1521-0111/88/6/1011–1023$25.00 http://dx.doi.org/10.1124/mol.115.100917MOLECULAR PHARMACOLOGY Mol Pharmacol 88:1011–1023, December 2015Copyright ª 2015 by The American Society for Pharmacology and Experimental Therapeutics

Long-Range Inhibitor-Induced Conformational Regulation ofHuman IRE1a Endoribonuclease Activity s

Nestor O. Concha, Angela Smallwood, William Bonnette, Rachel Totoritis, Guofeng Zhang,Kelly Federowicz, Jingsong Yang, Hongwei Qi, Stephanie Chen, Nino Campobasso,Anthony E. Choudhry, Leanna E. Shuster, Karen A. Evans, Jeff Ralph, Sharon Sweitzer,Dirk A. Heerding, Carolyn A. Buser, Dai-Shi Su, and M. Phillip DeYoungOncology R&D (K.F., J.Y., L.E.S., K.A.E., J.R., D.A.H., C.A.B., D.S.S, M.P.D.), Biological Sciences (R.T., G.Z., H.Q., S.C., A.E.C.,S.S.), and Chemical Sciences, GlaxoSmithKline Research and Development, Collegeville, Pennsylvania (N.O.C., A.S., W.B., N.C.)

Received July 20, 2015; accepted September 25, 2015

ABSTRACTActivation of the inositol-requiring enzyme-1 alpha (IRE1a) pro-tein caused by endoplasmic reticulum stress results in thehomodimerization of the N-terminal endoplasmic reticulumluminal domains, autophosphorylation of the cytoplasmic ki-nase domains, and conformational changes to the cytoplasmicendoribonuclease (RNase) domains, which render them func-tional and can lead to the splicing of X-box binding protein 1(XBP 1) mRNA. Herein, we report the first crystal structures of thecytoplasmic portion of a human phosphorylated IRE1a dimerin complex with (R)-2-(3,4-dichlorobenzyl)-N-(4-methylbenzyl)-2,7-diazaspiro(4.5)decane-7-carboxamide, a novel, IRE1a-selective kinase inhibitor, and staurosporine, a broad spectrumkinase inhibitor. (R)-2-(3,4-dichlorobenzyl)-N-(4-methylbenzyl)-2,7-diazaspiro(4.5)decane-7-carboxamide inhibits both the ki-nase and RNase activities of IRE1a. The inhibitor interacts withthe catalytic residues Lys599 and Glu612 and displaces the

kinase activation loop to the DFG-out conformation. Inactiva-tion of IRE1a RNase activity appears to be caused by aconformational change, whereby the aC helix is displaced,resulting in the rearrangement of the kinase domain-dimerinterface and a rotation of the RNase domains away from eachother. In contrast, staurosporine binds at the ATP-binding siteof IRE1a, resulting in a dimer consistent with RNase activeyeast Ire1 dimers. Activation of IRE1a RNase activity appearsto be promoted by a network of hydrogen bond interactionsbetween highly conserved residues across the RNase dimerinterface that place key catalytic residues poised for reaction.These data implicate that the intermolecular interactions be-tween conserved residues in the RNase domain are required foractivity, and that the disruption of these interactions can beachieved pharmacologically by small molecule kinase domaininhibitors.

IntroductionCellular stresses, such as accumulation of unfolded pro-

teins, hypoxia, glucose deprivation, depletion of endoplasmicreticulum (ER) calcium levels, and changes in ER redox statusactivate the unfolded protein response (UPR), an intracellularsignal transduction network involved in restoring proteinhomeostasis [reviewed by Walter and Ron (2011)]. To allevi-ate these types of stress responses, the UPR responds byhalting protein translation, activating transcription of UPR-associated target genes, and degrading misfolded proteins

(Harding, et al., 2002; Ron, 2002; Feldman et al., 2005). UPRsignaling also regulates cell survival bymodulating apoptosis andautophagy and can induce cell death under prolongedER stress ifthe misfolded protein burden is too high (Ma and Hendershot,2004; Rouschop et al., 2010; Woehlbier and Hetz, 2011).Three key ER membrane proteins have been identified as

primary effectors of the UPR: protein kinase R–like ER kinase(PERK), inositol-requiring enzyme-1 (IRE1) a/b, and activatingtranscription factor 6 (Schroder andKaufman, 2005). IRE1a is atransmembrane protein that functions both as an ER stresssensing receptor via its N-terminal ER luminal domain and as asignal transducer via its cytoplasmic C-terminal kinase andendoribonuclease (RNase) domains (Tirasophon et al., 1998).Upon sensing ER stress, the extracellular portion of the IRE1aprotein will homodimerize, allowing for transautophosphoryla-tion, which, in turn, induces a conformational change, resulting

All authors are past or present employees of GlaxoSmithKline. No potentialconflicts of interest were disclosed by the authors.

dx.doi.org/10.1124/mol.115.100917.s This article has supplemental material available at molpharm.

aspetjournals.org.

ABBREVIATIONS: APY29, 2-N-(3H-benzimidazol-5-yl)-4-N-(5-cyclopropyl-1H-pyrazol-3-yl)pyrimidine-2,4-diamine; BHQ-1, Black Hole quencher-1;ER, endoplasmic reticulum; FAM, 6-carboxyfluorescein fluorescent reporter; FBS, fetal bovine serum; GSK2850163, (R)-2-(3,4-dichlorobenzyl)-N-(4-methylbenzyl)-2,7-diazaspiro(4.5)decane-7-carboxamide; ID, identity; IRE1, inositol-requiring enzyme-1; KIRA6, 1-(4-[8-amino-3-tert-butylimidazo(1,5-a)pyrazin-1-yl]naphthalen-1-yl)-3-[3-(trifluoromethyl)phenyl]urea; PCR, polymerase chain reaction; PDB, Protein Data Bank; PERK, protein kinase R–likeendoplasmic reticulum kinase; pIRE1a, phosphorylated inositol-requiring enzyme-1 alpha; RNase, endoribonuclease; RT, real time; SAR, structureactivity relationship; STS, staurosporine; TEV, tobacco etch virus; UPR, unfolded protein response; UPRE, unfolded protein response element; XBP 1,X-box binding protein 1.

1011

http://molpharm.aspetjournals.org/content/suppl/2015/10/05/mol.115.100917.DC1Supplemental material to this article can be found at:

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

in activation of the RNase domains (Ali et al., 2011). Phosphor-ylation within the kinase activation loop is an essential step forRNase activation (Prischi et al., 2014). Mammalian IRE1aexcises a 26–base pair intron from the mRNA of X-box bindingprotein 1 (XBP 1), which causes a translational frame shiftdownstream of the splice site to produce XBP 1s, the active formof the transcription factor (Yoshida et al., 2001; Calfon et al.,2002; Lee et al., 2002). XBP 1 is responsible for the activation ofkey UPR target genes, including molecular chaperones andcomponents of the ER-associated protein degradation machin-ery (Lee at al., 2003). Activation of IRE1a is also reported toresult in the induction of regulated IRE1a-dependent degrada-tion of a subset of mRNAs encoding secretory proteins or theinduction of apoptosis via IRE1a signaling through its kinasedomain anddownstreameffectorsASK1, JNK1, andCaspase-12(Urano et al., 2000; Hollien and Weissman, 2006).Loss of ER homeostasis (i.e., loss or hyperactivation of

UPR signaling) has been attributed to a number of dis-eases, including cancer, diabetes, cardiovascular diseases,liver diseases, and neurodegenerative disorders, and the UPRis increasingly becoming an attractive pathway in drugdiscovery (Hetz et al., 2013; Maly and Papa, 2014). To thisend, an increasing body of work has been performed to identifypotent and selective molecules of IRE1a and better under-stand how these molecules bind and affect IRE1a activa-tion mechanisms. Crystal structures of the C-terminal regionof phosphorylated (active) yeast Ire1 were the first to be char-acterized and revealed that Ire1 forms dimers arranged ina “back-to-back” configuration, with the kinase active sitesfacing outward [Protein Data Bank (PDB) identity (ID) 2RIO;Lee at al., 2008; PDB ID 3FBV; Korennykh et al., 2009; PDBID 3LJ0; Wiseman et al., 2010). This dimer arrangement alsoformed the basis of a rod-shaped helical structure, represent-ing a high-order oligomeric Ire1 structure in complexwith the kinase inhibitor 2-N-(3H-benzimidazol-5-yl)-4-N-(5-cyclopropyl-1H-pyrazol-3-yl)pyrimidine-2,4-diamine (APY29)(3FBV) (Korennykh et al., 2009). These yeast back-to-backstructures contrast with the first reported human IRE1adimer: a dephosphorylated C-terminal IRE1a- Mg21-ADPcomplex, possibly representing an early state prior to phos-phoryl transfer (PDB ID 3P23; Ali et al., 2011). In thisstructure, the kinase active sites are facing each other andare in a suitable orientation and proximity for transautophos-phorylation, but the RNase domains are far from each otherand inactive. A similar “face-to-face” structure was recentlyreported for mouse IRE1a, but this structure was phosphor-ylated (PDB ID 4PL3; Sanches et al., 2014). More recently, afew other dephosphorylated human crystal structures werereported. One is of a cocrystal structure containing a kinasedomain inhibitor bound to an IRE1amonomer (PDB ID 4U6R;Harrington et al., 2014). Here, no dimers were found with theinhibitor-bound structure, suggesting that the compound mayeither prevent dimerization or stabilize a monomeric IRE1a.The second report presented two back-to-back dimers ofIRE1a, one in the apo form and one in an inhibitor-boundform (PDB ID 4Z7G and 4Z7H; Joshi et al., 2015). Thesestructures are consistent with yeast Ire1 dimers, but aredistinct in that the apo structure contains a twisted interfacebetween the dimers across the RNase domains, which mayrepresent an additional intermediate of IRE1a prior to fullactivation. The culmination of all these structures may depictthe IRE1 protein at various levels of activation and suggest

that this process is conserved evolutionarily. Here, we presenta proposed structure of the final state of a human phosphor-ylated (active) IRE1a dimer that is cocrystallized with twokinase inhibitors that have opposing effects to the RNaseactivity of the protein.

