Research Article

Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads

Stanford University, United States
University of California, San Francisco, United States
Chan Zuckerberg Biohub, United States

Jul 8, 2019

https://doi.org/10.7554/eLife.40499

Version of Record

Accepted for publication after peer review and revision.

Download
Cite
Share
CommentOpen annotations (there are currently 0 annotations on this page).

Version of Record published: September 5, 2019 (This version)
Accepted Manuscript published: July 8, 2019 (Go to version)
Accepted: July 3, 2019
Received: August 8, 2018

1. Of interest
Drug Discovery: Decoding the mechanisms of allostery

Saif Khan, Cornelius Gati

Insight Jun 2, 2023
Further reading

Article
Figures and data
Abstract
Introduction
Results
Discussion
Materials and methods
Data availability
References
Decision letter
Author response
Article and author information
Metrics

Abstract

Transient, regulated binding of globular protein domains to Short Linear Motifs (SLiMs) in disordered regions of other proteins drives cellular signaling. Mapping the energy landscapes of these interactions is essential for deciphering and perturbing signaling networks but is challenging due to their weak affinities. We present a powerful technology (MRBLE-pep) that simultaneously quantifies protein binding to a library of peptides directly synthesized on beads containing unique spectral codes. Using MRBLE-pep, we systematically probe binding of calcineurin (CN), a conserved protein phosphatase essential for the immune response and target of immunosuppressants, to the PxIxIT SLiM. We discover that flanking residues and post-translational modifications critically contribute to PxIxIT-CN affinity and identify CN-binding peptides based on multiple scaffolds with a wide range of affinities. The quantitative biophysical data provided by this approach will improve computational modeling efforts, elucidate a broad range of weak protein-SLiM interactions, and revolutionize our understanding of signaling networks.

https://doi.org/10.7554/eLife.40499.001

Introduction

In vivo, rapid regulation of weak, transient protein-protein interactions is essential for dynamically shaping cellular responses. Nearly 40% of these interactions are mediated by 3–10 amino acid Short Linear Motifs (SLiMs) interacting with protein globular domains (e.g. SH3, SH2, and PDZ domains) or enzymes (e.g. kinases and phosphatases) (Dinkel et al., 2016; Neduva and Russell, 2006; Tompa et al., 2014). The human proteome is estimated to contain more than 100,000 of these SLiMs, many of which are highly regulated by post-translational modifications (PTMs) such as phosphorylation (Tompa et al., 2014; Ivarsson and Jemth, 2019). As the weak affinities of these interactions ( $K_{d}$ values of ∼1 to 500 μM) are often close to the physiological concentrations of the interacting partners in vivo, subtle differences in concentrations or interaction affinities can have large effects on the fraction bound and downstream signaling output (Roy et al., 2007; Hein et al., 2017). Measuring and predicting not only which SLiMs a protein binds but also the affinities of SLiM-binding interactions is therefore essential for predicting signal strengths within signaling networks, understanding how these networks are perturbed by human disease, and identifying new therapeutic inhibitors (Uyar et al., 2014).

Calcineurin (CN), a conserved Ca²⁺/calmodulin-dependent phosphatase, is a prime example of a signaling protein that relies on weak SLiM-mediated interactions for substrate recognition. Although CN plays critical roles in the human immune, nervous, and cardiovascular systems and likely dephosphorylates hundreds of downstream targets, only ∼70 are known to date (Sheftic et al., 2016). These include the NFAT family of transcription factors, whose dephosphorylation by CN is required for T-cell activation and adaptive immunity (Jain et al., 1993). Consequently, CN is the target of the widely used immunosuppressants cyclosporin A (CysA) and FK506. CN dephosphorylates sites with little sequence similarity, instead recognizing substrates by binding to two characterized SLiMs (PxIxIT and LxVP) located at variable distances from the phosphosite (Roy and Cyert, 2009). Blocking SLiM binding to CN prevents dephosphorylation without altering its catalytic center: FK506 and CysA prevent LxVP docking, the viral inhibitor A238L blocks PxIxIT and LxVP binding, and the high-affinity peptide inhibitor PVIVIT blocks PxIxIT binding (Grigoriu et al., 2013; Aramburu et al., 1999). The affinities of PxIxIT motifs determine biological output by specifying the Ca²⁺ concentration-dependence of substrate dephosphorylation in vivo (Müller et al., 2009; Roy et al., 2007). However, the relationship between PxIxIT sequence and CN binding affinity has never been probed outside of the core motif. A comprehensive understanding of PxIxIT-CN binding would allow discovery of novel CN substrates and aid efforts to rationally design CN inhibitors with enhanced selectivity.

Unfortunately, the limited binding interfaces associated with SLiM-mediated interactions result in low to moderate affinities (with typical $K_{d}$ values in the range of 1–500 uM), high dissociation rates (Zhou, 2012; Dogan et al., 2015), and a rapid equilibrium (Gianni and Jemth, 2017; Bagshaw, 2017), complicating experimental efforts to measure affinities. Display-based methods such as combinatorial phage display (Tonikian et al., 2008) and ProP-PD (Ivarsson et al., 2014; Sundell and Ivarsson, 2014) allow screening for binding between a protein of interest and large libraries (up to 10¹⁰) of candidate SLiM-containing peptides. However, these methods typically identify only strong binders, cannot return negative information about residues that ablate binding critical for downstream target prediction in vivo (Gfeller et al., 2011), and do not directly probe effects of PTMs. Array-based methods (Fodor et al., 1991) allow quantification of binding of labeled proteins to 10s-100s (SPOT arrays) (Frank et al., 1990), tens of thousands (Atwater et al., 2018), or millions (ultra-high density arrays) (Buus et al., 2012; Forsström et al., 2014; Price et al., 2012; Carmona et al., 2015) of peptides chemically synthesized in situ, permitting direct incorporation of PTMs or unnatural amino acids at specific positions (Engelmann et al., 2014; Tinti et al., 2013; Filippakopoulos et al., 2012). However, it is difficult to evaluate the yield and purity of peptides in each spot, contributing to false positives and negatives (Tinti et al., 2013; Blikstad and Ivarsson, 2015). Most importantly, display- and array-based screening methods cannot measure quantitative interaction affinities ( $K_{d}$ s or $Δ Δ G$ s), as these measurements require washes that take the system out of thermodynamic equilibrium and lead to preferential loss of interactions with faster off-rates (Zhou, 2012; Dogan et al., 2015; Ivarsson and Jemth, 2019). As a result, candidate interactions identified through screening methods are typically validated and their affinities quantified via subsequent quantitative, low-throughput methods that require large amounts of material and are labor-intensive (e.g. isothermal calorimetry, fluorescence polarization, or surface plasmon resonance) (Gibson et al., 2015; Davey et al., 2017; Dinkel et al., 2016; Tonikian et al., 2008). The ability to measure binding affinities for hundreds of protein-peptide interactions in parallel would simultaneously remove this validation and quantification bottleneck and allow quantitative mapping of binding specificity landscapes to improve target prediction in vivo.

Here, we present a powerful bead-based technology for quantitatively measuring affinities for many SLiM-mediated protein-peptide interactions in parallel using very small amounts of material. Peptides are synthesized directly on spectrally encoded beads (MRBLEs, Microspheres with Ratiometric Barcode Lanthanide Encoding) (Gerver et al., 2012; Nguyen et al., 2017a) with a unique linkage between each peptide sequence and a given spectral code. MRBLE-pep libraries can then be pooled and assayed for protein binding in a single small volume before being imaged to identify the peptide sequence associated with each bead and quantify the amount of protein bound. On-MRBLE chemical synthesis allows for precise control of peptide density, incorporation of PTMs at known locations, and in-line assessment of peptide quality via mass spectrometry to identify amino acids that ablate and promote protein binding with equal confidence. Most importantly, MRBLE polymeric beads have been deliberately engineered to have slow on- and off-rates, thereby slowing dissociation and allowing quantitative measurement interaction strengths for of weak and transient interactions difficult to detect via other techniques. MRBLE-pep provides a critical complement to high-throughput screening approaches, enabling facile downstream screening of candidate motif hits as well as systematic mapping of how individual residues at each position within the motif contribute to binding affinity and specificity.

We use MRBLE-pep to systematically mutate residues at each position within and flanking three previously-characterized PxIxIT sequences and quantify their effects on affinity. We find that flanking amino acids and post-translational modifications play surprisingly critical and previously unappreciated roles in determining affinity and specificity. Through iterative cycles of mutagenesis, in vitro affinity measurements, and in vivo activity measurements, we map the CN-PxIxIT affinity landscape and identify several PxIxIT peptides of unprecedented binding affinity that could prove useful for future development of additional potent CN inhibitors. This approach can be applied to a broad range of protein-SLiM interactions to map binding energy landscapes, model signaling networks, and identify novel therapeutic inhibitors.

Results

MRBLE-pep experimental assay overview

We developed a new high-throughput protein-peptide interaction assay based on spectrally encoded hydrogel beads containing unique ratios of lanthanide nanophosphors (MRBLEs) (Gerver et al., 2012; Nguyen et al., 2017a). MRBLEs containing a given code are output to a filtered-tip in a 96-well format or two 48–2 mL reaction tubes (Figure 1A). These tips or tubes are then transferred to a peptide synthesizer for functionalization and solid phase peptide synthesis (SPPS), thereby uniquely linking peptide sequences with spectral codes (Figure 1A) (Hintersteiner et al., 2009; Lam et al., 1991; Lee et al., 2014; Liu et al., 2002; Meldal, 2002). Following SPPS, MRBLE-bound peptide libraries are pooled, incubated with an epitope-tagged protein of interest and a fluorescently-labeled antibody, washed, and imaged to reveal the peptide sequence associated with each bead (by quantifying lanthanide emissions) and the amount of protein bound (by quantifying fluorescence) (Figure 1B,C). Unlike commercially available spectrally encoded beads (e.g. Luminex), spectral codes can be ‘read’ using a single UV excitation source, preserving the ability to multiplex binding measurements using up to three detection antibodies (Nguyen et al., 2017b).

Figure 1 with 4 supplements see all

Download asset Open asset

MRBLE-pep experimental pipeline for high-throughput measurement of protein-peptide interactions.

(A) Encoded beads are synthesized in a microfluidic device and output to peptide synthesis tips arrayed in a 96 well plate format with one MRBLE code per tip. Peptides are then synthesized directly on MRBLEs via solid phase peptide synthesis with a unique 1:1 linkage between peptide sequence and spectral code. (B) MRBLE-pep libraries are pooled and incubated with an epitope-tagged protein and fluorescently-labeled antibody. (C) Following incubation and washing, peptide sequence and amount of bound protein are determined via imaging. (D) Example data showing images of MRBLE-pep beads coated with a CN-bound and negative control peptide, measured lanthanide ratios, and bound antibody intensities.

https://doi.org/10.7554/eLife.40499.002

Figure 1—source data 1 PVIVIT code ratios.: https://doi.org/10.7554/eLife.40499.007
Download elife-40499-fig1-data1-v2.csv
Figure 1—source data 2 ‘PVIAVT code ratios.: https://doi.org/10.7554/eLife.40499.008
Download elife-40499-fig1-data2-v2.csv

Integrated synthesis quality control ensures production of full-length, correct peptide sequences

Computational prediction of novel CN-binding SLiMs in vivo can be improved by including information about residues that ablate binding (Krystkowiak et al., 2018; Kaushansky et al., 2010; Gfeller et al., 2011). To accomplish this, we implemented in-line peptide quality assessment to ensure that any observed non-binding results from a true absence of interaction and not a failure to synthesize the correct peptide. MRBLEs were first functionalized with an acid-labile rink amide linker within the bead core and a non-labile glycine linker on the outer bead shell (Figure 1—figure supplement 1) (Liu et al., 2011); varying the ratio of extendible to non-extendible glycine linkers in the bead shell allows tuning of displayed peptide density (Liu et al., 2002; Chen et al., 2009). Next, peptides were synthesized on both linkers via standard Fmoc SPPS. Peptides coupled to the MRBLE core via the acid-labile linker were eluted during the final acid global deprotection step and can be verified via MALDI mass spectrometry. As peptides within the core are inaccessible to large proteins, elution allows sequence verification without reducing binding signal in downstream binding assays. Measured MRBLE lanthanide ratios before, during, and after SPPS remained constant, (Figure 1—figure supplement 2), establishing that embedded codes are unchanged by chemical exposures.

Measuring binding affinities from a single pooled assay requires reliable estimates of the effective concentration of each peptide, which could be skewed by sequence-dependent differences in SPPS efficiency. To detect any large differences, a portion of each MRBLE-pep library was biotinylated, incubated with labeled streptavidin (DyLight650-SA), washed, and imaged to quantify bound streptavidin. Bound DyLight650-SA intensities were relatively consistent between sequences (Figure 1—figure supplement 3) and saturated at ∼20 nM DyLight650-SA for ∼7100 beads in a 100 μL reaction volume (Figure 1—figure supplement 4), establishing a surface density of ∼2 × 10⁸ peptide molecules per MRBLE.

MRBLE-pep yields quantitative measurements of calcineurin binding affinities

Measuring CN-PxIxIT binding affinities represents a demanding test for a new protein-peptide interaction assay, as CN binds PxIxIT peptide with moderate-to-weak affinities (Figure 2A) via surface interactions (Figure 2B), and prior literature affinity estimates for a given substrate have varied over 50-fold (Figure 2A, NFATc1 and RCAN1). To determine MRBLE-pep assay sensitivity and resolve discrepancies in prior measurements, we performed a pooled assay in which MRBLEs bearing 10 peptides (each synthesized on three spectral codes) were incubated with varying concentrations of His-tagged CN complexed with fluorescently-labeled anti-His antibody. These 10 peptides (the ‘Triplicate low’ dataset) represent all calcineurin-binding SLiMs for which affinities have been measured, including the known NFATc1, NFATc2, AKAP79, and RCAN1 natural CN-interacting PxIxIT binding sites (Figure 2A; Supplementary file 1) (Aramburu et al., 1998; Grigoriu et al., 2013; Li et al., 2007; Mulero et al., 2009), a set of 5 PVIVIT peptide mutants previously characterized via competitive fluorescence polarization assays (Aramburu et al., 1999; Li et al., 2007), and a scrambled negative control sequence. Consistent with previous observations, the high-affinity PVIVIT and PVIVVT variants showed strong binding, a scrambled peptide showed no binding, and the PVIAVT and PVIVIN variants showed low to intermediate binding (Figure 2C, Figure 2—source data 1), with consistent results for the same peptide sequence synthesized on MRBLEs with different codes. NFATc2 and AKAP79 showed measurable binding and were therefore selected for further systematic mutagenesis. Although MALDI mass spectrometry confirmed successful synthesis of both RCAN1 and NFATc1 full-length peptides, binding was near the limit of detection in this assay (Figure 2C).

Figure 2 with 3 supplements see all

Download asset Open asset

MRBLEs allow quantitative measurement of CN-SLiM interactions in high-throughput.

(A) Sequence alignment for previously identified CN PxIxIT binding sites (see Supplementary file 1 for references). (B) Crystal structure showing CN heterodimer bound to the A238L inhibitor including both PxIxIT and LxVP binding sites (PDB ID: 4F0Z). (C) Concentration-dependent binding for CN interacting with either PVIVIT variants (top), natural PxIxIT peptides (bottom), or a scrambled PxIxIT motif (bottom). Each peptide was synthesized on three different MRBLE codes (light grey lines); affinities were determined from global Langmuir isotherm fits (black) to median values at each concentration (grey circles). (D) MRBLE-derived and literature reported $K_{d}$ values for various CN-PxIxIT interactions. Grey solid line indicates linear regression between data sets; dotted line indicates expected 1:1 agreement. (E) Langmuir isotherm fits to concentration-dependent binding behavior for CN interacting with candidate high-affinity PVIVIT core variants.

https://doi.org/10.7554/eLife.40499.009

Figure 2—source data 1 ‘Triplicate low’ affinity measurements.: https://doi.org/10.7554/eLife.40499.013
Download elife-40499-fig2-data1-v2.csv
Figure 2—source data 2 ‘Triplicate high’ affinity measurements.: https://doi.org/10.7554/eLife.40499.014
Download elife-40499-fig2-data2-v2.csv

To determine binding affinities for each peptide, we globally fit all data from a given assay to a single-site binding model with the assumption that while individual $K_{d}$ values may differ between peptides, the stoichiometry of binding remains constant (see Materials and methods). Although these measurements take place out of equilibrium due to the need to wash beads prior to imaging, we specifically engineered MRBLEs to have very slow on- and off-rates (with measured rates on the order of hours) (Figure 2—figure supplement 1, Figure 2—figure supplement 2), allowing measured intensities to approximate equilibrium values to yield apparent dissociation constants (Figure 2C, Figure 2—figure supplement 3). Slow on- and off-rates have previously been observed for hydrogel particles and are known to scale with particle radius, as proteins initially encounter and bind bead-bound ligands and then slowly diffuse into the particle via iterative dissociation and rebinding events (Shapiro et al., 2012). Resultant apparent $K_{d}$ s largely agree with previously published values (Figure 2D) and establish that MRBLE-pep can resolve differences in weak affinities spanning from ∼0.20 to ∼50 μM (a dynamic range > 2 orders of magnitude, comparable to SPR and FP; Jiang and Barclay, 2009). MRBLE-pep apparent $K_{d}$ s slightly underestimate affinities relative to published values while preserving rank order and relative differences; therefore, we provide relative affinity differences ( $Δ Δ G$ ) in all subsequent analyses, which are unaffected by this shift.

