LETTER mRNA Display Selection of a High-Affinity, Modification-Specific Phospho-I␬B␣-Binding Fibronectin C. Anders Olson†, Hsiang-I Liao§, Ren Sun§, and Richard W. Roberts‡,* †Biochemistry and Molecular Biophysics Option, California Institute of Technology, Pasadena, California 91125, ‡Departments of Chemistry, Chemical Engineering/Materials Science, and Biology, University of Southern California, Los Angeles, California 90089-1211, and §Department of Molecular and Medical Pharmacology, University of California, Los Angeles, California 90095 hereas the human genome con- pressed efficiently in bacteria and eukary- ABSTRACT The complexity of the human pro- tains less than 25,000 genes, the otic cells, and (iii) possess improved stabil- teome is greatly expanded by post-translational W proteome likely contains over ity (12). modifications. New tools capable of recognizing 1,000,000 different proteins when alterna- Intrinsically unstructured domains com- these modifications in a sequence-specific fash- tive splicing and post-translational modifica- pose a large fraction of the proteome, and ion provide a route to purify these modified pro- tions (PTM) are taken into account (1). Post- many are subjected to PTM (1, 13). An ex- teins, to alter protein trafficking, and to visualize translational modifications are essential to ample of such a domain is the N-terminus of signal transduction in real time. Here, we have cellular signaling, and the ability to detect I␬B. In the classical NF-␬B pathway, three evolved novel, modification-specific ligands that and quantitate modifications is challenging I␬B proteins, I␬B␣, ␤, and ⑀, regulate the ac- target phosphorylated I␬B␣. To do this, we em- but immensely valuable. Nucleic-acid-based tivity of the ubiquitous transcription factor ployed mRNA display-based in vitro selection us- microarrays are powerful proteomic tools, NF-␬B (14). NF-␬B plays an important role in ing a 30-trillion-member protein library based although they cannot detect the presence inflammation and is implicated in autoim- on the fibronectin type III domain. The selection and extent of PTMs and may not be accu- mune disease and cancer (15). Upon stimu- yielded one fibronectin molecule, 10C17C25, rate predictors of protein quantities (2). At- lation, I␬B kinase (IKK) immediately phos- that binds a phospho-I␬B␣ peptide with Kd ϭ tempts to generate open-access resources phorylates I␬B␣ at serines 32 and 36 (16, 18 nM and is over 1000-fold specific compared for modification-specific affinity reagents to 17). I␬B␣ is then ubiquitinated by the SCF- to the nonphosphorylated peptide. 10C17C25 all proteins are currently under way (3). In ␤TrCP E3 ligase complex and degraded by specifically recognizes endogenous phosphory- addition to detection in vitro, novel protein the proteasome, thereby freeing NF-␬B to di- lated I␬B␣ from mammalian cell extract and sta- affinity reagents that are functional in the in- rect transcription (18). Reagents that detect bilizes phospho-I B in vivo. We also incorpo- ␬ ␣ tracellular environment are also useful for or inhibit the degradation of I␬B␣ in vivo rated 10C17C25 into a FRET indicator that determining protein localization and for as- may be useful in order to understand the detects I␬B kinase (IKK) activity in vitro, demon- sessing protein function (4, 5). Compared to many complex pathways to NF-␬B activa- strating the utility of selecting designed adaptors gene and transcript knockout techniques, tion. Inhibition at this control point is ideal for kinase activity sensors. domain and state-specific protein binders since the functions of IKK and ␤TrCP are not that knock down activity may be better de- specific to the classical NF-␬B pathway (15, terminants of protein function (6). Increas- 19). ingly, in vitro selection techniques such as We employed in vitro selection by mRNA *Corresponding author, ribosome display and mRNA display are be- display to generate novel protein affinity re- [email protected]. ing implemented to generate novel protein- agents that recognize phosphorylated I␬B␣ based affinity reagents with alternative scaf- specifically. We utilized a combinatorial pro- folds (7–9). Nonimmunoglobulin scaffolds tein library based on the 10th fibronectin Received for review March 24, 2008 such as the ankyrin-based “DARPins” (10) type III domain of human fibronectin and accepted June 2, 2008. and fibronectin type III “monobodies” (11) (10FnIII) to generate protein molecules that Published online July 1, 2008 have advantages over antibodies for both in are functional both in vitro and in vivo (20). 