Materials and MethodsIRE1a Protein Expression and Purification. The cytosolic

domain of human IRE1a (NM_001433), encompassing amino acids547–977, was cloned into pENTR/tobacco etch virus (TEV)/D-TOPO(Life Technologies, Carlsbad, CA) and subsequently transferred to apDest8 (Life Technologies) vector backbone containing the N-terminalFlag epitope tag followed by the 6xHis tag and TEV cleavage site(ENLYFQG/S). Baculovirus generation was accomplished usingthe Bac-to-Bac baculovirus generation system (Life Technologies).Flag-His6-TEV-IRE1a (547–977) protein expression in baculovirus-infected insect cells was accomplished following established procedures(Wasilko and Lee, 2006). Briefly, a proprietary Sf9 insect cell linewas grown to the early log phase, infected with 1 � 107 baculovirus-infected insect cells/10 l culture, and incubated at 27°C. Cell paste washarvested at 66–72 hours postinfection. A human phosphorylated ordephosphorylated IRE1a protein containing N-terminal Flag-His6tags and a TEV protease cleavage site between the tags and an IRE1aprotein was purified from ∼150 g of cells from a 10-l culture [lysed in1.5 l of lysis buffer (50 mM Hepes, pH 7.5, 10% glycerol, and 300 mMNaCl) by the EmulsiFlex-C50 homogenizer (Avestin, Ottowa, ON,Canada)]. The protein in the clear supernatant from centrifugation at30,000g for 30minutes at 4°Cwas first captured in 20ml of NiNTA-SFbeads (Qiagen, Venlo, Netherlands) in batch mode for 4 hours at 4°C.The beads were poured into a column and washed with 20 mMimidazole in lysis buffer, and the IRE1a protein was eluted from thecolumn by 300 mM imidazole in 50 mM Hepes, pH 7.5, and 150 mMNaCl buffer. The Ni elution pool was concentrated using a 10-kDamolecular weight cutoff filter concentrator to about 25 ml, to which3 mg of TEV protease was added to remove the His6 tag. This mixturewas then transferred into dialysis tubing (8-kDa molecular weightcutoff) and dialyzed overnight against 3 l of MonoQ buffer A (50 mMHepes, pH 7.5, 50 mM NaCl, 5 mM DTT, and 1 mM EDTA),passed through a second 20-ml MonoQ column (GE Healthcare,Piscataway, NJ), and eluted with a 50–500 mM NaCl gradient over10 column volumes. The eluted samples were analyzed by liquidchromatography–mass spectrometry, and the major peak with a massconsistent with the expected molecular weight for the protein plusthree phosphates (80 mass units per phosphate) was purified in aHiload Superdex 200 sizing column (GE Healthcare), with a buffer of50 mM Hepes, pH 7.5, 200 mM NaCl, 5 mM DTT, and 1 mM EDTA.The eluted protein (2–3 mg protein in 1-ml aliquots) was stored at280°C and later used in assays and crystallography. The remainingfractions from the Mono Q column were pooled and treated withl-phosphatase to produce the fully dephosphorylated enzyme beforeit was further purified and stored in the same way as the triplyphosphorylated protein.

Phosphorylated IRE1a RNase Activity Assay. The nucleaseenzymatic activity of phosphorylated IRE1a (pIRE1a) was measuredusing a dual labeled 36-mer RNA substrate that contained the IRE1arecognition sequence, with a 6-carboxyfluorescein fluorescent reporter(FAM) at the 39 end, and the Black Hole quencher-1 (BHQ-1) at the59 end (59/6-FAM/rCrArG rUrCrC rGrCrA rGrCrA rCrUrG/BHQ-1/39;Integrated DNA Technologies, Coralville, IA). Upon cleavage, therelease of FAM results in an increase in fluorescent signal measuredat lex/lem 5 485/535 nm. A typical enzymatic reaction was carried outwith 10 nM pIRE1a and a 500 nM substrate in a buffer containing20 mM Hepes, pH 7.5, 5 mM MgCl2, 10 mM NaCl, 1 mM DTT, 0.05%Tween 20, and 0.02% heat-treated casein (heated for 20 minutes at60°C prior to use each time), and the fluorescence changewas followedusing an Envision plate reader (PerkinElmer, Waltham, MA).

1012 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

To identify inhibitors of IRE1a nuclease activity, pIRE1a wasscreened against the GlaxoSmithKline compound collection. The RNAoligomer substrate was added to the assay plates containing 10 mM ofcompound. The reaction was initiated immediately by the addition ofthe enzyme, and the plates were centrifuged for 1 minute at 500 rpm.The final reaction mixture contained 10 nM pIRE1a, 25 nM RNAoligomer, 20 mM Hepes, pH 7.5, 5 mM MgCl2, 1 mM DTT, 0.05%Tween 20, 10 mM NaCl, and 0.02% casein. The reaction plates wereincubated at room temperature for 90 minutes before the reaction wasterminatedwith 0.015%SDS in 20mMHepes, pH 7.5. The plateswerecentrifuged for 3 minutes at 1000 rpm prior to measuring productformation using the Viewlux imager (PerkinElmer).

pIRE1a Kinase Assays. In the ADP-Glo assay, a compound’spotency toward pIRE1a kinase activity wasmeasured as its inhibitionof an intrinsic, slow ATP hydrolysis activity. One hundred nanolitersof dimethylsulfoxide solution of (R)-2-(3,4-dichlorobenzyl)-N-(4-methylbenzyl)-2,7-diazaspiro(4.5)decane-7-carboxamide (GSK2850163)at various concentrations was added into a white Greiner low volume384-well plate. The reaction was carried out with 5 nM pIRE1a and60 mM ATP in 10 ml of 50 mM Hepes buffer, pH 7.5, containing 30 mMNaCl, 10 mMMgCl2, 1 mM DTT, 0.02% Chaps, and 0.01 mg/ml bovineserum albumin. The reaction was stopped after 2 hours by adding 5 mlof ADP-Glo reagent I, which also depletes the remaining ATP. Follow-ing a 1-hour incubation, 5 ml of ADP-Glo reagent II was addedinto the reaction, which converts the ADP product into ATP to serveas the substrate for the coupled luciferin/luciferase reaction. After30 minutes, the plate was read on a Wallac ViewLux microplate imager(PerkinElmer).

In the fluorescence polarization assay, a compound’s affinity wasmeasured by its ability to compete with a fluorescent-labeled ATP siteligand for binding to the kinase domain. The fluorescence polarizationassay was carried out in the same reaction buffer described above. A10-ml reaction contained 5 nM pIRE1a, 5 nM fluorescent ligand(GSK369716A), and GSK2851063 at various concentrations. Afterthe addition of reagents, the plate was incubated at room temperaturefor 15 minutes and then read on an Analyst GT multimode reader(Molecular Devices, Sunnyvale, CA) at an excitation/emission of 485/530 nm, with a dichroic of 505 nm. The IC50 values were calculatedusing the data fitting software GraFit (Erithacus Software, Horley,UK).

PERK Kinase Assay. Recombinant PERK protein (Life Technol-ogies) was purchased for use in a selectivity assay to test the pIRE1ainhibitors. A solution of PERK (0.40 nM) in assay buffer containing10mMHEPES, pH7.5, 2mMCHAPS, 5mMMgCl2, and1mMDTTwaspreincubated with the compound for 30 minutes at room temperature.The reaction was initiated by the addition of the substrate mixture,which contained 0.040 mMbiotinylated eukaryotic translation initiationfactor 2a peptide and 5 mM ATP. The reaction plates were incubated atroom temperature for 60 minutes, followed by termination with adetection/quench mixture. The detection/quench mixture contained 15 mMEDTA, 4 nM eukaryotic translation initiation factor 2a phosphoanti-body (Millipore, Billerica, MA), 4 nM europium-labeled anti-rabbitantibody (PerkinElmer), and 40 nM streptavidin APC (PerkinElmer).The plateswere incubated for 10minutes at room temperature prior tomeasuring the homogeneous time resolved fluorescence signal (exci-tation at 610–640 nm and emission at 660 nm) using the Viewluximager (PerkinElmer).