Examination of known CN PxIxIT motifs (NFATc1, NFATc2) and a viral inhibitor (A238L) suggested a potential preference for positively charged (R, K) or hydroxyl (S) residues in the degenerate PxIxIT position 2 (Figure 2A). To evaluate assay reproducibility and whether these substitutions enhance binding, we measured binding to these same 10 peptides plus an additional 3 PVIVIT variants (PSIVIT, PRIVIT, PKIVIT) (‘triplicate high’) (Figure 2E, Figure 2—source data 2). These data reproduce previously observed trends and additionally suggest that the PKIVIT peptide binds slightly more strongly than PVIVIT; therefore, we also selected PKIVIT for additional systematic mutagenesis.

High-throughput MRBLE-pep mapping of the CN-PxIxIT binding energy landscape

Using this new assay, we experimentally mapped portions of the CN-PxIxIT binding energy landscape by measuring CN binding to two MRBLE-pep libraries for each of the PVIVIT, PKIVIT, NFATc2, and AKAP79 sequences containing systematic mutations in either the ‘core’ (positions 1–6) or ‘flanking’ PxIxIT residues (positions −1 and 7–9) (368 peptides total) (Figure 3A, Figure 3—figure supplement 1, Figure 3—figure supplement 2, Figure 3—figure supplement 3, Figure 3—figure supplement 4, Figure 3—figure supplement 5, Figure 3—figure supplement 6, Figure 3—figure supplement 7, Figure 3—figure supplement 8, Figure 3—source data 1, Figure 3—source data 2, Figure 3—source data 3, Figure 3—source data 4, Figure 3—source data 5, Figure 3—source data 6, Figure 3—source data 7, Figure 3—source data 8). As the P and I residues in positions 1, 3, and 5 are known to be strongly conserved across substrates, we prioritized mutagenesis at other, less conserved positions. The optimal number of peptides to screen per assay depends on the competing effects of ligand depletion and competition, the fraction of peptides expected to bind, the range of interaction strengths, and the statistical robustness of each measurement (Pollard, 2010). To reduce error and maximize the ability to resolve subtle differences, we profiled 48 peptides per reaction, with ∼100 beads per sequence in each assay. To ensure that measured binding resulted from a true CN-PxIxIT interaction and not nonspecific binding, we repeated all experiments at a single, high concentration (250 nM) using a CN N330A/I1331A/R332A mutant defective in PxIxIT binding (CN NIR) (Li et al., 2004) and labeled antibody alone (Figure 3—figure supplement 9, Figure 3—figure supplement 10, Figure 3—figure supplement 11, Figure 3—figure supplement 12; Figure 3—source data 9, Figure 3—source data 10). Several peptides were strongly bound by anti-His antibody but did not show binding when the antibody was pre-incubated with His-tagged CN; peptides strongly bound by the CN NIR mutant (e.g. PRIRIT) were removed from downstream analysis. All PVIVIT, PKIVIT, and NFATc2 libraries and the AKAP79 ‘core’ mutation library were fit by a single-site binding model, permitting direct measurement of $Δ Δ G$ for each substitution. As measured intensities for the AKAP79 ‘flank’ mutation library never reached saturation, MRBLE-pep measurements provide only qualitative estimates of affinity effects.

Figure 3 with 23 supplements see all

Download asset Open asset

MRBLE-pep mapping of the CN-PxIxIT binding energy landscape.

(A) Langmuir isotherm fits to concentration-dependent binding behavior for CN interacting with PVIVIT variants containing systematic mutations within position 2. (B) Relative change in binding affinity for individual substitutions at position 2 (as compared to WT). (C) Heat map showing relative affinities per substitution per position for 4 PxIxIT motifs. (D) Langmuir isotherm fits to concentration-dependent binding behavior for CN interacting with a medium-affinity PVIVIT variant containing the wild type histidine, a serine, and a phosphoserine at position 9. (E) Comparison of the effect on affinity for a particular substitution at a particular position in the PKIVIT motif (y axis) vs. the PVIVIT motif (x axis). Inset shows relative effects on affinity for various substitutions at the V4 position in PKIVIT (top) and PVIVIT (bottom), highlighting differential effects of an I substitution.

https://doi.org/10.7554/eLife.40499.015

Figure 3—source data 1 Concentration-dependent binding data for PVIVIT core library.: https://doi.org/10.7554/eLife.40499.039
Download elife-40499-fig3-data1-v2.csv
Figure 3—source data 2 Concentration-dependent binding data for PVIVIT flank library.: https://doi.org/10.7554/eLife.40499.040
Download elife-40499-fig3-data2-v2.csv
Figure 3—source data 3 Concentration-dependent binding data for PKIVIT core library.: https://doi.org/10.7554/eLife.40499.041
Download elife-40499-fig3-data3-v2.csv
Figure 3—source data 4 Concentration-dependent binding data for PKIVIT flank library.: https://doi.org/10.7554/eLife.40499.042
Download elife-40499-fig3-data4-v2.csv
Figure 3—source data 5 Concentration-dependent binding data for NFATc2 core library.: https://doi.org/10.7554/eLife.40499.043
Download elife-40499-fig3-data5-v2.csv
Figure 3—source data 6 Concentration-dependent binding data for NFATc2 flank library.: https://doi.org/10.7554/eLife.40499.044
Download elife-40499-fig3-data6-v2.csv
Figure 3—source data 7 Concentration-dependent binding data for AKAP79 core library.: https://doi.org/10.7554/eLife.40499.045
Download elife-40499-fig3-data7-v2.csv
Figure 3—source data 8 Concentration-dependent binding data for AKAP79 flank library.: https://doi.org/10.7554/eLife.40499.046
Download elife-40499-fig3-data8-v2.csv
Figure 3—source data 9 WT and mutant binding data for PVIVIT core library.: https://doi.org/10.7554/eLife.40499.047
Download elife-40499-fig3-data9-v2.csv
Figure 3—source data 10 WT and mutant binding data for PVIVIT, PKIVIT, NFACTc2, and AKAP79 core and flank libraries.: https://doi.org/10.7554/eLife.40499.048
Download elife-40499-fig3-data10-v2.xlsx

To visualize the binding affinity landscape of CN-PxIxIT interactions, we generated graphs showing the relative change in affinity upon substitution to each mutant amino acid at each position (Figure 3B, Figure 3—figure supplement 13, Figure 3—figure supplement 14, Figure 3—figure supplement 15, Figure 3—figure supplement 16, Figure 3—figure supplement 17, Figure 3—figure supplement 18, Figure 3—figure supplement 19, Figure 3—figure supplement 20, Figure 3—figure supplement 21); information at all positions can be combined to produce a heat map of relative changes in binding energies ( $Δ Δ G$ ) (Figure 3C) or scaled logos (Figure 3—figure supplement 22) (Tareen and Kinney, 2019). Although considered to be degenerate in the consensus motif, many substitutions at positions 2 and 4 had unexpectedly strong effects on binding (Figure 3B). In particular, V2R and V2K mutations in the PVIVIT scaffold significantly enhanced binding in experimental assays, consistent with previous observations of high-affinity CN binding to the viral inhibitor (A238L) containing a positively-charged residue at this position (PKIIIT) (Grigoriu et al., 2013).

Flanking residues play a major role in defining binding affinity and specificity

The experimental impact of sequences flanking the core ‘PxIxIT’ residues on CN binding has never been probed systematically. At the −1 position, MRBLE-pep data revealed a clear CN preference for hydrophobic residues (V,L,I) and a weak preference for polar residues (T, Y, H, N, and R) for all three of the PVIVIT, PKIVIT, and NFATc2 scaffolds. By contrast, substitution to acidic (D, E) residues at this position strongly reduced binding (Figure 3C, Figure 3—figure supplement 1, Figure 3—figure supplement 2, Figure 3—figure supplement 3, Figure 3—figure supplement 4, Figure 3—figure supplement 5, Figure 3—figure supplement 6, Figure 3—figure supplement 13, Figure 3—figure supplement 14, Figure 3—figure supplement 15, Figure 3—figure supplement 16, Figure 3—figure supplement 17, Figure 3—figure supplement 18). At position 9, the majority of mutations in the PVIVIT scaffold increased affinity. Substitution of phosphomimetic residues (D, E) or unphosphorylated serine increased affinity slightly and not at all, respectively, but substitution of phosphoserine and phosphothreonine residues (X, Z) increased affinity nearly 50-fold (from $K_{d}$ = 170 + /- 40 nM to $K_{d}$ = 3.4 + /- 0.2 nM) (Figure 3C–D, Figure 3—figure supplement 2). This effect was specific to the PVIVIT sequence, with significantly less drastic increases and decreases in affinity observed for the same substitutions in the NFATc2 and AKAP79 scaffolds, respectively (Figure 3—figure supplement 6, Figure 3—figure supplement 8, Figure 3—figure supplement 23). These MRBLE-pep data establish that flanking residues make major contributions to affinity and the PxIxIT SLiM is significantly longer than previously thought.

Effects of individual substitutions show evidence of non-additivity

Understanding the degree to which CN-SLiM binding specificity can be explained by a linear additive model is critical for estimating the likely accuracy of downstream efforts to identify substrates or design therapeutics. Deviations from additivity can be quantified via double-mutant cycle (DMC) analysis, in which the effects of individual mutations are measured alone and in combination and then compared. DMC analysis between the PVIVIT and PKIVIT scaffolds revealed that the V4F substitution resulted in a significant loss of affinity within the PVIVIT context but had only minor effects on PKIVIT binding; by contrast, V4I mutation of this solvent-exposed position significantly increased binding only within the PKIVIT sequence (Figure 3E, Figure 3—figure supplement 23). This last substitution suggests that the high affinity of the PKIIIT sequence in the A238L inhibitor in particular relies on cooperativity between the position 2 and 4 residues. The magnitude of $Δ Δ G$ effects for the same substitution at the same position varies significantly between different backbones, further illustrating the importance of epistatic interactions in dictating overall binding strength. Although critical for improved prediction of in vivo substrates and outputs, these epistatic interactions are obscured in typical high-throughput screens and can only be observed in systematic pairwise measurements like those presented here.

Comparing experimental measurements with results from structure-based computational predictions

To test the degree to which computational methods can quantitatively predict effects of amino acid substitutions within SLiMs on protein-SLiM binding affinities, we leveraged high-resolution co-crystal structures and recently-developed computational modeling techniques (Smith and Kortemme, 2010; Barlow et al., 2018; Guerois et al., 2002; Schymkowitz et al., 2005) to estimate the effects of each mutation. Co-crystal structures of CN bound to PVIVIT (PDB: 2P6B) (Li et al., 2007) and AKAP79-derived PIAIIIT (PDB: 3LL8) (Li et al., 2012) reveal that the conserved P1, I3, and I5 residues are buried in hydrophobic pockets on the CN surface in all three structures; for AKAP79, an I1 residue occupies the pocket typically occupied by a P1 residue. In both structures, the less-conserved solvent-exposed (positions 2 and 4) and flanking (positions −1 and 7–9) residues adopt variable side chain orientations (Figure 4A).

Figure 4 with 7 supplements see all

Download asset Open asset

Structure-based modeling predictions.

(A) Co-crystal structures showing zoomed in overlay of substrates (PDB: 2P6B). (B) Heat maps showing predicted changes in binding energy from the Rosetta ‘flex_ddG’ protocol for each amino acid substitution at each position for PVIVIT and IAIIIT targets. (C) Heat maps showing FoldX-predicted changes in binding energy for each amino acid substitution at each position for PVIVIT and IAIIIT targets. (D) Fraction of residues correctly predicted to be stabilizing or destabilizing by both models as a function of position for PVIVIT (top) and IAIIIT (bottom).

https://doi.org/10.7554/eLife.40499.049

Figure 4—source data 1 Rosetta sequence tolerance protocol frequencies for PVIVIT.: https://doi.org/10.7554/eLife.40499.057
Download elife-40499-fig4-data1-v2.csv
Figure 4—source data 2 Rosetta sequence tolerance protocol frequencies for IAIIIT.: https://doi.org/10.7554/eLife.40499.058
Download elife-40499-fig4-data2-v2.csv
Figure 4—source data 3 Flex ddG-predicted $Δ Δ G$ values for PVIVIT.: https://doi.org/10.7554/eLife.40499.059
Download elife-40499-fig4-data3-v2.csv
Figure 4—source data 4 Flex ddG-predicted $Δ Δ G$ values for IAIIIT.: https://doi.org/10.7554/eLife.40499.060
Download elife-40499-fig4-data4-v2.csv
Figure 4—source data 5 FoldX ddG-predicted $Δ Δ G$ values for PVIVIT.: https://doi.org/10.7554/eLife.40499.061
Download elife-40499-fig4-data5-v2.csv
Figure 4—source data 6 FoldX ddG-predicted $Δ Δ G$ values for IAIIIT.: https://doi.org/10.7554/eLife.40499.062
Download elife-40499-fig4-data6-v2.csv

To predict mutational effects, we used computational methods implemented in the protein structure prediction and design programs Rosetta (Leaver-Fay et al., 2011) and FoldX (Guerois et al., 2002; Schymkowitz et al., 2005). The Rosetta Sequence Tolerance protocol (Smith and Kortemme, 2010) samples rotamers of different amino acid residues at each position and returns a qualitative estimate of effects of substitutions on binding (Figure 4—figure supplement 2, Figure 4—source data 1, Figure 4—source data 2). The Rosetta ‘flex_ddG’ protocol (Barlow et al., 2018) quantitatively estimates effects of mutations on binding free energies for systematic mutations at each position (Figure 4B, Figure 4—source data 3, Figure 4—source data 4), and has been shown to perform comparably to more computationally expensive alchemical free energy calculations for predicting mutational effects on protein-ligand affinities (Aldeghi et al., 2018). As expected, substitutions at invariant positions P1 and I3 were generally predicted to be destabilizing, while changes to the solvent-exposed, degenerate x2 and x4 positions were predicted to have minimal effects (Figure 4B). Intriguingly, mutations within upstream and downstream residues were frequently predicted to enhance binding. Mutations to more hydrophobic or aromatic residues (W,F,Y,I,V) in the −1 position for the PVIVIT motif (Figure 4B, left) or to most residues other than the native P-1 for the AKAP79 motif (Figure 4B, right) improved binding in silico, as did mutations at Position eight to aromatic (F,Y), aliphatic (I, V, L), or acidic (D, E) residues (Figure 4B). The FoldX protocol uses an empirical force field to allow rapid evaluation of the effects of mutations on the free energy of protein folding and complex formation (Guerois et al., 2002; Schymkowitz et al., 2005). These calculations revealed globally similar results (Figure 4C, Figure 4—figure supplement 2, Figure 4—figure supplement 3, Figure 4—source data 5, Figure 4—source data 6), with mutations to conserved hydrophobic residues again predicted to be destabilizing and mutations to other residues predicted to have little effect. However, Rosetta predicted less deleterious effects for substituting glycine and proline residues, perhaps due to better local sampling of backbone conformations.

To determine the degree to which Rosetta and FoldX successfully predict observed mutational effects on binding, we generated scatter plots showing relationships between measured affinities and predicted frequencies or changes in free energy of binding ( $Δ Δ G$ ), respectively, for each residue at each position (Figure 4—figure supplement 4; Figure 4—figure supplement 5, Figure 4—figure supplement 6). All methods correctly predicted that I3 and I5 mutations decrease affinity; however, modeling approaches predicted that variations at solvent-exposed positions (V2 and V4) would have little effect on affinity, in direct contrast to experimental observations (Figure 4—figure supplement 4, Figure 4—figure supplement 5, Figure 4—figure supplement 6). Calculations of the fraction of residues correctly predicted to be either stabilizing ( $Δ Δ G$ < 0) or destabilizing ( $Δ Δ G$ > 0) by each algorithm at each position within the PVIVIT and AKAP79 peptides confirm relatively strong performance for buried hydrophobic residues with variable performance at other positions (Figure 4D), with FoldX generally slightly outperforming Rosetta flex_ddG. These results highlight difficulties associated with properly modeling conformational changes in solvent-exposed and non-conserved positions, and show the utility of quantitative experimental measurements such as those presented here for improving binding models. Thermodynamic Integration and Free Energy Perturbation (TI/FEP) methods would likely enhance the ability to predict effects of mutating solvent-exposed residues, but are significantly more computationally expensive (Kilburg and Gallicchio, 2016; Gallicchio and Levy, 2011). Comparisons of MRBLE-pep measurements and Rosetta and FoldX predictions of the energetic effects of single site substitutions with orthogonal measurements are complicated by a lack of previously published affinity data. However, a direct comparison between MRBLE-pep measurements and computational predictions for two single-site substitutions in the PVIVIT motif reveal that only MRBLE-pep correctly discerns that a PVIVVT substitution has little effect while a PVIVIN substitution reduces binding (Figure 4—figure supplement 7).

