Mechanism-based inactivator of isocitrate 1 and 2 from Mycobacterium

Truc V. Phama, Andrew S. Murkinb, Margaret M. Moynihanb,1, Lawrence Harrisc, Peter C. Tylerc, Nishant Shettya,d, James C. Sacchettinia,e, Hsiao-ling Huange,f, and Thomas D. Meeka,2

aDepartment of Biochemistry and Biophysics, Texas A&M University, College Station, TX 77843; bDepartment of Chemistry, University at Buffalo, The State University of New York, Buffalo, NY 14260; cThe Ferrier Research Institute, Victoria University of Wellington, Wellington 5046, New Zealand; dFujifilm Diosynth Biotechnologies Texas, College Station, TX 77845; eDepartment of Chemistry, Texas A&M University, College Station, TX 77843; and fAlbany Molecular Research, Inc., Albany, NY 12203

Edited by Perry Allen Frey, University of Wisconsin, Madison, WI, and approved June 7, 2017 (received for review May 2, 2017) Isocitrate (ICL, types 1 and 2) is the first of the McFadden (16) demonstrated that the affinity label, 3-bromopyruvate glyoxylate shunt, an essential pathway for Mycobacterium tuberculosis (3BP) (Fig. 1), exerted time-dependent inactivation of Escherichia (Mtb) during the persistent phase of human TB infection. Here, we coli ICL. A crystal structure of Mtb ICL1 treated with 3BP report 2-vinyl-D-isocitrate (2-VIC) as a mechanism-based inactivator revealed that Cys191 had been S-pyruvoylated (10). One could of Mtb ICL1 and ICL2. The enzyme-catalyzed retro-aldol cleavage envision synthetic analogs of D-isocitrate that bear nascent elec- of 2-VIC unmasks a Michael substrate, 2-vinylglyoxylate, which then trophilic substituents at C-2, which are unmasked only following forms a slowly reversible, covalent adduct with the thiolate form of ICL-catalyzed retro-aldol cleavage. We speculated that 2-C-vinyl- active-site Cys191. 2-VIC displayed kinetic properties consistent with D-isocitrate (2-VIC) (numbering in Fig. 1 based on D-isocitrate) covalent, mechanism-based inactivation of ICL1 and ICL2 with high would provide such an inactivator. ICL-catalyzed cleavage of the efficiency (partition ratio, <1). Analysis of a complex of ICL1:2-VIC by C2–C3 bond would produce succinate as well as an enzyme-bound electrospray ionization and X-ray crystallography Michael acceptor, 2-vinylglyoxylate (2VG), which is poised to re- S confirmed the formation of the predicted covalent -homopyruvoyl act with the thiolate form of the proximal Cys191. Mechanism- adduct of the active-site Cys191. based inactivation of target to produce drug candidates

has a history of proven success (17–19). In this work, we describe BIOCHEMISTRY mechanism-based inactivation | isocitrate lyase | tuberculosis | 2-vinyl the preliminary kinetic and structural analysis of the mechanism- isocitrate | covalent adduct based inactivation of Mtb isocitrate lyases 1 and 2 by 2-VIC.

uberculosis (TB) is the leading cause of death from an in- Results Tfectious disease. The World Health Organization reported Time-Dependent Inactivation of Mtb ICL1 and ICL2 with 2-VIC. Pre- that, in 2014, an estimated 9.6 million people became infected with incubation of ICL1 (800 nM) with 0–40 μM2-VIC(5a) (chemistry TB, with 1.5 million deaths [World Health Organization report: described in SI Appendix) over a time course of 0–70 min, followed Global Tuberculosis Report 2015 (1)]. Infection by Mycobacterium by dilution into reaction mixtures containing high concentrations tuberculosis (Mtb), the causative agent of TB, may assume a latent of the substrates glyoxylate and succinate, resulted in a time- state when harbored within macrophage phagosomes during ex- dependent loss of ICL1 activity conforming to first-order kinetics tended, “persistent” stages (2). In this hypoxic environment, fatty (Fig. 2A). More than 95% of ICL1 was inactivated after a 65-min acids provide the primary carbon source, and mycobacteria activate the glyoxylate shunt, a functional abridgement of the tricarboxylic Significance acid cycle (3). Here, D-isocitrate is converted to glyoxylate and succinate by the action of two isocitrate lyases (ICLs) of 27% se- Tuberculosis, caused by Mycobacterium tuberculosis (Mtb)bacte- quence identity, ICL1 (428 aa) and ICL2 (766 aa), encoded by the ria, is the most prevalent infectious disease, affecting one-third of genes icl1 and aceA, respectively (4–7). Glyoxylate is subsequently the global population, especially in developing countries. First-line converted to L-malate by , encoded by the gene glcB therapies to treat this disease are losing efficacy due to the (8). Enzymes of the glyoxylate shunt are found in , lower emergence of drug resistance. Accordingly, new therapeutic , and , but are absent in mammals (6). Deletion of agents of novel mechanisms of action remain an urgent medical both ICL genes leads to growth impairment of Mtb in infected mice need. The isocitrate lyases (ICL1 and ICL2) comprise metabolically and rapid elimination of from the lungs (4, 5). The ICL essential enzymes of Mtb, are absent in mammals, and thereby inhibitor, 3-nitropropionate (3NP), inhibited Mtb in cell cultures (4, 9, provide therapeutically important drug targets for tuberculosis. 10). Collectively, these results demonstrated the essentiality of the Here, we describe the first example of a mechanism-based inac- ICLs in persistent-stage Mtb. Despite an absence of drug quality tivator of ICL1 and ICL2 that could provide a starting point for the inhibitors of Mtb ICLs, they remain validated targets for the devel- development of new drugs to treat tuberculosis. opment of new drugs to treat TB. Accordingly, the exploitation of the Author contributions: T.V.P., A.S.M., M.M.M., L.H., P.C.T., H.-l.H., and T.D.M. designed chemical mechanism to discover covalent inactivators could rein- research; T.V.P., A.S.M., M.M.M., L.H., N.S., H.-l.H., and T.D.M. performed research; vigorate drug discovery for the ICLs, particularly because their active T.V.P., A.S.M., L.H., P.C.T., N.S., J.C.S., H.-l.H., and T.D.M. analyzed data; and T.V.P., sites contain conserved, catalytic cysteines. Compounds that form A.S.M., M.M.M., J.C.S., H.-l.H., and T.D.M. wrote the paper. covalent bonds, especially reversible ones, with cysteine residues in or The authors declare no conflict of interest. near the enzymatic active sites have received renewed attention as a This article is a PNAS Direct Submission. strategy for the development of enzyme inactivators (11, 12). Freely available online through the PNAS open access option. The catalytic mechanism of Mtb ICL derived from both struc- Data deposition: The atomic coordinates and structure factors have been deposited in the tural data (10) and kinetic analysis (13–15)isdepictedinFig.1. , www.pdb.org (PDB ID code 5DQL). (2R,3S)-Isocitrate (1)[D-isocitrate (IC)] coordinates an active-site 1Present address: Janssen Research and Development, Malvern, PA 19355. magnesium ion and undergoes a base-catalyzed retro-aldol re- 2To whom correspondence should be addressed. Email: [email protected]. a 2 3 action ( ) to form glyoxylate ( )andtheaci-anion ( ) of succinate. This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. Cys191 then protonates C-2 of 3 to afford succinate (4). Ko and 1073/pnas.1706134114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1706134114 PNAS Early Edition | 1of6 Downloaded by guest on September 25, 2021 ICL1), but 2-VIC displayed no cytotoxicity in human dermal fi- broblasts upon treatment for 72 h at ≤400 μM.