RNase L Activity Assay. A solution of 0.1 nM RNase L, 1 mMATP, and 8 nM2-5A [29,59-linked oligoadenylate, p1–3(A29p59A)n$ 2] inassay buffer containing 20 mM Hepes, pH 7.5, 10 mM MgCl2, 1 mMTCEP, 0.5 mM CHAPS, 100 mM NaCl, and 0.02% casein wasincubated at room temperature for 30 minutes. A 0.2 mM solution ofa 36-mer RNA substrate (59/6-FAM rUrUrA rUrCrA rArArU rUrCrUrUrArU rUrUrG rCrCrC rCrArU rUrUrU rUrUrU rGrGrU rUrUrABHQ-1/39; Integrated DNA Technologies) was prepared in assaybuffer and added to reaction plates containing compounds. Thereaction was initiated with the addition of the RNase L-ATP-2-5Amixture, and the plates were incubated at room temperature for

90 minutes. Following the incubation, the reaction was terminatedwith 0.02% SDS solution in nuclease-free water. The plates werecentrifuged for 3 minutes at 1000 rpm prior to measuring productformation using the Viewlux imager (PerkinElmer).

Competitive Inhibition Studies. In single compound inhibitionstudies, the concentrations of the substrate or noncleavable substrate(analog in which the ribose linked to the guanine at the cleavage sitewas replaced by deoxyribose) and GSK2850163 were varied and theenzyme concentration was fixed at 10 nM. The modes of inhibitionand the inhibition constants were determined by fitting the initialvelocities to different models (competitive, uncompetitive, and non-competitive) using Grafit software (Erithacus Software). In a doubleinhibition experiment, the concentration of the first inhibitor wasvaried at several different concentrations of the second inhibitor, andthe concentrations of the enzyme and substrate were kept constant at10 and 100 nM, respectively. The reactions were monitored kineti-cally, and the initial reaction velocities were analyzed using theYonetani-Theorell equation.

Protein Crystallization and Structure Determination. Crys-tals of pIRE1a (547–977) with GSK2850163 (PDB ID 4YZ9) wereprepared bymixing pIRE1awith 0.5mMGSK2850163 and incubatingovernight on ice. The crystals were grown at 20°C by vapor diffusion insitting drops containing 2 ml of protein (13 mg/ml in 50 mMHepes, pH7.5, 200 mM NaCl, 5 mM DTT, 1 mM EDTA, 0.5 mM GSK2850163,and 0.25% dimethylsulfoxide) and 2 ml of reservoir solution containingpolyethylene glycol (PEG) 3350 (16–22%), 100 mMHepes, pH 7.0, and200mMCa21 acetate. The crystals were thick rods that appeared over2–5 days and reached full size (0.05 � 0.025 � 0.3 mm) in 2 weeks.Seeding was used to improve crystal quality. The pIRE1a-GSK2850163crystals were frozen in a solution of 20% ethylene glycol, 22%PEG 3350,and 0.2 M calcium acetate added in a stepwise manner to the proteindrop before mounting the crystal on the loop.

The crystals of pIRE1awithMg21-ADP (PDB ID 4YZD)were grownby mixing 2 ml of protein solution [10 mg/ml pIRE1a (547–977) in50 mMHepes, pH 7.5, 200 mMNaCl, 5 mM DTT, 1 mM EDTA, 1 mMADP (100 mM ADP stock was pH ∼7.0), and 1 mMMgCl2] with 2 ml ofreservoir solution (16% PEG 3350 and 200mMNa1malonate, pH 6.0)in sitting drops at room temperature. Seeding was used to initiatecrystal growth. Crystals appeared the next day and grew to full size in3 weeks. For data collection, the crystals were frozen in a solution of20% ethylene glycol, 22% PEG 3350, and 200 mM Na1 malonate, pH6.0, and added to the protein drop before mounting the crystals on theloop.

The complex of pIRE1a (547–977) with staurosporine (STS) (PDBID 4YZC) was prepared by mixing the protein with 0.5 mM staur-osporine and incubating overnight on ice. The crystals were grown at20°C by vapor diffusion in a sitting drop containing 2 ml of protein(13 mg/ml in 50 mM Hepes, pH 7.5, 200 mM NaCl, 5 mM DTT, and1 mM EDTA) and 2 ml of reservoir solution containing PEG 300 (30–40%), 100 mM Hepes, pH 7.5, and 200 mM KCl. Small hexagonalplates appeared over 5–10 days and reached full size (0.05 � 0.75 �0.15 mm) in ∼20 days. The crystals were flash frozen in liquid N2

directly from the crystallization drop. All diffraction data werecollected at the Advanced Photon Source, Argonne National Labora-tories, Life Sciences Collaborative Access Team, Sector 21.

All structures were determined by molecular replacement withPhaser (Afonine et al., 2010), as implemented in CCP4 (Winn et al.,2011), using the humandephosphorylated IRE1a- Mg21-ADP complex(3P23; Ali et al., 2011) as a model. Refinement was performed by acombination of Refmac5 (Murshudov et al., 1997) and Phenix (Afonineet al., 2010), with manual adjustments to the model in COOT (Emsleyet al., 2010). The quality of the model was monitored usingMolprobity(Chen et al., 2010). Figures were made with PYMOL (Schrödinger,LLC, New York, NY).

Detection of XBP 1 Splicing by Real-Time PolymeraseChain Reaction. Multiple myeloma cancer cell lines were obtainedfromATCC (Manassas, VA) or DSMZ (Braunschweig, Germany). Cellswere cultured in the appropriate culture medium supplemented with

Structure of Human Phosphorylated (Active) IRE1a Dimer 1013

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

10% fetal bovine serum (FBS) (Sigma-Aldrich, St. Louis, MO) at 37°Cin humidified incubators under 5% CO2. Cells were seeded into six-well plates at a density of 1.5� 106 cells/well in the appropriate mediacontaining 1% FBSmedia and were treated with 5 mg/ml tunicamycin(MPBiomedicals, Newport Beach, CA) for 1 hour before the addition ofGSK2850163 for 3 hours (4 hours total). For studies involvingstaurosporine (Sigma-Aldrich) treatment, cells were treated withstaurosporine for 30 minutes, followed by 5 mg/ml tunicamycin for1 hour. RNA was collected using the Qiagen RNEasy Mini Kit. RNAwas quantitated using a NanoDrop (Thermo Scientific, Philadelphia,PA) and stored at280°C. To generate cDNA, theHigh-Capacity cDNAReverse Transcription Kit (Life Technologies) was used. For polymer-ase chain reaction (PCR), a 25-ml reaction included 12.5 ml ofMaster Mix (Promega Go-Taq 2X Master Mix, Madison, WI), 1 mleach of forward and reverse primers (100 ng/ml), 9.5 ml of water, and1 ml of cDNA template. Primers for human XBP 1 were forward,59-CCTGGTTGCTGAAGAGGAGG-39, and reverse, 59-CCATGGGGA-GATGTTCTGGG-39. Real-time (RT) PCR conditions were 95°Cfor 5 minutes, 95°C for 30 seconds, 58°C for 30 seconds, 72°C for30 seconds, and 72°C for 5 minutes, with 35 cycles of amplification.After RT-PCR, reactions were run on a 3% agarose gel and visualizedusing SYBR Safe DNA gel stain (Life Technologies) and a BioRadImager (Hercules, CA).

Western Blot Analysis. RPMI 8226 cells were seeded into six-well plates at a density of 2.0 � 106 cells/well in RPMI 1640 mediacontaining 1% FBS. Cells were treated with the same conditions asdescribed above for XBP 1 splicing detection by RT-PCR. To harvestprotein lysates, cells were lysed with 60 ml of 1X cell lysis buffer (CellSignaling Technologies, Danvers, MA) containing protease and phos-phatase inhibitors. Cell lysates were quantified using the Pierce BCAProtein Assay Kit (Thermo Scientific), and samples were read on aSPECTRAmax 190 instrument (Molecular Devices, Sunnyvale, CA).Following quantitation, 40 mg of protein was run on 4–12% Bis-Trisgels (Life Technologies), and protein was transferred onto 0.45 mMnitrocellulose membranes (Life Technologies) using the BioRad semi-dry transfer blotting apparatus. Membranes were blocked for 1 hourusing Li-Cor Odyssey Blocking Buffer (Lincoln, NE) and then probedwith the following antibodies overnight: pIRE1a S724 (1:1000,#ab48187; Abcam, Cambridge, UK), total IRE1a (1:1000, #ab37073;Abcam), and GAPDH (1:2000, #A300-639A; Bethyl Laboratories,Montgomery, TX). After washing, blots were incubated with donkeyanti-rabbit IRDye-800CW secondary antibody (Li-Cor), and proteinswere visualized using the Odyssey Imaging System (Li-Cor).

XBP 1 Transcriptional Activity Assay. PANC-1 cells wereseeded into six-well plates at a density of 5.0 � 103 cells/well in RPMI1640 media containing 10% FBS. Cells were cotransfected with apGL3-5x unfolded protein response element (UPRE)–luciferase re-porter containing five repetitions of the XBP-1 DNA binding site(a kind gift from R. Prywes, Columbia University, New York, NY)and pRL-SV40 (Promega) using the FuGENE6 transfection reagent(Roche, Indianapolis, IN). Forty-eight hours later, cells were treatedwith 2.5 mg/ml tunicamycin for 1 hour, followed by GSK2850163treatment for 16 hours. Luciferase expression was measured usingDual-Glo Luciferase Assay kit (Promega) and normalized to Renillaexpression levels.