Absolute binding affinities confirm that mutations to flanking residues dramatically change affinities

To determine absolute affinities for peptides spanning a range of binding behaviors, we performed two experiments measuring concentration-dependent CN binding to an additional ‘calibration’ library containing ∼6 peptides from each core and flank library (Figure 5—figure supplement 1, Figure 5—figure supplement 2, Figure 5—source data 1, Figure 5—source data 2). These experiments represented full technical replicates in which peptides were synthesized de novo on MRBLEs produced at different times and assayed for binding using different batches of purified calcineurin. Measured $Δ Δ G$ values relative to the PVIVIT variant between experiments showed strong agreement (Pearson r² = 0.72 (Figure 5—figure supplement 3), demonstrating the robustness of the assay. Averaged absolute affinities estimated for all variants using these $Δ Δ G$ values and the known literature $K_{d}$ for PVIVIT confirm that hydrophobic residues at position −1 and phosphorylated residues at position nine significantly increase affinity by 10-fold and 100-fold, respectively (Figure 5—figure supplement 4).

Identification of high-affinity scaffolds with therapeutic potential

Although cyclosporin A and FK506 are routinely prescribed to transplant patients to inhibit CN-dependent immune response activation, both drugs are associated with adverse effects that likely result from inhibition of CN-substrate dephosphorylation in non-immune tissues. While peptides themselves are rarely used as drugs, high-affinity peptides with increased specificity for a subset of CN-SLiM interactions could serve as initial scaffolds for downstream medicinal chemistry efforts to identify improved inhibitors. To probe for high-affinity variants, we measured CN binding to a final MRBLE-pep library containing the set of 36 ‘calibration’ peptides along with 11 peptides containing combinations of mutations discovered to increase affinity (Figure 5—figure supplement 5, Figure 5—figure supplement 6, Figure 5—figure supplement 7, Figure 5—source data 3, Figure 5—source data 4). PVIVIT variants combining a phosphothreonine (Z) at position nine with hydrophobic residues (I, V) at position −1 showed the strongest binding, with measured $K_{d}$ values of ∼10 nM (50x stronger than the measured PVIVIT affinity), and multiple PVIVIT, PKIVIT, and AKAP79 variants also showed enhanced binding (Figure 5A). Measured $Δ Δ G$ s for these full technical replicates showed remarkable agreement (r² = 0.83) (Figure 5B), establishing the robustness and reproducibility of the assay.

Figure 5 with 7 supplements see all

Download asset Open asset

Calibrated *in vitro* affinities for a range of peptides.

(A) *In vitro* measurements of binding affinities for high-affinity binders identified by combining favorable residues. (B) Correlation between measured $Δ Δ G$ values for a full technical replicate (new bead synthesis, new peptide synthesis, and new calcineurin protein purification) for the library shown in (A). Solid red line shows expected 1:1 linear relationship, dotted grey line shows linear regression of log-transformed values.

https://doi.org/10.7554/eLife.40499.063

Figure 5—source data 1 Concentration-dependent binding for calibration library, replicate 1.: https://doi.org/10.7554/eLife.40499.071
Download elife-40499-fig5-data1-v2.csv
Figure 5—source data 2 Concentration-dependent binding for calibration library, replicate 2.: https://doi.org/10.7554/eLife.40499.072
Download elife-40499-fig5-data2-v2.csv
Figure 5—source data 3 Concentration-dependent binding for full calibration library, replicate 1.: https://doi.org/10.7554/eLife.40499.073
Download elife-40499-fig5-data3-v2.csv
Figure 5—source data 4 Concentration-dependent binding for full calibration library, replicate 2.: https://doi.org/10.7554/eLife.40499.074
Download elife-40499-fig5-data4-v2.csv

Quantitative comparisons between in vitro affinities and in vivo inhibition via calcineurin activity assays

The ultimate goal of mapping CN-SLiM binding energy landscapes is to improve understanding of CN target substrate recognition and enable rational design of in vivo inhibitors. To test the degree to which in vitro MRBLE-pep affinities predict in vivo inhibition, we used a previously developed dual luciferase reporter assay to assess the activity of the NFAT2 transcription factor in HEK293T cells (Figure 6A) (Grigoriu et al., 2013). NFAT2 must be dephosphorylated by CN to accumulate in the nucleus, where it activates transcription of its target genes, and this interaction can be inhibited by peptides or small molecules that binds to CN. Reduction of NFAT2-dependent transcriptional activity upon expression of a competing PxIxIT peptide therefore reflects, quantitatively, the affinity of the inhibitor peptide for CN. Candidate PxIxIT inhibitors were co-expressed as GFP-fusion proteins within cells to enhance their stability and facilitate direct measurement of inhibitor concentrations.

Figure 6

Download asset Open asset

*In vivo* assays to quantify CN inhibition and an *in vitro* search for a potent inhibitor.

(A) *In vivo* experimental assay schematic. (B) Reporter gene inhibition in the presence of an empty eGFP vector (control), an empty eGFP vector with topical application of FK506, and vectors expressing PVIVIT, PKIVIT, NFATc2, and AKAP79 PxIxIT peptides with a C-terminal eGFP tag. (C) Reporter gene inhibition in the presence of various PVIVIT-, AKAP79-, and NFATc2-eGFP variants. (D) Reporter gene inhibition in the presence of various PKIVIT-eGFP variants. (E) *In vitro* calibrated binding affinities ( $K_{d}$ , nM) plotted against relative *in vivo* activation for data shown in panels B (orange), C (green), and D (blue).

https://doi.org/10.7554/eLife.40499.075

Figure 6—source data 1 Initial in vivo experiment.: https://doi.org/10.7554/eLife.40499.076
Download elife-40499-fig6-data1-v2.csv
Figure 6—source data 2 PKIVIT in vivo experiment.: https://doi.org/10.7554/eLife.40499.077
Download elife-40499-fig6-data2-v2.csv
Figure 6—source data 3 All other peptides in vivo experiment.: https://doi.org/10.7554/eLife.40499.078
Download elife-40499-fig6-data3-v2.csv

We compared levels of NFAT-driven luciferase activity in the presence of empty vector, vectors driving expression of the 4 PxIxIT peptide scaffolds probed extensively in vitro (PVIVIT, PKIVIT, NFATc2, and AKAP79), or FK506, a small-molecule inhibitor of CN with a previously measured IC50 of ∼0.5 nM ( Figure 6B, Figure 6—source data 1, Figure 6—source data 2, Figure 6—source data 3) (Fruman et al., 1992). All 4 PxIxIT peptides substantially inhibited NFAT-driven luciferase activity, confirming competitive inhibition of CN-NFAT-(PxIxIT) binding in vivo. Consistent with MRBLE-pep measurements, PKIVIT was the most effective inhibitor, followed closely by PVIVIT. NFATc2 and AKAP79 showed weaker in vivo inhibition, consistent with MRBLE-pep results and in contrast with prior work suggesting that AKAP79 binds with equal affinity to PVIVIT ( $K_{d}$ ∼500 nM) (Li et al., 2012). Co-expression of the PKIVIT peptide resulted in greater inhibition than FK506, which interferes with CN-substrate interactions by occluding the LxVP-SLiM binding site (Rodríguez et al., 2009) and is a potent CN inhibitor.

We next examined the importance of PxIxIT flanking residues by creating substitutions in the −1 and 9 positions (Figure 6C,D). Consistent with in silico predictions and MRBLE-pep affinities, substitution of a hydrophobic residue (IPKIVIT or LPRIEIT) at the −1 position significantly increased inhibition, whereas incorporating an acidic residue (EPKIVIT or DPRIEIT) greatly diminished inhibition. As predicted computationally and observed in the MRBLE-pep assays, introduction of the phosphomimetic mutation H9D resulted in slightly increased inhibition; it was not possible to directly test the extremely high affinity phosphoserine or phosphothreonine in vivo. Taken together, in silico modeling, the MRBLE-pep affinities, and in vivo results suggest that phosphorylation at position nine may be a key determinant of modulating PxIxIT-CN affinity.

Finally, we sought to test whether MRBLE-pep in vitro affinity measurements could identify initial peptide scaffolds exhibiting strong inhibition in vivo for future optimization as peptide or small-molecule inhibitors. PKIVIT was a significantly stronger in vivo inhibitor than the previously patented PVIVIT peptide (Nakamura et al., 2007), with activity comparable to the PKIIIT site within the known A238L inhibitor. As predicted, both the IPKIVITGPH and HPKIVITGPD variants exhibited even stronger binding and inhibition (Figure 6D); similarly, mutating the I1 and A2 residues in the AKAP79 PIAIIIT motif to create a novel PxIxIT sites with adjacent prolines and a positively charged residue in the second position (PPKIIIT) greatly enhanced inhibition. Overall, MRBLE-pep in vitro measurements were strongly predictive of in vivo inhibition (Figure 6E).

Discussion

Here, we demonstrate a novel strategy to quantitatively profile parts of the binding specificity landscape for a peptide-protein interaction, thereby generating key insights required to map and model essential signaling networks in healthy and diseased cells. Our approach is complementary to current techniques, with the potential to provide new insights into SLiM recognition and function throughout signaling networks. Yeast and phage display can powerfully discriminate between ‘bound’ and ‘unbound’ peptide populations (Bazan et al., 2012; Cherf and Cochran, 2015) for > 10⁸ protein-peptide interactions, and array-based methods can return information about qualitative binding strengths for many protein-peptide interactions in parallel. However, neither method can return the quantitative thermodynamic constants (dissociation constants (K_ds) and relative binding energies ( $Δ Δ G$ s) essential for predicting protein occupancies in vivo and effects on downstream signaling responses. For display- and array-based screens, measurements take place far from equilibrium, biasing recovered interactions towards the strongest binding motifs and losing information about those interactions with $K_{d}$ s close to physiological concentrations most likely to be dynamically regulated in vivo.

The MRBLE-pep assay is faster and requires less in material than alternative low-throughput quantitative methods: measuring interaction affinities for 384 peptides using MRBLE-pep requires ∼400x, ∼700x, and ∼2000x less purified protein than surface plasmon resonance, fluorescence polarization, and isothermal calorimetry, respectively, with savings increasing with library size (Supplementary file 2). This reduction in material should allow future profiling of a wide range of biologically important SLiM-binding domains, including those that are uncharacterized, unstable, and/or difficult to express recombinantly. MRBLE-pep libraries can be synthesized, assayed, and imaged in days, facilitating iterative rounds of synthesis and measurement for efficient affinity landscape mapping as well as testing and refining computational prediction algorithms. After screening large sequence spaces for candidate interaction motifs, MRBLE-pep can be used to quickly and efficiently scan through thousands of potential SLiMs to validate true hits, measure affinities for these hits in high-throughput, and systematically perturb amino acids within motifs to derive a position-specific affinity matrix that can be used to predict additional candidate downstream targets in vivo. Moreover, the ability to tune the density of applied peptides enables fast optimization to probe the affinity range of interest as well as test for surface density-induced binding artifacts. An elegant ‘hold-up’ assay that couples high-throughput affinity purification to microfluidic gel electrophoresis recently demonstrated a similar ability to measure relative affinities for 400 domain-peptide pairs (Vincentelli et al., 2015). However, this method requires considerably more infrastructure than the MRBLE-pep assay and has not yet been used in subsequent work. In future work, the MRBLE-pep platform could be combined with a simple and portable imaging device to allow labs to employ the assay with only standard laboratory equipment (e.g. pipettes and a benchtop centrifuge).

MRBLE-pep measurements can also be combined with data from high-throughput in vivo methods to decipher the structure and wiring of cellular signaling networks. Yeast two-hybrid (Y2H) assays and affinity purification coupled to mass spectrometry (AP-MS) techniques have revealed tens of thousands of new interactions within the human interactome (Rolland et al., 2014; Huttlin et al., 2015; Huttlin et al., 2017). However, SLiM-mediated binding interactions remain statistically underrepresented in these datasets (Neduva and Russell, 2006; Van Roey et al., 2014; Davey et al., 2012), likely due to fast dissociation rates, and AP-MS results cannot distinguish between binary interactions and higher-order complexes, and provide no direct information about the motif involved. MRBLE-pep measurements of peptides tiled across candidate interacting proteins could confirm candidate interactors, identify the SLiMs responsible, and determine whether a candidate interaction is direct or indirect. In particular, the ability to return information about residues that ablate binding can significantly improve the ability to predict true targets in vivo (Krystkowiak et al., 2018). To test candidate inhibitor specificities, MRBLE-pep could profile binding of a broad set of cellular proteins against high-affinitiy inhibitors with unnatural amino acids returned from recently developed high-throughput mass spectrometry-based (LC/MS/MS) approaches (Gates et al., 2018; Vinogradov et al., 2017).

Beyond the method itself, our findings reveal biologically significant insights into CN-PxIxIT specificity and illustrate how systematic analyses can shed light on elusive SLiM-protein interactions. The MRBLE-pep data and in vivo validation presented here are the first to reveal the impact of PxIxIT flanking residues on affinity, allowing identification of several peptides whose affinity for CN is two orders of magnitude higher than that of the highest affinity known substrate (PVIVIT) and potently inhibit CN-NFAT signaling in vivo. We reveal the presence of significant epistasis between residues, including one interaction important for the high affinity of a known CN inhibitor (A238L) (Figure 3E). Finally, we provide affinity measurements for a collection of peptides based on different known PxIxIT scaffolds with a wide variety of affinities, yielding a tool kit of peptide sequences that can be used to probe CN recognition and function in vivo.

In addition to identifying candidate peptides for manipulation of CN signaling, this first demonstration that residues both upstream and downstream contribute to PxIxIT-CN affinity redefines the motif itself. Current computational strategies attempt to identify novel CN substrates by searching for sequences within the proteome that match a consensus for the six core residues alone. The expanded 10-residue definition of the PxIxIT motif derived here will significantly enhance future computational efforts to comprehensively map the CN signaling network, reducing the number of both false positive and false negative substrates. In addition, MRBLE-pep provides critical Information about residues that reduce affinity (e.g. acidic residues at the −1 position reduces affinity) that is never revealed by positive screening methods and is rarely collected due to the resource-intensive nature of such analyses using established techniques. The ability to predict affinity based on SLiM sequence is an essential step toward accurate modeling of signaling dynamics in vivo, as the extent of in vivo dephosphorylation has previously been shown to depend on primary PxIxIT amino acid sequence (and the associated interaction affinity) (Roy et al., 2007). Even subtle (∼2-fold) differences in binding could have profound effects on downstream signaling under physiologic conditions given the low affinities of CN-PxIxIT interactions.

Establishing the impact of post translational modifications on SLiM-protein affinity is critical for understanding cross talk between signaling networks in vivo. Consistent with our findings that acidic residues at the −1 position decrease affinity, JNK kinase regulates CN signaling by phosphorylating a serine that immediately precedes the PxIxIT of NFAT4 (Chow et al., 2000). Similarly, our demonstration that phosphorylated residues downstream of the core PxIxIT sequence at position nine enhance PVIVIT affinity echoes observations that phosphorylation of a threonine in this position is required for binding of CN to a PxIxIT site in C16ORF74, and for its ability to promote invasiveness of pancreatic ductal adenocarcinoma (PDAC) (Nakamura et al., 2017). This positive effect on binding is context-dependent, increasing PVIVIT binding 50-fold but having little or no effect on other PxIxIT sequences, reinforcing the importance of systematic analyses for generating predictive information.

Beyond improving the ability to reconstruct downstream CN signaling networks in vivo, these results also have direct relevance to precision medicine. Current annotations of 2RQ, 7 PA, and 7PL missense mutations within the NFATc2 PxIxIT site recovered from sequenced lung and breast adenocarcinomas describe them as unlikely to have functional effects in vivo (cBioPortal, 2018). While the experiments presented here did not directly test the functional consequences of a R2Q mutation, all mutations away from the native residue at this position resulted in a dramatic loss of binding (Figure 3—figure supplement 17), and the P7A substitution led to a nearly complete loss of binding (Figure 3—figure supplement 6, Figure 3—figure supplement 17, Figure 3—figure supplement 18). Thus, binding specificity maps like those obtained here could both help clinicians identify functionally significant missense mutations, and refine the computational algorithms that predict the impact of such alleles. Overall, the approaches outlined here will help decipher the language of SLiM-domain binding events, and of the cellular signaling networks they define in both healthy and diseased cells.

Materials and methods

Reagents for peptide synthesis were purchased and used without further purification from NovaBiochem, AnaSpec (Fremont, CA), and Sigma-Aldrich (St. Louis, MO). All other solvents and chemical reagents were purchased from Sigma-Aldrich.

MRBLE synthesis and collection

Request a detailed protocol

MRBLEs were synthesized using a previously published microfluidic device (Gerver et al., 2012; Nguyen et al., 2017a; Nguyen et al., 2017b) with each code collected into wells of a 96-well plate using an open-source in-house fraction collector. Briefly, all lanthanide input mixtures contained double-distilled water (ddH₂O), 42.8% v/v 700 MW PEG-diacrylate (PEG-DA) (Sigma-Aldrich), 19.6 mg mL⁻¹ lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP), and 5.0% v/v YVO4:Eu (at 25 mg mL⁻¹). Individual input mixtures contained 16.3% v/v of either YVO4:Sm (25 mg mL⁻¹), YVO4:Dy (25 mg mL⁻¹), or YVO4:Tm (12.5 mg mL⁻¹). Lanthanide solutions were mixed using a herringbone mixer channel and forced into droplets using a T-junction with a continuous stream of HFE7500 (3M Novec) containing 2% w/w modified ionic Krytox 157FSH (Miller Stephenson, Danbury, CT) (DeJournette et al., 2013). Droplets were then photopolymerized with UV light from a full-spectrum 200W Xenon arc lamp (Dymax, Torrington, CT, USA). With this setup, ∼3000 beads containing each spectral code were synthesized in 70 s (as described previously Gerver et al., 2012; Nguyen et al., 2017a); for the 48-plex MRBLE library used here, each code was produced 10 times to yield ∼30,000 beads per code.

