).
Corresponding author(s): Patrick Zarrinkar
Initial submission Revised version Final submission
Life Sciences Reporting Summary
Nature Research wishes to improve the reproducibility of the work that we publish. This form is intended for publication with all accepted life science papers and provides structure for consistency and transparency in reporting. Every life science submission will use this form; some list items might not apply to an individual manuscript, but all fields must be completed for clarity.
For further information on the points included in this form, see Reporting Life Sciences Research. For further information on Nature Research policies, including our data availability policy, see Authors & Referees and the Editorial Policy Checklist.
ti Experimental design
1.Sample size

Describe how sample size was determined.
None of the conclusions in the paper are based on formal statistical calculations, therefore no formal sample size calculations were performed for any of the experiments. The number of replicates for each experiment is provided, and in almost all cases is between two to four, chosen to provide confidence that the consistency and reproducibility of the results is sufficient to support the conclusions, based on the absolute variability observed across replicates and provided in the Tables and Figures.

2.Data exclusions

Describe any data exclusions.
For biochemical engagement assays data points below the limit of detection or quantitation were excluded.

3.Replication

Describe whether the experimental findings were reliably reproduced.
All data in Table 1 and shown in Figure 1, Supplementary Figure 1, and Supplementary Figure 4 are based on three independent replicate experiments, except for the kinact/Ki, Ki and kinact values for ARS-853, which are based on four independent experiments. Results were consistent each time these experiments were performed once methods development was complete for each type of experiment.
All data in Table 2 and shown in Supplementary Figure 3a are also based on three independent replicate experiments, except for kintrinsic for the RFAACAA peptide, which is based on four independent replicate experiments. Results were consistent each time these experiments were performed once methods development was complete for each type of experiment.
The gain of nucleotide exchange inhibition shown in Figure 2a and the engagement competition shown in Figure 2c each were from a single experiment and were not repeated once methods development was complete.
The comparison of engagement rates at pH 7.5 and 10 shown from a single experiment in Supplementary Figure 3b was performed two additional times, with similar results. The determination of cysteine pKa values with N-benzylacrylamide shown in Supplementary Figure 3c was performed once to confirm the values already obtained with iodoacetamide, and was not repeated.
The kinact/Ki experiments with ARS-853 at pH 9.5 and with the D92C mutant each were performed twice, as shown. No additional replicates were performed. Crystal structures were obtained once from one crystal each.

4.Randomization

Describe how samples/organisms/participants were allocated into experimental groups.
All experiments described are biochemical or biophysical experiments in which all samples are of defined composition and not subject to biological variability. Randomization was therefore not performed.

5.Blinding

Describe whether the investigators were blinded to group allocation during data collection and/or analysis.
All biochemical experiments described, except for the nucleotide exchange experiments, utilized a mass spectrometry readout, where sample analysis is largely automated. The experiments themselves, however, are not blinded as a

single researcher designs and executes the experiment, and it was not practical to do otherwise for biochemical experiments.
Note: all studies involving animals and/or human research participants must disclose whether blinding and randomization were used.
6.Statistical parameters
For all figures and tables that use statistical methods, confirm that the following items are present in relevant figure legends (or in the Methods section if additional space is needed).

n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement (animals, litters, cultures, etc.) A description of how samples were collected, noting whether measurements were taken from distinct samples or whether the same
sample was measured repeatedly
A statement indicating how many times each experiment was replicated
The statistical test(s) used and whether they are one- or two-sided (note: only common tests should be described solely by name; more complex techniques should be described in the Methods section)
A description of any assumptions or corrections, such as an adjustment for multiple comparisons
The test results (e.g. P values) given as exact values whenever possible and with confidence intervals noted
A clear description of statistics including central tendency (e.g. median, mean) and variation (e.g. standard deviation, interquartile range) Clearly defined error bars
See the web collection on statistics for biologists for further resources and guidance.
ti Software
Policy information about availability of computer code
7.Software

Describe the software used to analyze the data in this study.
Kinetic analysis and fitting was performed in GraphPad Prism.
Mass spectrometry raw data files were processed with Protein Deconvolution Software (Thermo Scientific) and Skyline open source software (University of Washington).
Crystallography diffraction data were analyzed with iMosflm, structures solved by molecular replacement using Phaser crystallographic software, and refined with Refmac, all part of the CCP4 software suite [Winn et al. Acta. Cryst. D67, 235 (2011)].

For manuscripts utilizing custom algorithms or software that are central to the paper but not yet described in the published literature, software must be made available to editors and reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). Nature Methods guidance for providing algorithms and software for publication provides further information on this topic.

ti Materials and reagents
Policy information about availability of materials
8.Materials availability

Indicate whether there are restrictions on availability of unique materials or if these materials are only available for distribution by a for-profit company.
9.Antibodies
Protein reagents and compounds may be generated as described in the Methods. Compounds not commercially available and protein reagents may be requested, in reasonable amounts, from the authors.

Describe the antibodies used and how they were validated for use in the system under study (i.e. assay and species).
No antibodies were used.

10.Eukaryotic cell lines
a.State the source of each eukaryotic cell line used. No eukaryotic cell lines were used.
b.Describe the method of cell line authentication used. No eukaryotic cell lines were used.

c.Report whether the cell lines were tested for mycoplasma contamination.

No eukaryotic cell lines were used.

d.If any of the cell lines used are listed in the database of commonly misidentified cell lines maintained by ICLAC, provide a scientific rationale for their use.
No eukaryotic cell lines were used.ARS853

ti Animals and human research participants
Policy information about studies involving animals; when reporting animal research, follow the ARRIVE guidelines
11.Description of research animals

Provide details on animals and/or animal-derived materials used in the study.
No animals were used in this study.
Policy information about studies involving human research participants
12.Description of human research participants

Describe the covariate-relevant population characteristics of the human research participants.
This study did not involve human research participants.