Hydrogen-Deuterium Exchange. pIRE1a (547–977, 78 mM)was incubated with 250 mM GSK2850163 or 1 mM staurosporine forat least 18 hours. Two microliters of protein-inhibitor mixture wasmixedwith 18ml of D2O at room temperature for 1minute, after which20 ml of 4 M guanidinium chloride in 1 M glycine buffer, pH 2.5, and120ml of formic acid were added and immediately transferred to an icecold bath. All subsequent treatment and analysis was done at 2–4°C.Fifty microliters of this solution was injected into a Waters Enzymateethylene bridged hybrid pepsin 2.1 � 30 mm column for digestion,then to a C18 column, and into an LTQ XL Orbitrap mass spectrom-eter (Thermo Scientific). The mass-to-charge values were calculatedwith XCalibur (Thermo Scientific) and compared with sequences inthe MASCOT database. The analysis of the hydrogen-deuterium

exchange was done with HDExaminer. Tryptic digestion of all thesamples gave a sequence coverage of at least 98% of the amino acidsequence.

Preparation and Characterization of Compounds 1–24. De-tails are provided in the Supplemental Methods.

ResultsCrystallization of the Novel Inhibitor GSK2850163

Bound to pIRE1a. Because of the relevance of the IRE1a/XBP 1 pathway in human disease, we sought to identifysmall molecules that would inhibit IRE1a RNase activity.GSK2850163 was discovered as a result of a high-throughputscreening campaign to identify IRE1a-selective inhibitors ofXBP 1 splicing. It is a highly selective inhibitor with dualactivity: it inhibits IRE1a kinase activity (IC50 5 20 nM)and RNase activity (IC50 5 200 nM) (Fig. 1A; SupplementalTable 1). In competition kinetic studies, GSK2850163 and anoncleavable RNA substrate demonstrated mutually exclu-sive binding to activated IRE1a (Ki 5 200 6 20 nM)(Supplemental Fig. 1). We hypothesized that this was due tobound GSK2850163 altering the preferred enzyme struc-ture for RNA substrate binding (and vice versa) and not dueto a physical overlapping of binding sites.To investigate the mode of binding and enable structure-

guided optimization of GSK2850163, we determined the coc-rystal structure of GSK2850163 with the C-terminal portion ofpIRE1a (residues 547–977; PDB ID 4YZ9) (Fig. 1B; Supple-mental Fig. 2; Table 1). The structure of pIRE1a-GSK2850163is of a back-to-back dimer, with one inhibitor molecule bound tothe kinase domain of each protomer in a pocket next to thekinase aChelix, approximately 12Å from the hinge region (Fig.1C). GSK2850163 adopts a U-shaped conformation, with thetolyl and dichlorophenyl groups facing the inside of the proteinand the spirodecane core partially solvent exposed with thepiperidine ring (A-ring) in a chair conformation. Two keyinteractions with conserved kinase catalytic residues, Glu612and Lys599, are observed. The urea nitrogen and the pyrroli-dine nitrogen of GSK2850163 forma hydrogen bond interactionwith the side chain of Glu612, and the carbonyl oxygen of theurea of GSK2850163 interacts through a hydrogen bond withthe side chain of Lys599.GSK2850163 displaces the kinase activation loop of

pIRE1a, such that the DFG motif is in the “out” conformationand is flipped by nearly 180°, occupying the ATP binding site(Fig. 1C). This clearly contrasts with our resolved structureof pIRE1a-ADP-Mg21 (PDB ID 4YZD), where the DFG motifis found in the “in” conformation and Phe712 occupies thesame pocket where the GSK2850163 binds in the pIRE1a-GSK2850163 structure (Fig. 1D; Table 1). Phe712 makesa p interaction with Tyr628 in a pocket lined by Val613,Leu616, Vals625, and Leu679. Superposition of the ADP-Mg21 and GSK2850163-bound pIRE1a complexes shows thatGSK2850163 does not overlap with the ATP-binding site(data not shown). The structure of pIRE1a in complex withMg21-ADP forms a “face-to-face” dimer across the symmetryplanes with neighboring molecules (Supplemental Fig. 3). Con-sistent with previously published structures, the kinase activesites are facing eachother andare inanorientationandproximityfavorable for transautophosphorylation [3P23 (Ali et al., 2011)and 4PL3 (Sanches et al., 2014)]. The present structure maypossibly represent an early postphosphoryl-transfer dimer,

1014 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

while the face-to-face structure of dephosphorylated humanIRE1amay represent a state just prior to phosphoryl transfer.Structure Activity Relationship of GSK2850163. To

provide supportive evidence for the binding mode ofGSK2850163 observed in the crystal structure, we performedstructure activity relationship (SAR) studies, whereby wesystematically modified the structure of the ligand andfollowed the effect by measuring pIRE1a RNase inhibition.First, we observed that while the S-enantiomer is inactive andits conformation does not fit in the electron density map, theR-enantiomer inhibited the RNase activity in vitro, with anIC50 of 0.2 mM (Fig. 2A). The R-enantiomer is refined in the

structure of the complex. Since the activity of theR-enantiomeris equally potent to the racemic compoundand theS-enantiomeris not active, the initial SAR study was performed with the ra-cemic analogs.Second, the NH of urea nitrogen and the pyrrolidine

nitrogen are required to make an H bond with Glu612(2.98 Å), and any changes in the GSK2850163 (compound 1)molecule disrupting these specific H-bond interactionsresulted in loss of activity. For example, replacement of theurea nitrogen (compound 2) or capping the NH with a methylgroup (compound 3) caused a loss of potency (Fig. 2A).Similarly, conversion of a basic amine in the pyrrolidine ring

Fig. 1. GSK2850163 binds to human pIRE1a and inhibits XBP 1 splicing. (A) GSK2850163 is a selective inhibitor of pIRE1a kinase activity and RNaseactivity. It exhibits no or weak inhibition toward a wide range of the other kinases (shown here is the UPR kinase PERK) and nucleases (shown here isRNase L, its closest homolog). Refer toMaterials andMethods for assay details. (B) Overall structure of the pIRE1a-GSK2850163 (PDB ID 4YZ9) back-to-back dimer. Chemical structure of GSK2850163 (inset). (C) GSK2850163 (yellow) binds next to the aC helix and displaces the activation loop (DFG-out),such that it is occupying the ATP-binding site represented by its surface. GSK2850163 makes H-bond interactions with the key kinase catalytic residuesLys599 (K599) and Glu612 (E612). (D) For comparison, the DFG loop is in the in conformation in the structure of pIRE1a-Mg2+-ADP (PDB ID 4YZD), andPhe712 (F712) occupies the same pocket where GSK2850163 would bind. For both (C) and (D), * indicates the C-terminal end of the visible portion of theactivation loop.

Structure of Human Phosphorylated (Active) IRE1a Dimer 1015

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

to a nonbasic amide yielded an inactive analog (compound 4;Fig. 2A).Third, in the binding mode of GSK2850163 in complex with

pIRE1a described here, the lipophilic groups at both ends ofthe molecule (the tolyl and dichlorophenyl groups) are accom-modated within the lipophilic pockets observed in the corecrystal structure (Supplemental Fig. 2B) and are critical forthe inhibitor’s activity. All analogs with polar functionality,such as pyridines, were not active (compounds 5 and 6; Fig.2B) or much less active (compounds 15–19; Fig. 2C). On theother hand, the lipophilic groups were well tolerated (com-pounds 8–14 and 20–24) in the hydrophobic pocket. Deletion ofone of the chlorines to form 3- or 4-monochloro analogs (7 and8, respectively) revealed that the 4-chloro substitution wasmore tolerated than the 3 position in terms of potency (Fig.2B). Replacement of 3-chloro with other groups, such as fluoro(11), methyl (12), or methoxy (13), provided equally potentanalogs, but the trifluoromethyl group (14) caused a signifi-cant loss of potency. The tolyl group at the A-ring ureapreferred lipophilic substitutions, and polar functionalities[e.g., pyridines (15–17) and substituted pyridines (18 and 19)]were not tolerated. Deletion of the methyl group gave thesimple phenyl analog 20 that maintained RNase activity(IC50 5 0.40 mM) (Fig. 2B). Replacement of the methyl groupat the 4-position with other lipophilic groups, such as chlorine(21), methoxy (22), and fluorine (23), was well tolerated. Thebenzodioxole analog 24 exhibited an IC50 value of 100 nM.These SAR study results are consistent with the bindingmodeof GSK2850163 to pIRE1a observed in the crystal structure,and indicate that GSK2850163 functions as a kinase andRNase inhibitor of pIRE1a.Cellular Activity of GSK2850163. We used a panel of

eightmultiplemyelomacell lines to test the effects ofGSK2850163on IRE1a RNase activity. Increased IRE1a activity has been

observed in primary multiple myeloma specimens, and clini-cal studies have associated high levels of spliced XBP 1mRNAwith poor patient survival (Reimold et al., 2001; Nakamuraet al., 2006; Bagratuni, et al., 2010). To recapitulate an ERstress-induced environment in cell culture, tunicamycin, aninhibitor of N-linked glycosylation, was used (Yoshida et al.,2001). XBP 1 mRNA is found primarily in the unspliced formunder basal conditions. Upon ER stress stimulation, all cellsinduced varying degrees of XBP 1 splicing, which could bereversed following treatment with GSK2850163 (Fig. 3A). ERstress also induced increased autophosphorylation of IRE1a,which could be reduced in a dose-dependent manner byGSK2850163 (Fig. 3B). To determine if GSK2850163 couldaffect the transcriptional function of XBP 1, cells expressing areporter plasmid under the control of five tandem repeats ofthe UPRE motif were treated with tunicamycin, followed byincreasing doses of GSK2850163 (Wang et al., 2000). TheUPRE sequence contains the binding site found at the pro-moter of XBP 1 target genes. Induction of ER stress bytunicamycin significantly increases XBP 1 transcriptional activ-ity in these cells, yet increasing concentrations of GSK2850163are capable of reducing this activity (Fig. 3C). Thesedata supportour biochemical and structural evidence that GSK2850163 is aninhibitor of IRE1a, which is capable of inhibiting both kinase andRNase activities of IRE1a.Crystallization of Staurosporine Bound to pIRE1a. To

further understand the mechanism by which the RNaseactivity of IRE1a could be modulated pharmacologically withkinase inhibitors, we sought to investigate the effect of thebroad-spectrum kinase inhibitor STS on IRE1a. It has beenpreviously shown that STS inhibits IRE1a kinase activity(Ali et al., 2011) and is likely to bind to the ATP-binding siteof IRE1a since it competed with the irreversible IRE1a inhib-itor 4m8C, preventing the formation of a Schiff base with the