Bead functionalization

View detailed protocol

MRBLE hydrogel beads were functionalized with terminal amine handles via Michael addition by reacting available acrylates in the MRBLEs with a solution of cysteamine (50 eq) containing pyridine (50 eq) in H2O:DMF (1:3) for 18 hr at ambient temperature. Next, to selectively functionalize outer and inner regions, MRBLEs were swelled in water overnight and drained using a manifold before the addition of a solution containing Fmoc-N-hydroxysuccinimide (0.2 eq) and diisopropylethylamine (DIPEA) (0.8 eq) in dichloromethane:diethyl ether (55:45) with vigorous shaking (1600 rpm for 15 s followed by 30 s at rest) for 30 min (Liu et al., 2002). To selectively functionalize MRBLE inner core regions with an acid-labile rink amide linker, MRBLEs were then treated with 4-[(2,4-Dimethoxyphenyl)(Fmoc-amino)methyl]phenoxyacetic acid (rink amide) (5.0 eq), N,N′-Diisopropylcarbodiimide (DIC) (5.0 eq), and DIPEA (10 eq) in dimethylformamide (DMF) for 1 hr and repeated twice. After removal of the Fmoc protecting group using 20% 4-methylpiperidine (4-MP) in DMF for 20 min, the effective on-bead peptide concentration was reduced by reacting the bead with a mixture of Fmoc-glycine-OH:Ac-N-glycine-OH (1:9) (5.0 eq), DIC(5.0 eq), and DIPEA (10 eq) for 14 hr. Following Fmoc deprotection, MRBLEs were transferred to an automated peptide synthesizer for solid phase peptide synthesis (SPPS).

On-bead peptide synthesis

View detailed protocol

Peptide synthesis was performed using a Biotage Syro II automated peptide synthesizer following instructions from the manufacturer (Biotage, Charlotte, NC). During coupling steps, Fmoc-protected amino acids (10 eq) were activated with HCTU (9.8 eq) and NMM (20 eq) with coupling times of 8 min for standard amino acids and 25 min for phosphorylated amino acids. Each coupling round comprised of 2 sequential coupling reactions with the addition of fresh amino acid and coupling reagents. Deprotection was performed initially with 40% 4-MP for 2 min and then followed with 20% 4-MP for an additional 6 min. MRBLEs were then washed thoroughly with 6 rounds of DMF (0.4 mL) before the next coupling step. Before global deprotection with TFA, 50 µL from a volume of 400 µL for each code was saved for biotin conjugation (as described below). To perform the global deprotection step, MRBLEs were washed with DCM and dried under vacuum before the addition of 0.5 mL of TFA cocktail (Reagent B, TFA:phenol:ddH2O:triisopropylsilane, 88:5:5:2, v/m/v/v) was added to each reaction tube and reacted with shaking (15 s shaking and 1 min rest) for 1.5 hr. After TFA deprotection, MRBLEs containing deprotected peptides were washed with TFA (~0.5 mL, and collected for MALDI analysis), DCM (~2 mL), neutralized with 10% DIPEA in DMF twice (~1 mL), and then finally washed with storage buffer (1 mL, 0.1% TBST with 0.02% NaN3) 3 times (Kumaresan et al., 2011). To assess peptide quality, the TFA solution containing cleaved/unprotected peptide was transferred to 15 mL falcon tubes using the provided liquid transfer system from Biotage. Peptides were triturated with cold diethylether (~1 mL), pelleted, decanted, and repeated these steps 3 times before preparing for MALDI analysis using a general protocol described below.

Peptide concentration estimation via biotin conjugation assay

View detailed protocol

A small aliquot (~50 μL from 400 μL) of each code was transferred and pooled into a separate fritted reaction tube (Biotage 2 mL reaction tubes for SPPS) for biotin (50 eq) conjugation using DIC (50 eq), DIPEA (100 eq), and DMF overnight. After washing the MRBLEs (DMF, MeOH, and DCM), another round of coupling using the same conditions mentioned above was mixed for 2 hr and then washed. MRBLEs with terminal biotin were then globally deprotected using a cocktail of TFA/phenol/H2O/TIPS (87.5:5:5:2.5 v/v) at ambient temperature for 1.5 hr. After global deprotection, MRBLEs were neutralized with 10% DIPEA in DMF, washed with DCM, washed with 3 X PBS containing 0.1% TWEEN 20 (0.1% PBST), and stored in 0.1% PBST containing 0.02% NaN3. After biotin conjugation, a 40 μL aliquot from a 600 μL suspension of beads was passivated with 0.1% PBST containing 5% BSA in a 150 μL PCR strip tube on a rotator overnight at ~5 °C. The beads were then washed with 0.1% PBST containing 2% BSA 3 times (~100 μL), followed by incubation with 1 μL of labeled streptavidin (final concentration = ~189 nM, Abcam, ab134241) in 0.1% PBST containing 2% BSA (99 μL) for 30 min on a rotator at ambient temperature. After incubation, MRBLEs were pelleted and washed with 0.1% PBST once and then imaged to obtain binding data.

MALDI quality control

View detailed protocol

After global deprotection, supernatants collected in 15 mL Falcon tubes for each code were placed into a freezer for 1 hr; these tubes were then centrifuged (4000 g for 20 min at 4 °C) and decanted (repeated 3 times). Peptides were then dissolved with 60% ACN/H2O (20 μL, phosphopeptides were dissolved with 50% acetic acid) for MALDI analysis using THAP (250 mM in ACN) as the matrix. To prepare the MALDI plate (microScout Target MSP 96 target polished steel BC, part #8280800), 0.5 μL of sodium citrate (250 mM in H2O containing 0.1% TFA) was spotted on to the plate surface and allowed to dry. After drying, 1 μL of a 1:1 mixture of the peptide solution with a solution of sodium citrate:THAP (1:1) was spotted onto the plate and allowed to dry again before analysis. Data was obtained using a Bruker microflex MALDI-TOF (Billerica, MA, USA). The instrument was run on positive-ion reflector mode with a laser setting of 1,810 V and data averaged over 100 scans. Raw data was analyzed using FlexAnalysis and mMass (ver. 5.5, http://www.mmass.org/).

Purification of calcineurin

Request a detailed protocol

N-terminally, 6-His-tagged human calcineurin A isoforms (truncated at residue 400), either wild-type or containing the mutation 330NIR333-AAA, were expressed in tandem with the calcineurin B subunit in E. coli BL21 (DE3) cells (Invitrogen) and cultured in LB medium containing carbenicillin (50 mg/ml) at 37°C to mid-log phase. Expression was induced with 1 mM isopropyl 1-thio-b-galactosidase at 16°C for 18 hr. Cells were pelleted, washed and frozen at −80°C for at least 12 hr. Thawed cell pellets were re-suspended in lysis buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.1% Tween 20, 1 mM $β$ -mercapto ethanol, protease inhibitors) and lysed by sonication using four 1 min pulses at 40% output. Extracts were clarified using two rounds of centrifugation (20,000 X g, 20 min) and then bound to 1 ml of Ni-NTA beads in lysis buffer containing 5 mM imidazole. Beads were washed with lysis buffer containing 20 mM imidazole and eluted with lysis buffer containing 300 mM imidazole. Purified calcineurin heterodimer were dialyzed in buffer (50 mM Tris-HCl pH 7.5, 150 mM NaCl, 1 mM $β$ -mercapto ethanol) and stored in 10–15% glycerol at −80°C.

Bead and antibody imaging

View detailed protocol

MRBLEs were imaged by transferring 20 μL of suspended beads (entire volume) onto a quartz microscope slide (Electron Microscopy Sciences, quartz microscope slide, 75 mm x 25 mm, 1 mm thick, cat. #72250–03), placing an additional quartz coverslip (Electron Microscopy Sciences, quartz coverslip, 25 mm x 25 mm, 1 mm thick, cat. #72256–02) onto the droplet, and then depositing mineral oil around the edges of the coverslip to prevent the sample from drying out during imaging. MRBLEs were imaged largely as described previously (Nguyen et al., 2017a) in 11 channels: a bright field channel, a Cy5 fluorescence channel (using a SOLA light engine for excitation), and nine additional lanthanide channels (435, 474, 536, 546, 572, 620, 630, 650, 780 nm, with exposure times of 500, 1000, 500, 500, 375, 150, 75, 225, 2000 ms, respectively; Semrock, Rochester, NY) using excitation illumination generated by a Xenon arc lamp (Lambda SL, Sutter Instrument, Novato, CA) and directed through a 292/27 nm bandpass filter (Semrock, Rochester, NY) via a UV-liquid light guide (Sutter Instrument, Novato, CA) mounted in place of the condenser using a custom 3D printed holder.

Image processing (code calling)

Request a detailed protocol

Bead images were processed using a custom-built open-source Python software package freely available through PyPI and GitHub and regularly maintained (Harink et al., 2019; Harink, 2018). Briefly, MRBLE boundaries were identified from bright field images, followed by quantification of median lanthanide intensities in each channel to identify embedded MRBLE codes. CN binding was then quantified for each MRBLE by calculating the median fluorescence intensity for bound protein associated with the outer shell of the MRBLE.

Calcineurin binding assays (time series and dilution series)

Request a detailed protocol

Pre-incubation of CN with anti-His antibody significantly reduced observed background binding to sequences containing multiple basic residues due to cross-reactivity of anti-His antibody (Figure 3—figure supplement 9, Figure 3—figure supplement 10, Figure 3—figure supplement 11, Figure 3—figure supplement 12). Therefore, CN:α−6xHis antibody-DyLight-650 complex (2.5 μM, CN:αHisAb650, Abcam ab117504) was prepared by pre-incubating 6x-His-tagged CN with equal concentration of αHisAb650 in CN buffer at ∼5 °C for 1 hr. MRBLEs were prepared by transferring a 20 μL aliquot from a 600 μL suspension (∼3000–8000 beads depending on pellet size) to a 100 μL PCR strip (20 μL for each CN concentration), exchanging buffer via 3 cycles of iterative pelleting, decanting, and resuspension with 0.1% PBST containing 5% BSA, pH = 7.5 (100 μL), and then left mixing at ∼5 °C on a rotator in the same wash buffer overnight (∼14 hr). MRBLEs were then buffer exchanged once again with CN binding buffer (50 mM Tris pH = 7.5, 150 mM NaCl, 0.1% TWEEN 20) via three iterative cycles of pelleting, decanting, and resuspension.

To perform binding assays, the CN:αHisAb650 complex (at 2.5 μM) was serially diluted into tubes containing BSA-passivated MRBLEs and CN buffer (2 μM, 1μM, 500 nM, 250 nM, 125 nM, 62.5 nM) and incubated for 5 hr on a rotator at ∼5 °C. After incubation, MRBLEs were pelleted, decanted, and washed once with 0.1% PBST (100 μL). After a final round of pelleting and decanting, 20 μL of 0.1% PBST was added and the beads were transferred to a quartz microscope slide for imaging. To confirm equilibrium conditions, we measured CN:MRBLE interactions for six peptides after incubation times ranging from 30 min to 24 hr. Binding appeared to reach equilibrium after ∼5 hr (Figure 2—figure supplement 1), so all following experiments were performed after an incubation time of 5–6 hr.

Affinity modeling (competitive binding assay)

Request a detailed protocol

The affinity ( $K_{d}$ ) for each peptide in a MBRLE library was determined via a two-step nonlinear regression using a single-site binding model:

I_{i} = (I_{m a x} [C N_{t o t a l}]) / (K_{d, i} + [C N_{t o t a l}])

where $I_{i}$ is the median measured fluorescence intensity for each peptide, $K_{d, i}$ is the dissociation constant for that interaction, [ $C N_{t o t a l}$ ] is total CN concentration, and $I_{m a x}$ is a global variable representing the fluorescence intensity once MRBLE-pep beads are saturated. First, we determined the global saturation value ( $I_{m a x}$ ), by globally fitting the top 80% highest-intensity MRBLE codes. Next, we globally fit the entire data set using this $I_{m a x}$ value to yield an estimated absolute affinity ( $K_{d}$ ) for all peptides in the library. Given that the estimated peptide concentration in these assays (20 nM/x, where x represents the number of species probed) is significantly lower than estimated $K_{d}$ values of these interactions, we make the approximation $[C N_{t o t a l}] \approx [C N_{f r e e}]$ . Differences in binding affinity were calculated relative to a reference peptide using the standard equation:

Δ Δ G = R T l n (K_{d, i} / K_{d, r e f})

To calibrate measured affinities across multiple assays, we first determined an absolute affinity for a single high-affinity reference peptide (MAGPHPVIVITGPHEE) using the mean of the ‘Triplicate low’ value (Figure 2C), ‘Triplicate high’ value (Figure 2—figure supplement 2), and value from a binding assay in which this peptide appeared alone (data not shown, $K_{d}$ = 980 nM). Next, we used this reference value and calculated differences in binding affinity ( $Δ Δ G$ ) to estimate absolute $K_{d}$ values for each peptide within the calibration and full calibration libraries (Figure 2—source data 1, Figure 2—source data 2, Figure 5—source data 1, Figure 5—source data 2, Figure 5—source data 3, Figure 5—source data 4).

Rosetta-based ‘Sequence Tolerance' method

Request a detailed protocol

For estimating tolerated amino acid substitutions at each position in the two available CN-PxIxIT co-crystal structures with well-defined electron density (positions −1 to 9 of the PxIxIT motif) we used the generalized sequence tolerance module of the Rosetta Backrub server available at https://kortemmeweb.ucsf.edu/backrub (Smith and Kortemme, 2011). Briefly, the protocol samples amino acid residues in different rotameric conformations on backbone ensembles generated using Rosetta Backrub simulations and records low energy amino acid sequences, from which amino acid frequencies of tolerated substitutions are derived. In contrast to the published protocol, we modified only one position in the peptide at a time, to more closely mimick the experimental measurements. We used ensembles of 100 backbone structures with a kT value of 0.228; self-energies and interchain interaction energies were reweighted using the default scaling factors of 0.4 and 1, respectively.

Rosetta-based method for estimating the energetic effects of amino acid substitutions (‘flex_ ddG’)

Request a detailed protocol

To estimate changes in binding energy upon mutation ( $Δ Δ G$ ), we used the Rosetta flex_ddG protocol (Barlow et al., 2018). We systematically substituted all twenty natural amino acids at each position (−1:9) for the two peptides with available crystal structures bound to calcineurin (CN-PVIVIT, CN-IAIIIT), with the remaining positions restricted to the wild-type amino acid residues in each crystal structure. Briefly, the flex_ddG protocol uses the Rosetta Talaris 2014 energy function, minimization with harmonic restraints, Rosetta Backrub simulations to generate conformational ensembles, mutation and optimization of side chain conformations, and another final retrained minimization step. We followed the protocol published in Barlow et al. (2018), except using 10,000 Backrub steps instead of the default value of 35,000 steps.

FoldX calculations to estimate the energetic effects of amino acid substitutions

Request a detailed protocol

To prepare systems for analysis with FoldX, we obtained the PDB entries for calcineurin bound to a PVIVIT peptide (PDB entry 2P6B) and bound to a IAIIIT peptide (PDB entry 3LL8). We then deleted chains C and D, as well as accessory ions and waters, from both structures. In the 2P6B structure, we also removed residues Pro4, Glu16 and the -NH2 cap to match the resolved portions of the peptide present in the 3LL8 structure. We used two FoldX commands to perform the FoldX calculations. First, we ran the RepairPDB command on each system with default options to generate structures in a suitable format for the subsequent step. Second, for each mutation, we ran the BuildModel command for 10 iterations. We then extracted delta-delta Gs from the automatically generated output file that reported differences in the delta Gs between wild type and mutant. Data values in Figure 4 correspond to median value of the delta-delta Gs for each of the ten iterations for a given mutant.

In vivo calcineurin activity assay

Request a detailed protocol

PxIxIT peptides were fused to eGFP in pEGFPc1 vector. HEK293T cells were transfected with pEGFPc1 clones, pNFAT-Luc and CMV-Renilla in a 6-well plate format. 18 hr post transfection, 1μM FK506 or DMSO (vehicle) was added to the media as needed. 36 hr post transfection, cells were treated with 1 mM Ionomycin and 1 mM phorbol 12,13 di-butyrate to activate calcineurin and AP-1 (via PKC) respectively. 6 hr after pathway activation, cells were collected, washed in PBS and re-suspended in DMEM media. 80% of the cells were used to measure luciferase activity and renilla using the Dual-Glo assay system (Promega) with three technical replicates. The remaining cells were frozen and stored at −80°C. Cell lysates were prepared in RIPA buffer. 15–20 μg of lysate was analyzed by Western for expression of GFP. GFP signal was normalized to either actin or tubulin. Luciferase activity (normalized to renilla expression) was further normalized to eGFP expression. Data represent at least three experimental replicates.

Data availability

All data generated or analysed during this study are included in the manuscript and supporting files. In addition, all data generated or analyzed during this study are available in an associated public OSF repository (https://doi.org/10.17605/OSF.IO/FPVE2).

