Changing the Apoptosis Pathway through Evolutionary Protein … · 2019. 2. 27. · Changing the Apoptosis Pathway through Evolutionary Protein Design David Shultis1, Pralay Mitra1,

Article

Changing the ApEvolutionary Pro

David Shultis1, Pralay Mitra1, Xiaoqiang1 2

0022-2836/© 2019 Elsevie

optosis Pathway throughtein Design

Huang1, Jarrett Johnson1,Naureen Aslam Khattak , Felicia Gray , Clint Piper1, Jeff Czajka1, Logan Hansen1,Bingbing Wan3, Krishnapriya Chinnaswamy4, Liu Liu5, Mi Wang5, Jingxi Pan6,Jeanne Stuckey4, Tomasz Cierpicki 2, Christoph H. Borchers6,Shaomeng Wang5, Ming Lei3 and Yang Zhang1, 3

1 - Department of Computational Medicine and Bioinformatics, University of Michigan, 100 Washtenaw Avenue, Ann Arbor,MI 48109, USA2 - Department of Pathology, University of Michigan, 1150 W. Medical Center Drive, Ann Arbor, MI 48109, USA3 - Department of Biological Chemistry, University of Michigan, 1150 West Medical Center Drive, Ann Arbor, MI 48109, USA4 - Life Sciences Institute, University of Michigan, 210 Washtenaw Avenue, Ann Arbor, MI 48109, USA5 - Department of Internal Medicine, University of Michigan, 1500 East Medical Center Drive, Ann Arbor, MI 48109, USA6 - Department of Biochemistry & Microbiology, The University of Victoria–Genome BC Proteomics Centre, Victoria, BC,Canada V8Z 7X8

Correspondence to Yang Zhang: Department of Computational Medicine and Bioinformatics, University of Michigan,100 Washtenaw Avenue, Ann Arbor, MI 48109, USA. [email protected]://doi.org/10.1016/j.jmb.2018.12.016Edited by Amy Keating

Abstract

One obstacle in de novo protein design is the vast sequence space that needs to be searched through toobtain functional proteins. We developed a new method using structural profiles created from evolutionarilyrelated proteins to constrain the simulation search process, with functions specified by atomic-level ligand–protein binding interactions. The approach was applied to redesigning the BIR3 domain of the X-linkedinhibitor of apoptosis protein (XIAP), whose primary function is to suppress the cell death by inhibitingcaspase-9 activity; however, the function of the wild-type XIAP can be eliminated by the binding of Smacpeptides. Isothermal calorimetry and luminescence assay reveal that the designed XIAP domains can bindstrongly with the Smac peptides but do not significantly inhibit the caspase-9 proteolytic activity in vitrocompared with the wild-type XIAP protein. Detailed mutation assay experiments suggest that the bindingspecificity in the designs is essentially determined by the interplay of structural profile and physicalinteractions, which demonstrates the potential to modify apoptosis pathways through computational design.

© 2019 Elsevier Ltd. All rights reserved.

Introduction

The computational design of functional macro-molecules useful for disease model systems, diag-nostics, therapeutics, and industrial applications isbecoming a viable protein engineering method, butsuccess has been hindered by the complex atomicinteraction graph that yields such diverse function-ality and specificity [1–5]. Here we report a hybridcomputational protein–peptide design method usingstructure-based evolutionary profiles to reduce theinherent complexity of the design simulation search

r Ltd. All rights reserved.

through the identification of protein evolutionaryfingerprints, with the biological ligand-binding inter-action specified by the physics-based force field.The method is applied to create protein domains tomodulate programmed cell death, or apoptosis.Regulating apoptosis is a powerful medicinal

approach, as it can be used to either protect cellsfrom death or cull them. In cancer, for example,promoting apoptosis in oncogenic cells is beneficialto remove them from the body; but blocking apoptosisin cardiovascular disease, such as following re-duced blood flow to the heart, or ischemia, may be

Journal of Molecular Biology (2019) 431, 825–841

826 Evolutionary Design of Apoptosis Proteins

cardioprotective, and thus delay or reduce myocar-dial infarction [6]. The BIR3 domain of X-linkedinhibitor of apoptosis (XIAP) is an attractive proteindesign target because it is an inhibitor of caspase-9-dependent cellular apoptosis and has a compact,well-characterized fold that is the subject of activedrug discovery [7–9]. The XIAP BIR3 domain inhibitscaspase-9 activity through the formation of a stableheterodimeric complex, which blocks caspase-9from forming a homodimeric proteolytically activestate. Caspase-9 is an initiator of the caspaseproteolytic cascade involving caspase-3, caspase-6,and caspase- 7 that directly cause cell death andthus completes apoptosis [10]. Interestingly, theprotein Smac also binds the XIAP BIR3 domain andsubsequently blocks the XIAP-caspase-9 interaction,thus, freeing caspase-9 to homodimerize and initiateapoptosis [7,9,11]. Smac and caspase-9 bind tothe same surface region on XIAP and compete foran N-terminal tetrapeptide binding pocket that ulti-mately governs whether XIAP associates with Smacor caspase-9. The primary goal of this study was todesign de novo XIAP “like” protein sequences thatwere capable of binding the N-terminal Smac tetra-peptide with equal or better affinity than WT-XIAP.These designed proteins were intended to function as“Smac sinks” to remove free cytosolic Smac from thecell, or Smac-like therapeutic compounds, and thusbe anti-apoptotic in nature and useful as a reagent inan apoptotic disease model system.One challenge of computational protein design can

be attributed to the fact that the sequence searchspace is vast compared to the available computationalpower (20101 permutations for the 101 residue XIAPBIR3 domain) [5]. The problem is exacerbated byimperfect force fields, which cannot accurately de-scribe atomic interactions, or correctly recognizeprotein folds of given sequences.Borrowing the criticallessons from protein structure prediction experiments,where evolutionary references and fingerprints de-rived from the ensemble of known protein structureshave been the major driving force for the success ofhigh-resolution structure modeling [12,13], we pro-pose to exploit the evolutionary sequenceprofiles frommultiple homologous structures in the PDB to improvethe energy landscape of physics-base force fields andthe sequence space search of protein design.In fact, the idea of using evolutionary information

to specify the fold of the target protein is not new incomputational protein design. For instance, Socolichand coworkers [14] successfully designed a stablefold of WW proteins using constraints from multiplesequencealignments collected byPSI-BLAST search.In recent studies, we proposed a method called

Fig. 1. Flowchart of the extended EvoDesign for hybrid evolution-based protein design. The protocol consists of threeodules: (1) structure profile construction by threading the scaffold structure through the PDB using TM-align [24],) Monte Carlo sequence search guided by evolutionary structural profiles combined with physics-based bindingotentials, (3) sequence selection by clustering with distance matrix defined from BLOSUM62 substitution scores.

m(2p

EvoDesign [15,16], which utilized the sequenceprofiles collected from structural alignments to rede-sign 330 protein domains (87 from the PDB and 243from theMycobacterium tuberculosis genome), where3 out of 5 tested proteins have well-ordered tertiarystructure [15]. In particular, the crystal structure of thedesigned Phox homology domain from the cytokine-independent survival kinases was found to be veryclose (with 1.32 Å) to the target model of the designedsequence predicted by the I-TASSER-based proteinstructure prediction [17–19].Despite the power of the evolutionary design in

specifying protein folds, most of the designed proteinsare assumed to be non-functional since no biologicalinformation (e.g., binding, catalysis, etc.) was incorpo-rated. Here we examine the possibility of introducingfunction into the evolution-based design simulations bycoupling the evolutionary profiles with specific ligand-binding interactions. The binding potentials can beeither physical [20] or evolutionary based [21,22].When applying the method for XIAP, we focused onthe use of an atomic ligand–protein interaction potentialextended from FoldX [20] to enhance the bindingspecificity of the XIAP–peptide interactions, whichcontains van der Waals, solvation, hydrogen-bonding,atomic clash, and entropic interaction terms. Inaddition, an empirical equation designed for enhancingthe association rate of complex formation [23] wasintroduced to count for the electrostatic contributionbetween atoms of the interacting molecules (see Fig. 1for the hybrid pipeline extended from EvoDesign).A variety of biophysical experiments are designed to

examine the folding and ligand-binding affinity of thedesigned XIAP BIR3 domains. Of particular interest isthe novel use of high-resolution hydrogen–deuteriumexchange (HDX) mass spectroscopy (MS) in conjunc-tion with a new HDX prediction algorithm to examinethe tertiary fold by the comparison of HDX data withI-TASSER-based protein structure prediction [18,19].Furthermore, the binding specificity of XIAP with twocognate N-terminal Smac motif peptides and theinhibition of caspase-9 function are quantitativelycharacterized through isothermal calorimetry (ITC)and in vitro luminescence inhibition assay, respec-tively. The data should provide useful insight intowhether and how the physics-based binding poten-tials can be used to introduce biological activity andspecificity into evolutionary protein designs.