Protection of ICL from 2-VIC Inactivation by D-Malate, Glyoxylate, Succinate, and Added . Concentrations of 0.7 mM succinate or 0.1 mM glyoxylate afforded protection from inactivation by 30 μM 2-VIC. D-Malate, a competitive inhibitor of isocitrate (Ki = 310 ± 30 μM; SI Appendix,Fig.S1), served as an isocitrate surrogate to demonstrate that the inactivation effected by 2-VIC occurs at the . In preincubation studies of ICL1 with 50 μM2-VIC,the presence of 0.1–3.0 mM D-malate exhibited a concentration- Fig. 1. Chemical mechanism of isocitrate lyase and structures of inactivators. dependent ablation of ICL1 inactivation by 2-VIC (SI Appendix, Proposed two-step chemical mechanism of Mtb ICL1 based on structural (10) Fig. S2). The inactivation was almost entirely eliminated at 3 mM and kinetic (13–15) data. Structures of 3-bromopyruvate, 2-vinyl D-isocitrate, D-malate, which is 10-fold higher than its inhibition constant. The and 2-vinylglyoxylate. concentration dependence of protection from inactivation by D-malate was evaluated by fitting data (SI Appendix,Eq.S7), resulting in an apparent dissociation constant of K = 360 ± 60 μM, ≥ μ d preincubation with 2-VIC at concentrations of 20 M. Data were nearly equal to its value of K , and a Hill coefficient n = 0.56 (SI 1 i fitted globally to Eq. , resulting in a maximal rate constant of in- Appendix,Fig.S2, Inset). = ± −1 activation, kinact 0.080 0.006 min , and a concentration of Addition of either DTT or glutathione to preincubation mixtures = ± μ 2-VIC leading to half-maximal inactivation, Kinact 22 3 M. diminished inactivation of ICL1 by 2-VIC, presumably by in- Using the method of Kitz and Wilson (20), the apparent first-order 2 terception of an enzyme-generated electrophilic species by reaction rate constant of inactivation, kobs,obtainedfromEq. demon- with the added thiols. Addition of micromolar concentrations of strated a hyperbolic dependence on the concentration of the inac- 3 either DTT or glutathione to preincubation mixtures of ICL1 and tivator upon fitting to Eq. (Fig. 2C), indicating saturation behavior 2-VIC demonstrated concentration-dependent protection from of inactivation. The (2R,3R)-diastereomer of 2-VIC (5b) displayed – μ 2-VIC inactivation of ICL1. As shown in SI Appendix,Fig.S3A, no inactivation of ICL1 when preincubated with enzyme at 1 80 M increasing fixed concentrations of 10–1,000 μM DTT in prein- concentrations for as long as 1,200 min. 2-VIC effected no appre- cubation mixtures afforded increasing protection from inactivation, ciable time-dependent inhibition of the coupling enzymes isocitrate and indicated apparent, saturable binding of DTT to ICL1. Fitting of dehydrogenase or . A sample of 2-VIC– the fractional protection versus [DTT] (SI Appendix,Fig.S3A, Inset) inactivated ICL1 that retained 7% residual ICL1 activity [(v /v ) = i 0 as described provided an apparent binding constant of DTT of K = 0.07 ± 0.02] was then subjected to spin filtration (Amicon Ultra) d 8.4 ± 0.1 μM. Similar results were found from protection studies comprising a 100,000-fold dilution of 2-VIC. The fraction of using glutathione (K = 3.8 ± 0.2 μM; SI Appendix,Fig.S3B). DTT remaining activity, compared with a control sample, was v /v = d i 0 has been shown to reduce the ability of 3BP to inactivate ICL from 0.08 ± 0.01, invariant from the input sample, suggesting that the inactivation was apparently irreversible. However, 50% activity was restored following 24-h dialysis (room temperature) of replicate – 0.06 samples of 2-VIC inactivated ICL1 against 50 mM Hepes (pH 7.5) 2-VIC A (µM) C 0.05 and 10 mM MgCl2; dialysis for 24 and 48 h in the same buffer 1.0 ) ICL1 0.0 -1 containing 10 mM DTT resulted in activity recoveries of 50% and 0.04 0.03 nearly 100%, respectively. Accordingly, the formation of an ap- 5.0 (min v /v 0.02 i 0 10.0 obs k parently covalent adduct of 2-VIC and ICL1 was reversible at room 0.01 −1 0.1 temperature, albeit at a slow rate of reactivation (0.0002 min ). 20.0 0.00 0 102030405060 2-VIC also demonstrated saturable, time-dependent inactivation 2-VIC (µM) ICL1 30.0 = 40.0 of ICL2, from which we obtained kinetic parameters of kinact v/v −1 D 0.019 ± 0.001 min and K = 420 ± 70 μM, upon global fitting 020406080 ) 0.020

inact -1 1 Preincubation Time (Min) ICL2 ofthedatatoEq. (Fig. 2B). As with ICL1, replotting values of 0.015 (min kobs versus [2-VIC] indicated a hyperbolic dependence of enzyme

obs 0.008 k inactivation by 2-VIC (Fig. 2D). Kinetic parameters for B and mechanism-based inactivation are summarized in SI Appendix, 2-VIC 0.000 (mM) 0.0 1.0 2.0 3.0 Table S1. Briefly, initial velocity data of the forward reaction 1.0 0.0 2VIC (mM) (isocitrate cleavage) resulted in kinetic parameters of kcat = 1,000 ± − − − 0.2 E