The following data sets were generated

(2018) Open Science Framework
Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads.
https://doi.org/10.17605/OSF.IO/FPVE2

References

(2018) Accurate estimation of ligand binding affinity changes upon protein mutation
ACS Central Science 4:1708–1718.
https://doi.org/10.1021/acscentsci.8b00717
- PubMed
- Google Scholar
(1998) Selective inhibition of NFAT activation by a peptide spanning the calcineurin targeting site of NFAT
Molecular Cell 1:627–637.
https://doi.org/10.1016/S1097-2765(00)80063-5
- PubMed
- Google Scholar
(1999) Affinity-driven peptide selection of an NFAT inhibitor more selective than cyclosporin A
Science 285:2129–2133.
https://doi.org/10.1126/science.285.5436.2129
- PubMed
- Google Scholar
(2018) Combinatorial synthesis of macromolecular arrays by microchannel cantilever spotting (µCS)
Advanced Materials 30:1801632.
https://doi.org/10.1002/adma.201801632
- Google Scholar
Book
1. Bagshaw CR
(2017)
Biomolecular Kinetics: A Step-by-Step Guide

CRC Press.
- Google Scholar
(2018) Flex ddG: rosetta Ensemble-Based estimation of changes in Protein-Protein binding affinity upon mutation
The Journal of Physical Chemistry B 122:5389–5399.
https://doi.org/10.1021/acs.jpcb.7b11367
- PubMed
- Google Scholar
(2012) Phage display—A powerful technique for immunotherapy
Human Vaccines & Immunotherapeutics 8:1817–1828.
https://doi.org/10.4161/hv.21703
- Google Scholar
1. Blikstad C
2. Ivarsson Y
(2015) High-throughput methods for identification of protein-protein interactions involving short linear motifs
Cell Communication and Signaling 13:38.
https://doi.org/10.1186/s12964-015-0116-8
- PubMed
- Google Scholar
(2012) High-resolution mapping of linear antibody epitopes using ultrahigh-density peptide microarrays
Molecular & Cellular Proteomics 11:1790–1800.
https://doi.org/10.1074/mcp.M112.020800
- PubMed
- Google Scholar
1. Carmona SJ
2. Nielsen M
3. Schafer-Nielsen C
4. Mucci J
5. Altcheh J
6. Balouz V
7. Tekiel V
8. Frasch AC
9. Campetella O
10. Buscaglia CA
11. Agüero F
(2015) Towards High-throughput immunomics for infectious diseases: use of Next-generation peptide microarrays for rapid discovery and mapping of antigenic determinants
Molecular & Cellular Proteomics 14:1871–1884.
https://doi.org/10.1074/mcp.M114.045906
- PubMed
- Google Scholar
1. cBioPortal
(2018) cBioPortal for Cancer Genomics Online
cBioPortal for Cancer Genomics Online, http://www.cbioportal.org/.

http://www.cbioportal.org/
- Google Scholar
1. Chen X
2. Tan PH
3. Zhang Y
4. Pei D
(2009) On-bead screening of combinatorial libraries: reduction of nonspecific binding by decreasing surface ligand density
Journal of Combinatorial Chemistry 11:604–611.
https://doi.org/10.1021/cc9000168
- PubMed
- Google Scholar
1. Cherf GM
2. Cochran JR
(2015) Applications of yeast surface display for protein engineering
Methods in Molecular Biology 1319:155–175.
https://doi.org/10.1007/978-1-4939-2748-7_8
- PubMed
- Google Scholar
1. Chow CW
2. Dong C
3. Flavell RA
4. Davis RJ
(2000) c-Jun NH(2)-terminal kinase inhibits targeting of the protein phosphatase calcineurin to NFATc1
Molecular and Cellular Biology 20:5227–5234.
https://doi.org/10.1128/MCB.20.14.5227-5234.2000
- PubMed
- Google Scholar
1. Davey NE
2. Van Roey K
3. Weatheritt RJ
4. Toedt G
5. Uyar B
6. Altenberg B
7. Budd A
8. Diella F
9. Dinkel H
10. Gibson TJ
(2012) Attributes of short linear motifs
Mol. BioSyst. 8:268–281.
https://doi.org/10.1039/C1MB05231D
- PubMed
- Google Scholar
1. Davey NE
2. Seo MH
3. Yadav VK
4. Jeon J
5. Nim S
6. Krystkowiak I
7. Blikstad C
8. Dong D
9. Markova N
10. Kim PM
11. Ivarsson Y
(2017) Discovery of short linear motif-mediated interactions through phage display of intrinsically disordered regions of the human proteome
The FEBS Journal 284:485–498.
https://doi.org/10.1111/febs.13995
- PubMed
- Google Scholar
1. DeJournette CJ
2. Kim J
3. Medlen H
4. Li X
5. Vincent LJ
6. Easley CJ
(2013) Creating biocompatible oil-water interfaces without synthesis: direct interactions between primary amines and carboxylated perfluorocarbon surfactants
Analytical Chemistry 85:10556–10564.
https://doi.org/10.1021/ac4026048
- PubMed
- Google Scholar
1. Dinkel H
2. Van Roey K
3. Michael S
4. Kumar M
5. Uyar B
6. Altenberg B
7. Milchevskaya V
8. Schneider M
9. Kühn H
10. Behrendt A
11. Dahl SL
12. Damerell V
13. Diebel S
14. Kalman S
15. Klein S
16. Knudsen AC
17. Mäder C
18. Merrill S
19. Staudt A
20. Thiel V
21. Welti L
22. Davey NE
23. Diella F
24. Gibson TJ
(2016) ELM 2016--data update and new functionality of the eukaryotic linear motif resource
Nucleic Acids Research 44:D294–D300.
https://doi.org/10.1093/nar/gkv1291
- PubMed
- Google Scholar
(2015) Binding rate constants reveal distinct features of disordered protein domains
Biochemistry 54:4741–4750.
https://doi.org/10.1021/acs.biochem.5b00520
- PubMed
- Google Scholar
1. Engelmann BW
2. Kim Y
3. Wang M
4. Peters B
5. Rock RS
6. Nash PD
(2014) The development and application of a quantitative peptide microarray based approach to protein interaction domain specificity space
Molecular & Cellular Proteomics 13:3647–3662.
https://doi.org/10.1074/mcp.O114.038695
- PubMed
- Google Scholar
1. Filippakopoulos P
2. Picaud S
3. Mangos M
4. Keates T
5. Lambert JP
6. Barsyte-Lovejoy D
7. Felletar I
8. Volkmer R
9. Müller S
10. Pawson T
11. Gingras AC
12. Arrowsmith CH
13. Knapp S
(2012) Histone recognition and large-scale structural analysis of the human bromodomain family
Cell 149:214–231.
https://doi.org/10.1016/j.cell.2012.02.013
- PubMed
- Google Scholar
1. Fodor SP
2. Read JL
3. Pirrung MC
4. Stryer L
5. Lu AT
6. Solas D
(1991) Light-directed, spatially addressable parallel chemical synthesis
Science 251:767–773.
https://doi.org/10.1126/science.1990438
- PubMed
- Google Scholar
1. Forsström B
2. Axnäs BB
3. Stengele KP
4. Bühler J
5. Albert TJ
6. Richmond TA
7. Hu FJ
8. Nilsson P
9. Hudson EP
10. Rockberg J
11. Uhlen M
(2014) Proteome-wide epitope mapping of antibodies using ultra-dense peptide arrays
Molecular & Cellular Proteomics 13:1585–1597.
https://doi.org/10.1074/mcp.M113.033308
- PubMed
- Google Scholar
(1990)
Facile and rapid “spot-synthesis” of large numbers of peptides onmembrane sheets

Peptides 2:151–152.
- Google Scholar
(1992) Calcineurin phosphatase activity in T lymphocytes is inhibited by FK 506 and cyclosporin A
PNAS 89:3686–3690.
https://doi.org/10.1073/pnas.89.9.3686
- PubMed
- Google Scholar
1. Gallicchio E
2. Levy RM
(2011)
Advances in Protein Chemistry and Structural Biology

27–80, Recent theoretical and computational advances for modeling protein–ligand binding affinities, Advances in Protein Chemistry and Structural Biology, 85, Elsevier, 10.1016/B978-0-12-386485-7.00002-8.
- Google Scholar
1. Gates ZP
2. Vinogradov AA
3. Quartararo AJ
4. Bandyopadhyay A
5. Choo ZN
6. Evans ED
7. Halloran KH
8. Mijalis AJ
9. Mong SK
10. Simon MD
11. Standley EA
12. Styduhar ED
13. Tasker SZ
14. Touti F
15. Weber JM
16. Wilson JL
17. Jamison TF
18. Pentelute BL
(2018) Xenoprotein engineering via synthetic libraries
PNAS 115:E5298–E5306.
https://doi.org/10.1073/pnas.1722633115
- PubMed
- Google Scholar
(2012) Programmable microfluidic synthesis of spectrally encoded microspheres
Lab Chip 12:4716–4723.
https://doi.org/10.1039/C2LC40699C
- PubMed
- Google Scholar
1. Gfeller D
2. Butty F
3. Wierzbicka M
4. Verschueren E
5. Vanhee P
6. Huang H
7. Ernst A
8. Dar N
9. Stagljar I
10. Serrano L
11. Sidhu SS
12. Bader GD
13. Kim PM
(2011) The multiple-specificity landscape of modular peptide recognition domains
Molecular Systems Biology 7:484.
https://doi.org/10.1038/msb.2011.18
- PubMed
- Google Scholar
1. Gianni S
2. Jemth P
(2017) How fast is Protein-Ligand association?
Trends in Biochemical Sciences 42:847–849.
https://doi.org/10.1016/j.tibs.2017.08.007
- PubMed
- Google Scholar
(2015) Experimental detection of short regulatory motifs in eukaryotic proteins: tips for good practice as well as for bad
Cell Communication and Signaling 13:42.
https://doi.org/10.1186/s12964-015-0121-y
- PubMed
- Google Scholar
1. Grigoriu S
2. Bond R
3. Cossio P
4. Chen JA
5. Ly N
6. Hummer G
7. Page R
8. Cyert MS
9. Peti W
(2013) The molecular mechanism of substrate engagement and immunosuppressant inhibition of calcineurin
PLOS Biology 11:e1001492.
https://doi.org/10.1371/journal.pbio.1001492
- PubMed
- Google Scholar
(2002) Predicting changes in the stability of proteins and protein complexes: a study of more than 1000 mutations
Journal of Molecular Biology 320:369–387.
https://doi.org/10.1016/S0022-2836(02)00442-4
- PubMed
- Google Scholar
Software
1. Harink B
(2018) Decoding and analysis software for mrbles (microspheres with ratiometric barcode lanthanide encoding)
Github.

https://github.com/FordyceLab/MRBLEs
1. Harink B
2. Nguyen H
3. Thorn K
4. Fordyce P
(2019) An open-source software analysis package for microspheres with ratiometric barcode lanthanide encoding (MRBLEs)
PLOS ONE 14:e0203725.
https://doi.org/10.1371/journal.pone.0203725
- PubMed
- Google Scholar
1. Hein JB
2. Hertz EPT
3. Garvanska DH
4. Kruse T
5. Nilsson J
(2017) Distinct kinetics of serine and threonine dephosphorylation are essential for mitosis
Nature Cell Biology 19:1433–1440.
https://doi.org/10.1038/ncb3634
- PubMed
- Google Scholar
1. Hintersteiner M
2. Kimmerlin T
3. Kalthoff F
4. Stoeckli M
5. Garavel G
6. Seifert JM
7. Meisner NC
8. Uhl V
9. Buehler C
10. Weidemann T
11. Auer M
(2009) Single bead labeling method for combining confocal fluorescence on-bead screening and solution validation of tagged one-bead one-compound libraries
Chemistry & Biology 16:724–735.
https://doi.org/10.1016/j.chembiol.2009.06.011
- PubMed
- Google Scholar
1. Huttlin EL
2. Ting L
3. Bruckner RJ
4. Gebreab F
5. Gygi MP
6. Szpyt J
7. Tam S
8. Zarraga G
9. Colby G
10. Baltier K
11. Dong R
12. Guarani V
13. Vaites LP
14. Ordureau A
15. Rad R
16. Erickson BK
17. Wühr M
18. Chick J
19. Zhai B
20. Kolippakkam D
21. Mintseris J
22. Obar RA
23. Harris T
24. Artavanis-Tsakonas S
25. Sowa ME
26. De Camilli P
27. Paulo JA
28. Harper JW
29. Gygi SP
(2015) The BioPlex network: a systematic exploration of the human interactome
Cell 162:425–440.
https://doi.org/10.1016/j.cell.2015.06.043
- PubMed
- Google Scholar
1. Huttlin EL
2. Bruckner RJ
3. Paulo JA
4. Cannon JR
5. Ting L
6. Baltier K
7. Colby G
8. Gebreab F
9. Gygi MP
10. Parzen H
11. Szpyt J
12. Tam S
13. Zarraga G
14. Pontano-Vaites L
15. Swarup S
16. White AE
17. Schweppe DK
18. Rad R
19. Erickson BK
20. Obar RA
21. Guruharsha KG
22. Li K
23. Artavanis-Tsakonas S
24. Gygi SP
25. Harper JW
(2017) Architecture of the human interactome defines protein communities and disease networks
Nature 545:505–509.
https://doi.org/10.1038/nature22366
- PubMed
- Google Scholar
1. Ivarsson Y
2. Arnold R
3. McLaughlin M
4. Nim S
5. Joshi R
6. Ray D
7. Liu B
8. Teyra J
9. Pawson T
10. Moffat J
11. Li SS
12. Sidhu SS
13. Kim PM
(2014) Large-scale interaction profiling of PDZ domains through proteomic peptide-phage display using human and viral phage peptidomes
PNAS 111:2542–2547.
https://doi.org/10.1073/pnas.1312296111
- PubMed
- Google Scholar
1. Ivarsson Y
2. Jemth P
(2019) Affinity and specificity of motif-based protein-protein interactions
Current Opinion in Structural Biology 54:26–33.
https://doi.org/10.1016/j.sbi.2018.09.009
- PubMed
- Google Scholar
1. Jain J
2. McCaffrey PG
3. Miner Z
4. Kerppola TK
5. Lambert JN
6. Verdine GL
7. Curran T
8. Rao A
(1993) The T-cell transcription factor NFATp is a substrate for calcineurin and interacts with fos and jun
Nature 365:352–355.
https://doi.org/10.1038/365352a0
- PubMed
- Google Scholar
1. Jiang L
2. Barclay AN
(2009) New assay to detect low-affinity interactions and characterization of leukocyte receptors for collagen including leukocyte-associated Ig-like receptor-1 (LAIR-1)
European Journal of Immunology 39:1167–1175.
https://doi.org/10.1002/eji.200839188
- PubMed
- Google Scholar
1. Kaushansky A
2. Allen JE
3. Gordus A
4. Stiffler MA
5. Karp ES
6. Chang BH
7. MacBeath G
(2010) Quantifying protein-protein interactions in high throughput using protein domain microarrays
Nature Protocols 5:773–790.
https://doi.org/10.1038/nprot.2010.36
- PubMed
- Google Scholar
1. Kilburg D
2. Gallicchio E
(2016)
Advances in Protein Chemistry and Structural Biology

27–57, Recent advances in computational models for the study of protein–peptide interactions, Advances in Protein Chemistry and Structural Biology, 105, Elsevier.
- Google Scholar
(2018) PSSMSearch: a server for modeling, visualization, proteome-wide discovery and annotation of protein motif specificity determinants
Nucleic Acids Research 46:W235–W241.
https://doi.org/10.1093/nar/gky426
- PubMed
- Google Scholar
1. Kumaresan PR
2. Wang Y
3. Saunders M
4. Maeda Y
5. Liu R
6. Wang X
7. Lam KS
(2011) Rapid discovery of death ligands with one-bead-two-compound combinatorial library methods
ACS Combinatorial Science 13:259–264.
https://doi.org/10.1021/co100069t
- PubMed
- Google Scholar
1. Lam KS
2. Salmon SE
3. Hersh EM
4. Hruby VJ
5. Kazmierski WM
6. Knapp RJ
(1991) A new type of synthetic peptide library for identifying ligand-binding activity
Nature 354:82–84.
https://doi.org/10.1038/354082a0
- PubMed
- Google Scholar
1. Leaver-Fay A
2. Tyka M
3. Lewis SM
4. Lange OF
5. Thompson J
6. Jacak R
7. Kaufman KW
8. Renfrew PD
9. Smith CA
10. Sheffler W
(2011)
Methods in Enzymology