Results

Twosequences, Dynamic-InterfaceXIAP (DI-XIAP)and Fixed-Interface XIAP (FI-XIAP), were designed

Fig. 1 (legend on previous page )

827Evolutionary Design of Apoptosis Proteins

828 Evolutionary Design of Apoptosis Proteins

by the extended EvoDesign pipeline (Fig. 1), usingthe X-ray structure of the Human XIAP BIR3 domainco-crystallized with a high affinity N-terminus Smac

Fig. 2. Sequence and predicted structure of designed XIAP pSecondary structural elements from WT-XIAP X-ray structureresidues betweenWT-XIAP and the “AVPF” tetrapeptide or caspspheres. Red blocked residues indicate differences in the peptidresidues are the mutations outside the N-terminal binding p(b) Superposition of I-TASSER models of DI- and FI-XIAP on thsequence (PDB IDs: 1F9XA, 1G3FA, 1NW9A, 1TFQA, 1TFTA,1G73D, 2OPYA, 3CM7A, 3G76A, 3CM2A). The wild-type PDBshown in cartoons (blue to red running fromN- to C-terminals). Tinterface residues are highlighted by red and blue sticks for WTthe “AVPF” tetrapeptide from 2OPZ. (c) Complex structure of dby superposing of the designed XIAPmodels on theWT-XIAP BXIAP/Caspase-9 interface, where mutations (G326Q/N, H343are highlighted [8]. Blue and red are side-chain conformations fr

tetrapeptide “AVPF” (PDBID: 2OPZ) [25] as thescaffold. Ten non-homologous proteins with a TM-score N0.5 and the sequence identity b80% to the

roteins. (a) Sequence alignments ofWT-, FI- andDI-XIAPs.(2OPZ) are displayed above the alignments. Interfacialase-9 are shownbelow the alignments and colored in blacke-binding site betweenWT- andDI-XIAPs. Orange blockedocket known to result in loss of caspase-9 inhibition [8].e 17 PDB structures all solved for the same wild-type XIAP2JK7A, 2OPZA, 2VSLA, 3CLXA, 3EYLA, 4EC4A, 4HY0A,structures are in thin lines with the DI- and FI-XIAP modelswo arrowsmark the borders of the disordered tails. Mutated-XIAP and DI-XIAP, respectively. Yellow spheres indicateesigned XIAPs with Caspase-9 crystal structure generatedIR3 domain of the complex X-ray structure. Insets show theQ/K and L344G/A) known to abolish caspase-9 inhibitionsom designed andWT-XIAP proteins for the threemutations.

Table 1. Parameter summary of the designed proteins withstandard deviations

Parameters FI-XIAP DI-XIAP WT-XIAP

Sequence identity towild-typea

52% (100%) 47% (53%)

RMSD of I-TASSERmodel to WT (Å)b

1.12 1.01 1.28

HDX correlationc 0.55 0.74Kds with “AVPF” (nM)d 352 ± 79 167 ± 61 80 ± 25Kds with “AVPIAQKSEKY”

(nM)d971 ± 191 554 ± 93 428 ± 72

a Sequence identity between the designed and wild-typesequences. Values in parenthesis are that in the binding pocket.

b Average RMSD of the I-TASSER models in the core region[Y265–Q336] for the designed and wild-type sequences to the 17PDB structures solved for the same wild-type sequence; whilethe average RMSD between the 17 PDB structure is 1.29 Å in thecore region.

c PCCs between the observed and predicted HDX rates on thedesigned sequences.

d Average dissociation constants (Kds) from five repeated ITCexperiments. Errors are the average of the standard error fromeach of the repeated ITC experiments.

829Evolutionary Design of Apoptosis Proteins

scaffold were identified by TM-align [24], which havethe pairwise protein sequence and structure align-ments and the similarity scores listed in Table S1 inSupporting Information (SI, see also Text S1). Thepairwise structure alignments were used to constructa profile (Fig. S1A) to guide the sequence designsimulations, where the physics-based ligand–proteinbinding potential from FoldX was extended toconstrain the Smac-XIAP interactions (see Materialsand Methods). Here, the profile is specified by thesubstitution scoring matrix derived from the multiplesequence alignments of the templates that arecollected based on structural alignments (see Eq. (1)in Materials and Methods), which is termed “structuralprofile” afterward. In DI-XIAP, multiple low-energysequences were generated by the extensive replica-exchange Monte Carlo (REMC) simulation, with thesequence of the global minimum free energy selectedby sequence clustering. In FI-XIAP, similar REMCsearch was implemented but the interface residues incontact with the peptide were taken from the wild-typesequence and kept frozen during the simulation;this choice of two designs is made for examining theimpact of extensive versus constrained interfacesearch on the final designs.The design simulation and selection procedures

were fully automated. Only one sequence wasselected for each protein from the center of the largestsequence cluster and no experimental optimizationwas conducted. The DNA and protein sequencesdesigned are listed in Table S2 (see also Text S2).Fig. 2a shows the sequence alignment of the threeproteins (WT-XIAP, FI-XIAP, and DI-XIAP) withthe functional sites bound with the N-terminal Smacpeptide motif or the full-length caspase-9 labeledbelow the sequences. The overall sequence identityof FI- and DI-XIAP is 52% and 47% to the wild-typeXIAP protein, which is higher than all the templatesthat were used to construct the structure profile(except for 3T6P that has a sequence identity 48.5%by the sequence-basedNW-align butwith a sequenceidentity 41.6%by TM-align; see Table S1). Among the15 (or 30) residues bound with Smac (or caspase-9),7 (or 14) in DI-XIAP differ from that in WT-XIAP,showing that nearly half of the interface residueswere redesigned, with a mutation rate similar to theglobal sequences. A parameter summary of the FI-,DI-, and WI-XIAP sequences is listed in Table 1.The sequence identity between DI- and FI-XIAP is

51%, which seems indicating that the freezing of afew interface residues in FI-XIAP could result ina dramatic change on the global sequence designsince nearly half of the sequences are different.In fact, the change is largely due to the sequenceselection process, since a number of DI- and FI-XIAP sequence pairs in the designed sequencetrajectories have a high sequence identity (N80%)but SPICKER clustering does not select them sincethey were not located at the center of the largest

cluster. Meanwhile, the majority of the sequencevariations are located in the tail regions, suggestingthat many of the difference in the final DI- and FI-XIAP selections are not essential to their functions,except for the residues at the core regions.

I-TASSER structure predictions match with HDXdata

Prior to gene synthesis, we examined the fold-ability of the designed XIAPs using I-TASSERstructure folding simulations [18,19]. In a large-scale experiment to examine the folding of designedsequences [15], it was shown that there is a strongcorrelation between the confidence score (C-score)of I-TASSER simulations and the folding rate ofdesign proteins, and 80% (or 100%) of designedsequences are foldable for the sequence with anI-TASSER C-score N0 (or N0.8). Such correlationwas also confirmed in another design study for the PXdomain from cytokine-independent survival kinase, inwhich the I-TASSERmodel of the designed sequencewith a C-score = 1.31 has a TM-score = 0.91 (orRMSD 1.31 Å) to the finally solved X-ray structure[17]. Here, although all homologous templates witha sequence identity N30% to the target or detected byPSI-BLAST with E-value b0.5 were excluded, thetrajectories of the I-TASSER simulations on DI-XIAP(or FI-XIAP) are highly converged with 86% (or 82%)of conformations accumulated in the first SPICKERcluster [26] at an RMSD cutoff of 3.5 Å; this results ina high C-score of folding 0.82 and 0.8 for the DI- andFI-XIAPs, respectively, both being above the thresh-old of confident folding based on previous benchmarkdata [15].

830 Evolutionary Design of Apoptosis Proteins

In Fig. 2b, we show the first I-TASSER models ofDI- and FI-XIAP, superimposed on the wild-typeXIAP structures that were solved in 17 PDB entriesall for the same sequence. Although nearly halfof the designed sequences differ from WT-XIAP, theI-TASSER models are close to the wild-type XIAPs,where the average RMSDs of DI- and FI-XIAPsto the 17 PDB structures are 1.01 Å and 1.12 Å,respectively, in the core region (E16–Q87 onDI-XIAP or Y265–Q336 onWT-XIAP) after removingthe tails that are disordered. These distances areeven closer than the average distance among the17 PDB structures (RMSD = 1.29 Å), although noneof the PDB structures have been used as templatein the I-TASSER simulations. This result is under-standable because the DI- and FI-XIAP sequenceshave been designed with constraints from structuralprofiles and therefore have the structural featuresand folding pattern close to the consensus of theXIAP family. When we applied I-TASSER on theWT-XIAP sequence, the average RMSD of theI-TASSER model was 1.28 Å to the 17 PDBstructures, which is slightly higher than that of thedesigned XIAP sequences but lower than theaverage RMSD between the PDB structures ofWT-XIAP (Table 1). For further confirmation, wealso submitted the designed sequences to four otherstate-of-the-art structure prediction programs, in-cluding QUARK [27], Rosetta [28], RaptorX [29],and Phyre2 [30], which are built on ab initio and