1 1 1 M)

120 min , KIC = 37 ± 8 μM, and kcat/KIC = 28 ± 5 μM ·min for 1.0 µ 30 −1 0.4 20 ICL1, and kcat = 102 ± 3min , KIC = 100 ± 12 μM, and kcat/KIC = vi/v0 0.6 v /v − − i 0 10 Slope= 1 1 0.8 1.24±0.04

± μ · Released 1.0 0.2 M min for ICL2, demonstrating that, although the 1.8 0.5 0

2.0 Succinate ( 0 102030 affinity of isocitrate binding to either isozyme differs by threefold, the 0.1 Inactivated ICL1 (µM) ICL2 turnover of ICL2 is only 10% that of ICL1. The kinetic parameter 0.0 0 50 100 150 200 250 300 0246 kinact/Kinact is analogous to the specificity constant kcat/KIC (that is, the Preincubation Time (Min) 2-VIC (µM) higher the value, the more efficient the inactivator) (19). Values of −3 −1 −1 kinact/Kinact for 2-VIC of (3.5 ± 0.7) × 10 μM ·min (ICL1) and Fig. 2. Inactivation of ICL1 and ICL2 by 2-VIC. Residual activity (vi/v0)of(A)ICL1 − − − (4.6 ± 0.7) × 10 5 μM 1·min 1 (ICL2) are 7,800- and 22,400-fold (800 nM) following preincubation with 0–40 μM2-VICand(B)ICL2(15.0μM) with 0–3mM2-VIC.C and D are replots of rates of inactivation, (kobs) vs. [2-VIC], lower, respectively, than kcat/KIC, despite the fact that comparable values of K and K were observed. Ratios of k /k for from A and B, respectively. (E) Residual ICL1 activity (vi/v0) after preincubation IC inact cat inact of 1.0 μMICL1with0–5.0 μM 2-VIC. The line drawn through the experimental ICL1 and ICL2 were 13,000 and 5,000, respectively, indicating that data points results from data fitting to Eq. 4, resulting in a value of p,the reaction steps of inactivation are considerably slower than for partition coefficient, of 0.24 ± 0.04. The curve drawn through the data results turnover of isocitrate. 2-VIC was an inhibitor of human isocitrate from fitting of the data to Eq. 5.(Inset) Concentrations of 2-VIC–inactivated dehydrogenase (IC50 of 10 ± 1.3 μM, similar to its binding to Mtb ICL1 versus product succinate formed during inactivation (slope = 1.24).

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1706134114 Pham et al. Downloaded by guest on September 25, 2021 S315 H193 W283 _ S315 - R228 O O S317 H193 O O k _ H 3 O _ k 2VIC T347 _ _ O O H193 1 2-VIC C191 S315 O O _ S317 O k O C191 HS-Cys 4 2+ 191 k Mg2+ Mg H C191 2 - S91 B O H T347 D153 O 6 S Mg2+ 2+ 2+ Mg E - 2VIC O Mg Cys 191 S91 closed D153 H K B E closed E open k 10 2VG E closed -2VG-aci-Succ k _ 9 - k k O 6 5 O S317 - O S Cys191 S317 S315 H193 S315 H193 W283 E-2VG W283 SUCC R228 _ k k7 _ O O 11 OO _ k R228 O S 12 T347 T347 C191-HP 8 Cys191 k O 8 Succ/ 2-VG C191 O - Thiols O O S W93 6 Cys191 2+ 2+ Mg Mg S91 2+ 2+ S91 Mg O Mg H H D153 B B D153 E closed-HP Eclosed-2VG-Succ

Fig. 3. Proposed inactivation mechanism of ICL1 by 2-vinyl isocitrate. Crystal structure of (Upper Left) unliganded Mtb ICL1 (Eopen; blue) showing closure (Eclosed; BIOCHEMISTRY orange) of the active-site loop upon substrate binding (k1). 2-VIC docked into ICL1 (Eclosed–2VIC), for which the Inset shows the base-catalyzed, retro-aldol cleavage of 2-vinyl isocitrate (k3) forming aci-succinate (E–2VG–aci-Succ; green) and 2VG (6) (magenta). Protonation of aci-succinate by Cys191 forms succinate (k5) and Cys191 − thiolate. Desorption of succinate (k7) provides steric freedom for Cys191-S in Eclosed–2VG to proceed to either reaction with enzyme-bound 2VG (k11)(8,Eclosed–HP) or desorption (k9)of2VG.Inthek8 step, either succinate (succ) or DTT (thiols) may enter the active site.