TABLE 1Data collection and refinement statistics (molecular replacement)Values in parentheses are for the highest-resolution shell. Each dataset was collected from a single crystal.

pIRE1a + GSK2850163 pIRE1a + ADP pIRE1a + STS

Data collectionSpace group P21212 C2 P212121Cell dimensions

a, b, c (Å) 244.0, 77.8, 88.3 191.8, 122.5, 77.9 49.3, 155.3, 155.6a, b, g (°) 90, 90, 90 90.0, 107.1, 90.0 90, 90, 90

Molecules/a.u. 3 3 2Resolution (Å) 2.45 (2.54–2.45) 3.14 (3.25–3.10) 2.62 (2.71–2.62)Rsym or Rmerge 0.073 (0.592) 0.087 (0.569) 0.118 (0.565)I/sI 22.9 (3.1) 15.2 (2.4) 22.7 (5.6)Completeness (%) 96.7 (89.7) 100.0 (100.0) 100.0 (100.0)Redundancy 6.3 (6.3) 3.8 (3.8) 11.8 (11.6)RefinementResolution (Å) 30.0–2.45 45.6–3.14 40.0–2.62Number of reflections 59,654 60,654 41,900Rwork/Rfree 0.24/0.29 0.20/0.25 0.25/0.29Number of atoms

Protein 17,853 9.822 12,041Ligand/ion 90 90 70Water 81 35 62

B factorsProtein 57 31 55Ligand/ion 57 24 23Water 51 10 48

R.m.s. deviationsBond lengths (Å) 0.007 0.008 0.002Bond angles (°) 0.93 1.09 1.15

a.u., asymmetric unit.

1016 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

kinase active site residue Lys599 (Cross et al., 2012). Wefound that STS inhibited pIRE1a kinase enzymatic activity(IC50 5 3 nM), but had no effect on pIRE1a RNase activity invitro (Fig. 4A). STS exhibited a similar binding affinitytoward phosphorylated IRE1a (IC50 5 30 nM) and dephos-phorylated IRE1a (IC50 5 50 nM) in a competitive bindingassay (data not shown). Interestingly, incubation of STSwithdephosphorylated (inactive) IRE1a was capable of activat-ing IRE1a RNase activity (Fig. 4B). To test the effects ofSTS on IRE1a RNase activity in a cellular context, RPMI

8226multiple myeloma cells were treated with STS, followedby tunicamycin, to measure XBP 1 splicing. STS treat-ment alone was sufficient to induce XBP 1 splicing simi-lar to levels achieved with tunicamycin (Fig. 4C). Despiteinducing XBP 1 splicing, STS did not increase IRE1aautophosphorylation. Instead, STS inhibited tunicamycin-induced autophosphorylation (Fig. 4D). STS treatment mayalso inhibit IRE1a autophosphorylation below basal levels,but it was difficult to measure this convincingly by Westernblot analysis. These observations, both biochemical and in

Fig. 2. Selected SAR results of compound 1 (GSK2850163). (A) Replacement of the urea nitrogen with a carbon atom (compound 2) or capping the NHwith a methyl group (compound 3) caused a loss of potency. Conversion of basic amine in pyrrolidine ring to a nonbasic amide yielded an inactive analog(compound 4). (B andC) The lipophilic groups at the tolyl and dichlorophenyl groupswere critical for the compound’s activity. On the other hand, the polarfunctionalities, such as pyridines, were not tolerated.

Structure of Human Phosphorylated (Active) IRE1a Dimer 1017

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

cells, suggest that the effects of STS on IRE1a phosphor-ylation and XBP 1 splicing could be due to direct bind-ing of STS to IRE1a and not just due to general stressthat could occur due to the broad-spectrum nature of thecompound.To understand the structural basis of the activation of the

RNase activity of IRE1a by staurosporine, we determined thecrystal structure of the human pIRE1a-STS complex (PDB ID4YZC; Table 1). The pIRE1a-STS complex forms a back-to-back dimer, with one STS molecule bound per protomer (Fig.4E; Supplemental Fig. 4A). STS binds in the ATP-binding siteof the kinase domain and interacts with the hinge residuesGlu643, Cys645, and His692 (Supplemental Fig. 4B). Theactivation loop containing the phosphorylated Ser residues(pSer724, pSer726, and pSer729) that was previously disor-dered in both the pIRE1a-GSK2850163 and pIRE1a-ADPcomplexes is well defined in the pIRE1a-STS complex (Sup-plemental Fig. 4C). pSer724 interacts with Asn750 via two Hbonds, pSer726 is involved in an H-bond interaction withArg722 (3.1 Å) and Arg728 (3.3 Å), and pS729 is interactingthrough an H bond with Lys716 (2.7 Å) and Arg687 (2.5 Å). Anetwork of H bonds extends from the serines in the activationloop to the aChelix and into the active site DFGmotif. Bindingof STS to pIRE1a locks the kinase domain in a conformation,in which the 1) activation loop is in the DFG (711–713)-inconformation; 2) the DFG-aspartate interacts with His689in theHRDmotif in the catalytic loop; 3) the conserved Leu616in the aC helix interacts with the DFG phenylalanine in theregulatory spine of the kinase; and 4) Arg687 of the HRDmotifinteracts with pSer729 (Supplemental Fig. 4C). These are thesignatures of a kinase in the active conformation (Kornev and

Taylor, 2010). Attempts to crystallize dephosphorylated IRE1awith STS were not successful.Comparison of pIRE1a Dimer Interactions and

Protein Dynamics. To test the pIRE1a dimer interactionsobserved in the structures, we determined the hydrogen-deuterium exchange of pIRE1a in the presence or absence ofeither GSK2850163 or STS in solution (Supplemental Fig. 5).Changes in the labeling pattern can be related to changes inthe protein structure or dynamics resulting from a ligand-binding event (Englander et al., 2003; Percy et al., 2012). Bymapping the observed labeling onto the two cocrystal struc-tures, we observed three areas where most of the structuralchanges occurred. The first region is in the vicinity of theligand-binding site (Supplemental Fig. 5A). Increased solventexposure of the activation loop and disruption of the conservedsalt bridge Arg627-Asp620 by GSK2850163 are both consis-tent with the displacement of the activation loop to the DFG-out conformation and the changes in the aC helix to theinactive conformation (Supplemental Fig. 5B). Second, in thepresence of STS, the residues at the kinase-RNase intra-molecular domain interface show increased solvent exposure(Supplemental Fig. 5C). The loosening of the intramolecularinteractions is consistent with the reorientation of the pIRE1adimer that leads to the dimerization of the RNase domains.Simultaneously, the decrease in overall labeling of the RNasedomains occurs in concert with the conditions under whichRNase domains dimerize (Fig. 5, A and D). Third, in thepresence of bound STS, there is an increase in solventexposure of the residues 900–916 (Supplemental Fig. 5D).These residues form a helix-loop element that interact directlywith the RNA substrate and include the catalytically essential

Fig. 3. Inhibition of IRE1a kinase and RNase activity in cells treated with GSK2850163. (A) Multiple myeloma cell lines were left untreated or treatedwith tunicamycin for 1 hour. GSK2850163 (GSK163) was then added for 3 hours at the indicated doses. Total RNA was harvested, and RT-PCR wasperformed using human-specific XBP 1 primers flanking the splice site that distinguish between unspliced (XBP 1u) and spliced (XBP 1s) XBP 1 mRNA.(B) RPMI 8226 cells were left untreated or treated with tunicamycin and/or GSK2850163 as described in (A). Changes in IRE1a phosphorylation weredetected by Western blot using a phosphospecific antibody raised against Ser 724 in the kinase activation loop. (C) XBP 1 transcriptional activity isreduced in cells treated with GSK2850163. PANC-1 cells were transfected with a 5X UPRE luciferase reporter and pRL-SV40 Renilla constructs. Forty-eight hours later, cells were left untreated or treated with tunicamycin for 1 hour, followed by GSK2850163 treatment of 16 hours. UPRE luciferaseexpression was measured and normalized to Renilla expression levels.