Fig. 3. HDX data of the designed XIAP proteins in the coredown-triangles indicate observed data from c and z⁎ fragmentstructure-based HDX predictions. The dashed and solid lines cthe figure denotes secondary structure assignments based on

template-based modeling, respectively. As shownin Tables S3 and S4, the models predicted bythe different methods are highly consistent with theI-TASSER models, which are all close to the wild-type structure with a TM-score above 0.8 and RMSDbelow 3.85 Å. These initial computational foldingtests gave us confidence on the foldability of thedesigned sequences; that is, they should probablyadopt a similar fold to the wild type despite the low-sequence identity.To further examine the 3D fold of the designed

sequences, we subjected the designed sequencesto the HDX experiments [31]. The proteins, purifiedfrom bacteria, were incubated briefly in deuteriumoxide and the level of backbone amide deuteriumincorporation was determined through electron cap-ture dissociation (ECD) MS. The HDX experimentswere repeated three times for each design. In Fig. 3,we present the average HDX rate data for both DI-and FI-XIAP proteins. Because the N- and C-tails ofthe BIR3 domains are disordered as observed in thePDB structures (Fig. 2b), only the HDX levels inthe core region (E16–Q87) are presented. From theHDX profile, the loop regions generally have a higherdeuterium exchange rate (values approaching 1),indicating that these residues are largely exposed tobulk solvent. In contrast, strand and helix regionshave lower scores indicative of being more buried(values approaching 0). However, there are alsoseveral loop residues (e.g., 25–30, 50–55, etc.) having

region (E16–Q87 or Y265–Q336 on WT-XIAP). Up- andions, respectively, while open circles are from I-TASSERonnect the data points to guide the eye. The cartoon aboveDI-XIAP model by DSSP. (a) DI-XIAP. (b) FI-XIAP.

831Evolutionary Design of Apoptosis Proteins

low deuterium exchange rates and other regularsecondary structure regions (e.g., 30–33) with highdeuterium exchange rates, which are not consistentwith the simple secondary structure assignments.The open circles in Fig. 3 show the estimated HDX

rates for each residue based on the I-TASSERmodel. The predicted HDX score is made using anempirical model calculation based on the solventaccessibility of the backbone amide group (seeEq. (8) in Materials and Methods). Despite thesimplicity of the model, the estimation is largelyconsistent with the HDX data, partly confirming theI-TASSER models. The Pearson correlation coeffi-cients (PCCs) between the observed and predictedHDX rates are 0.74 and 0.55, respectively, for DI- andFI-XIAP proteins (Table 1). These correlations ap-proach the limit of the systematical errors of theexperimental data. In fact, we compared two sets ofHDX profiles on the same ubiquitin protein, onefrom top-down mass spectrometry [32] and anotherfrom NMR spectroscopy [33], and obtained a PCC of0.72 which is only slightly higher than the I-TASSER-basedmatch for FI-XIAP, but lower than DI-XIAP. Theleave-one-out cross-validation on the 394 trainingdata points showed an average PCC of 0.75 that is

Fig. 4. NMR chemical shift perturbation assays on designeratios. (a) DI-XIAP with peptide “AVPF.” (b) FI-XIAP with the pechanges with the three small polygons labeling distinct chempeptides.

also consistent with observation on the designedXIAPs.The same type of top-down ECD experiments

was also tried on WT-XIAP. However, the poor ECDfragmentation prevented us from obtaining enoughfragments to make figures as for the designedsequences. It is known that the ECD cleavage ishighly dependent on the sequenceof specific proteins.Although we did not have the HDX data for WT-XIAP,the comparison of structures determined by the top-down HDX-ECD to that determined by NMR has beenmade on many proteins in our previous experimentsin which excellent agreement was achieved [31,34],demonstrating the reliability of the methods. Here, wehave used the same conditions as previously used,including sample infusion setup, mass spectrometer,and instrumental settings, for the DI- and FI-XIAPs toensure that there was no hydrogen/deuterium scram-bling during the measurements.

Binding affinity of XIAP with the Smac peptidesdetected by 2D NMR and ITC assays

To examine the folding and binding ability ofthe designed proteins with the target peptides, we

d XIAP and Smac complexes with different stoichiometricptide “AVPIAQKSEKY.” Inset in panel b highlights spectraical shift perturbations witnessed upon the addition of the

832 Evolutionary Design of Apoptosis Proteins

conducted protein–peptide 2D NMR chemical shiftperturbation experiments on FI- and DI-XIAP usingtwo different native Smac peptides (i.e., FI-XIAP with“AVPIAQKSEKY” and DI-XIAP with “AVPF”), whichcover both the “best” and “worse” binding affinity withthe designs in the ITC experiments (see below). Theratios of peptide to protein were varied from 0:1, 0.5:1,4:1, and 5:1 to ensure that the proteins were saturatedwith the peptides. As shown in Fig. 4, the 15N–1HHSQC spectra of the designed proteins present twosets of resonances at sub-stoichiometric ratios ofpeptide to protein (0.5:1), and saturation was clearlyreached bya 4:1 ratio of peptide to protein. Thereweremore than 10 peaks associated with strong chemicalshift differences for each of the experiments betweenthe unsaturated and saturated samples (see, e.g., theinset of Fig. 4b). These data are consistent with slow-exchange kinetics of binding and high-affinity interac-tions. Meanwhile, the significant peak dispersion ofthe spectra also confirms the well-folded characteris-tics of the designed sequences.To further quantify the binding affinity, ITC exper-

iments were performed on the XIAP proteins withboth peptides of “AVPF” and “AVPIAQKSEKY” [25].The experiment was repeated five times for eachsample and all proteins were shown to have a ~1:1stoichiometry with the peptides. Fig. 5 shows atypical example of the ITC results obtained from the

Fig. 5. ITC binding assay on DI-XIAP and “AVPF”peptide complex. The top panel is the corrected heat rateper injection, and bottom is the heat per mole of injection.Protein concentrations ranged from 60 to 90 μM andpeptide from 0.7 to 1.1 mM. Peptide injection volumeswere 2 μL, and conditions were 30 mM NaPO4 (pH 7.5)and 150 mM NaCl at 298 K.

exothermic DI-XIAP/“AVPF”–peptide interaction,with the dissociation constant (Kd) = 105 nM, pep-tide to protein stoichiometry (n) = 0.87, enthalpychange (ΔH) = −3.2 kcal/mol, and entropy change(ΔS) = 21.3 cal/mol. A summary of all the ITCexperiments repeated for the WT-, DI-, and FI-XIAPswith the peptide “AVPF” is listed in Fig. S2, wherethe average Kds were found to be 80 ± 25 nM forWT-XIAP, 167 ± 61 nM for DI-XIAP, and 352 ±79 nM for FI-XIAP.For the peptide of “AVPIAQKSEKY,” the binding

affinity is general lower, with the average Kds being428 ± 72 nM forWT-XIAP, 971 ± 191 nM for FI-XIAP,and 554 ± 93 nM for DI-XIAP (Table 1). The lowerKds of the designed proteins with “AVPIAQKSEKY”are probably due to the fact that the designs wereoptimized for binding with “AVPF” because the co-crystallized XIAP/“AVPF” complex structure has beenused as the design scaffold. However, the Kd valueof the wild-type XIAP with “AVPIAQKSEKY” is alsonearly 5-fold lower than that with “AVPF”; these dataare consistent with the results obtained by otherexperiments on the WT-XIAP with the same peptides[35,36], which suggests that the peptide “AVPIAQK-SEKY” is probably more difficult to be associated withthe XIAP proteins.Overall, the magnitudes of the binding affinities are

roughly similar for the three proteins, with WT-XIAPhaving a slightly greater affinity than DI-XIAP, andDI-XIAP with a stronger affinity than FI-XIAP. Thebinding affinity on “AVPF” is generally stronger thanthat on “AVPIAQKSEKY” but the relative orderingof affinities is retained. The difference betweenDI-XIAP and FI-XIAP binding affinity is interestingbut understandable, because the sequence spacesearch in the design simulations, as guided by theatomic binding interactions, is more extensive inDI-XIAP (with all residues dynamically changed);therefore, the DI-XIAP design could identify the statesof a lower binding free-energy basin compared to theFI-XIAP in which part of the residues in the interfaceis frozen and the match of the interface design to theglobal structural profile is probably suboptimal.