E. coli, as a result of its reaction with 3BP during preincubation with analysis of ICL2 resulted in P = 0.6 ± 0.1 (SI Appendix,Fig.S4), the enzyme (16). To determine whether DTT reacts with intact wherein the titration curve significantly deviated from linearity. 2-VIC in solution in the absence of enzyme, we combined 0.5–1.0 mM Correction for the reversibility of the covalent adduct is pro- 2-VIC with 1 mM DTT for 60 min, which was then diluted 10-fold vided by plotting of data in Fig. 2E and SI Appendix,Fig.S4to into reaction mixtures containing ICL1, yielding final concentra- Eq. 5,forwhichKd is the dissociation constant for the covalent tions of 100 μMDTTand100μM 2-VIC. Both DTT-treated 2-VIC adduct formed from 2-VIC and ICL1/2. Fitting yielded values of and untreated 2-VIC in the presence of 100 μM DTT effected P = k9/k11 = 0.11 ± 0.04 (ICL1) and 0.13 ± 0.08 (ICL2) and Kd = −1 complete inactivation of ICL1 at identical rates of 0.009 min . k9k12/k10k11 = 0.04 ± 0.02 μM (ICL1) and 4 ± 2 μM (ICL2) (rate These results indicated that 2-VIC does not react with DTT in the constants as in Fig. 3). Note that when none of the covalent adduct time frame of the inactivation studies. Accordingly, either the decomposes to form free enzyme, k12 = Kd = 0, and Eq. 5 is compounds form covalent adducts with an electrophilic species identical to Eq. 4.ValuesofKd, although poorly determined −7 outside the active site of ICL (which would not demonstrate satu- from fitting, suggested that, for ICL1 and ICL2, k12/k10 ∼ 10 M − ration behavior), thereby preventing its rebinding and reaction with and ∼ 10 5 M, respectively, indicating a more rapid recovery of ICL, or both thiol compounds are capable of gaining access to an enzyme activity from the ICL2-adduct than that of ICL1. enzyme-bound electrophile generated from 2-VIC or its putative The partition ratio was also evaluated by a second approach. The covalent enzyme-bound adduct. The latter is more likely due to the ICL-catalyzed generation of succinate from 2-VIC compared with small partition ratios observed (see below), indicating that very little the amount of inactivated enzyme provides a measure of the stoi- of the enzyme-generated electrophilic species escapes from the chiometry for inactivation. We ascertained the concentration of active site before the formation of a covalent adduct with ICL1. succinate produced from 2-VIC during the mechanism-based in- activation of ICL1 with saturating concentrations of 2-VIC. Here, Partition Ratio. To determine the efficiency of 2-VIC inactivation, concentration of free enzyme is negligible for rebinding of 2VG (k10 we measured its partition ratio: the fraction of bound 2-VIC that step,Fig.3)andthepreincubationtime is sufficiently short (90 min) dissociates versus that which forms a covalent adduct with ICL1 that the reversal of covalent adduct formation (k12 step) was in- (19). The residual catalytic activity (vi/v0)of1.0μMICL1wasde- significant. The stoichiometric ratio of quantifiable succinate versus termined after a 20-h preincubation with 0–5.0 μM 2-VIC (Fig. 2E). inactivated ICL1 (Fig. 2E, Inset) indicated that 1.24 molecules of The value of (vi/v0) failed to reach zero when [2-VIC] exceeded succinate were released per inactivation event, thereby providing an ∼ [ICL1], such that a constant level of 10% ICL1 activity remained at independent assessment of the partition ratio (k9/k11 = 0.24). A [2-VIC] ≤ 5.0. Because inactivated ICL1 recovered 50% activity at similar analysis of ICL2 demonstrated that 1.4 ± 0.1 molecules of room temperature after 24 h, conditions similar to this study, we succinate were released per inactivation event of ICL2 (P = 0.4 ± attribute this loss of linearity to the slow reversibility of the covalent 0.1; SI Appendix,Fig.S4, Inset), in good agreement with results from adduct. Fitting of the data plotted as activity (vi/v0)versus[2-VIC] fitting to Eqs. 4 and 5. These results indicate that 2-VIC is a highly (Eq. 4;[2-VIC]= 0–5.0 μM, [ICL1] = 1.0 μM) resulted in an ex- efficient inactivator of both ICL1 and ICL2, where the partition trapolated x interceptof1.24± 0.04, from which we obtained a ratios, averaged from all three evaluations above, are, respectively, partition ratio (P = k9/k11,Fig.3)of0.24± 0.04. This indicates that P = 0.2 ± 0.08 and 0.4 ± 0.2. The rate of succinate formation (ksucc) 1.2 mol of 2-VIC was required to inactivate 1 mol of ICL1. A similar from 2-VIC (0.5 mM) catalyzed by ICL1 (20 μM) was also measured

