The LeEIX2 receptor-EIX effector interaction can be reproduced in ribosome display
In order to study the LeEIX2-EIX interaction, this direct binding event had to be reproduced in a cell-free context. We hypothesized that the LeEIX2 ectodomain should be sufficient, as it excluded the signal peptide, transmembrane, and cytosolic domains, which are unlikely to bind EIX but may adversely impact functional protein expression on the ribosome. Notably, the displayed amino acid sequence includes the N-terminal and C-terminal “caps” of the LeEIX2 ectodomain, which bury the hydrophobic cores of the first and last LRRs, mitigating the risk that unnaturally exposed hydrophobic residues may hinder proper folding or drive nonspecific interactions in binding assays. Each of the LeEIX2 protein domains and LRR units were defined in accordance with the study that first identified the receptor13.
The LeEIX2 ectodomain was expressed in RD format, and the resulting ribosomal complexes were panned against immobilized EIX either without or with excess EIX at a concentration of 1.25 µM in solution to produce total and nonspecific binding in the assay, respectively (Fig. 1B). After washing to remove unbound complexes, RT-PCR and subsequent gel electrophoresis of the remaining bound complexes provided a measure of mRNA retained due to binding of the LeEIX2 complex to EIX. A distinct signal difference between uncompeted and competed wells indicated specific binding to EIX (Fig. 1C, Supplementary Fig. 1). As additional validation, the LeEIX2 ribosomal complexes were separately panned against an immobilized unrelated target protein, maltose-binding protein (MBP), which did not generate a significant signal (Supplementary Figs. 2 and 3).
Notably, the full LeEIX2 ectodomain (31 LRRs in addition to the terminal caps and loopout domain) is ~ 115 kDa, which to our knowledge is the largest protein functionally displayed on the ribosome. This required numerous technical advances to overcome the inherently poor signal-to-noise ratio in RD for large proteins14. First, we used the reconstituted PURExpress cell-free translation system (New England Biolabs) to enhance the signal since it is a minimal expression system that, in contrast to cell lysates, significantly reduces undesired components such as nucleases and proteases. Second, codon optimization of the LeEIX2 template also boosted the signal by improving the efficiency of translation in the E. coli-based system15. Third, the use of a protein-free blocking buffer, SynBlock (Bio-Rad), reduced nonspecific interactions that contribute to the noise compared to less inert but commonly used blocking agents such as bovine serum albumin or casein. Finally, Mfold was used to optimize the mRNA sequence for the reverse transcription step to maximize product yield16 (Supplementary Fig. 4A,B). These collective advances facilitated clear detection of the full ectodomain of LeEIX2 on the ribosome.
LeEIX2 segmentation reveals two disjoint EIX-interacting hotspots
Having recreated the full LeEIX2 ectodomain interaction with EIX using RD, we sought to identify the subset of LeEIX2 LRRs that drive binding. Given the putative globular fold of EIX, our initial hypothesis was that the binding paratope in LeEIX2 would lie within a contiguous set of LRRs, so we employed a pseudo-branch and bound algorithm that leveraged the modularity of the repeat unit architecture. The 31-LRR LeEIX2 was split into three 13-LRR segments with partial overlap (Fig. 1D), and each segment was assayed in RD for the signal difference between uncompeted and competed wells to identify specific binding. Each segment shared 4 LRRs with its neighboring segment(s) to lessen the impact of potential misfolding of the outermost LRRs due to imperfect repeat stacking with the LeEIX2 N- and C-terminal caps, which were appended to all constructs. These initial RD experiments showed that both the first (LRRs 1–13) and the third (LRRs 19–31) segments exhibited specific binding to EIX (Fig. 1E, Supplementary Fig. 1).
Model-informed design of receptor sub-segmentation
To refine the location of these two paratopes within LeEIX2, we sought to further partition the first and third LRR segments for additional RD experimentation. Given that there are numerous ways to further subdivide these segments for experimental testing, we sought to develop a strategy that maximized the information gained from these experiments while maintaining the binding properties of the repeat units.
In subdividing segment 1, we focused on the 11-LRR region of LRRs 1–11, since LRRs 12 and 13 were considered unlikely interactors based on the non-interaction of the second LeEIX2 segment with EIX. Likewise, in subdividing segment 3, we focused on the region from LRRs 21–31, as LRRs 19 and 20 were unlikely EIX interactors. To account for the additional loopout domain in segment 3, we treated it as a 12-LRR region that needed to be subdivided.
Initially, a biology-agnostic approach was employed to determine how much information could be obtained from an experiment in which a segment of set length L (number of repeats) was partitioned into sub-segments of length n (number of consecutive repeats). The experiment would thus consist of L – n + 1 unique sub-segments of equal length. Each individual repeat unit in the segment, i, would be included in some tested constructs and excluded in others, with Pi describing the fraction of constructs that would include repeat i in EIX binding assays. The information entropy for each possible experimental setup was calculated as
$$entropy= -sum_{i=1}^{L}{P}_{i},{mathrm{log}}_{2},{P}_{i}$$
This calculation was performed for both 11- and 12-LRR segments (Fig. 2A,B). For LeEIX2 segment 1, maximal entropy occurs at n = 3, whereas the entropy for segment 3 is maximized when n = 4. We set an entropy tolerance of 90% of the maximal entropy, such that sub-segments with n = 2, 3, 4, 5 were considered for segment 1 and sub-segments with n = 3, 4, 5 were considered for segment 3. This algorithm is generalizable to inform unbiased segmentation of a repeat protein of any length.