Interplay of evolutionary profile and physicalpotential drives the interface design

The interface design of DI-XIAP is mainly driven bythe profile conservation score and the FoldX bindingforce field. To examine the specific roles of thesedriving forces, we list in Table 2 the conservationscores of all Smac binding-site residues (a completelist of the conservation scores for all residues isgiven in Table S5). Here, a conservation scorewas calculated as the average of the BLOSUM62substitution scales between the wild-type residueand the residues of all homologous templates ateach position of the multiple structure alignment(MSA) built by TM-align as shown in Fig. S1A, where

Table 2. Conservation scores (CS) and frequencies for the interface residues in the MSA used to guide EvoDesign

Residueposition

Amino acid inWT (DI) XIAP

CS of WT (DI)amino acid

Frequency of WT (DI)amino acid

Highest frequency(amino acid) in MSA

Highest CS(amino acid)

292 L (T) −1.0 (3.5) 0.0 (0.7) 0.7 (T) 3.5 (T)297* K (K) 1.5 (1.5) 0.3 (0.3) 0.3 (K) 1.5 (K)298 V (V) 3.6 (3.6) 0.9 (0.9) 0.9 (V) 3.6 (V)299 K (K) 2.9 (2.9) 0.4 (0.4) 0.4 (K) 2.9 (K)306* G (G) 1.2 (1.2) 0.4 (0.4) 0.4 (G) 1.2 (G)307 L (L) 3.5 (3.5) 0.8 (0.8) 0.8 (L) 3.5 (L)308 T (A) −1.2 (−1.0) 0.0 (0.0) 0.2 (D/G/R) 0.7 (Q)309 D (S) 0.0 (0.9) 0.1 (0.2) 0.3 (N) 1.5 (N)310 W (W) 11.0 (11.0) 1.0 (1.0) 1.0 (W) 11.0 (W)311 K (E) 1.7 (3.3) 0.2 (0.6) 0.6 (E) 3.3 (E)314 E (D) 2.0 (6.0) 0.0 (1.0) 1.0 (D) 6.0 (D)319 Q (E) 2.1 (3.8) 0.1 (0.7) 0.7 (E) 3.8 (E)322 K (K) 3.7 (3.7) 0.7 (0.7) 0.7 (K) 3.7 (K)323 W (W) 5.7 (5.7) 0.6 (0.6) 0.6 (W) 5.7 (W)324 Y (F) 1.5 (2.5) 0.1 (0.5) 0.5 (F) 2.5 (F)

The bold font indicates the locations that were mutated in DI-XIAP (DI), which all have a conservation score ≤2.1. “*” labels the positionsthat have a conservation score below 2.1 but were kept un-mutated in DI-XIAP.

833Evolutionary Design of Apoptosis Proteins

a higher mutation score indicates a higher degree ofconservation in evolution at the position. Ashighlightedin bold font in Table 2, all the binding residues thatweremutated in DI-XIAP have a relatively low conservationscore (≤2.1), where most of the un-mutated residueshave a high conservation score, suggesting thatEvoDesign tends to select the evolutionally variablesites to mutate due to the constraints from theevolutionary structural profile. However there are afew exceptions, where two residues (K297 and G306)have a conservation score below 2.1 but were keptun-mutated in DI-XIAP.To experimentally examine the relevance of these

positions to the binding affinity, we made a mutationat G306D that has the lowest conservation scoreamong all the un-mutated binding residues. Here, wechose the aspartate partly because of the fact thatthe aspartate has a medium size but with a negativecharge, which may result in an energetic changethat is balanced between the steric and Coulombinteractions compared to glycine that has no side-chain and neutral in charge, while a mutation to alarge-sized residue could make the steric violationdominate the energetic changes. Also, G306 is closeto a lysine K299 where a salt bridge might form whenmutated to aspartic acid, which may potentiallyenhance the binding between XIAP and the peptide.However, Fig. 6a shows that this mutation drasticallyreduced the binding affinity by 36 folds with the sameSmac peptide of “AVPF.” In Fig. 6b, we presentthe 3D structure model of the DI-XIAP and Smaccomplex built from I-TASSER, where the mutatedaspartate is sterically overlapped with the Smacpeptide atoms, despite the medium size, whichprobably explains the reduction of the binding affinity.In addition, we also made a single-point saturationmutagenesis analysis of G306 using FoldX to checkthe binding affinity of all the mutations. The binding

affinity changes formutationsG306A,G306C,G306D,G306E, G306F, G306H, G306I, G306K, G306L,G306M, G306N, G306P, G306Q, G306R, G306S,G307T, G306V, G306W, and G306Y are 3.5, 3.8, 5.0,4.5, 3.6, 3.1, 4.1, 2.0, 4.1, 1.5, 3.5, 4.2, 2.3, 2.3, 4.0,4.1, 3.8, 5.7, and 3.1 kcal/mol, respectively, comparedto G306, which indicates that none of the mutationsis favorable to binding in FoldX. Thus, considering thatG306 is themost common amino acid at the position inMSA (despite of the low conservation score), thesedata suggest that the driving force of the DI-XIAPinterface design should be attributed to the interplayof both evolutionary profile and the physics-basedprotein/peptide interactions.The important impact of the evolutionary profile

on the interface design can also be seen by theobservation that five (L292 T, K311E, E314D, Q319E,and Y324F) out of the seven mutated interfaceresidues in DI-XIAP have the highest MSA frequencyfor the mutant residue among all the amino acid types(Table 2). In other two interface mutants (T308A andD309S), however, the residues were not mutated tothe amino acids that have the highest MSA frequency,that is, T308 mutated into alanine, which does notappear in the MSA at all and D309 into serine thathas a lower frequency (0.2) than asparagine with thehighest frequency of 0.3; these data are againconsistent with the fact that the interface design ofDI-XIAP is governed by both the profile conservationscore and the FoldX binding force field.Since the mutations in the designed sequences

were made mainly on the evolutionarily variableresidues in the structural profile, one relevantquestion is if the mutations selected by EvoDesignin the interface involve any critical residues in thebinding pocket. Figure S3 presents the 3D structuremodel of the DI-XIAP/Smac–peptide complex withthe mutated interface residues highlighted in red.

Fig. 6. Impact of interface mutation on the binding affinity of DI-XIAP. (a) ITC binding assay on mutated DI-XIAP and“AVPF” peptide complex. The top panel is the corrected heat rate per injection, and the bottom is the heat per mole ofinjection. (b) Complex structure DI-XIAP and “AVPF” peptide created by I-TASSER, where G306D mutation results in asteric overlap with the peptide molecule.

834 Evolutionary Design of Apoptosis Proteins

Compared to the un-mutated interface residues thatare shown in blue, there is no obvious tendencywhere the mutations are positioned. In fact, exceptfor K311E and D309S that are obviously at theborder of the binding pocket, most of the mutations inDI-XIAP have the side-chain interacting directly withand/or oriented toward the ligand.To have a more quantitative estimation of the

locations of the mutations related to the bindingpocket, we calculated and compared in Table S6 therelative accessible surface area (rASA) of theinterface residues in the monomer (rASAm) andcomplex (rASAc) structures. Based on the classifi-cation of Levy [37], three out of the seven mutatedresidues in DI-XIAP (292L, 309D, 311K) haverASAm N25% and are categorized as “rim,” two(314E, 319Q) have rASAc b25% and are catego-rized as “support,” and two (308T and 324Y) haverASAm N25% and rASAc b25% and belong to “core”interface residues that are usually more essential tothe binding interactions due to the higher portion ofthe area involved in the interactions. The numbersof “rim,” “support” and “core” residues in the eightresidues that were not mutated are four, three, andone, respectively. These numbers further confirmthe fact that there are no specific locations on theinterface that EvoDesign tends to mutate.Thus, although we could not conclude that the

design in DI-XIAP has changed the bindingmode, it isclear that several “core” residues, whose side-chainsare in close contact with the peptide, have beenchanged. This is technically understandable becausethe homologous proteins for the profile constructionhave been collected by fold similarity rather thanfunctional similarity. Most of the homologous proteins

do not have the same binding pocket/mode asWT-XIAP. Therefore, the conservation score fromthe resultant structural profile does not necessarilycorrelate with importance of the binding residueswith Smac. Consequently, the mutations in DI-XIAP,which are mainly selected by the conservation score,can be located at both critical and less-critical bindingsites.

Inhibition of caspase-9 enzymatic activity relieson the specificity of interface design

Since the XIAP sequences were designed using acognate Smac peptide as the binding partner, it is ofinterest to examine if the designed XIAP proteinswould bind caspase-9 and inhibit its function since thelatter was not involved in the design simulations. Theinhibition of caspase-9 enzymatic activity of FI- andDI-XIAP was tested and compared with WT-XIAPusing a commercially available in vitro luminescenceXIAP/caspase-9 inhibition assay (Caspase-Glo® 9Assay). Catalysis of the commercial luminogenicsubstrate by an active caspase-9 enzyme releases asubstrate for luciferase (aminoluciferin), resulting inthe luciferase reaction and a detectable luminescenceemission in vitro; the luminescence signal generatedis proportional to the amount of caspase activitypresent, and thus, luminescence can be used as amarker for caspase-9 activity. To confirm the data, werepeated the caspase-9 inhibition experiments inde-pendently three times. In Fig. 7, we present theaverage percentage of inhibition of caspase-9 activity,converted from the relative light units (RLU) byInhibition% = 100 * [1 − (RLU − RLUp)/(RLUn −RLUp)], where RLUn is the luminescence of negative

Fig. 7. Inhibition assay of wild-type and designed XIAPson caspase-9 proteolytic activity by the Caspase-Glo® 9Assay kit from Promega. The percent inhibitions areconverted from the relative light units at different concen-tration of XIAP proteins. Lines connect data points to guidethe eye.