Pham et al. PNAS Early Edition | 3of6 Downloaded by guest on September 25, 2021 and compared with its corresponding and apparent rate of enzyme the Cα of His193), to yield a closed, solvent-inaccessible active site −1 – inactivation for this concentration of 2-VIC; ksucc = 0.06 min and (10). A crystal structure of 2-VIC treated ICL1 at 1.8-Å resolution −1 kinact = 0.04 min . The rate of succinate cleavage from 2-VIC is displayed the electron density of a covalent, thioether-linked, −1 much slower than full turnover of isocitrate (kcat = 1,000 min ), homopyruvoyl (HP) moiety attached to the active-site Cys191 in indicating that carbon–carbon bond cleavage is very slow for 2-VIC all four subunits of the asymmetric unit (Fig. 4A). Occupancy of and that the reaction of 2-vinylglyoxylate with a residue on ICL1 the covalent adduct was different in the four monomers, which cannot be the sole rate-limiting step of inactivation. likely reflects the observed reversibility of the homopyruvoyl ad- duct during crystallization. Chain A exhibited the highest occupancy Mass Spectrometry of Mtb ICL1 Treated with 2-VIC. Electrospray ioni- for the electron density of all atoms, and all of the interactions de- zation time-of-flight mass spectrometry was performed to affirm the scribed below refer to those of chain A. Although the α-hydroxy- existence of an enzyme–inactivator covalent adduct. An untreated carboxylate substituent of 2-VIC would be expected to bind initially + sample of ICL1 enzyme demonstrated a peak of an average mass of to the Mg2 ion to undergo retro-aldol cleavage to form 2VG, the + 48,788 ± 1 Da, consistent with the theoretical molecular weight of observed homopyruvoyl adduct is not coordinated to this Mg2 ion. ICL1 monomers (48,787 Da) (SI Appendix,Fig.S5A). A 2-VIC– Instead, the covalently tethered keto acid (electron density was + treated ICL1 sample displayed none of this peak, but instead a major modeled as an α-keto acid) has turned away from the Mg2 ion peak at 48,887 ± 1 Da was observed (SI Appendix,Fig.S5B). The toward the succinate , as in the case of 3BP (10). The apparent increased mass of the enzyme by 99 ± 1 Da is consistent closed active site provides adequate sequestration of the reactive with the addition of C4H3O3 to an ICL1 monomer resulting from 2VG for its eventual migration to and reaction with Cys191. One of formation of a covalent adduct between ICL1 and 2-vinylglyoxylate its carboxylate oxygens forms apparent hydrogen bonds with His193 (2VG). Treatment of the 2-VIC–inactivated ICL1 with 10 mM DTT and Ser315 (both at distances of 2.5 Å), whereas the other carbox- (90 min; 37 °C) demonstrated an approximate 60:40 ratio of the ylate oxygen is within hydrogen-bonding distance to Asn313,Ser317, 48,887-Da and 48,788-Da (2-VIC–ICL1 and ICL1) species, in- and Thr347 (3.3-, 2.6-, and 2.8-Å distances, respectively; Fig. 4B). dicating that treatment with DTT had removed nearly 40% of the The α-keto oxygen forms no apparent hydrogen bonds with the covalent adduct from the inactivated enzyme (SI Appendix,Fig. protein, and it is oriented toward the indole ring of Trp93;Trp93 S5C). In concert with these results, kinetic analysis demonstrated possibly prevents the recoordination of the cysteine-tethered keto 2+ that, upon addition of 10 mM DTT (37 °C), 2-VIC–treated acid to the Mg ion. In the crystal structure of the C191S ICL1 was reactivated to 30% or more of its initial activity before ICL1 mutant complexed with glyoxylate and 3NP, the bound 3NP is treatment with 2-VIC (SI Appendix,Fig.S6). structurally indistinguishable from succinate, and this structure pro- vides a model for the ICL1–glyoxylate–succinate complex (10). When Crystallographic Analysis of 2-VIC–Treated Isocitrate Lyase. As kinetic the 3NP-bound structure is overlaid with the HP-ICL structure, the and mass-spectrometric characterization of the inactivation of HP moiety appears to bind preferentially in the succinate binding ICL1 by 2-VIC is consistent with alkylation of the enzyme, we pocket (Fig. 4C). In order for the homopyruvoyl group to adopt these sought a crystallographic assessment of the mechanism of interactions, the succinate formed from 2-VIC would need to dis- + inactivation. The active site of the ICL1–Mg2 enzyme assumes an sociate from enzyme. Additionally, the homopyruvoyl group attached open, solvent-accessible conformation (10). The ICL1–glyoxylate– to C191 in ICL1 assumes a binding mode that is very similar to that 3NP complex of the C191S mutant of ICL1 was characterized by of the S-pyruvoylated form of 3BP-treated ICL1 (10). Fig. 4D shows the overlay of the structures of 3BR-treated ICL1 (gray; PDB ID movement of the K189KCGH193 loop by about 15 Å (measured as code 1F8M) and 2-VIC–treated ICL1 (purple). All of the residues in the active site are nearly superimposable. In both covalently modified + structures, the binding site in which glyoxylate is chelated to the Mg2 A T347 B S315 ion is vacant, as both S-pyruvoyl and S-homopyruvoyl groups bind in N313 N313 the succinate binding pocket, and interact with the same five residues D153 T347 2.6 Å W93 2.7 Å 3.0 Å via apparent hydrogen bonds. Unlike the S-homopyruvoyl structure, S317 3.3 Å only one of the carboxylate oxygens from the S-pyruvoyl group is in S315 2.5 Å adequate proximity to Thr347,Ser315,andAsn313 for hydrogen H193 H193 bonding. The α-keto group of the S-pyruvoyl complex is within D153 G192 hydrogen-bonding distance to Ser317 and His193. W93 C191 C191 S317 Effects of Ligand Binding on the Intrinsic Protein Fluorescence of ICL1. The intrinsic protein fluorescence of Mtb ICL1 (λex = 290 nm, λem = S315 S315 CDN313 N313 320 nm) is diminished by ligand binding, and by covalent reaction with T347 W93 T347 3BP and 2-VIC (Fig. 5A). Tyrosine and tryptophan residues proximal W93 S317 S317 totheactivesiteofICL1[e.g.,Tyr89 and Trp93 (Figs. 3 and 4)] likely

dd contribute to the observed changes in fluorescence upon ligand binding H193 D153 H193 in the active site. In comparison with the unliganded enzyme, 3BP- D153 G192 G192 inactivated ICL1 exhibits the largest decrease in protein fluorescence C191 − − C191/S191 ( 60%) followed by ICL1 plus Glx ( 50%), ICL1 plus isocitrate (−55%), and 2-VIC-inactivated enzyme (−40%). Because an equilib- rium between isocitrate and the products of enzymatic turnover is Fig. 4. X-ray structural characterization of 2-VIC–treated ICL1. (A) Density map established within minutes of mixing ICL1 and isocitrate at our ex- structure of ICL1 displaying the S-homopyruvolyated Cys191 residue following perimental concentrations, the nearly identical fluorescence emission treatment with 2-VIC. (B)Cys191 modified with 2-VIC (purple). Potential hydro- spectra of ICL1 plus isocitrate and ICL1 plus Glx suggests that the gen bonds (dashed lines) and interatomic distances are shown. (C)Superposi- binary ICL1–Glx complex, and not ICL1–isocitrate, would be the tion of crystal structures of ICL1 bound to glyoxylate and 3-NP (gold) or after predominant species in the ICL1 plus isocitrate sample. The decrease treatment with 2-VIC (purple). (D) Superposition of active-site structures of in intrinsic protein fluorescence accompanies a significant change in ICL1 including Cys191 (yellow) treated with 3-bromopyruvate (S-pyruvoylated enzyme, gray) (10) and 2-VIC (S-homopyruvoylated, purple). The labeling ICL1 conformation, as observed in crystallographic analysis of ICL1, scheme is as follows: nitrogen, blue; oxygen, red; sulfur, yellow; water, red arising in part from the large (15-Å) movement of the aforementioned 2+ spheres; Mg ion, green sphere; and hydrogen bonds, dashed lines. active-site loop (containing Cys191) upon ligand binding (10). Notably,