10.1021/cb800069c CCC: $40.75 vitro and in vivo applications because they The target used for this selection was a syn- © 2008 American Chemical Society (i) lack disulfide bonds, (ii) can be ex- thetic phospho-peptide corresponding to 480 ACS CHEMICAL BIOLOGY • VOL.3 NO.8 www.acschemicalbiology.org LETTER was evolved using the GFP reporter screen a b 60 developed by Waldo et al. (22). Three 60 rounds of evolution (A, B, and C) were per- 50 50 formed using error-prone PCR so that one or 40 − Target 40 two mutations were generated per gene. 30 + Phospho-IκBα 30 Representative sequences obtained from 20 20 each of the three rounds of evolution per- Percent bound Percent bound formed are listed in Figure 2, panel a. Ap- 10 10 proximately 2000 colonies were screened 0 0 α α in each round, and the brightest 24 colonies rget κB κB a Pool 4 Pool 5 Pool 6 Pool 7 Pool 8 Pool 9 T + I Pool 10 − were picked and assayed. The improve- ment in expression was analyzed by com- Phospho-Elk1 Phospho-I + + paring total cell fluorescence of 10C17-GFP variants relative to that of WT 10FnIII Figure 1. Selection of phospho-I␬B␣ binders. a) Enrichment of target binders was monitored by radiolabeled fusion binding assays. b) Pool 10 binding is sequence-specific and phospho- (Figure 2, panel b). For the first two rounds specific. of evolution, expression was induced with IPTG for 4 h (screen 1). We found that al- amino acid residues 22Ϫ41 of human I␬B␣. 380Ϫ394 containing two phosphorylated though the initial mutations improved ex- This peptide contains the E3 ligase recogni- serines. pression, the result of the long induction tion sequence including phosphorylated Pool 10 was cloned, and 11 sequences was that fluorescent aggregates were able serines 32 and 36, which are essential for were obtained. All 11 sequences were to accumulate. This resulted in obtaining degradation. The target peptide contained highly similar, suggesting that all sequences variants that performed better than wild- an N-terminal biotin spaced by aminohex- share a common ancestor (see Supporting type (WT) 10FnIII in the assay but did not ex- anoic acid for specific immobilization to ei- Information). One representative sequence, press as well. We found that decreasing ther neutravidin agarose or streptavidin ac- 10C17, is illustrated in Figure 2, panel a. The the length of expression to 1.5 h improved rylamide beads. wild-type sequence shown represents a the sensitivity of the screen (Figure 2, The first round of selection was carried 7-residue deletion mutant previously de- panel b, screen 2). In order to ensure that out so that the first affinity enrichment step scribed as 10FnIII (⌬1Ϫ7) and has equiva- function was retained, binding was mea- contained at least three copies of the 30- lent expression, solubility, and folding sta- sured for each of the variants using SDS- trillion-member 10FnIII library (21). Enrich- bility as full-length 10FnIII (20). Essentially, PAGE quantitated by autoradiography ment was monitored by measuring radiola- only one solution to the molecular recogni- (Figure 2, panel c). beled fusion binding (Figure 1, panel a). In tion of phospho-I␬B␣ was obtained. This From round one of the evolution, one vari- order to reduce the enrichment of bead may be due to an intrinsic rarity of binders ant containing two mutations was obtained binders, a negative selection step was em- in the library, as obtaining binders to non- (A21) that improved both solubility and ployed at round 6. After 10 rounds of selec- structured targets may be more difficult than binding (Figure 2). Two sequences were ob- tion, pool 10 was determined to have ap- obtaining binders to structured targets. Al- tained from round two that improved solu- proximately 50% binding efficiency at RT ternatively, it is possible that the pool bility and did not affect binding. B11 was a with relatively low background binding reached an artificial bottleneck due to overly unique clone, whereas the B01 sequence (1%). Figure 1 (panel b) demonstrates that stringent selection conditions or that the im- was found three times out of the 24 clones pool 10 binding is both sequence-specific mobilization scheme on streptavidin beads assayed. We created 10C17B25 by combin- and phospho-specific. Binding to a non- limited accessibility of the target. ing the two beneficial point mutations se- phosphorylated I␬B␣ peptide otherwise The ability for the binders to express in lected, L13Q and A10T, from B11 and B01, identical to the target was equal to back- bacteria is important for their utility as affin- respectively. 10C17B25 displays an addi- ground. Likewise, phospho-serines alone ity reagents. Both 10C17 and another pool tive improvement in solubility. The third are not sufficient for binding. No binding 10 clone, 10C19, do not express well in E. round of evolution again generated two vari- was detected above background to a pep- coli BL21(DE3) (data not shown). In order to ants both with improved expression (C21, tide corresponding to Elk1 residues improve the expression of 10C17, solubility which was found twice, and C22, which was www.acschemicalbiology.org VOL.3 NO.8 • 480–485 • 2008 481 a BC FG βA βB loop βC βD βE βF loop βG WT10FnIII ML E V V A A T P T S L L I SWDA P A V T VRY YR I T YGE TGGNS P VQE F T V PGS K S T A T I SG L K PGVDY T I T V Y A V TGRGDS P A S S K P I S I NYR T 10C17 ML E V V V A T P A S L L I SWP A S SRWRRY YR I T YGE TGGNS P VQE F T V PGS K T T A T I SG L K PGVDY T I T V S A V TQQ L HQP KWRWP I S I NYR T 10C17A21 .