835Evolutionary Design of Apoptosis Proteins

control (no inhibition) and RLUp is the luminescenceof positive control (100% inhibition by caspase-9inhibitor Ac-LEHD-CHO) of that specific experiment.The data show that WT-XIAP strongly inhibitscaspase-9 activity, as demonstrated by the increasedinhibition% with increasing XIAP concentration. Incontrast, inhibition of the designed XIAP domains onthe function of the caspase-9 enzyme is significantlyreduced,where nearly 60%and 80%of the caspase-9protease activity remained even when the FI-XIAPand DI-XIAP concentration increases up to 10 k nM,but the caspase-9 activity reduces below 5% at thesame concentration of WT-XIAP (Fig. 7).The significantly reduced suppression of caspase-

9 activity by the new designs is expected, as severalkey residues involved in the WT-XIAP/caspase-9interaction were not constrained in the bindinginteractions in the design process (outside of theSmac/caspase-9 N-termini peptide-binding site,Fig. 2a). As shown in Fig. 2c, three residues, whichare known to be critical to the WT-XIAP/caspase-9binding interactions [8], have been mutated, includ-ing G326(Q/N), H343(K/Q), and L344(A/G), where(X/X) represents the (FI-XIAP/DI-XIAP) mutationsat those positions. These mutations introduce largepolar residues into a non-polar interaction surfacearea,which disrupt/clash the normal packing (G326Q/N andH343Q/K) or remove the interface contact surface(L344G/A). There are also other mutations in theseregions, as highlighted by black dots on caspase-9interface row in Fig. 2a, which may disrupt theinteraction further. These results illustrate that aphysiological function not restrained will likely beattenuated or lost during the design process.It should be mentioned that several studies have

used the point mutation technique to identify thesingle mutants that may interfere the binding inter-

action of XIAP with caspase-9 [8,38,39]. Dependingon the locations, some mutations, for example,E314S, were found to impair binding affinity of XIAPto both caspase-9 and Smac [38]. This residuewas also mutated in DI-XIAP but with a differentamino acid, that is, E314D. Due to the restraint of thebinding force field, this mutation does not impair theinteraction with Smac in our case, which partlyhighlights the specificity of the EvoDesign. Althoughthere are other point mutations that may impaircaspase-9 but not Smac, most of which are outsidethe Smac binding groove [38], we want to emphasizethat the principle of the EvoDesign process isfundamentally different from that of the single-pointmutation studies. While the point mutation is de-signed to manually select one or a few residues tochange, the de novo design algorithm allows for acomplete redesign of the sequences based onautomated and comprehensive search simulationsguided with specific profiling and binding force field,which has resulted in more than half of the residuechanged in the case of DI-XIAP. Among the 30amino acids interacting with caspase-9, 16 of themdo not interact with the Smac peptide, where halfof them (G326, E337, I339, N340, N341, H343, L344,T345) were mutated in DI-XIAP. Again, all of theeight mutated residues have a relatively low conser-vation score ≤2.1, where the majority of the un-mutated residues have a conservation score N2.1(see Table S7).

Discussion

We have extended the evolution-based method,EvoDesign, for functional protein design, whereevolutionary profiles constructed from analogousstructures in the PDB have been used as a foldingfingerprint to constrain the sequence search simu-lations, with the physical potentials extended fromFoldX for describing the ligand-binding interactions.Compared to the existing evolution-based designs[16] that focus mainly on specifying stable proteinfolds [14,15], the major technical extension of thiswork is the incorporation of physics-based ligand–protein binding interactions from FoldX [20], allowingfor the introduction of biological functions intodesigned macromolecular “chassis.”When applied to the X-linked inhibitor of apoptosis

protein (XIAP), two sequences were created bythe new binding-specific EvoDesign pipeline, onewith all residues dynamically changed (DI-XIAP)and another with binding interface residue frozen(FI-XIAP). To assess the tertiary structure fold,we used five state-of-the-art methods to fold thedesigned sequences, which generated models allwith a close similarity to the consensus of multiplesolved structures for the wild-type XIAP sequence inthe PDB. An empirical formula estimating solvent

836 Evolutionary Design of Apoptosis Proteins

accessibility of the backbone amide showed that thecomputationally predicted models from I-TASSERare in close agreement with the HDX data from thehigh-resolution ECD MS experiments for both DI-and FI-XIAP proteins.To examine the function of the design proteins, we

incubated DI- and FI-XIAP with two cognate Smacpeptides of “AVPF” and “AVPIAQKSEKY,” respective-ly. At different concentration of the peptides, the15N–1H NMR chemical shift perturbation assaysshowed a clear shift of resonance peaks betweenthe unsaturated and saturated samples, illustratingthe binding interaction between the peptides andthe design proteins. Meanwhile, the excellent peakdispersion in the NMR spectra indicates well-foldedfeature of the designed proteins. Furthermore, ITCassays showed that the binding affinity of DI-XIAP isstronger than the FI-XIAP to the peptides, both beingslightly lower than thewild-typeXIAP, but all in a similarmid/high nanomolar magnitude range. The data partlydemonstrated the advantage of dynamic interfacedesign procedure in generating tighter ligand-bindinginteractions. Physically, this is probably due to the factthat the interface residues are liberated in DI-XIAPduring the sequence search simulations, which allowsthe design simulations to identify optimized sequenceconformations with a lower folding and binding freeenergy, compared to FI-XIAP in which the constraintfrom the frozen interface can limit the optimal matchbetween the interface and the global structure profile.The binding interactions of the designed XIAPs with

caspase-9 were examined by the in vitro lumines-cence inhibition assays, where dramatically reducedinhibition of the caspase-9 activity was observed incomparison to thewild-typeXIAPprotein. Thedata areexpected because the caspase-9 binding interactionwas not considered in the design simulations andseveral key residues that areoutside theSmacbindingpocket but are involved in WT-XIAP/caspase-9interactions have been mutated. These mutationsintroduce non-physical polar and steric overlaps,which block the binding interactions with caspase-9proteins. Overall, the results showed the possibility tointroduce biological function into well-designed stablefolds by incorporating physics-based ligand-bindinginteractions into the evolutionary-based designprocedure. Apparently a higher-resolution binding-interaction potential with improved accuracy [21,22]will be essential to further enhance the specificity of thefunctional design. There is clearly room to evolve.

Materials and Methods

Pipeline of evolution-based protein design

The computational design of XIAP BIR3 domainis performed by an extended version of EvoDesign

[15,16], which consists of three modules: structuralprofile construction, Monte Carlo sequence search,and sequence selection. A flowchart of the proce-dure is depicted in Fig. 1.

Structural profile construction

The recently solved XIAP structure (PDBID:2OPZ) is a structure of XIAP bound to Smac peptide,which was used as the scaffold to model both boundand apo structures. Ten non-redundant proteins,including 1C9Q, 1E31, 1JD5, 1OXQ, 1QBH, 1SE0,2QRA, 2VM5, 3M0A, and 3T6P, which have thesame fold to the scaffold with a TM-score N0.5 and asequence identity b80% to the target, were identifiedfrom the PDB using the structure alignment program,TM-align [24]. A MSA matrix is then constructedbased on the pairwise TM-align alignments, wherethe designed DI-/FI-XIAP sequences were added tothe bottom of two MSAs for reference comparison(see Fig. S1). Here, the bound zinc in 2OPZ hasbeen removed, but it does not affect much of thesubsequent design simulations as the cysteine/histidine package is well conserved in the MSA.Next, an L × 20 profile matrix, M(p,a), was calcu-lated from the MSA, which denotes the mutationprobability of the amino acid a at the pth positionalong the sequence, where L is length of the scaffold.The element of the profile matrix is given by

M p; að Þ ¼X20x¼1

B a; xð Þ � w p; xð Þ ð1Þ

where B(a,x) is the BLOSUM62 substitution matrixand w(p,x) = ∑k

fxp

H(k). Here fxp is the frequency of the

amino acid x appearing at the pth position of theMSAand H(k) is the Henikoff weight [40] of the kthtemplate sequence in the MSA. The target scaffoldis represented by the structural profile in the follow-updesign simulations.