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1706134114 Pham et al. Downloaded by guest on September 25, 2021 −1 A B value of kcat (60 min ), which is 5% that of isocitrate (7), suggesting 1.2 1.2 No Ligands t = 7 min that the substitution of a methyl or vinyl group on carbon-2 of iso- 1.0 2-VIC 1.0 t = 13 min citrate may further reduce the ability of ICL to cleave the C2–C3 bond 0.8 Glx 0.8 t = 27 min Glx+Succ t = 42 min of 2-VIC, and render k3 a slow step in inactivation (kinact). We pro- 0.6 D,L-Isocitrate 0.6 t = 50 min pose that the E–2VG complex (vacant succinate subsite) is the 0.4 Bromopyruvate 0.4 No Ligands complex at which covalent reaction with enzyme is interdicted by 0.2 0.2 added thiols (k8). As shown in the dialysis studies that the ICL1-HP Fluorescence Fluorescence 0.0 0.0 320 340 360 380 400 320 340 360 380 400 adduct was slowly reversible by a presumed retro-Michael reaction, Wavelength (nm) Wavelength (nm) DTT likely intercepts and reacts with reformed 2-VG upon its release from Cys191. Although we observed that extended incubation of DTT Fig. 5. Intrinsic protein fluorescence changes arising from ligand binding to with ICL1-HP resulted in reactivation of enzyme, both kinetically and μ λ = ICL1. (A) Free 1 M ICL1 (blue line) displayed fluorescence maxima at max by mass spectrometric analysis, incubation of ICL1-3BP with DTT 325 nm, which is diminished upon binding of 1 mM D,L-isocitrate (orange), does not restore ICL1 activity (SI Appendix,Fig.S6). This suggests 1 mM glyoxylate (yellow), 0.1 mM 2-VIC (red), 0.1 mM 3BP (black), and that the apparent restoration of free Cys191 from S-homopyruvoylated 10 mM glyoxylate and succinate (purple). (B) Time-dependent fluorescence ICL1, possibly mediated by DTT, likely occurs by a retro-Michael change upon incubation of 1 μM ICL1 with 0.1 mM 2-VIC. reaction, a reaction not available to the S-pyruvoylated enzyme. Kinetic, mass spectrometric, protein fluorescence, molecular mod- the apparent covalent inactivation arising from added 0.1 mM 2-VIC eling, and X-ray crystallographic data are all consistent with the pro- (red) or 0.1 mM 3BP (black) led to a diminution of ICL1 fluorescence, posed chemical mechanism of inactivation of ICL by 2-VIC shown in which is consistent with the closed conformation of the active site for Fig. 3. Our results demonstrate that 2-VIC comprises a true the covalently modified ICL1 structure. We also observed a time- mechanism-based inactivator of both ICL1 and ICL2, effectively pro- dependent shift and decrease of the relative fluorescence of ICL1, viding an electrophile similar to the affinity label 3BP, namely, from 1.0 to an invariant value of 0.6, during a 20-min incubation of ICL1 2-vinylglyoxylate, almost exclusively confined to the active site of Mtb with 100 μM2-VIC(Fig.5B). An apparent, first-order rate constant of ICL, and which forms a covalent bond with its conserved active-site −1 cysteine. We believe that this type of mechanism-based inactivation change in fluorescence (kobs = 0.01 min ) was comparable to kinact obtained from kinetic studies. To evaluate the structural changes oc- may be the first of its kind for an isocitrate lyase. As a drug discovery curring during 2-VIC inactivation, 2-VIC and its reaction products strategy for TB, mechanism-based enzyme inactivation of target en- Mycobacteria BIOCHEMISTRY 2VG and succinate were docked into the closed conformation of the zymes in offers significant advantages over competitive inhibition, which include the following: (i)a“precision” covalent re- ICL1 crystal structures using AutoDock Vina (21) based on these action confined to active-site residues of the target enzyme; (ii)irre- previous crystallographic structures, and the results are discussed below. versible, or slowly reversible, covalent inactivation of the target enzyme; Discussion and (iii) the targeting of an invariant active-site catalytic group, Cys191, which should be less subject to the development of resistance due to A proposed chemical mechanism, accompanied with docked struc- mutation. Forthcoming analogs of 2-VIC, although they may bind to tures of the enzyme species and ligands formed during ICL1 in- other human enzymes, should exert little toxicity in mammalian cells activation by 2-VIC, is depicted in Fig. 3. The unliganded form of lacking an isocitrate lyase owing to their inability to generate 2-vinyl- ICL1 preferentially maintains the open conformation (Fig. 3, E ), open glyoxylate. Recently, benzothiazinones have been shown to be potent in which the active-site loop (residues 189–193) is distal to the bound 2+ antimycobacterial agents, and are mechanism-based inactivators of Mg ion. Molecular docking of 2-VIC into the active site of C191S DprE1, an enzyme located in the cell wall of Mtb. The benzothiazi- ICL1, for which the 1-carboxylate and 2-hydroxyl groups are co- + nones undergo reduction of a nitro substituent, resulting in a nitroso ordinated to Mg2 (E–2-VIC), suggests that 2-VIC binds to ICL1 – group, which subsequently reacts with a cysteine residue in the active (step k1) in the same manner as isocitrate (Eclosed 2VIC). Closing of site of DprE1 (22). The inactivation kinetic parameters of these the active site upon ligand binding (Eclosed) results in movement of this 2+ compounds are comparable to those of 2-VIC. Our future efforts will loop as well as slight movement of the Mg ion. This brings Cys191 concern the development of new analogs of 2-VIC that likewise pro- into position for catalysis. Base-catalyzed cleavage of the C2–C3 bond 2+ vide mechanism-based inactivation as well as antimycobacterial activity (k3) produces Mg -bound 2VG (green) and the aci-anionic form of in cell culture and in in vivo models of TB infection. succinate (E–2VG–aci-Succ). Protonation of aci-succinate by Cys191 (k5) then produces succinate and the thiolate form of Cys191 (E– Materials and Methods – – – 2VG Succ). Modeling of the ICL1 2VG succinate complex is con- DL-Isocitrate (trisodium salt), D-threo-isocitrate (potassium salt), NaCl, DMSO, 2+ sistent with the coordination of 2VG to the Mg ion, and its in- sodium glyoxylate (monohydrate), , D-, glutathione (re- teraction with active-site residues is similar to that of unsubstituted duced form), DTT, and all other chemicals were obtained from Sigma-Aldrich. + glyoxylate. It is noteworthy that succinate blocks access of Mg2 -bound (IDH) and lactate dehydrogenase (LDH) from E. coli were prepared as described (13). 2VG to the thiolate ion of Cys191, which is consistent with the ob- served protection from 2-VIC inactivation afforded by added succi- Enzymes. Two constructs of recombinant ICL1 from Mtb were used in these nate (k8). We propose that the expulsion of succinate from the – – studies: a tag-free form (ICL1-TF) and a hexahistidine-tagged form (ICL1), and partially or fully opened active site of E 2VG Succ (k7)allows2VGto their expression and purification are described in SI Appendix. A truncated rotate and migrate to the succinate binding site, forming the covalent form of Mtb ICL2 (amino acids 1–605) with a C-terminal His6 tag was expressed complex ICL–HP (8) via Michael addition of the Cys191 thiolate to in E. coli BL21(DE3) as described in SI Appendix. 2VG (k11). Desorption of succinate from a partially opened active site also allows 2VG either to escape from the active site (k9), or to access X-Ray Crystallography of 2-VIC–Treated ICL. ICL1:2-VIC complexes were prepared ∼ added thiols for preemptive reaction with enzyme-bound 2VG (k8). by incubating ICL-TF ( 10 mg/mL) in 50 mM Tris (pH 8.0), 10 mM MgCl2,and The release of 2VG from ICL1–2VG occurs at a rate that is 20% 0.5 mM DTT with 3 mM 2-VIC overnight at 17 °C. A concentration of 0.5 mM DTT did not impede covalent inactivation of ICL1-TF but was sufficient to prevent that of its reaction with Cys191 to form E-HP, as indicated by the oxidation of the cysteine residues of ICL1. The hanging drop vapor diffusion partition analysis from which k9/k11 = 0.20. Because little 2VG de- = method was used to produce crystals. A 1:1 volume ratio of the protein with a sorbs from the active site before formation of ICL-HP (k9/k11 0.2), it solution of 0.1 M Tris·HCl (pH 8.0), 0.2 M sodium , and 20–30% (wt/vol) PEG is unlikely that k11 is the rate-limiting step in 2-VIC inactivation, as is 4000 was used to crystalize the protein. Crystals appeared in 2–3 wk after setting supported by the measured rate of succinate formation from 2-VIC ∼3-μL drops against 500 μL of mother liquor in a closed reservoir at 17 °C. Fresh equal to 1.5kinact. ICL1-catalyzed cleavage of 2-methyl-isocitrate has a mother liquor was used as the cryoprotectant, and crystals were flash-frozen in