Information entropy for different LRR segmentation lengths. Information entropy for experimentation involving sub-segments of length n (number of repeats) from a segment of length (A) 11 or (B) 12. The grey bar indicates a level above which at least 90% of maximal information is obtained.
Bringing in the biological context, we expected that non-native repeat unit-cap interfaces would be imperfect matches. Accordingly, repeats further from a mismatched interface are more likely to have native biological structure and function. There was also a concern that using too small of an n might reduce the interfacial contact area so much as to prevent detectable binding. Therefore, by testing sub-segments of n = 5, we were able to obtain near-maximal information while allowing for a more true-to-native biological environment for the repeats in each sub-segment.
Modular sub-segmentation refines the interacting repeat units
RD of these 5-LRR sub-segments against EIX confirmed the presence of two disjoint EIX-binding hotspots, with LRRs 6–10 and LRRs 22–26 showing the greatest signal-to-noise differences (Fig. 3A,B, Supplementary Fig. 5 and 6). These results were further validated by qRT-PCR, from which a mean ΔCq was calculated as the Cq of the competed wells minus the Cq of the uncompeted wells for each mRNA input tested (Fig. 3C, Supplementary Fig. 7). Again, LRRs 6–10 and 22–26 were found to have statistically significant signal differences between competed and uncompeted wells. The strong signals and high signal-to-noise ratios provide evidence for these two EIX-interacting sites. Although RD has been effectively used to express a number of functional proteins that are otherwise intractable11,12, low signals are more difficult to interpret, as there is no clear method to differentiate lack of function from protein misfolding on the ribosome.
5-LRR sub-segment mapping. Gel electrophoresis results for RD technical duplicates of (A) LRRs 1–5 to 7–11 or (B) LRRs 21–25 to loopout (l/o)-31 panning against immobilized EIX target in the absence or presence of excess competitor EIX. An unrelated RT-PCR sample (†) serves as a loading control so that signal intensities across gels can be compared. (C) ΔCq means calculated from normalized qRT-PCR replicates measured in 5-LRR sub-segment RD experiments. Error bars represent replicate standard deviations. •p ≤ 0.1 for Tukey’s method pairwise comparison of ΔCq following single-factor ANOVA. A high-variance data point (x) was identified as an outlier by Tukey’s heuristic and excluded from statistical analysis.
Further experimentation for hotspot refinement was conducted by branching of neighboring 4-LRR segments (Fig. 4A, Supplementary Fig. 8), with these results generally in agreement with those from 5-LRR sub-segments, suggesting that proper folding of these smaller protein constructs was largely maintained on the ribosome. Constructs containing either LRRs 6–9 or LRRs 21–24 from LeEIX2, which had high signal-to-noise ratios, were used to show that signal in an EIX competition assay is dose-dependent (Fig. 4B, Supplementary Fig. 8). An additional binding assay in RD format demonstrated that sub-segment binding is specific to EIX by comparing to non-specific interactions with casein as an immobilized target or as a competitor in solution (Fig. 4C, Supplementary Fig. 8). Collectively, these results provide a body of evidence for two disjoint EIX binding hotspots on LeEIX2.
4-LRR sub-segment mapping. (A) RD results of 4-LRR LeEIX2 sub-segment/EIX binding signal in the absence or presence of competitor EIX in solution (‘l/o’ refers to loopout domain). (B) RD results of dose-dependent LeEIX2 sub-segment/EIX binding signal for sub-segments 6–9 (top) and 21–24 (bottom). Lanes are for varying EIX competitor concentration in solution. (C) RD results of 4-LRR LeEIX2 sub-segment binding signal against immobilized EIX, immobilized unrelated target casein, immobilized EIX in the presence of excess EIX competitor, and immobilized EIX in the presence of excess casein competitor.
Analogous binding hotspots are found in the decoy receptor LeEIX1
LeEIX1 is posited to be a decoy receptor in the EIX response network; this receptor is known to bind EIX but cannot signal a hypersensitive response upon target recognition17. The unexpected result of a disjoint paratope in LeEIX2 led us to test a new hypothesis that LeEIX1 may lack one of the two hotspots for binding to EIX, leading to the observed phenotype. We first compared the amino acid sequences of both receptors at the identified LeEIX2 hotspot LRRs (Fig. 5A). Although there were differences at this level, even at the likely target-facing residues based on similar LRR receptor interactions, it remained unclear what role these mutations may have. We therefore performed analytical RD using the analogous LRRs 6–10 and 22–26 of LeEIX1 (Fig. 5B, Supplementary Fig. 9). The results indicate that both LeEIX1 hotspot sites appear to be functional in binding EIX.
Hotspot comparison between LeEIX1 and LeEIX2. (A) Amino acid sequence comparison between LeEIX1 and LeEIX2 for sub-segment LRRs 6–10 (left) and 22–26 (right). Residue comparison is indicated as fully conserved (*), strongly similar (:), or weakly similar (.), with emphasis on target-facing residues through coloring as fully conserved (blue) or not fully conserved (orange). (B) RD results of LeEIX1 and LeEIX2 sub-segments LRRs 6–10 (left) or 22–26 (right) panning against immobilized EIX in the absence or presence of excess competitor EIX.