Monte Carlo sequence search

Starting from a random sequence, REMC simula-tions are performed to create a trajectory of artificialsequences (called sequence decoys), where ran-dom mutations are made on a set of randomlyselected residues at each step of the movements.The energy function of the MC simulation consists ofthree parts. The first part contains knowledge-basedevolutionary terms, which match the ith residue ofthe decoy sequence with the jth position of thestructural profile of the target by a score of

S i ; jð Þ ¼ M j ;Aið Þ þ w1δ ssi ; ss j� �þ w2 1−2 sai ; saj

�� ��� �

þw3 1−2 ϕi−ϕ j

�� ��� �þ 1−2 ψi−ψ j

�� ��� �� �ð2Þ

where Ai, ssi, sai, ϕi, and ψi are, respectively,the amino acid, secondary structure (SS), solvent

837Evolutionary Design of Apoptosis Proteins

accessibility (SA), and torsion angles (Φ/Ψ) of the ithresidue of the decoy sequence, and sj, saj, ϕj, and ψjare those at the jth position of the scaffold structure.The SS, SA, and Φ/Ψ features of the target are pre-assigned by DSSP [41] based on the scaffoldstructure. However, predictions on SS, SA, andΦ/Ψ are needed for the sequence decoy at each stepof the movements since the sequence and thereforethe corresponding secondary features change aftereachmutation. A quick single-sequence based neural-network predictor was developed,which ismuch faster(takes bb1 s) than the traditionally used PSI-BLASTbased predictors but with a comparable predictionaccuracy (72.6% for SS, 70.5% for SA, and 28°/46°for Φ/Ψ).Based on S(i, j), an optimal alignment path between

the design and target sequences is obtained by theNeedleman–Wunsch dynamic programming [42] withthe maximum score assigned as Eevolution, that is,

Eevolution ¼Xmax

k

S k ; kð Þ ð3Þ

where k denotes the residue index along the optimalpath of dynamic programming alignments. Indepen-dent from the sequence alignment, side-chain rota-mers of all the residues for each decoy sequence arereconstructed bySCWRL [43] based on the backboneof the scaffold structure and therefore the design doesnot incorporate indels with respect to the structuraltemplate (see Fig. S4 for illustration). We note that thereconstruction of side-chain conformations is per-formed at each Monte Carlo step when the sequencedecoy is updated. The side-chain repacking isimplemented on both chains of the bound complexstructure, during which the backbone structure is keptfrozen. No further relaxation/refinement is performedafter SCWRL modeling.The second energy function,Efoldx(XIAPapo), counts

for the physics-based atomic interactions in the apo-form XIAP monomer structure. Efoldx(XIAPapo) con-tains nine empirical terms accounting for van derWaals interaction, solvation energy, water bridgehydrogen bonding, intra-molecule hydrogen bonding,Coulomb interaction, entropy costs for fixing main-chain and side-chain atoms, and the penalty fromatomic steric overlaps [20].The third energy term counts for the ligand–protein

interactions, converted from FoldX:

E foldx interfaceð Þ ¼ E foldx complexð Þ− E foldx XIAPapo

� �þ E foldx Smacapo� �� �

ð4Þwhere Efoldx(complex) counts for the XIAP-Smaccomplex energy by FoldX. Efoldx(XIAPapo) and Efoldx(Smacapo) are the apo-form monomer energiesfor XIAP and Smac conformations, respectively. In

FoldX, ligand–protein interactions include the inher-ent contributions of complex structures from van derWaals, solvation, hydrogen-bonding, atomic clash,and entropic interactions, which are similar to theapo-form monomer, but calculated for atom pairsacross inter-chains. In addition, an empirical equa-tion that was designed to enhance the associationrate of complex formation [23] was introduced tocount for the additional electrostatic contributionbetween atoms of opposite chains, that is,

Eele interfaceð Þ ¼ Eele complexð Þ− Eele XIAPapo

� �þ Eele Smacapo� �� �

ð5Þwhere the electrostatic energy is calculated throughthe Debye–Huckel equation of

Eele ¼ 12

Xi ; j

qiq j

4π�0�r ij

e−k r ij−αð Þ1þ κα

ð6Þ

Here, qi and qj are the charge of the ith and jthcharged atoms and rij is the distance; �0 is thepermittivity of vacuum. Following Selzer et al. [23], αis set to 6 Å, κ = 0.488, and � = 80.To balance the energy terms from different

resources, Monte Carlo simulations were guidedby the sum of the Z-score of three parts of energies,that is,

EMC ¼ w4Eevolution− Eevolutionh i

δEevolution

þ w5E foldx XIAPapo

� �− E foldx XIAPapo

� �� �δE foldx XIAPapo

� �

þ w6E foldx interfaceð Þ− E foldx interfaceð Þh i

δE foldx interfaceð Þ

ð7Þ

where hEi and δE are the average and standarddeviations of the energy scores calculated from 1000random sequences. It is noted that the standarddeviations are not a constant and recalculatedin each protein design simulation. Because FoldXcontains tolerance to large steric clashes, theadoption of the random sequences does notdramatically affect the stability of the standarddeviation calculations. As shown in Fig. S5, thestandard deviations of different energy terms canquickly converge with the increase of the number ofrandom decoys.Because the average values do not affect ΔEMC =

EMC,new − EMC,old between two MC simulation steps,the actual energy weights for the three terms arew4/δEevolution, w5/δE(XIAPapo), and w6/δE(interface),respectively. The optimized parameters in Eqs. (2)and (7) are as follows: w1 = 1.58, w2 = 2.45, w3 = 1,w4 = −0.5, w5 = 1.22, and w6 = 1.22, which were

838 Evolutionary Design of Apoptosis Proteins

decided on 625 non-redundant training proteins,that is,w1,w2, andw3were proportional to the relativeaccuracy of the SS, SA and Φ/Ψ feature predictors;w4 and w5 were adjusted so that the average con-tribution from evolutional terms and physical termsare comparable based on the design simulations onthe training proteins; and w6 is set to be equal to w5since the terms have the same origin from FoldX.Similarly, the weight parameter for the cross-chainelectrostatic contribution (Eq. (5)) is set to be sameto that of the Coulomb interaction in FoldX. Thetarget protein XIAP was not included in the trainingprotein set.Compared with the previous EvoDesign protocol

[15], the major difference in scoring function designfor the Smac/XIAP complex design is the explicitcalculations of the binding interaction with the Efoldx(interface) term in Eq. (7). As shown in Table S8,the average value of δEfoldx(interface) is muchsmaller than that of δEfoldx(XIAPapo), which canresult in neglecting of the binding term in theprevious protocol due to the dominant variation ofthe monomer energy term. On the other hand, thenew protocol allows for the appropriate renormaliza-tion of different energy terms according to theirown deviations and therefore increases the relativeweight of physics-based binding interactions. Oursimulations show that the change can slightlyincrease the mutation rate of the interface residues(Fig. S6). It should be also noted that there is a slightinconsistency between the force field of side-chainpacking from SCWRL and the physical componentof design score from FoldX. However, this inconsis-tency is largely relieved by the involvement of theevolutionary profiles in EvoDesign. Meanwhile, theextensive REMC simulations cover huge sequencespace, which helps to identify the optimal designmatch even if the force fields are from differentorigins, given that both tools are well benchmarkedand represent reasonable approaches to proteindesign applications.

Sequence selection

The sequence decoys generated by the REMCsimulations are clustered using SPICKER [26] withthe distance scale defined by the sum of BLOSUM62substitution scores overall all the residue pairs that arealigned between the two sequences. All sequenceswith a distance below a threshold are counted intothe same cluster. The choice of distance scaleby mutation score instead of sequence identitycan help group more homology-related sequences.The threshold parameter is initially set to zero andgradually increased until 40% of the sequences areincluded in the primary cluster [44]. The sequencewiththe most neighbors in the primary cluster is chosenas the final design sequence, which represents thelowest free energy state in the MC simulation [26].

Computational time

We use SCWRL for side-chain repacking andFoldX for design scoring, both of which are not veryfast. In XIAP/“AVPF” PPI design, it takes about 48 hfor a typical 300,000-step REMC simulation runon 20 2.5-Hz Intel (R) Xeon (R) CPU cores in theXSEDE comet server [45].

Biophysical characterizations

Peptides

Two Smac peptide variants were used in thebinding assays. The first peptide was the N-terminaltetrapeptide “AVPF” from the 2OPZ crystal structure,and the second consisted of a longer version“AVPIAQKSEKY” (the last two residues are artifacts)from the NMR and crystal structures [7,46].

Expression constructs

DNA sequences of designed FI- and DI-XIAP wereoptimized based on E. coli K-12 frequent codonusage. The genes were synthesized by IntegratedDNA technologies and cloned into an MCSG-7 over-expression vector containing an N-terminal Histag and rTEV protease site via ligation-independentcloning. The following N-terminal artificial cloningresidues, “SNA,” remain after rTEV protease cleav-age during purification extending the length of thepurified proteins from 101 to 104 amino acids. Thecontrol WT-XIAP protein, consisting of 116 residues(241–356), was previously cloned into a pet28B(N-terminal 6 × HIS Tag) expression vector [35]. TheWT-XIAP expression construct is 139 residues long,which has a C-terminal Cys residue that formsintermolecular dimers in vitro via a disulfide bridge.However, the presence of the disulfide bridge doesnot affect Smac-XIAP interactions [35]. In our design,a 6-residue segment in the C-terminal containingthe cysteine was truncated to create a monomericprotein that simplifies the biophysical characteriza-tion of the domains. In Fig. S7, we present thegel filtration results of original WT-XIAP, truncatedWT-XIAP, and DI-XIAP, showing that the truncationindeed converts the dimer (original WT-XIAP) intomonomer domains (DI-XIAPand truncatedWT-XIAP).

Hydrogen/deuterium exchange

Pulsed HDX was conducted using a three-syringe,two-stage continuous-flow setup as described pre-viously [31]. Syringe 1 contains 50 μM XIAP in10 mM ammonium acetate at pH 7.0. Syringe 2contains 10 mM ammonium acetate in D2O. Theflow rates of the two syringes were 2 and 8 μL/min.After a labeling time of 10 s, the solution wasquenched by mixing with the outflow (20 μL/min)from syringe 3, which contained 80% D2O with 0.4%

839Evolutionary Design of Apoptosis Proteins

formic acid. The final solution, after the secondmixing tee, flows directly into the mass spectrometer.The residence time of the labeled protein underquench conditions was 1.4 s. This short quenchingtime results in an amide back-exchange level of lessof than 1%.