Pham et al. PNAS Early Edition | 5of6 Downloaded by guest on September 25, 2021 liquid nitrogen for X-ray data collection, and maintained chilled under a stream of vi = ½− liquid nitrogen (100 K) throughout data collection. High-resolution diffraction exp kobst , [2] v0 data for ICL–2-VIC cocrystals were collected on an ADSC Quantum 315 detector at

a wavelength of 0.9793 Å from beamline 19ID of the Structural Biology Center kinact½I kobs = + kbgd. [3] (Advanced Photon Source, Argonne National Laboratory). Detailed data trans- Kinact + ½I formation and structure refinement are described in SI Appendix. For the partition analysis of mechanism-based inactivation, data were fitted to Enzyme Assays. Unless otherwise specified, all assays were conducted in reaction Eqs. 4 and 5; vi and v0 are, respectively, ICL activity measured after pre-

mixtures of 50 mM Hepes (pH 7.5), 5 mM MgCl2, and 1 mM DTT at 37 °C. In the incubation with or without 2-VIC, [2-VIC] and [Et] are the micromolar direction of isocitrate lysis, product glyoxylate was either converted to glycolate concentrations of inactivator and enzyme, respectively, p is the partition co- « = −1 in a coupled-enzyme assay using E. coli lactate dehydrogenase ( 340 6,220 M efficient (k9/k11)andK′d = Kd/(1 + p) = (k9k12/k10k11)/(1 + k9/k11): ·cm−1) or reacted with phenylhydrazine-HCl to form a phenylhydrazone prod- « = −1· −1 v ½2-VIC=½E uct ( 324 17,000 M cm ). In the direction of isocitrate synthesis, isocitrate i = 1 − t , [4] product was converted to α-ketoglutarate in a coupled-enzyme assay using v0 ð1 + pÞ « = −1· −1 E. coli isocitrate dehydrogenase ( 340 6,220 M cm ). rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 0  1 ½ + ½ + ′ − ½ + ½ + ′ 2 − ½ ½ B Et 2-VIC Kd Et 2-VIC Kd 4 Et 2-VIC C Quantification of Product Succinate in ICL Inactivation Studies. Succinic acid vi B C = 1 − @ A. [5] formed from ICL1-catalyzed cleavage of 2-VIC was detected using the succinate v0 2ð1 + pÞ½Et colorimetric assay kit and protocol of Biovision.

Preincubation Inactivation Studies of ICL1 and ICL2. The time-dependent inactivation Protein Fluorescence. ICL1 (1 μM monomers) in a buffer mixture of 50 mM of ICL1 by 2-VIC (5a) was assessed by preincubation studies, in which residual en- Hepes (pH 7.5) and 10 mM MgCl was incubated at room temperature for zyme activity was measured as initial velocityofeithertheforwardorreverse 2 50 min with 1 mM DL-isocitrate, 100 μM2-VIC,or100μM3BP Reaction mix- reactions after preincubation of enzyme with 2-VIC. For the reverse reaction of ICL, . tures (200 μL) were excited at λ = 290 nm, and the intrinsic fluorescence 0.6-mL reaction mixtures containing 50 mM Hepes (pH 7.5), 5 mM MgSO ,0.8μM ex 4 λ = – ICL, and 0–40 μM inactivator were incubated in sealed Eppendorf tubes. Fifty- emission spectra ( em 315 400 nm) were measured at room temperature in a microliter samples were withdrawn from the preincubation mixtures (0–300 min) Biotek Synergy plate reader using 96-well Greiner black, clear-bottom plates. and diluted 50-fold, and residual ICL1 activity was measured by spectrophotometry The intrinsic fluorescence of ICL1 was also recorded in the presence of 10 mM (Cary 100 spectrophotometer; 340 nm) of the absorbance of NADPH formation glyoxylate and/or 10 mM succinate. using the isocitrate dehydrogenase assay. For measurement of residual ICL activity using the forward reaction, 1 μMICL1or15μM ICL2 was incubated with 0–3mM Molecular Docking. Molecular models of ICL1 binary and tertiary complexes