545–574, Rosetta3: an object-oriented software suite for the simulation and design of macromolecules, Methods in Enzymology, 487, Elsevier, 10.1016/B978-0-12-381270-4.00019-6.
- Google Scholar
1. Lee J
2. Bisso PW
3. Srinivas RL
4. Kim JJ
5. Swiston AJ
6. Doyle PS
(2014) Universal process-inert encoding architecture for polymer microparticles
Nature Materials 13:524–529.
https://doi.org/10.1038/nmat3938
- PubMed
- Google Scholar
1. Li H
2. Rao A
3. Hogan PG
(2004) Structural delineation of the calcineurin-NFAT interaction and its parallels to PP1 targeting interactions
Journal of Molecular Biology 342:1659–1674.
https://doi.org/10.1016/j.jmb.2004.07.068
- PubMed
- Google Scholar
1. Li H
2. Zhang L
3. Rao A
4. Harrison SC
5. Hogan PG
(2007) Structure of calcineurin in complex with PVIVIT peptide: portrait of a low-affinity signalling interaction
Journal of Molecular Biology 369:1296–1306.
https://doi.org/10.1016/j.jmb.2007.04.032
- PubMed
- Google Scholar
1. Li H
2. Pink MD
3. Murphy JG
4. Stein A
5. Dell'Acqua ML
6. Hogan PG
(2012) Balanced interactions of calcineurin with AKAP79 regulate Ca2+-calcineurin-NFAT signaling
Nature Structural & Molecular Biology 19:337–345.
https://doi.org/10.1038/nsmb.2238
- PubMed
- Google Scholar
1. Liu R
2. Marik J
3. Lam KS
(2002) A novel peptide-based encoding system for "one-bead one-compound" peptidomimetic and small molecule combinatorial libraries
Journal of the American Chemical Society 124:7678–7680.
https://doi.org/10.1021/ja026421t
- PubMed
- Google Scholar
1. Liu T
2. Qian Z
3. Xiao Q
4. Pei D
(2011) High-throughput screening of one-bead-one-compound libraries: identification of cyclic peptidyl inhibitors against calcineurin/NFAT interaction
ACS Combinatorial Science 13:537–546.
https://doi.org/10.1021/co200101w
- PubMed
- Google Scholar
1. Meldal M
(2002) The one-bead two-compound assay for solid phase screening of combinatorial libraries
Biopolymers 66:93–100.
https://doi.org/10.1002/bip.10229
- PubMed
- Google Scholar
(2009) Inhibiting the calcineurin-NFAT (nuclear factor of activated T cells) signaling pathway with a regulator of calcineurin-derived peptide without affecting general calcineurin phosphatase activity
Journal of Biological Chemistry 284:9394–9401.
https://doi.org/10.1074/jbc.M805889200
- PubMed
- Google Scholar
1. Müller MR
2. Sasaki Y
3. Stevanovic I
4. Lamperti ED
5. Ghosh S
6. Sharma S
7. Gelinas C
8. Rossi DJ
9. Pipkin ME
10. Rajewsky K
11. Hogan PG
12. Rao A
(2009) Requirement for balanced ca/NFAT signaling in hematopoietic and embryonic development
PNAS 106:7034–7039.
https://doi.org/10.1073/pnas.0813296106
- PubMed
- Google Scholar
Patent
(2007) Vivit polypeptides, therapeutic agent comprising the same, and method of screening for anti-cancer agent (WO2007013358A2)
IFI CLAIMS Patent Services. United States.

http://www.google.com/patents/WO2007013358A2
1. Nakamura T
2. Katagiri T
3. Sato S
4. Kushibiki T
5. Hontani K
6. Tsuchikawa T
7. Hirano S
8. Nakamura Y
(2017) Overexpression of C16orf74 is involved in aggressive pancreatic cancers
Oncotarget 8:50460–50475.
https://doi.org/10.18632/oncotarget.10912
- PubMed
- Google Scholar
1. Neduva V
2. Russell RB
(2006) Peptides mediating interaction networks: new leads at last
Current Opinion in Biotechnology 17:465–471.
https://doi.org/10.1016/j.copbio.2006.08.002
- PubMed
- Google Scholar
1. Nguyen HQ
2. Baxter BC
3. Brower K
4. Diaz-Botia CA
5. DeRisi JL
6. Fordyce PM
7. Thorn KS
(2017a) Programmable microfluidic synthesis of over one thousand uniquely identifiable spectral codes
Advanced Optical Materials 5:1600548.
https://doi.org/10.1002/adom.201600548
- PubMed
- Google Scholar
Conference
1. Nguyen HQ
2. Brower K
3. Harink B
4. Baxter B
5. Thorn KS
6. Fordyce PM
(2017b) Peptide library synthesis on spectrally encoded beads for multiplexed protein/peptide bioassays
SPIE 10061, Microfluidics, BioMEMS, and Medical Microsystems XV. 100610Z.
https://doi.org/10.1117/12.2254908
- Google Scholar
1. Pollard TD
(2010) A guide to simple and informative binding assays
Molecular Biology of the Cell 21:4061–4067.
https://doi.org/10.1091/mbc.e10-08-0683
- PubMed
- Google Scholar
1. Price JV
2. Tangsombatvisit S
3. Xu G
4. Yu J
5. Levy D
6. Baechler EC
7. Gozani O
8. Varma M
9. Utz PJ
10. Liu CL
(2012) On silico peptide microarrays for high-resolution mapping of antibody epitopes and diverse protein-protein interactions
Nature Medicine 18:1434–1440.
https://doi.org/10.1038/nm.2913
- PubMed
- Google Scholar
(2009) A conserved docking surface on calcineurin mediates interaction with substrates and immunosuppressants
Molecular Cell 33:616–626.
https://doi.org/10.1016/j.molcel.2009.01.030
- PubMed
- Google Scholar
1. Rolland T
2. Taşan M
3. Charloteaux B
4. Pevzner SJ
5. Zhong Q
6. Sahni N
7. Yi S
8. Lemmens I
9. Fontanillo C
10. Mosca R
11. Kamburov A
12. Ghiassian SD
13. Yang X
14. Ghamsari L
15. Balcha D
16. Begg BE
17. Braun P
18. Brehme M
19. Broly MP
20. Carvunis AR
21. Convery-Zupan D
22. Corominas R
23. Coulombe-Huntington J
24. Dann E
25. Dreze M
26. Dricot A
27. Fan C
28. Franzosa E
29. Gebreab F
30. Gutierrez BJ
31. Hardy MF
32. Jin M
33. Kang S
34. Kiros R
35. Lin GN
36. Luck K
37. MacWilliams A
38. Menche J
39. Murray RR
40. Palagi A
41. Poulin MM
42. Rambout X
43. Rasla J
44. Reichert P
45. Romero V
46. Ruyssinck E
47. Sahalie JM
48. Scholz A
49. Shah AA
50. Sharma A
51. Shen Y
52. Spirohn K
53. Tam S
54. Tejeda AO
55. Trigg SA
56. Twizere JC
57. Vega K
58. Walsh J
59. Cusick ME
60. Xia Y
61. Barabási AL
62. Iakoucheva LM
63. Aloy P
64. De Las Rivas J
65. Tavernier J
66. Calderwood MA
67. Hill DE
68. Hao T
69. Roth FP
70. Vidal M
(2014) A proteome-scale map of the human interactome network
Cell 159:1212–1226.
https://doi.org/10.1016/j.cell.2014.10.050
- PubMed
- Google Scholar
1. Roy J
2. Li H
3. Hogan PG
4. Cyert MS
(2007) A conserved docking site modulates substrate affinity for Cal-cineurin, signaling output, and in vivo function
Molecular Cell 6:889–901.
https://doi.org/10.1016/j.molcel.2007.02.014
- Google Scholar
1. Roy J
2. Cyert MS
(2009) Cracking the phosphatase code: docking interactions determine substrate specificity
Science Signaling 2:re9.
https://doi.org/10.1126/scisignal.2100re9
- PubMed
- Google Scholar
1. Schymkowitz J
2. Borg J
3. Stricher F
4. Nys R
5. Rousseau F
6. Serrano L
(2005) The FoldX web server: an online force field
Nucleic Acids Research 33:W382–W388.
https://doi.org/10.1093/nar/gki387
- PubMed
- Google Scholar
1. Shapiro ND
2. Mirica KA
3. Soh S
4. Phillips ST
5. Taran O
6. Mace CR
7. Shevkoplyas SS
8. Whitesides GM
(2012) Measuring binding of protein to gel-bound ligands using magnetic levitation
Journal of the American Chemical Society 134:5637–5646.
https://doi.org/10.1021/ja211788e
- PubMed
- Google Scholar
1. Sheftic SR
2. Page R
3. Peti W
(2016) Investigating the human calcineurin interaction network using the πɸlxvp SLiM
Scientific Reports 6:38920.
https://doi.org/10.1038/srep38920
- PubMed
- Google Scholar
1. Smith CA
2. Kortemme T
(2010) Structure-based prediction of the peptide sequence space recognized by natural and synthetic PDZ domains
Journal of Molecular Biology 402:460–474.
https://doi.org/10.1016/j.jmb.2010.07.032
- PubMed
- Google Scholar
1. Smith CA
2. Kortemme T
(2011) Predicting the tolerated sequences for proteins and protein interfaces using RosettaBackrub flexible backbone design
PLOS ONE 6:e20451.
https://doi.org/10.1371/journal.pone.0020451
- PubMed
- Google Scholar
1. Sundell GN
2. Ivarsson Y
(2014) Interaction analysis through proteomic phage display
BioMed Research International 2014:1–9.
https://doi.org/10.1155/2014/176172
- Google Scholar
Preprint
1. Tareen A
2. Kinney JB
(2019) Logomaker: beautiful sequence logos in Python
bioRxiv.
https://doi.org/10.1101/635029
- Google Scholar
1. Tinti M
2. Kiemer L
3. Costa S
4. Miller ML
5. Sacco F
6. Olsen JV
7. Carducci M
8. Paoluzi S
9. Langone F
10. Workman CT
11. Blom N
12. Machida K
13. Thompson CM
14. Schutkowski M
15. Brunak S
16. Mann M
17. Mayer BJ
18. Castagnoli L
19. Cesareni G
(2013) The SH2 domain interaction landscape
Cell Reports 3:1293–1305.
https://doi.org/10.1016/j.celrep.2013.03.001
- PubMed
- Google Scholar
1. Tompa P
2. Davey NE
3. Gibson TJ
4. Babu MM
(2014) A million peptide motifs for the molecular biologist
Molecular Cell 55:161–169.
https://doi.org/10.1016/j.molcel.2014.05.032
- PubMed
- Google Scholar
1. Tonikian R
2. Zhang Y
3. Sazinsky SL
4. Currell B
5. Yeh JH
6. Reva B
7. Held HA
8. Appleton BA
9. Evangelista M
10. Wu Y
11. Xin X
12. Chan AC
13. Seshagiri S
14. Lasky LA
15. Sander C
16. Boone C
17. Bader GD
18. Sidhu SS
(2008) A specificity map for the PDZ domain family
PLOS Biology 6:e239.
https://doi.org/10.1371/journal.pbio.0060239
- PubMed
- Google Scholar
1. Uyar B
2. Weatheritt RJ
3. Dinkel H
4. Davey NE
5. Gibson TJ
(2014) Proteome-wide analysis of human disease mutations in short linear motifs: neglected players in Cancer?
Molecular BioSystems 10:2626–2642.
https://doi.org/10.1039/C4MB00290C
- PubMed
- Google Scholar
1. Van Roey K
2. Uyar B
3. Weatheritt RJ
4. Dinkel H
5. Seiler M
6. Budd A
7. Gibson TJ
8. Davey NE
(2014) Short linear motifs: ubiquitous and functionally diverse protein interaction modules directing cell regulation
Chemical Reviews 114:6733–6778.
https://doi.org/10.1021/cr400585q
- PubMed
- Google Scholar
1. Vincentelli R
2. Luck K
3. Poirson J
4. Polanowska J
5. Abdat J
6. Blémont M
7. Turchetto J
8. Iv F
9. Ricquier K
10. Straub ML
11. Forster A
12. Cassonnet P
13. Borg JP
14. Jacob Y
15. Masson M
16. Nominé Y
17. Reboul J
18. Wolff N
19. Charbonnier S
20. Travé G
(2015) Quantifying domain-ligand affinities and specificities by high-throughput holdup assay
Nature Methods 12:787–793.
https://doi.org/10.1038/nmeth.3438
- PubMed
- Google Scholar
(2017) Library Design-Facilitated High-Throughput sequencing of synthetic peptide libraries
ACS Combinatorial Science 19:694–701.
https://doi.org/10.1021/acscombsci.7b00109
- PubMed
- Google Scholar
1. Zhou HX
(2012) Intrinsic disorder: signaling via highly specific but short-lived association
Trends in Biochemical Sciences 37:43–48.
https://doi.org/10.1016/j.tibs.2011.11.002
- PubMed
- Google Scholar

Decision letter

Michael A Marletta

Senior and Reviewing Editor; University of California, Berkeley, United States
Toby Gibson

Reviewer; European Molecular Biology Laboratory, Germany

In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.

Thank you for submitting your work entitled "Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The following individual involved in review of your submission has agreed to reveal his identity: Toby Gibson (Reviewer #2).

Our decision has been reached after consultation within the editorial board. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife. In particular, there was uniform agreement that the issues raised by reviewer 1 are highly unlikely to be addressed in revision.

Reviewer #1:

In this study, the authors present a new technique to measure the affinity of protein-peptide interaction in parallel. They apply their method to the well-known Calcineurin-PxIxIT interaction, obtain more data on this interaction, and use it to identify high-affinity peptides. They validate some of these peptides in a cell-based assay.

This work is based around the bead-encoded peptide-synthesis and subsequent affinity measurements using the spectral encoding. This all strikes me as technically state-of-the-art and quite elegant. However, from the paper as written, it is a bit difficult to get a true sense of the advantages of the new method over existing ones, and no direct comparison is attempted in the manuscript. For instance, mapping of the affinities of substitutions (as in Figure 4) surely would be possible using SPOT (to a slightly different degree, also using Phage). The shown r² of 0.68 is pretty good, but not all that much higher than what has been claimed for SPOT (though a little higher than for phage – see e.g., Ivarsson et al., where, of course, a much higher number of peptides is measured). One should also note that this is for the same protein – incidentally, modern computational methods (such as alchemy-based ones, e.g., TI or CGI) achieve these sort of correlations, and usually perform better than rosetta. As high-quality structures are available, they would be applicable in this case, and may perform similar to the new bead-based method.

Similar things hold for the high-affinity variants obtained here (Figure 5 or later). While obtaining low-nanomolar binding peptides is really quite good, this domain is (as the authors themselves state) what phage display really is optimized for. While their new method can incorporate non-canonical amino acids, that is similarly true of novel synthetic library methods as what recently came out of the Pentelute lab.

Finally, while the authors seem to emphasize ability to detect "weak" affinities, I only see convincing data for affinities of up to maybe 50uM, which is all still detectable using techniques such as AP-MS or phage.

I should note that I do not really want to sound too negative. I see a new technique that brings substantial technical innovation that could be very exciting in this field. However, I find it difficult to ascertain its advantages over the variety of existing methods in the manuscript as presented. I would strongly suggest to include more direct comparisons with existing methods (such as SPOT, ProP-PD, hold-up, etc.) as well a more detailed discussion thereof.

A few additional points:

1) The authors state "CN binds PxIxIT peptides with weak affinities.… estimates for known substrates have varied over a wide range.…". These statements are somewhat contradictory in itself and, in any case, the affinities listed in Figure 2A seem to be all in the single digit uM, which is neither a wide range, nor particularly weak (roughly the expected range for domain-motif interactions).

2) Given the relatively length of the current paper, it would be good scholarship to discuss existing methods (including the variety of phage-display approaches, SPOT arrays and the recently developed holdup assay) – currently, this all happens in part of a single paragraph, with some of the methods I point out above not even mentioned.

3) In a similar vein, the citations (or lack thereof) are a bit puzzling. A 12-year old paper (Neduva and Russell) is cited regarding motif-interactions in Y2H/AP-MS, when most of the relevant work was published in the last decade. For Proteomic phage display, only a review paper is cited.

Reviewer #2:

This paper introduces MRBLE-pep, a medium-scale throughput protein-peptide affinity tool and applies it to calcineurin and variants of the PxIxIT docking motif. Using MRBLEs – colour-coded microbeads, each position in the original peptide can be substituted with all other amino acids. Systematic evaluation of binding determinants can then be rapidly determined using small amounts of reagents. Efficient methods for examining SLiM specificities are vey much needed given the abundance of protein:motif interactions in cell regulation and therefore MRBLE could see widespread adoption. Although the PxIxIT motif is well known from several calcineurin substrates, there has been uncertainty regarding the motif specificity determinants as well as the cooperative involvement of the second LxVP docking motif.

Weight matrices (PSSMs/HMMs) are often used to represent SLiMs in sequence searches but assume that each position in the sequence can be treated independently of the others. Regular expressions are also in widespread use but are overdetermined and insufficiently flexible regarding sequence space. The results here can be used to improve bioinformatic PxIxIT detection in candidate substrate proteins. However, the results also emphasise that positions in the PxIxIT motif are not independent and also that flanking residues affect the binding affinity (which has been seen for a number of other linear motifs too). The data from MRBLE-pep could allow a different computational approach whereby candidate motifs in protein sequences could be compared to and ranked by the total landscape affinity data.

No substantive concerns were identified.

[Editors’ note: what now follows is the decision letter after the authors submitted for further consideration.]

Thank you for resubmitting your work entitled "Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads" for further consideration at eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below in the first review below. Normally, we try to avoid second revisions, but we feel reviewer #1's comments should be addressed and we hope you will be able to do this easily and quickly.