Mass spectrometry

AllMSdatawereacquired onaBruker 12 TApex-Qehybrid FT-ICR mass spectrometer (Bruker Daltonics,Billerica, MA, USA). The parameters for the ionsampling interface and the ion transfer were kept thesame as described previously [31] to ensure that nocollisional activation-induced H/D scrambling occurs.An ion accumulation time of 0.2 s in the collisioncell was used for the acquisition of survey-scan massspectra, while 0.3 s was used for obtaining ECD data.TheECDexperimentswere performedon the entire ionpopulation of XIAPwithin the ICR cell without precursorselection. Top-down ECD experiments on unlabeledXIAP were done by infusing the protein (2.5 μM) in anaqueous solution containing 0.1% formic acid. TheECD parameters are as follows: electron pulse length,11 ms; electron beam bias, 1.4 V; grid potential, 12 V;and cathode filament heater current, 1.2 A. Up to 600scans were accumulated for each ECD spectrum overthe m/z range 250–2600; this corresponds to anaccumulation time of 10 min. Mass calibration wasperformed using the ECD fragments of ubiquitin.

NMR spectroscopy

ABruker 600MHZspectrometerwith cryoprobewasused for NMRexperiments at 20 mMNaPO4 (pH 7.5),150 mM NaCl, and 298 K with protein concentrationsranging from70 to 150 μM.For 2DNMRchemical shiftperturbation assays, Smac peptide was added to 15Nisotopically labeled XIAP-designed proteins in 0.5:1,4:1, and 5:1 peptide to protein ratios. Saturation wasachieved by a 4:1 ratio of peptide to protein. HSQCexperiments were performed with 32 scans, 80increments in the indirect dimension, and 15 Nspectral width of 1400 Hz, with offset =118 ppm.

Isothermal calorimetry

ITC assays were conducted in 30 mM NaPO4(pH 7.5) and150 mMNaCl at 298 KusingTAsystemsand MicroCal calorimeters using degassed samples.Experiments were conducted in triplicate and theresults averaged. Protein concentrations ranged from60 to 90 μM, peptide concentration ranged from 0.7to 1.1 mM, and peptide injection volumes were 2 μL.

Cell-free caspase-9 functional assay

The enzymatic activity of active recombinantcaspase-9 (Enzo Life Sciences) was evaluated by

the Caspase-Glo® 9 Assay kit from Promega, inwhich catalysis of a substrate by caspase-9 releasesa substrate for luciferase (aminoluciferin), resultingin the luciferase reaction and a detectable lumines-cence emission in vitro. Ten microliters of serialdilutions of designed protein in caspase assay buffer[50 mM of Hepes, 100 mM of NaCl, 1 mM of EDTA,1 mM DTT with 0.1% of CHAPS and 10% of glycerol(pH 7.4)] were mixed with 2.5 μL of active caspase-9solution in caspase assay buffer. This mixture wasincubated at room temperature for 15 min. Lumino-genic Z-LEHD substrate was added with 1:1 ratio togive final caspase-9 concentration of 2.5 unit/reaction(according to the manufacturer's instructions). Thismixture was incubated at 295 K for 1 h without light,and luminescence from substrate cleavage was thendetermined by a Tecan Infinite M-1000 multimodeplate reader.

Structure-based estimation of HDX rate

From a given structure model, the HDX rate of ithresidue is estimated by Di = 1 − Si, where Si iscalculated by a linear combination of six scoringterms counting for the solvent buried status ofhydrogen bonds associated with the amide groups,that is,

Si ¼ c1SiSS þ c2S

iSA þ c3S

iNW þ c4S

iNH þ c5S

iNHv

þ c6SiCont: ð8Þ

Here

SiSS¼

1; if i th residue in strand H−bondedð Þor helix non−terminalð Þ0:25; if i th residue in strand but not H−bonded0:33; if i th residue in terminal helix0; otherwise

8><>:

ð9ÞSSAi = 1 − fSA with fSA

i being the fraction of solventaccessibility of ith residue assigned by DSSP [41];SNWi = 1 − [0.6(NNWs

i /NNWs,m)2 + 0.4(NNWb

i /NNWb,mi )2]

counts for the solvent accessibility of the amidegroups, where nNWs

i and nNWbi are the numbers

of water molecules accessible to the amide groupwith a distance cutoff 3.5 and 4.7 Å, respectively, andnNWb, si (=10) and nNWb, m

i (=50) are the maximumnumber of nNWs

i and nNWbi , respectively; SNH

i = nNH isthe number of hydrogen-bonds associated with theamide group as assigned by HBplus [47] and SNHv

i =∑nNH cos θ i counts for the feature of amide hydrogen-bonding vector, where θ i is angle between theamide proton vector (N → H) and the vector pointingfrom the amide to the protein center of mass; SCont

i isthe number of residues that have a distance below3.7 Å to the ith residue divided by the maximumof contacts for a given residue (=14). The weights inEq. (8) are selected to be c1 = 20, c2 = 15, c3 = 20,c4 = 5, c5 = 5, and c6 = 30, which were decided on a

840 Evolutionary Design of Apoptosis Proteins

training set of 394NMR-basedHDXdata points fromalpha lactalbumin (PDB ID: 1HML), kinase interact-ing forkhead-associated domain (KI-FHA) (PDB ID:1MZK), ubiquitin (PDB ID: 1UBQ), CopK (PDB ID:2K0Q), dihydrofolate reductase (PDB ID: 2L28),small archaeal modifier protein 1 (Samp1) (PDB ID:2L52), and staphylococcal nuclease (PDB ID:2KQ3), by maximizing the correlation between thepredicted and experimental HDX values. None ofthe training proteins are homologous to the XIAPprotein that was tested in this study. As part of themethod verification, we made a leave-one-outcross-validation test on the 394 HDX data points,where an average PCC of 0.75 was achievedbetween the predicted and observed HDX rates.This value was slightly higher than but consistentwith the application on the XIAP proteins, suggest-ing that the weighting parameters selected arereasonably robust for different protein sets.

Acknowledgments

We are grateful to Dr. Jeffrey R. Brender andDr. Yang Cao for insightful comments and discus-sions. The study is supported in part by the NationalInstitute of General Medical Sciences (GM083107and GM084222) and National Science Foundation(1564756). The protein design process was per-formed on the Extreme Science and EngineeringDiscovery Environment (XSEDE) clusters [45].

Appendix A. Supplementary data

Supplementary data to this article can be foundonline at https://doi.org/10.1016/j.jmb.2018.12.016.

Received 12 September 2018;Received in revised form 12 December 2018;

Accepted 28 December 2018Available online 6 January 2019

Keywords:protein design;

evolutionary profile;apoptosis pathway;

XIAP;isothermal calorimetry

Abbreviations used:XIAP, X-linked inhibitor of apoptosis protein; HDX,

hydrogen–deuterium exchange; MS, mass spectroscopy;DI-XIAP, Dynamic-Interface XIAP; FI-XIAP, Fixed-InterfaceXIAP; REMC, replica-exchangeMonteCarlo; ECD, electron

capture dissociation; PCCs, Pearson correlationcoefficients; ITC, isothermal calorimetry; MSA, multiple

structure alignment.

References

[1] N. Koga, R. Tatsumi-Koga, G. Liu, R. Xiao, T.B. Acton, G.T.Montelione, D. Baker, Principles for designing ideal proteinstructures, Nature 491 (2012) 222–227.

[2] D.N. Bolon, S.L. Mayo, Enzyme-like proteins by computa-tional design, Proc. Natl. Acad. Sci. U. S. A. 98 (2001)14274–14279.

[3] F. Yu, V.M.Cangelosi, M.L. Zastrow,M. Tegoni, J.S. Plegaria,A.G. Tebo, C.S. Mocny, L. Ruckthong, H. Qayyum, V.L.Pecoraro, Protein design: toward functional metalloenzymes,Chem. Rev. 114 (2014) 3495–3578.

[4] D.J. Mandell, T. Kortemme, Computer-aided design offunctional protein interactions, Nat. Chem. Biol. 5 (2009)797–807.

[5] I. Samish, C.M. MacDermaid, J.M. Perez-Aguilar, J.G.Saven, Theoretical and computational protein design,Annu. Rev. Phys. Chem. 62 (2011) 129–149.

[6] M.I. Oerlemans, S. Koudstaal, S.A. Chamuleau, D.P. deKleijn, P.A. Doevendans, J.P. Sluijter, Targeting cell death inthe reperfused heart: pharmacological approaches forcardioprotection, Int. J. Cardiol. 165 (2013) 410–422.

[7] G. Wu, J. Chai, T.L. Suber, J.-W. Wu, C. Du, X. Wang, Y. Shi,Structural basis of IAP recognition by Smac/DIABLO, Nature408 (2000) 1008–1012.