1018 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

His910 (Dong et al., 2001; Korennykh et al., 2009). Thisapparent increased dynamic state of the helix-loop catalyticelement may be related to the catalytic readiness favored bySTS but opposed by GSK2850163.Comparison of pIRE1a Cocrystal Structures. A com-

parison between the pIRE1a-GSK2850163 and pIRE1a-STScomplexes shows that both structures are back-to-back dimers(Fig. 5A). Binding of GSK2850163 causes the kinase aC helix(residues 610–619) to move to the inactive conformation by an

average of 4.46 1.0 Å (range of 3.03–6.40 Å on superposition ofthe aC helix carbon atoms) as compared with the positionobserved in the STS complex (Fig. 5B). While the centers ofmass of the kinase N domains (21.5–21.9 Å) and C domains(43.1–43.6 Å) remain constant, a nearly 20° rotation of thedomains relative to one another is observed in the twostructures. A differential gap of nearly 4 Å between the RNasedomains in the structures of pIRE1a-GSK2850163 (29.7 Å)and pIRE1a-STS (26.1 Å) is reflected in a different set of

Fig. 4. Structure and activity of the human pIRE1a-STS complex. (A) STS inhibits pIRE1a kinase activity, but not RNase activity. (B) Titration curveshowing that the nuclease activity of dephosphorylated IRE1a [IRE1a (-P)] is stimulated by STS in the RNase assay. (C) STS induces XBP 1 splicing inRPMI 8826 cells. Cells were treated with STS for 30 minutes, followed by tunicamycin (tuni) (5 mg/ml) treatment in half of the cells for 1 hour. Total RNAwas harvested, and RT-PCRwas performed using human-specific XBP 1 primers flanking the splice site that distinguish between unspliced (XBP 1u) andspliced (XBP 1s) XBP 1mRNA. (D) STS inhibits tunicamycin-induced IRE1a phosphorylation. RPMI 8226 cells were treated as described in (C). Changesin IRE1a phosphorylation were detected by Western blot using a phosphospecific antibody raised against Ser 724 in the kinase activation loop. (E)pIRE1a-STS dimer (PDB ID 4YZC) in the back-to-back configuration, with STS bound in the ATP-binding site.

Structure of Human Phosphorylated (Active) IRE1a Dimer 1019

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

interactions across the RNase dimer interfaces. Hence, theRNase dimer interface in the pIRE1a-GSK2850163 complex isreduced by 58% (269 Å2) compared with the pIRE1a-STScomplex (642 Å2) (Fig. 5E). In pIRE1a-GSK2850163, the RNasedomains are noticeably separated, such that the network ofinteractions across the dimer is no longer possible. The interac-tions between the protomers of the pIRE1a-GSK2850163 dimeroccur mostly between the kinase domains (1734 Å2), with lesscontact occurring between the RNase domains (269 Å2). Thisseparation of RNase domains is consistent with the biochemicaland cellular data, showing that GSK2850163 is a potentinhibitor of pIRE1a RNase activity.

In the pIRE1a-STS complex, the RNase domains form adimer, with two sets of interdigitating H-bond interactionsamong His909, Asp847, and Arg905 (3.2–3.5 Å) and amongArg955, Glu836, and Asp927 (3.1–3.5 Å), which allow the keynuclease catalytic residues Tyr892, Arg905, Asn906, andHis910 to be poised for action (Fig. 5D). Asp847, His909, andArg905 are highly conserved residues across multiple species(Dong et al., 2001). These interactions across the RNase dimerinterface are also analogous to those observed previouslyfor the activation of yeast Ire1 RNase activity by quercetin(3LJ0; Wiseman et al., 2010). In the structure of pIRE1a-GSK2850163, the RNase domains are separated; thus, the

Fig. 5. Conformational changes to pIRE1a upon ligand binding. (A) Increased interactions along the kinase dimer interface and decreased surface areacontact of the RNase domains in pIRE1a-GSK2850163 (left panel). pIRE1a2STS structure, where the interactions are tighter between the RNasedomains (right panel). A cartoonmodel depicts a rockingmotion between inactive (left) and active (right) pIRE1a dimers. (B) Residues 610–619 thatmakeup the aC helix move 4.4 6 1.0 Å between pIRE1a-STS (green) and pIRE1a-GSK2850163 (salmon). The structures of the kinase domains weresuperimposed usingCa atoms of the residues forming the hinge region. (C) Bottom-up view of the pIRE1a-GSK2850163RNase domains. There is one pairof residues interacting across the RNase dimer interface, which covers a contact area of 269 Å2. Glu913 on chain A (green) and Arg905 on chain B (cyan)form a salt bridge, but the equivalent interaction (i.e., Glu913 of chain B and Arg905 of chain A) is not possible because the side chain of Glu913 in chain Bis disordered. (D) Bottom-up view of the pIRE1a-STS RNase domains. There are two sets of interdigitating H-bond interactions. The first set involvesHis909, Asp847, and Arg905 (3.2–3.5 Å), and the second is between Arg955, Glu836, and Asp927 (3.1–3.5 Å). (E) Calculated buried surface area of contactfor the two pIRE1a dimers and individual domains.

1020 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

network of H bonds between the RNase domains observed inthe active pIRE1a-STS dimer does not exist (Fig. 5C). Hence,the data from the cocrystal structures presented here suggestthat the formation of interdigitating H bonds and the stabi-lization of the RNase dimer are critical for the active RNaseconformation and its enzymatic activity.

DiscussionRecent work has implicated the UPR in a number of

diseases, most notably in cancer, where UPR activation hasbeen shown to function as a survival mechanism in cancer,promoting tumor growth, regulating angiogenesis, and facil-itating adaptation to hypoxia (Ma and Hendershot, 2004;Feldman et al., 2005; Chen et al., 2014). Specifically, theIRE1a/XBP 1 pathway has shown to be overexpressed in avariety of human cancers, and activation of the pathway hasbeen shown to be essential for the survival of highly secretorymultiplemyeloma cells, where the protein load is high (Feldmanet al., 2005; Koong et al., 2006; Carrasco et al., 2007). Thesekey findings, coupled with the possibility to modulate IRE1aactivity via targeting two potentially druggable domains, hasmade IRE1a a very attractive target in drug discovery (Hetzet al., 2013; Harrington et al., 2014; Maly and Papa, 2014;Sanches et al., 2014; Joshi et al., 2015).An increasing body of evidence indicates that IRE1a

inhibitors bound to the kinase domain invariably inhibit itsATPase activity, yet some kinase inhibitors inhibit the RNaseactivity, while others activate its RNase activity. GSK2850163was discovered in an attempt to identify IRE1a-selectiveinhibitors of XBP 1 splicing that could regulate multiplemyeloma cancer cells. It is a highly selective kinase inhibitor;a panel of 284 kinases was assayed to determine the specificityof GSK2850163, and only two additional kinases were weaklyinhibited by GSK2850163: Ron (IC50 5 4.4 mM) and FGFR1V561M (IC50 5 17 mM) (Supplemental Table 1). GSK2850163inhibits IRE1a RNase activity due to the unique way themolecule binds in the kinase domain active site. GSK2850163binds deep in a pocket next to the kinase aC helix, approxi-mately 12 Å from the hinge region, which is clearly distinctfrom the ADP binding site. The molecule adopts a U-shapeconformation when bound to pIRE1a that could only beachieved with the R-stereoisomer (Supplemental Fig. 2A).Modeling of the S-stereoisomer places the difluorobenzeneoutside the electron density and clashes with the protein.Similarly, flipping the U-shaped molecule to the reverseorientation places the spirodecane within the electron density,but the four-bond length between the spirodecane and thetolyl group leads to a clash with Leu616, and the two-bondlength is too short and causes a clash between the difluor-obenzene and Lys599 instead of the H bond when in themodeled orientation. Based on this and the SAR studiesperformed, it is clear that only the R-stereoisomer in themodeled conformation and orientation is capable of fitting theelectron density, avoiding clashes, and engaging with Glu612and Lys599.The GSK2850163 mode of binding differs from classic

ATP-competitive inhibitors [reviewed by Dar and Shokat(2011)]. Inhibitors exemplified by APY29 (DFG-in and theaC helix in the active conformation) activate IRE1a RNaseactivity (Korennykh et al., 2009; Wang et al., 2012). In con-trast, inhibitors like 1-(4-[8-amino-3-tert-butylimidazo(1,5-a)

pyrazin-1-yl]naphthalen-1-yl)-3-[3-(trifluoromethyl)phenyl]urea(KIRA6) and several close analogs inhibit IRE1a kinase andRNase activities and possibly prevent dimer association (Wanget al., 2012; Ghosh et al., 2014). While cocrystal structures forKIRA6 are not available, a close analog was recently cocrystal-lized with the c-Src kinase domain, demonstrating a shift in theaC helix out to the inactive conformation (PDB ID 3QLF). Thus,it is possible that KIRA6 invokes a similar conformationalchange when bound to IRE1a. Similarly, another inhibitorrecently described by Amgen (Thousand Oaks, CA) potentlyinhibited IRE1a kinase and RNase activity and was cocrystal-lized with a dephosphorylated IREa monomer (Harringtonet al., 2014). The lack of IRE1a dimer structure here could bedue to the compound either preventing dimerization or stabiliz-ing a monomeric form of IRE1a. While it was previouslysuggested that dimerization/oligomerization occurs after auto-phosphorylation, recent crystal structures of dephosphorylatedhuman IRE1a dimers demonstrate that dimerization can pre-cede phosphorylation in both face-to-face and back-to-backconfigurations (Ali et al., 2011; Joshi et al., 2015). It should alsobe noted that this molecule shifted the aC-helix out. Hence, onecommonality between this molecule, GSK2850163, and theKIRA6 analogs is that all these inhibitors shift the aC helix toan inactive conformation, which may likely be a requirement forIRE1a RNase inactivation.Aside from kinase domain inhibitors, two other classes of