Reviewer #1:

Essentially, the revisions by the authors haven't changed my opinions drastically. I still see this is as an innovative technology that will likely be of interest; however, the paper is still hampered by some shortcomings that I think could be improved pretty easily. In a way I find it a bit frustrating that the authors choose to write a full two pages of rebuttal to a simple point (correlations to measured affinity - compare these R²s between SPOT and MRBLE-pep) when they could have simply provided two numbers. Also, little of these two pages of rebuttal seems to have made it to the text (or was in there to begin with). Why not provide some kind of comparison of correlation to measured affinity to SPOT in the paper? I'm reasonably certain MRBLE-pep would come out on top.

Similarly, the authors add calculations using FoldX (that I didn't ask for). Not sure why they include comparisons of FoldX to Rosetta in the main text/figure (these would be of interest in computational papers and indeed have been done a bit, but I doubt the main readership of this paper will be interested). Also, I find it a bit frustrating that while they added a many new figures, the one figure that I think would have added to the paper (again a simple direct comparison of ddGs derived from MRBLE-pep to ones derived from rosetta OR FoldX vs. measured ddGs, i.e. combining the panels of Figure 4—figure supplement 4 into one panel and comparing it to the appropriate one for MRLBLE-pep). It's fine if the authors don't want to do any TI/FEP calculations (again, this isn't a computational paper), though if they do want to make the point that FoldX/Rosetta don't perform well on solvent exposed residues (not surprising), they should also mention that this is something that TI/FEP would, at least in theory, do better at.

They did add a much more appropriate introduction with discussion of the literature.

Reviewer #2:

The original MRBLE-pep manuscript was rather let down by superficial treatment of other PPI methods and how MARBL-pep compares to them. The resubmitted manuscript is greatly improved in this regard. This is important because SLiM researchers need to be able to understand which methods are suitable for the projects they have in mind. These range from the low throughput "gold standard" ITC up to whole proteome screens of the "disorderome" with phage display to identify novel SLiM candidates. I believe that MARBL-pep will see significant adoption and will help to define many SLiM motif patterns. The data obtained for the chosen target, the Calcineurin-binding PxIxIT motif, will hopefully now be applied in the identification of novel substrates for this medically important phosphatase. I have no further concerns.

https://doi.org/10.7554/eLife.40499.085

Author response

[Editors’ note: the author responses to the first round of peer review follow.]

We are writing to follow up on the recent eLife review of our manuscript “Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads”. We believe that the criticisms are readily addressed and stem largely from the fact that we didn't clearly explain critical differences between our novel assay and current approaches. To frame this discussion, we provide an overview of the reviewers' comments below and attach a detailed response to each of the comments and suggestions.

Reviewer #2 (Toby Gibson) was very positive, noting that “efficient methods for examining SLiM specificities are very much needed given the abundance of protein:motif interactions in cell regulation”, that “MRBLE could see widespread adoption”, and that “the data from MRBLE-pep could allow a different computational approach whereby candidate motifs in protein sequences could be compared to and ranked by the total landscape affinity data”. We note that Toby Gibson is a world expert in protein-SLiM interactions, and are flattered by his positive reception to the work.

Reviewer #1 stated that the work was “technically state-of-the-art and quite elegant” and that they “see a new technique that brings substantial technical innovation that could be very exciting in this field”. However, they state that they found it “difficult to ascertain its advantages over the variety of existing methods in the manuscript as presented” and would “strongly suggest to include more direct comparisons with existing methods (such as SPOT, ProP-PD, hold-up, etc)”. This helpful suggestion is easily addressed in the text of the manuscript. To briefly summarize advantages of MRBLE-pep:

1) MRBLEs have been engineered to have very slow on- and off-rates (half-lives of ~ 5 hours), allowing equilibrium measurement of even weak and transient SLiM interactions to yield quantitative thermodynamic constants (K_D and ΔΔG). This is simply not possible with SPOT arrays or ProP-PD, as off-rates for domain-SLiM interactions are on the order of seconds, much shorter than typical washing steps such that interactions with fast off-rates are preferentially lost.

2) MRBLE-pep incorporates in-line quality control to provide critical negative information about residues that ablate or reduce binding that is not available via SPOT or ProP-PD.

3) MRBLE-pep allows direct assessment of the effects of post-translational modifications (not possible with ProP-PD).

4) MRBLE-pep requires significantly less material than current low-throughput quantitative techniques (~60000x less protein and ~6000x less peptide than micro/nano ITC).

Finally, we note that we apply this new technology towards studying a system of great biological interest and reveal new information about determinants of specificity. As Toby Gibson notes, “the results here could be used to improve bioinformatic PxIxIT in candidate substrate proteins” and “emphasize that positions in the PxIxIT motif are not independent and also that flanking residues affect the binding affinity”.

Reviewer #1:

In this study, the authors present a new technique to measure the affinity of protein-peptide interaction in parallel. They apply their method to the well-known Calcineurin-PxIxIT interaction, obtain more data on this interaction, and use it to identify high-affinity peptides. They validate some of these peptides in a cell-based assay.

This work is based around the bead-encoded peptide-synthesis and subsequent affinity measurements using the spectral encoding. This all strikes me as technically state-of-the-art and quite elegant.

We appreciate the reviewer’s kind opinion that the method is “technically state-of-the-art and quite elegant”!

However, from the paper as written, it is a bit difficult to get a true sense of the advantages of the new method over existing ones, and no direct comparison is attempted in the manuscript. For instance, mapping of the affinities of substitutions (as in Figure 4) surely would be possible using SPOT (to a slightly different degree, also using Phage). The shown r² of 0.68 is pretty good, but not all that much higher than what has been claimed for SPOT (though a little higher than for phage – see e.g., Ivarsson et al., where, of course, a much higher number of peptides is measured).

The reviewer makes an excellent point. We did not clearly explain the advantages and disadvantages of MRBLE-pep relative to current techniques, and we also did not emphasize why quantitative measurements of interaction affinities are so important! We appreciate the helpful suggestions for how to improve the manuscript. In response, we have extensively revised the text in the Introduction and Discussion to include several paragraphs comparing and contrasting MRBLE-pep to current technologies.

To ensure that we clearly communicate these points, we also discuss the advantages and disadvantages of these approaches here, where we are not constrained by length.

Array-based methods:

Array-based methods provide the ability to probe for interactions between a protein of interest and 10s^-100s (SPOT arrays) or up to millions (ultra-high density arrays) peptides in a single experiment (Frank et al., 1990; Atwater et al., 2018; Buus et al., 2012; Forsstrom et al., 2014; Price et al., 2012; Carmona et al., 2015). In most cases, peptides are synthesized in situ via chemical synthesis approaches, making it possible to profile interactions between proteins and peptides containing unnatural amino acids or post-translational modifications at known positions. However, array-based methods cannot return quantitative interaction affinities (dissociation constants (K_Ds) or relative affinities (ΔΔGs)) because the measurements do not take place at thermodynamic equilibrium. Following incubation of proteins of interest with a peptide of interest, arrays must be washed extensively prior to imaging to quantify bound material; when a fluorescently-labeled antibody is used to detect bound protein, two incubations and two washing steps are required. In either case, the long washing times (minutes to hours) relative to the interaction dissociation rates (seconds) (Zhou et al., 2012; Dogan et al., 2015; and discussion in Ivarsson et al., 2019) mean that only the most persistent interactions are retained and interactions with faster off-rates are preferentially lost. Multiple comparisons between bound protein intensities and measured affinities for array-based experiments have established that while intensities typically correlate well with affinities for high-affinity interactions, arrays fail to recover weaker interactions. Finally, array-based methods are associated with high numbers of false positive and false negative measurements because it is difficult to evaluate the yield and purity of peptides present in each spot (Blikstad et al., 2015).

MRBLE-pep offers the following advantages relative to array-based methods:

1) Equilibrium measurements to yield affinities. As shown in Figure 2—figure supplements 1 and 2, we have deliberately engineered MRBLE polymer beads to have extremely slow on- and off-rates. This property makes it possible to remove unbound material and image to quantify bound protein with the washing steps taking place over a period of time that is much smaller than the half-life of the reactions, thereby approximating equilibrium and allowing direct measurement of interaction affinities.

2) In-line quality control for yield negative binding information. The use of different chemical linkers in the MRBLE core and periphery make it possible to cleave synthesized peptides from the MRBLE core only and directly assess their quality via MALDI-TOF mass spectrometry while preserving peptides presented at the periphery for binding measurements. If quality control reveals issues with synthesis, particular MRBLE-peptide combinations can simply be re-synthesized, re-tested, and added to the pool.

3) Statistical replicates. Each pooled measurement assesses binding of protein to ~ 30 beads containing the same embedded code and displayed sequence, allowing robust discrimination of even subtle differences in binding between sequences.

Display-based methods: Display-based methods have the advantage of profiling interactions between SLiM-binding proteins and large numbers (up to 10¹⁰) of candidate SLiMs. While approaches using combinatorial peptide libraries were biased towards identifying hydrophobic substrates, Ylva Ivarsson’s lab has pioneered new approaches for profiling interactions between SLiM-binding proteins and peptides tiled across intrinsically disordered regions of the proteome (Pro- PD) that remove this bias, as well as a newer approach that profiles peptides containing either serines and threonines or phosphomimetic residues at the same position to examine phosphorylation-dependent binding specificities.

However, a primary disadvantage of display-based methods is that they cannot return thermodynamic interaction affinities. Display-based techniques require multiple rounds of selection and washing, and as a result, take place out of equilibrium and provide only semi-quantitative estimates of binding strength biased towards SLiMs with the slowest dissociation rates. While computational approaches for extracting binding affinities from all rounds of selection have recently been developed for and applied towards high-throughput measurements of transcription factor-DNA affinities via SELEX approaches (e.g. No Read Left Behind, Rastogi et al., PNAS 2018), current Pro-PD and ProP-PD experiments typically consider NGS reads from a single round of enrichment (e.g. the fourth round of selection) only. In addition, display-based techniques cannot return information about the effects of post-translational modifications. While ProP-PD incorporates phosphomimetic residues in place of native serines and threonines to attempt to probe the effects of phosphorylation, we demonstrate here that phosphomimetic residues do not always fully mimic the effects of phosphorylation (here, phosphorylation at position 9 increases affinities 50-fold while phosphomimetic residues have essentially no effect). Finally, display-based techniques return a high number of false positives and false negatives and must therefore be validated via an alternate approach such as isothermal calorimetry. Finally, all selection-based methods, including Pro-PD, return only positive information (i.e. identify motifs that interact with the domain in question), but do not directly test if a particular sequence binds. Therefore, critical information about residues that prevent binding is not identified.

In fact, MRBLE-pep is a perfect complement to Pro-PD. After identifying selected sequences via Pro-PD, MRBLE-pep could be used to validate binding to identified peptides, determine binding affinities and test effects of amino substitutions on affinity for comprehensive motif characterization. We have modified the manuscript to make this point more explicit, and appreciate the reviewer’s suggestion.

Other methods:

Y2H and AP-MS: A recent large-scale yeast-two-hybrid (Y2H) screen identified 14000 potential binary protein-protein interactions to reveal the human interactome network, representing a milestone in knowledge of this network (Rolland et al., 2014). Over the following years, other groups have expanded the knowledge of this network via AP-MS assays to detect interactions (Huttlin et al., 2015; Huttlin et al., 2017), revealing ~23,000 and 56,000 interactions, respectively. However, interactions between SLiMs and SLiM-binding proteins remain particularly underrepresented (Van Roey et al., 2014; Davey et al., 2017). Underrepresentation in AP-MS is likely due to the low affinities and fast dissociation rates for SLiM-mediated interactions and the need for washing steps in this technique. In addition, neither Y2H or AP-MS methods claim to return quantitative interaction affinities. Moreover, AP-MS reports only on the presence of an interaction between 2 protein components, without information about whether this interaction is binary or what protein fragment/motif is responsible.

Hold-up assay: The hold-up assay represents the most direct competitor to the work presented here. This method was first described in an elegant and thorough Nature Methods paper from 2015 by Vincentelli and colleagues in which they profiled interactions between 210 PDZ-peptide pairs with known affinities and then mapped the specificity landscape between 2 viral PDZ-binding motifs from human papillomavirus E6 oncoproteins and 209 PDZ domains. While this assay detects complexes binding at equilibrium (in contrast to other high-throughput methods mentioned above), in its standard form, the assay returns only ranked lists of likely interaction strengths (not actual affinities). Deriving quantitative affinity information for complexes with K_D > 8 µM requires repetition of the assay at higher protein concentrations in order to ensure that available soluble protein is not significantly depleted within the unbound fraction. As a result, the material requirements for the hold-up assay significantly exceed those for the assay presented here. Finally, we note that the hold- up assay has never been cited in subsequent experimental work, suggesting that there are significant technical barriers to widespread adoption.

Because most high-throughput technologies take place out of thermodynamic equilibrium and therefore cannot measure quantitative interaction affinities (equilibrium constants), the vast majority of SLiMs have been characterized by low- throughput experimentation (see Davey et al., 2017; Dinkel et al., 2016; Toniian et al., 2008). Rather than viewing high-throughput screening methods as competing technologies for MRBLE-pep, we believe that the two approaches are highly complementary. The large sequence spaces accessible to display-based methods make it possible to scan large sequence spaces for candidate interaction motifs. After such a scan, MRBLE-pep can be used to: (1) quickly and efficiently scan through thousands of potential SLiMs to validate true hits, (2) directly measure SLiM interaction affinities, and then (3) systematically perturb identified motifs to map the surrounding binding affinity landscape, assess binding specificity, and determine a position specific affinity matrix that can be used to more accurately predict candidate downstream targets across the proteome. MRBLE-pep has the potential to streamline tedious downstream validation and affinity measurement process relative to traditional quantitative biotechniques such as fluorescence polarization, surface plasmon resonance, and isothermal calorimetry. We have amended the Discussion to more clearly emphasize that we view MRBLE-pep as a method that is complementary to (rather than competitive with) high-throughput discovery techniques, and hope that this addresses this reviewer’s concerns.

One should also note that this is for the same protein – incidentally, modern computational methods (such as alchemy-based ones, e.g., TI or CGI) achieve these sort of correlations, and usually perform better than rosetta. As high-quality structures are available, they would be applicable in this case, and may perform similar to the new bead-based method.

We agree with the reviewer that including other computational methods will likely improve the ability to accurately predict the energetic effects of SLiM mutations. In particular, the reviewer suggests that alchemical free energy calculations may improve these predictions. However, a previous paper (Aldeghi et al., 2018) directly compared performance between these calculations and Rosetta flex_ddG and concluded:

“We show that both the free energy calculations and Rosetta are able to quantitatively predict changes in ligand binding affinity upon protein mutations, yet the best predictions are the result of combining the estimates of both methods.”

They also note the difference in computational cost:

“It is important to put the performances of the calculations in the context of their computational cost. With the free energy calculations, each ΔΔG estimate took between 2 and 5 days on a node with 20 CPU threads and 1 GPU (Intel Xeon E5-2630 v4; GTX 1080 Ti), depending on the size of the system (from ~30000 to ~100000 atoms). With the flex_ddG protocol, each ΔΔG estimate took up to a day on a single CPU core.”

To address the reviewer’s suggestion, we worked with Naomi Latorroca from Ron Dror’s lab at Stanford to use the FoldX server to predict the likely change in binding energy upon mutating different residues within PxIxIT peptides on calcineurin- PxIxIT binding affinities (see Figure 4 and associated supplementary figures and data sets). In many cases, the FoldX predictions outperformed those of the Rosetta flex_ddG protocol when assessing the ability to classify predicted substitutions as either stabilizing (ΔΔG < 0) or destabilizing (ΔΔG > 0), and we think that the manuscript has been improved as a result. We note, however, that the prediction accuracy remains somewhat low, consistent with the prevailing sentiment that predicting mutational effects for protein-peptide interactions remains a difficult problem due to the large number of possible conformations of the peptide when free in solution. MRBLE-pep therefore provides a unique opportunity to iterate quickly between computational predictions and experimental measurements, as high-throughput peptide synthesizers allow synthesis of up to 576 peptides on MRBLEs in parallel in about a week and downstream binding assays require only a few additional days. The inclusion of this analysis has substantially improved the paper and we appreciate the reviewer’s suggestion!

Similar things hold for the high-affinity variants obtained here (Figure 5 or later). While obtaining low-nanomolar binding peptides is really quite good, this domain is (as the authors themselves state) what phage display really is optimized for.

We appreciate the reviewer’s point that identifying super-binders is easily accomplished via a variety of other methods, including phage display, that allow searching of extremely large combinatorial libraries (up to 1010). However, the current approach demonstrated the ability to identify peptides with affinities 2 orders of magnitude stronger than those previously identified via phage display using only 400 measurements. This brings up an interesting question: what is the most efficient way to scan a potential sequence space? For peptides that bind as linear motifs, the results presented here suggest that systematically testing a single mutation at each position and then combining those results may provide an alternative method to obtain the same results via a much smaller number of experimental measurements.

While their new method can incorporate non-canonical amino acids, that is similarly true of novel synthetic library methods as what recently came out of the Pentelute lab.

We appreciate the discussion of the exciting new technologies that have emerged from the Pentelute lab. We regret failing to cite these in our initial submission and have amended this by including several citations in the current text. Again, however, we emphasize that these technologies are screening technologies that, while able to test a wide variety of potential interactions (including a wide variety of unnatural and modified amino acids), do not return quantitative interaction affinities. As with phage display, we view the current MRBLE-pep technology as complementary to, rather than competitive with, these screening technologies. We have amended the text to include references to 2 recent relevant publications from the Pentelute lab.