[8] E.N. Shiozaki, J. Chai, D.J. Rigotti, S.J. Riedl, P. Li, S.M.Srinivasula, E.S. Alnemri, R. Fairman, Y. Shi, Mechanism ofXIAP-mediated inhibition of caspase-9, Mol. Cell 11 (2003)519–527.

[9] S. Wang, Design of small-molecule Smac mimetics as IAPantagonists, Curr. Top. Microbiol. Immunol. (2010) 89–113.

[10] Q.L. Deveraux, N. Roy, H.R. Stennicke, T. Van Arsdale, Q.Zhou, S.M. Srinivasula, E.S. Alnemri, G.S. Salvesen, J.C.Reed, IAPs block apoptotic events induced by caspase-8and cytochrome c by direct inhibition of distinct caspases,EMBO J. 17 (1998) 2215–2223.

[11] C. Du, M. Fang, Y. Li, L. Li, X. Wang, Smac, a mitochondrialprotein that promotes cytochrome c-dependent caspaseactivation by eliminating IAP inhibition, Cell 102 (2000) 33–42.

[12] D. Baker, A. Sali, Protein structure prediction and structuralgenomics, Science 294 (2001) 93–96.

[13] Y. Zhang, Progress and challenges in protein structureprediction, Curr. Opin. Struct. Biol. 18 (2008) 342–348.

[14] M. Socolich, S.W. Lockless, W.P. Russ, H. Lee, K.H.Gardner, R. Ranganathan, Evolutionary information forspecifying a protein fold, Nature 437 (2005) 512–518.

[15] P. Mitra, D. Shultis, J.R. Brender, J. Czajka, D. Marsh, F.Gray, T. Cierpicki, Y. Zhang, An evolution-based approachto de novo protein design and case study on Mycobacte-rium tuberculosis, PLoS Comput. Biol. 9 (2013),e1003298.

[16] P. Mitra, D. Shultis, Y. Zhang, EvoDesign: de novo proteindesign based on structural and evolutionary profiles, NucleicAcids Res. 41 (2013) W273–W280.

[17] D. Shultis, G. Dodge, Y. Zhang, Crystal structure of designedPX domain from cytokine-independent survival kinaseand implications on evolution-based protein engineering,J. Struct. Biol. 191 (2015) 197–206.

[18] J. Yang, R. Yan, A. Roy, D. Xu, J. Poisson, Y. Zhang, TheI-TASSER suite: protein structure and function prediction,Nat. Methods 12 (2015) 7–8.

[19] A. Roy, A. Kucukural, Y. Zhang, I-TASSER: a unified platformfor automated protein structure and function prediction, Nat.Protoc. 5 (2010) 725–738.

841Evolutionary Design of Apoptosis Proteins

[20] J. Schymkowitz, J. Borg, F. Stricher, R. Nys, F. Rousseau, L.Serrano, The FoldX web server: an online force field, NucleicAcids Res. 33 (2005) W382–W388.

[21] J.R. Brender, Y. Zhang, Predicting the effect of mutations onprotein–protein binding interactions through structure-basedinterface profiles, PLoS Comput. Biol. 11 (2015), e1004494.

[22] P. Xiong, C. Zhang, W. Zheng, Y. Zhang, BindProfX:assessing mutation-induced binding affinity change byprotein interface profiles with pseudo-counts, J. Mol. Biol.429 (2017) 426–434.

[23] T. Selzer, S. Albeck, G. Schreiber, Rational design of fasterassociating and tighter binding protein complexes, Nat.Struct. Mol. Biol. 7 (2000) 537–541.

[24] Y. Zhang, J. Skolnick, TM-align: a protein structure alignmentalgorithm based on the TM-score, Nucleic Acids Res. 33(2005) 2302–2309.

[25] A.D. Wist, L. Gu, S.J. Riedl, Y. Shi, G.L. McLendon,Structure–activity based study of the Smac-binding pocketwithin the BIR3 domain of XIAP, Bioorg. Med. Chem. 15(2007) 2935–2943.

[26] Y. Zhang, J. Skolnick, SPICKER: a clustering approachto identify near-native protein folds, J. Comput. Chem. 25(2004) 865–871.

[27] D. Xu, Y. Zhang, Ab initio protein structure assembly usingcontinuous structure fragments and optimized knowledge-based force field, Proteins 80 (2012) 1715–1735.

[28] D.E. Kim, D. Chivian, D. Baker, Protein structure predictionand analysis using the Robetta server, Nucleic Acids Res. 32(2004) W526–W531.

[29] M. Källberg, H. Wang, S. Wang, J. Peng, Z. Wang, H. Lu, J.Xu, Template-based protein structure modeling using theRaptorX web server, Nat. Protoc. 7 (2012) 1511–1522.

[30] L.A. Kelley, S. Mezulis, C.M. Yates, M.N. Wass, M.J.Sternberg, The Phyre2 web portal for protein modeling,prediction and analysis, Nat. Protoc. 10 (2015) 845–858.

[31] J. Pan, J. Han, C.H. Borchers, L. Konermann, Hydrogen/deuterium exchange mass spectrometry with top-downelectron capture dissociation for characterizing structuraltransitions of a 17 kDa protein, J. Am. Chem. Soc. 131 (2009)12801–12808.

[32] G. Wang, R.R. Abzalimov, C.E. Bobst, I.A. Kaltashov,Conformer-specific characterizationof nonnativeprotein statesusing hydrogen exchange and top-down mass spectrometry,Proc. Natl. Acad. Sci. U. S. A. 110 (2013) 20087–20092.

[33] C. Bougault, L. Feng, J. Glushka, E. Kupče, J. Prestegard,Quantitation of rapid proton-deuteron amide exchange usinghadamard spectroscopy, J. Biomol. NMR 28 (2004) 385–390.

[34] J. Pan, S. Zhang, C.E. Parker, C.H. Borchers, Subzerotemperature chromatography and top-down mass spectrom-etry for protein higher-order structure characterization:

method validation and application to therapeutic antibodies,J. Am. Chem. Soc. 136 (2014) 13065–13071.

[35] Z. Nikolovska-Coleska, R. Wang, X. Fang, H. Pan, Y. Tomita,P. Li, P.P. Roller, K. Krajewski, N.G. Saito, J.A. Stuckey,Development and optimization of a binding assay for theXIAP BIR3 domain using fluorescence polarization, Anal.Biochem. 332 (2004) 261–273.

[36] R.A. Kipp, M.A. Case, A.D. Wist, C.M. Cresson, M. Carrell, E.Griner, A. Wiita, P.A. Albiniak, J. Chai, Y. Shi, Moleculartargeting of inhibitor of apoptosis proteins based on smallmolecule mimics of natural binding partners, Biochemistry 41(2002) 7344–7349.

[37] E.D. Levy, A simple definition of structural regions in proteinsand its use in analyzing interface evolution, J. Mol. Biol. 403(2010) 660–670.

[38] J. Silke, C.J. Hawkins, P.G. Ekert, J. Chew, C.L. Day, M.Pakusch, A.M. Verhagen, D.L. Vaux, The anti-apoptoticactivity of XIAP is retained uponmutation of both the caspase3- and caspase 9-interacting sites, J. Cell Biol. 157 (2002)115–124.

[39] C. Sun, M. Cai, R.P. Meadows, N. Xu, A.H. Gunasekera, J.Herrmann, J.C. Wu, S.W. Fesik, NMR structure and mutagen-esis of the third Bir domain of the inhibitor of apoptosis proteinXIAP, J. Biol. Chem. 275 (2000) 33777–33781.

[40] S. Henikoff, J.G. Henikoff, Position-based sequence weights,J. Mol. Biol. 243 (1994) 574–578.

[41] W. Kabsch, C. Sander, Dictionary of protein secondarystructure: pattern recognition of hydrogen-bonded andgeometrical features, Biopolymers 22 (1983) 2577–2637.

[42] S.B. Needleman, C.D. Wunsch, A general method applicableto the search for similarities in the amino acid sequence oftwo proteins, J. Mol. Biol. 48 (1970) 443–453.

[43] G.G. Krivov, M.V. Shapovalov, R.L. Dunbrack Jr., Improvedprediction of protein side-chain conformations with SCWRL4,Proteins 77 (2009) 778–795.

[44] A. Bazzoli, A.G. Tettamanzi, Y. Zhang, Computationalprotein design and large-scale assessment by I-TASSERstructure assembly simulations, J. Mol. Biol. 407 (2011)764–776.

[45] J. Towns, T. Cockerill, M. Dahan, I. Foster, K. Gaither, A.Grimshaw, V. Hazlewood, S. Lathrop, D. Lifka, G.D.Peterson, XSEDE: accelerating scientific discovery, Comput.Sci. Eng. 16 (2014) 62–74.

[46] Z. Liu, C. Sun, E.T. Olejniczak, R.P. Meadows, S.F. Betz, T.Oost, J. Herrmann, J.C. Wu, S.W. Fesik, Structural basis forbinding of Smac/DIABLO to the XIAP BIR3 domain, Nature408 (2000) 1004–1008.

[47] I.K. McDonald, J.M. Thornton, Satisfying hydrogen bondingpotential in proteins, J. Mol. Biol. 238 (1994) 777–793.