IRE1a inhibitors have been characterized that target otherpotential drug pockets in the IRE1a protein: quercetin andsalicylaldehyde-based inhibitors. Quercetin binds uniquely toyeast Ire1 in a pocket at the enzyme-dimer interface termedthe Q site, thereby stabilizing Ire1 in an active conformationthat augments RNase activity (Wiseman et al., 2010). Whilequercetin has been shown to be a broad spectrum kinaseinhibitor and may potentially target the nucleotide bindingsite of Ire1, it did not inhibit Ire1 autophosphorylation. In-dependently, a considerable amount of work has been con-ducted to target IRE1aRNase activity directly (Maly andPapa,2014; Sanches et al., 2014). These salicylaldehyde-based inhib-itors contain a reactive electrophile that most likely covalentlymodifies Lys907 at the IRE1a RNase active site. It is believedthat these inhibitors should not affect IRE1a autophosphor-ylation or dimerization. GSK2850163 is the first molecule thatinhibits both kinase and RNase activity by binding to phos-phorylated IRE1a and inducing a unique long-range conforma-tion change that alters the preferred enzyme structure for RNAsubstrate binding.The wealth of data on the dimeric/oligomeric state of IRE1

as a function of phosphorylation and the effects of variousclasses of kinase inhibitors do not fully explain the details ofthe conformational changes required to cause modulation ofIRE1a RNase activity. The hydrogen-deuterium exchangerate, which is dependent on both structural and conforma-tional changes, is very well suited to reveal these detailsand has the resolution necessary to pinpoint the requiredconformational changes in the solution state. Our hydrogen-deuterium exchange data are consistent with the model ofRNase inhibition suggested by the pIRE1a-GSK2850163cocrystal structure. The differences observed in labeling over-lap with the changes observed in the crystal state. The criticalobservation of the conserved salt bridge between Arg627 andAsp620 in the kinase domain being disrupted by GSK2850163is highly consistent with the changes in the aC helix to the out

Structure of Human Phosphorylated (Active) IRE1a Dimer 1021

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

(inactive) conformation and the reorientation of the pIRE1adimer to the inactive form. By the same analysis, the reverse istrue in the activation of the RNase.A comparison of the yeast Ire1 back-to-back dimers with the

human pIRE1a-STS dimer demonstrates that the kinasedomain DFG-loop is in the in conformation, the aC helix isin the active conformation, and the RNase domains areengaged through a network of H-bond interactions in thesestructures, which coincide with conditions of RNase activa-tion. Minor differences are observed between the pIRE1a-STSdimer and the three yeast Ire1 structures (PDB ID 2RIO,3FBV, and 3LJ0) (r.m.s. on Ca 5 1.9–2.1 Å); however, thedifferences are even greater when comparisons are made withthe pIRE1a-GSK2850163 dimer (r.m.s. on Ca 5 3.3–3.4 Å).GSK2850163 binding results in shifting the structure of thekinase domain to the DFG-out, aC helix inactive conforma-tion, and the RNase domains of the dimer are rotated awayfrom each other. This indicates that the RNase active forms ofyeast and human IRE1 aremarkedly different from theRNaseinactive pIRE1a-GSK2850163 dimer. While the key interac-tions in the RNase domains are not mediated by identicalresidues across species, they are essential. Mutagenesis stud-ies have shown that these interfacial residues are critical formaintaining RNase activity (Lee et al., 2008).Structural similarities are also observed between the human

phosphorylated (4YZD) and dephosphorylated (3P23) IRE1a-ADP-Mg21 structures (r.m.s. on Ca 5 0.5 Å). In both cases, thedimers are face to face with the RNase domains separated in aninactive conformation. Likewise, pIRE1a-ADP-Mg21 andmouseIRE1a complexed with the RNase domain inhibitor MKC9989(4PL3) are similar in dimeric structure (r.m.s on Ca 5 0.8 Å).The largest differences are located in a loop formed by residues654–659 in the kinase domain. This stretch of amino acids isengaged in intermolecular contacts within the neighboringmolecule of the asymmetric unit in the ADP complex. It is worthnoting that previous in vitro studies have shown that ADPbound to IRE1a can affect dimerization and/or reduce RNaseactivity of both mouse and human IRE1a (Prischi et al., 2014;Sanches et al., 2014).The pIRE1a-GSK2850163 and pIRE1a-STS cocrystal struc-

tures were also compared with two recently reported humanback-to-back dimers of IRE1a: one in the apo form (4Z7G) andone in the inhibitor-bound form (4Z7H). In these two struc-tures, the RNase domains are not engaging with each otheracross the dimer interface and more closely resemble thepIRE1a-GSK2850163 dimer (overall dimer r.m.s. on Ca5 1.6 Å)than the pIRE1a-STS dimer (overall dimer r.m.s. onCa5 2.7 Å).Despite this, the imidazopyridine molecule, compound 3, inter-acts with the hinge and the activation loop (DFG-in) in a similarfashion as staurosporine (r.m.s. on Ca5 0.9 Å) and behaves as atype I kinase inhibitor (Joshi et al., 2015). GSK2850163 andcompound 3 do not occupy overlapping pockets. This structuralobservation is at oddswith the other type I inhibitor–bound IRE1structures. A comparison of the IRE1a apo and IRE1a-imidazo-pyridine complex indicates that they are nearly identical (r.m.s.on Ca 5 0.6 A). This discrepancy may be due to the way theexperiments were performed, namely, that the inhibitor-boundform was resolved by soaking into the preformed apo crystals asopposed to performing cocrystallization.The characterization of the first cocrystal structure of

human phosphorylated IRE1a and its complex with a newclass of kinase inhibitors has revealed a novel mode of action for

the inhibition of pIRE1a kinase and RNase activity. Thecomparisons made between known structures across variousspecies and with inhibitors that have differential effects onIRE1a RNase activity should provide insights into the molecu-lar mechanisms responsible for activation and inhibition ofIRE1a RNase activity. These new data should also provide afeasible path to enable the design and use of pharmacologicalagents that differentially affect IRE1a/XBP 1 signaling.Coordinates for the pIRE1a-GSK2850163, pIRE1aADP-

MG2+, and pIRE1a-STS cocrystal structures have been de-posited to the Protein Data Bankwith the respective accessioncodes: PDB ID 4YZ9, 4YZD, and 4YZC.

Acknowledgments

We would like to thank the members of the IRE1a drug discov-ery program and the Computational and Structural Chemistry De-partment for fruitful discussions and insightful comments on themanuscript.

Authorship Contributions

Participated in research design: Evans, Heerding, Buser, Su,DeYoung.

Conducted experiments: Concha, Smallwood, Bonnette, Totoritis,Federowicz, Campobasso, Choudhry, Shuster, Evans, Su, DeYoung.

Contributed new reagents or analytic tools: Qi, Chen, Sweitzer,Shuster, Evans, Ralph.

Performed data analysis: Concha, Smallwood, Totoritis, Zhang,Yang, Choudhry, Heerding, Su, DeYoung.

Wrote or contributed to the writing of the manuscript: Concha,Buser, Su, DeYoung.

References

Ali MM, Bagratuni T, Davenport EL, Nowak PR, Silva-Santisteban MC, HardcastleA, McAndrews C, Rowlands MG, Morgan GJ, and Aherne W et al. (2011) Structureof the Ire1 autophosphorylation complex and implications for the unfolded proteinresponse. EMBO J 30:894–905.

Afonine PV, Mustyakimov M, Grosse-Kunstleve RW, Moriarty NW, Langan P,and Adams PD (2010) Joint X-ray and neutron refinement with phenix.refine. ActaCrystallogr D Biol Crystallogr 66:1153–1163.

Bagratuni T, Wu P, Gonzalez de Castro D, Davenport EL, Dickens NJ, Walker BA,Boyd K, Johnson DC, Gregory W, and Morgan GJ et al. (2010) XBP1s levels areimplicated in the biology and outcome of myeloma mediating different clinicaloutcomes to thalidomide-based treatments. Blood 116:250–253.

Calfon M, Zeng H, Urano F, Till JH, Hubbard SR, Harding HP, Clark SG, and Ron D(2002) IRE1 couples endoplasmic reticulum load to secretory capacity by process-ing the XBP-1 mRNA. Nature 415:92–96.

Carrasco DR, Sukhdeo K, Protopopova M, Sinha R, Enos M, Carrasco DE, Zheng M,Mani M, Henderson J, and Pinkus GS et al. (2007) The differentiation and stressresponse factor XBP-1 drives multiple myeloma pathogenesis. Cancer Cell 11:349–360.

Chen VB, Arendall WB, 3rd, Headd JJ, Keedy DA, Immormino RM, Kapral GJ,Murray LW, Richardson JS, and Richardson DC (2010) MolProbity: all-atomstructure validation for macromolecular crystallography. Acta Crystallogr D BiolCrystallogr 66:12–21.

Chen X, Iliopoulos D, Zhang Q, Tang Q, Greenblatt MB, Hatziapostolou M, Lim E,Tam WL, Ni M, and Chen Y et al. (2014) XBP1 promotes triple-negative breastcancer by controlling the HIF1a pathway. Nature 508:103–107.