2-VIC in preincubation buffer containing 50 mM Hepes (pH 7.5), 5 mM MgCl2,in were built by docking of 2-VIC and 2VG plus succinate, respectively, into the the presence or absence of DTT. Samples were withdrawn from preincubation crystal structure of the C191S mutant Mtb ICL1 (PDB ID code 1F61) (10) using mixtures, diluted 50-fold, and glyoxylate product from lysis of isocitrate was Chimera/AutoDock Vina (21), which performs fitting of small-molecule li- measured via the phenylhydrazine assay as described in SI Appendix. Initial rates gands with freely rotatable bonds separated by three consecutive covalent were normalized to that of an enzyme sample with no inactivator added. bonds or fewer. Its training set allows small-molecule rotational freedom for up to 35 atoms while fixing the macromolecular receptors in rigid Analysis of Kinetic Data. Data analysis was conducted by global fitting of nonlinear conformations. 2-VIC was posed in the closed conformation of ICL1 based regression using SigmaPlot (Systat). Time-dependent inactivation of ICL was assessed on the structure of the C191S ICL1–glyoxylate complex (10), in which its by fitting of the preincubation data to Eq. 1 in which v and v are the initial rates at i 0 1-carboxylate and 2-hydroxyl groups of 2-VIC are coordinated to the mag- time t and 0, respectively, t is preincubation time, [I] is the micromolar concen- nesium ion in a manner analogous to glyoxylate. The model of the ternary tration of inactivator, k is the maximal rate constant of inactivation, K is the inact inact – – – – concentration of inactivator at which the observed rate constant of inactivation is ICL1 2VG succinate complex was based on C191S ICL1 glyoxylate 3NP complex (10). one-half that of kinact,andkbgd is the observed rate constant for background loss of ICL activity over the experimental time courses, independent of the inactivator: ACKNOWLEDGMENTS. We thank GlaxoSmithKline Pharmaceuticals for sup- vi kinact½I port of the synthesis of 2-VIC and for materials to prepare ICL2. We thank the = exp − + kbgd t . [1] v0 Kinact + ½I Argonne National Laboratory for crystallographic data. We thank Kimberly Loesch for analysis of cellular toxicity. We thank Texas A&M AgriLife Research, In addition, we fitted the time courses at each concentration of inactivator to the Welch Foundation (Grant A-0015), and the NIH (Grant P011AI095208) for Eq. 2, where resulting values of kobs were then replotted using Eq. 3: providing funding for this research.

1. World Health Organization (2015) Global Tuberculosis Report 2015 (World Health 12. Bradshaw JM, et al. (2015) Prolonged and tunable residence time using reversible Organization, Geneva), Report No. WHO/HTM/TB/2015.22. covalent kinase inhibitors. Nat Chem Biol 11:525–531. 2. Wayne LG, Lin KY (1982) Glyoxylate metabolism and adaptation of Mycobacterium 13. Moynihan MM, Murkin AS (2014) Cysteine is the general base that serves in catalysis by tuberculosis to survival under anaerobic conditions. Infect Immun 37:1042–1049. isocitrate lyase and in mechanism-based inhibition by 3-nitropropionate. Biochemistry 3. Wheeler PR, Ratledge C (1988) Use of carbon sources for lipid biosynthesis in Myco- 53:178–187. bacterium leprae: A comparison with other pathogenic mycobacteria. J Gen Microbiol 14. Schloss JV, Cleland WW (1982) Inhibition of isocitrate lyase by 3-nitropropionate, a 134:2111–2121. reaction-intermediate analogue. Biochemistry 21:4420–4427. 4. McKinney JD, et al. (2000) Persistence of Mycobacterium tuberculosis in macrophages 15. Quartararo CE, Hadi T, Cahill SM, Blanchard JS (2013) Solvent isotope-induced equi- and mice requires the glyoxylate shunt enzyme isocitrate lyase. Nature 406:735–738. librium perturbation for isocitrate lyase. Biochemistry 52:9286–9293. 5. Muñoz-Elías EJ, McKinney JD (2005) Mycobacterium tuberculosis isocitrate lyases 16. Ko YH, McFadden BA (1990) Alkylation of isocitrate lyase from by 1 and 2 are jointly required for in vivo growth and virulence. Nat Med 11:638–644. 3-bromopyruvate. Arch Biochem Biophys 278:373–380. 6. Kondrashov FA, Koonin EV, Morgunov IG, Finogenova TV, Kondrashova MN (2006) 17. Silverman RB, Holladay MW (2014) The Organic Chemistry of Drug Design and Drug Evolution of glyoxylate cycle enzymes in Metazoa: Evidence of multiple horizontal Action (Elsevier, Amsterdam), 3rd Ed, pp 247–261. transfer events and pseudogene formation. Biol Direct 1:31. 18. Walsh CT (1984) Suicide substrates, mechanism-based enzyme inactivators: Recent de- 7. Gould TA, van de Langemheen H, Muñoz-Elías EJ, McKinney JD, Sacchettini JC (2006) velopments. Annu Rev Biochem 53:493–535. Dual role of isocitrate lyase 1 in the glyoxylate and methylcitrate cycles in Mycobac- 19. Silverman RB (1995) Mechanism-based enzyme inactivators. Methods Enzymol 249: terium tuberculosis. Mol Microbiol 61:940–947. 240–283. 8. Dunn MF, Ramírez-Trujillo JA, Hernández-Lucas I (2009) Major roles of isocitrate lyase and 20. Kitz R, Wilson IB (1962) Esters of methanesulfonic acid as irreversible inhibitors of malate synthase in bacterial and fungal pathogenesis. Microbiology 155:3166–3175. acetylcholinesterase. J Biol Chem 237:3245–3249. 9. Eoh H, Rhee KY (2014) Methylcitrate cycle defines the bactericidal essentiality of isocitrate lyase 21. Trott O, Olson AJ (2010) AutoDock Vina: Improving the speed and accuracy of docking for survival of Mycobacterium tuberculosis on fatty acids. Proc Natl Acad Sci USA 111:4976–4981. with a new scoring function, efficient optimization, and multithreading. J Comput Chem 10. Sharma V, et al. (2000) Structure of isocitrate lyase, a persistence factor of Myco- 31:455–461. bacterium tuberculosis. Nat Struct Biol 7:663–668. 22. Trefzer C, et al. (2012) Benzothiazinones are suicide inhibitors of mycobacterial 11. Singh J, Petter RC, Baillie TA, Whitty A (2011) The resurgence of covalent drugs. Nat decaprenylphosphoryl-β-D-ribofuranose 2′-oxidase DprE1. J Am Chem Soc 134: Rev Drug Discov 10:307–317. 912–915.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1706134114 Pham et al. Downloaded by guest on September 25, 2021