Finally, while the authors seem to emphasize ability to detect "weak" affinities, I only see convincing data for affinities of up to maybe 50uM, which is all still detectable using techniques such as AP-MS or phage.

We respectfully disagree with the reviewer’s suggestion that affinities of up to 50 µM are detectable using AP-MS, as prior analyses have demonstrated that low-affinity SLiM-mediated interactions are significantly underrepresented within these datasets, as described above. However, even if these interactions were not underrepresented within AP-MS datasets, there would still be a need to directly and quantitatively measure the interaction affinities to predict biological responses and downstream outputs, as well as to acquire in vitro data to distinguish between direct and indirect binding interactions. For this reason, we again suggest that MRBLE-pep may represent an ideal companion technology to quantify affinities for candidate AP-MS interactions to decode cellular signaling interaction networks and for use in whole-cell modeling efforts.

I should note that I do not really want to sound too negative. I see a new technique that brings substantial technical innovation that could be very exciting in this field. However, I find it difficult to ascertain its advantages over the variety of existing methods in the manuscript as presented. I would strongly suggest to include more direct comparisons with existing methods (such as SPOT, ProP-PD, hold-up, etc.) as well a more detailed discussion thereof.

We appreciate the reviewer’s encouraging comment that the review is not meant to sound too negative, as well as the sentiment that this is “a new technique that brings substantial technical innovation that could be very exciting in this field”! We believe that the reviewer’s suggestions to include more direct comparisons between MRBLE-pep and these existing techniques as well as to explore additional computational methods for predicting mutational effects on protein-peptide affinities have strengthened the paper considerably. We have significantly revised the Introduction in response to these helpful suggestions.

A few additional points:

1) The authors state "CN binds PxIxIT peptides with weak affinities.… estimates for known substrates have varied over a wide range.…". These statements are somewhat contradictory in itself and, in any case, the affinities listed in Figure 2A seem to be all in the single digit uM, which is neither a wide range, nor particularly weak (roughly the expected range for domain-motif interactions).

We agree with the reviewer that single digit µM Kd values are typical for SLiM-mediated binding interactions. However, SLiM-mediated binding interactions are typically referred to as moderate-to-weak affinity and transient (for example, please see Davey et al., 2012). To address this suggestion, we have amended the text to read “CN binds PxIxIT peptides with moderate-to-weak affinities…”. When we suggest that estimates for known substrates have varied over a wide range, we mean to say that published estimates for the same specific interaction between CN and a particular SLiM have varied over a wide range. To clarify this, we have amended the text to read: “… prior literature affinity estimates for a given substrate have varied over 50-fold.”

2) Given the relatively length of the current paper, it would be good scholarship to discuss existing methods (including the variety of phage-display approaches, SPOT arrays and the recently developed holdup assay) – currently, this all happens in part of a single paragraph, with some of the methods I point out above not even mentioned.

We appreciate the suggestion and have revised the Introduction extensively to include a more detailed discussion of other methods, as detailed above. We sincerely regret the oversight in the initial submission.

3) In a similar vein, the citations (or lack thereof) are a bit puzzling. A 12-year old paper (Neduva and Russell) is cited regarding motif-interactions in Y2H/AP-MS, when most of the relevant work was published in the last decade. For Proteomic phage display, only a review paper is cited.

The reviewer makes an excellent point. In re-reading the paper, it’s clear that we were overly focused on presenting the data and did an embarrassingly poor job of referencing much of the recent literature. We again regret this oversight and have extensively rewritten the Introduction and amended the paper to include a much more comprehensive discussion of recently published work. This is another case where we feel that the reviewer’s suggestions have significantly improved the paper by making comparisons to other methods more clear!

Reviewer #2:

[…] Weight matrices (PSSMs/HMMs) are often used to represent SLiMs in sequence searches but assume that each position in the sequence can be treated independently of the others. Regular expressions are also in widespread use but are overdetermined and insufficiently flexible regarding sequence space. The results here can be used to improve bioinformatic PxIxIT detection in candidate substrate proteins. However, the results also emphasise that positions in the PxIxIT motif are not independent and also that flanking residues affect the binding affinity (which has been seen for a number of other linear motifs too). The data from MRBLE-pep could allow a different computational approach whereby candidate motifs in protein sequences could be compared to and ranked by the total landscape affinity data.

No substantive concerns were identified.

We appreciate the reviewer’s kind words and positive assessment of the work! We particularly appreciate that the reviewer correctly noted the ability of MRBLE-pep to provide quantitative information about interaction epistasis. We have added an additional sentence to the manuscript to emphasize this point.

[Editors’ note: the author responses to the re-review follow.]

The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below in the first review below. Normally, we try to avoid second revisions, but we feel reviewer #1's comments should be addressed and we hope you will be able to do this easily and quickly.

Reviewer #1:

Essentially, the revisions by the authors haven't changed my opinions drastically. I still see this is as an innovative technology that will likely be of interest; however, the paper is still hampered by some shortcomings that I think could be improved pretty easily. In a way I find it a bit frustrating that the authors choose to write a full two pages of rebuttal to a simple point (correlations to measured affinity – compare these R²s between SPOT and MRBLE-pep) when they could have simply provided two numbers. Also, little of these two pages of rebuttal seems to have made it to the text (or was in there to begin with). Why not provide some kind of comparison of correlation to measured affinity to SPOT in the paper? I'm reasonably certain MRBLE-pep would come out on top.

The reviewer makes an excellent suggestion, and we fully agree that this would be a very interesting and worthwhile comparison. Unfortunately, this is simply not possible with current published data, largely due to the technical challenges associated with measuring quantitative binding affinities. Prior literature includes estimates of measured affinities for only a handful of calcineurin-PxIxIT interactions, all 8 of which we also measured via MRBLE-pep (Figure 2C). The strong agreement between MRBLE-pep affinities and these 8 prior measurements is shown in Figure 2D (r2 = 0.68), with only NFATc2 (PRIEIT) showing significant deviation. To clarify that this data set represents all known prior published affinities, we have modified a sentence in the “MRBLE-pep yield quantitative measurements of calcineurin binding affinities” to read: “These 10 peptides (the “Triplicate low” dataset) represent all calcineurinbinding SLiMs for which affinities have been measured, including the known NFATc1, NFATc2, AKAP79, and RCAN1 natural CN-interacting PxIxIT binding site (citations), a set of 5 PVIVIT peptide mutants previously characterized via competitive fluorescence polarization assays (citations), and a scrambled negative control sequence.”

These peptides have not been measured via SPOT arrays, preventing the desired comparison. The only published paper using arrays to quantify binding to calcineurin assessed binding of tiled calcineurin fragments to a single PxIxIT motif (the NFATc2 SLiM SGPSPRIEITPSH) and was unable to detect any binding between calcineurin fragments and the known PVIVIT substrate (Erdmann et al., ChemBioChem 2018). To fully address the reviewer’s request without relying on previously published literature, we attempted to perform ITC for a series of calcineurin-peptide interactions for approximately 3 months but were ultimately unsuccessful.

Similarly, the authors add calculations using FoldX (that I didn't ask for). Not sure why they include comparisons of FoldX to Rosetta in the main text/figure (these would be of interest in computational papers and indeed have been done a bit, but I doubt the main readership of this paper will be interested). Also, I find it a bit frustrating that while they added a many new figures, the one figure that I think would have added to the paper (again a simple direct comparison of ddGs derived from MRBLE-pep to ones derived from rosetta OR FoldX vs. measured ddGs, i.e. combining the panels of Figure 4—figure supplement 4 into one panel and comparing it to the appropriate one for MRLBLE-pep).

Figure 4—figure supplement 5 and Figure 4—figure supplement 6 compare MRBLE-pep measurements to the predictions of FoldX and the Rosetta flex_ddG protocol, respectively. Of the prior published measurements of calcineurin- PxIxIT affinities, only a single study considered how systematic substitutions within the PVIVIT site affect measured affinities by performing fluorescence polarization measurements in the presence of competing unlabeled peptides to estimate affinities (Li and Hogan, 2007). The FoldX and Rosetta computational analyses estimate effects on affinity for all possible single-site substitutions within the PVIVITGPH site, and therefore require orthogonal measurements of single-site substitutions for comparison. The Li and Hogan experiments measure relative affinities for only the wildtype PVIVIT peptide and 2 examples of single-site substitutions from the PVIVIT motif (PVIVVT and PVIVIN); the other measurement includes 2 amino acid mutations (PVIAVT).

The only possible comparison between MRBLE-pep, FoldX, and Rosetta and prior published results therefore assesses their performance on 2 WT-like peptides (PVIVIT and PVIVVT) and one weak affinity variant (PVIVIN), resulting in a 3-point comparison for which 2 points are experimentally identical. The MRBLE-pep measurements alone (and not the FoldX and Rosetta computational predictions) correctly recapitulate that: (1) PVIVIT and PVIVVT have the same affinities) and (2) the PVIVIN substitution substantially reduces binding.

In response to the reviewer suggestion, we have included this figure as Figure 4—figure supplement 7, and also added the following to the text: “Comparisons of MRBLE-pep measurements and Rosetta and FoldX predictions of the energetic effects of single site substitutions with orthogonal measurements are complicated by a lack of previously published affinity data. However, a direct comparison between MRBLE-pep measurements and computational predictions for 2 single-site substitutions in the PVIVIT motif reveal that only MRBLE-pep correctly discerns that a PVIVVT substitution has little effect while a PVIVIN substitution reduces binding.”

It's fine if the authors don't want to do any TI/FEP calculations (again, this isn't a computational paper), though if they do want to make the point that FoldX/Rosetta don't perform well on solvent exposed residues (not surprising), they should also mention that this is something that TI/FEP would, at least in theory, do better at.

We thank the reviewer for this suggestion and have added the following sentence to the section “Comparing experimental measurements with results from structure-based computational predictions”: “Thermodynamic Integration and Free Energy Perturbation (TI/FEP) methods would likely enhance the ability to predict effects of mutating solvent-exposed residues, but are significantly more computationally expensive (Kilburg and Gallichio, 2016; Gallichio and Levy, 2011).”

They did add a much more appropriate Introduction with Discussion of the literature.

We thank the reviewer for this suggestion, agree that the discussion of the literature is much improved, and apologize for the initial oversight.

https://doi.org/10.7554/eLife.40499.086

Article and author information

Author details

Huy Quoc Nguyen

Department of Genetics, Stanford University, Stanford, United States

Present address
Genentech, South San Francisco, United States

Contribution
Conceptualization, Resources, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-5252-0062
Jagoree Roy

Department of Biology, Stanford University, Stanford, United States

Contribution
Resources, Investigation, Writing—original draft, Writing—review and editing

Contributed equally with
Björn Harink and Nikhil P Damle

Competing interests
No competing interests declared
Björn Harink

Department of Genetics, Stanford University, Stanford, United States

Contribution
Data curation, Software, Formal analysis, Visualization

Contributed equally with
Jagoree Roy and Nikhil P Damle

Competing interests
No competing interests declared
Nikhil P Damle

Department of Biology, Stanford University, Stanford, United States

Present address
OSTHUS GmbH, Eisenbahnweg, Aachen, Germany

Contribution
Conceptualization, Formal analysis, Investigation

Contributed equally with
Jagoree Roy and Björn Harink

Competing interests
No competing interests declared
Naomi R Latorraca

Biophysics Program, Stanford University, Stanford, United States

Contribution
Formal analysis, Performed all FoldX analysis

Competing interests
No competing interests declared
Brian C Baxter

Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, United States

Contribution
Resources, Synthesized all lanthanide nanophosphors

Competing interests
No competing interests declared
Kara Brower

Department of Bioengineering, Stanford University, Stanford, United States

Contribution
Resources, Writing—review and editing

Competing interests
No competing interests declared
Scott A Longwell

Department of Bioengineering, Stanford University, Stanford, United States

Contribution
Resources, Expressed and purified tagged calcineurin

Competing interests
No competing interests declared
Tanja Kortemme
1. Department of Bioengineering and Therapeutic Sciences, University of California, San Francisco, San Francisco, United States
2. Chan Zuckerberg Biohub, San Francisco, United States
Contribution
Software, Supervision, Writing—review and editing

Competing interests
No competing interests declared
Kurt S Thorn

Department of Biochemistry and Biophysics, University of California, San Francisco, San Francisco, United States

Present address
Zymergen, Inc, Emeryville, United States

Contribution
Conceptualization, Formal analysis, Supervision, Project administration

Competing interests
No competing interests declared
Martha S Cyert

Department of Biology, Stanford University, Stanford, United States

Contribution
Conceptualization, Supervision, Funding acquisition, Writing—original draft, Project administration, Writing—review and editing

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0003-3825-7437
Polly Morrell Fordyce
1. Department of Genetics, Stanford University, Stanford, United States
2. Department of Bioengineering, Stanford University, Stanford, United States
3. Chan Zuckerberg Biohub, San Francisco, United States
4. ChEM-H Institute, Stanford University, Stanford, United States
Contribution
Conceptualization, Formal analysis, Supervision, Funding acquisition, Visualization, Writing—original draft, Project administration, Writing—review and editing

For correspondence
pfordyce@stanford.edu

Competing interests
No competing interests declared

"This ORCID iD identifies the author of this article:" 0000-0002-9505-0638

Funding

National Institute of General Medical Sciences (DP2GM123641)

Polly Morrell Fordyce

National Institute of General Medical Sciences (R01GM107132)

Kurt S Thorn

National Institute of General Medical Sciences (R01GM119336)

Martha S Cyert

National Institute of General Medical Sciences (R01GM117189)

Tanja Kortemme

National Institute of General Medical Sciences (R01GM110089)

Tanja Kortemme

Chan Zuckerberg Biohub

Tanja Kortemme
Polly Morrell Fordyce

Alfred P. Sloan Foundation

Polly Morrell Fordyce

Arnold and Mabel Beckman Foundation

Polly Morrell Fordyce

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

This work was supported by NIH/NIGMS grants DP2GM123641 and R01GM107132 to PMF and R01GM119336 to MSC. In addition, PMF is a Chan Zuckerberg Biohub Investigator and acknowledges support from the Beckman Foundation, the Sloan Research Foundation, and PMF and MSC acknowledge the support of a joint Bio-X Interdisciplinary Initiatives Fund seed grant. TK is a Chan Biohub investigator and supported by NIH/NIGMS grants R01GM117189 and R01 GM110089. We thank Prof. Justin Kinney (Cold Spring Harbor Laboratory) for helpful discussions and software used to generate logos; we thank Prof. Dan Herschlag (Stanford) for essential feedback on the manuscript.

Senior and Reviewing Editor

Michael A Marletta, University of California, Berkeley, United States

Reviewer

Toby Gibson, European Molecular Biology Laboratory, Germany

Publication history

Received: August 8, 2018
Accepted: July 3, 2019
Accepted Manuscript published: July 8, 2019 (version 1)
Version of Record published: September 5, 2019 (version 2)

Copyright

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

3,323

Page views
530

Downloads
34

Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Mendeley

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

Huy Quoc Nguyen
Jagoree Roy
Björn Harink
Nikhil P Damle
Naomi R Latorraca
Brian C Baxter
Kara Brower
Scott A Longwell
Tanja Kortemme
Kurt S Thorn
Martha S Cyert
Polly Morrell Fordyce

(2019)

Quantitative mapping of protein-peptide affinity landscapes using spectrally encoded beads

eLife 8:e40499.

https://doi.org/10.7554/eLife.40499

Categories and tags

Research organism

Human

Share this article

Cite this article

MRBLE-pep experimental pipeline for high-throughput measurement of protein-peptide interactions.

Figure 1—source data 1

Figure 1—source data 2

MRBLEs allow quantitative measurement of CN-SLiM interactions in high-throughput.

Figure 2—source data 1

Figure 2—source data 2

MRBLE-pep mapping of the CN-PxIxIT binding energy landscape.

Figure 3—source data 1

Figure 3—source data 2

Figure 3—source data 3

Figure 3—source data 4

Figure 3—source data 5

Figure 3—source data 6

Figure 3—source data 7

Figure 3—source data 8

Figure 3—source data 9

Figure 3—source data 10

Structure-based modeling predictions.

Figure 4—source data 1

Figure 4—source data 2

Figure 4—source data 3

Figure 4—source data 4

Figure 4—source data 5

Figure 4—source data 6

Calibrated in vitro affinities for a range of peptides.

Figure 5—source data 1

Figure 5—source data 2

Figure 5—source data 3

Figure 5—source data 4

In vivo assays to quantify CN inhibition and an in vitro search for a potent inhibitor.

Figure 6—source data 1

Figure 6—source data 2

Figure 6—source data 3

Author details

Huy Quoc Nguyen

Present address

Contribution

Competing interests

Jagoree Roy

Contribution

Contributed equally with

Competing interests

Björn Harink

Contribution

Contributed equally with

Competing interests

Nikhil P Damle

Present address

Contribution

Contributed equally with

Competing interests

Naomi R Latorraca

Contribution

Competing interests

Brian C Baxter

Contribution

Competing interests

Kara Brower

Contribution

Competing interests

Scott A Longwell

Contribution

Competing interests

Tanja Kortemme

Contribution

Competing interests

Kurt S Thorn

Present address

Contribution

Competing interests

Martha S Cyert

Contribution

Competing interests

Polly Morrell Fordyce

Contribution

For correspondence

Competing interests

Downloads (link to download the article as PDF)