Cross BC, Bond PJ, Sadowski PG, Jha BK, Zak J, Goodman JM, Silverman RH,Neubert TA, Baxendale IR, and Ron D et al. (2012) The molecular basis for se-lective inhibition of unconventional mRNA splicing by an IRE1-binding smallmolecule. Proc Natl Acad Sci USA 109:E869–E878.

Dar AC and Shokat KM (2011) The evolution of protein kinase inhibitors from an-tagonists to agonists of cellular signaling. Annu Rev Biochem 80:769–795.

Dong B, Niwa M, Walter P, and Silverman RH (2001) Basis for regulated RNAcleavage by functional analysis of RNase L and Ire1p. RNA 7:361–373.

Emsley P, Lohkamp B, Scott WG, and Cowtan K (2010) Features and development ofCoot. Acta Crystallogr D Biol Crystallogr 66:486–501.

Englander JJ, Del Mar C, Li W, Englander SW, Kim JS, Stranz DD, Hamuro Y,and Woods VL, Jr (2003) Protein structure change studied by hydrogen-deuteriumexchange, functional labeling, and mass spectrometry. Proc Natl Acad Sci USA100:7057–7062.

Feldman DE, Chauhan V, and Koong AC (2005) The unfolded protein response: anovel component of the hypoxic stress response in tumors. Mol Cancer Res 3:597–605.

Ghosh R, Wang L, Wang ES, Perera BG, Igbaria A, Morita S, Prado K, Thamsen M,Caswell D, and Macias H et al. (2014) Allosteric inhibition of the IRE1a RNase pre-serves cell viability and function during endoplasmic reticulum stress. Cell 158:534–548.

1022 Concha et al.

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from

Harding HP, Calfon M, Urano F, Novoa I, and Ron D (2002) Transcriptional andtranslational control in the Mammalian unfolded protein response. Annu Rev CellDev Biol 18:575–599.

Harrington PE, Biswas K, Malwitz D, Tasker AS, Mohr C, Andrews KL, DellamaggioreK, Kendall R, Beckmann H, and Jaeckel P et al. (2014) Unfolded protein response incancer: IRE1a inhibition by selective kinase ligands does not impair tumor cell via-bility. ACS Med Chem Lett 6:68–72.

Hetz C, Chevet E, and Harding HP (2013) Targeting the unfolded protein response indisease. Nat Rev Drug Discov 12:703–719.

Hollien J and Weissman JS (2006) Decay of endoplasmic reticulum-localized mRNAsduring the unfolded protein response. Science 313:104–107.

Joshi A, Newbatt Y, McAndrew PC, Stubbs M, Burke R, Richards MW, Bhatia C,Caldwell JJ, McHardy T, and Collins I et al. (2015) Molecular mechanisms of humanIRE1 activation through dimerization and ligand binding. Oncotarget 6:13019–13035.

Koong AC, Chauhan V, and Romero-Ramirez L (2006) Targeting XBP-1 as a novelanti-cancer strategy. Cancer Biol Ther 5:756–759.

Korennykh AV, Egea PF, Korostelev AA, Finer-Moore J, Zhang C, Shokat KM,Stroud RM, and Walter P (2009) The unfolded protein response signals throughhigh-order assembly of Ire1. Nature 457:687–693.

Kornev AP and Taylor SS (2010) Defining the conserved internal architecture of aprotein kinase. Biochim Biophys Acta 1804:440–444.

Lee AH, Iwakoshi NN, and Glimcher LH (2003) XBP-1 regulates a subset of endo-plasmic reticulum resident chaperone genes in the unfolded protein response. MolCell Biol 23:7448–7459.

Lee K, Tirasophon W, Shen X, Michalak M, Prywes R, Okada T, Yoshida H, Mori K,and Kaufman RJ (2002) IRE1-mediated unconventional mRNA splicing and S2P-mediated ATF6 cleavage merge to regulate XBP1 in signaling the unfolded proteinresponse. Genes Dev 16:452–466.

Lee KP, Dey M, Neculai D, Cao C, Dever TE, and Sicheri F (2008) Structure of thedual enzyme Ire1 reveals the basis for catalysis and regulation in nonconventionalRNA splicing. Cell 132:89–100.

Ma Y and Hendershot LM (2004) The role of the unfolded protein response in tumourdevelopment: friend or foe? Nat Rev Cancer 4:966–977.

Maly DJ and Papa FR (2014) Druggable sensors of the unfolded protein response.Nat Chem Biol 10:892–901.

Murshudov GNG, Vagin AA, and Dodson EJ (1997) Refinement of macromolecularstructures by the maximum-likelihood method. Acta Crystallogr D Biol Crystallogr53:240–255.

Nakamura M, Gotoh T, Okuno Y, Tatetsu H, Sonoki T, Uneda S, Mori M, Mitsuya H,and Hata H (2006) Activation of the endoplasmic reticulum stress pathway isassociated with survival of myeloma cells. Leuk Lymphoma 47:531–539.

Percy AJ, Rey M, Burns KM, and Schriemer DC (2012) Probing protein interactionswith hydrogen/deuterium exchange and mass spectrometry-a review. Anal ChimActa 721:7–21.

Prischi F, Nowak PR, Carrara M, and Ali MM (2014) Phosphoregulation of Ire1RNase splicing activity. Nat Commun 5:3554.

Reimold AM, Iwakoshi NN, Manis J, Vallabhajosyula P, Szomolanyi-Tsuda E,Gravallese EM, Friend D, Grusby MJ, Alt F, and Glimcher LH (2001) Plasma celldifferentiation requires the transcription factor XBP-1. Nature 412:300–307.

Ron D (2002) Translational control in the endoplasmic reticulum stress response. JClin Invest 110:1383–1388.

Rouschop KM, van den Beucken T, Dubois L, Niessen H, Bussink J, SavelkoulsK, Keulers T, Mujcic H, Landuyt W, and Voncken JW et al. (2010) The un-folded protein response protects human tumor cells during hypoxia throughregulation of the autophagy genes MAP1LC3B and ATG5. J Clin Invest 120:127–141.

Sanches M, Duffy NM, Talukdar M, Thevakumaran N, Chiovitti D, Canny MD, LeeK, Kurinov I, Uehling D, and Al-awar R et al. (2014) Structure and mechanism ofaction of the hydroxy-aryl-aldehyde class of IRE1 endoribonuclease inhibitors. NatCommun 5:4202.

Schröder M and Kaufman RJ (2005) The mammalian unfolded protein response.Annu Rev Biochem 74:739–789.

Tirasophon W, Welihinda AA, and Kaufman RJ (1998) A stress response pathwayfrom the endoplasmic reticulum to the nucleus requires a novel bifunctionalprotein kinase/endoribonuclease (Ire1p) in mammalian cells. Genes Dev 12:1812–1824.

Urano F, Wang X, Bertolotti A, Zhang Y, Chung P, Harding HP, and Ron D (2000)Coupling of stress in the ER to activation of JNK protein kinases by trans-membrane protein kinase IRE1. Science 287:664–666.

Walter P and Ron D (2011) The unfolded protein response: from stress pathway tohomeostatic regulation. Science 334:1081–1086.

Wang L, Perera BG, Hari SB, Bhhatarai B, Backes BJ, Seeliger MA, Schürer SC,Oakes SA, Papa FR, and Maly DJ (2012) Divergent allosteric control of the IRE1aendoribonuclease using kinase inhibitors. Nat Chem Biol 8:982–989.

Wang Y, Shen J, Arenzana N, Tirasophon W, Kaufman RJ, and Prywes R (2000)Activation of ATF6 and an ATF6 DNA binding site by the endoplasmic reticulumstress response. J Biol Chem 275:27013–27020.

Wasilko DJ and Lee SE (2006) TIPS: titerless infected-cells preservation and scale-up. BioProc J 5:29–32.

Winn MD, Ballard CC, Cowtan KD, Dodson EJ, Emsley P, Evans PR, Keegan RM,Krissinel EB, Leslie AG, and McCoy A et al. (2011) Overview of the CCP4 suite andcurrent developments. Acta Crystallogr D Biol Crystallogr 67:235–242.

Wiseman RL, Zhang Y, Lee KP, Harding HP, Haynes CM, Price J, Sicheri F, and RonD (2010) Flavonol activation defines an unanticipated ligand-binding site in thekinase-RNase domain of IRE1. Mol Cell 38:291–304.

Woehlbier U and Hetz C (2011) Modulating stress responses by the UPRosome: amatter of life and death. Trends Biochem Sci 36:329–337.

Yoshida H, Matsui T, Yamamoto A, Okada T, and Mori K (2001) XBP1 mRNA isinduced by ATF6 and spliced by IRE1 in response to ER stress to produce a highlyactive transcription factor. Cell 107:881–891.

Address correspondence to: M. Phillip DeYoung, GlaxoSmithKline,Oncology R&D, 1250 S. Collegeville Road, UP1450, Collegeville, PA 19426.E-mail: maurice.p.deyoung@gsk.com or Nestor O. Concha, GlaxoSmithKline,Oncology R&D, 1250 S. Collegeville Road, UP12-205, Collegeville, PA 19426.E-mail: Nestor.O.Concha@gsk.com

Structure of Human Phosphorylated (Active) IRE1a Dimer 1023

at ASPE

T Journals on M

arch 1, 2021m

olpharm.aspetjournals.org

Dow

nloaded from