Streamlined phage display library protocols for identification of insect gut binding peptides highlight peptide specificity

Highlights

• •Next generation sequencing increases sensitivity of biopanning.
• •Bioinformatics analysis for true insect gut binding peptides optimized.
• •Insect gut binding peptides unique to insects from three different orders.

Abstract

Phage display libraries have been used to isolate insect gut binding peptides for use as pathogen transmission blocking agents, and to provide artificial anchors for increased toxicity of bacteria-derived pesticidal proteins. Previously, phage clones displaying enriched peptides were sequenced by Sanger sequencing. Here we present a streamlined protocol for identification of insect gut binding peptides, using insect-appropriate feeding strategies, with next generation sequencing and tailored bioinformatics analyses. The bioinformatics pipeline is designed to eliminate poorly enriched and false positive peptides, and to identify peptides predicted to be stable and hydrophilic. In addition to developing streamlined protocols, we also sought to address whether candidate gut binding peptides can bind to insects from more than one order, which is an important consideration for safe, practical use of peptide-modified pesticidal proteins. To this end, we screened phage display libraries for peptides that bind to the gut epithelia of two pest insects, the Asian citrus psyllid, Diaphorina citri (Hemiptera) and beet armyworm, Spodoptera exigua (Lepidoptera), and one beneficial insect, the western honey bee, Apis mellifera (Hymenoptera). While unique peptide sequences totaling 13,427 for D. citri, 89,561 for S. exigua and 69,053 for A. mellifera were identified from phage eluted from the surface of the insect guts, final candidate pools were comprised of 53, 107 and 1423 peptides respectively. The benefits of multiple rounds of biopanning, along with peptide binding properties in relation to practical use of peptide-modified pesticidal proteins for insect pest control are discussed.

Graphical abstract

1. Introduction

The phage-display library technology allows for the use of a large library of phages that display peptides with random amino acid sequences on their surface to identify peptides that bind immobilized ligand molecules (Smith, 1985). This approach has been widely used for many clinical and therapeutic applications in biomedical research (Smith and Petrenko, 1997; Bazan et al., 2012), with increasing use in recent years for insect-related applications. Specifically, peptides that bind to proteins on the surface of the insect gut, can be used to interfere with the binding of insect vectored pathogens of humans, notably malaria (Azzazy and Highsmith, 2002; Benhar, 2001), and also of plants. Gut binding peptides selected from a phage display library inhibit the entry of Plasmodium into the midgut of the mosquito vector (Vega-Rodríguez et al., 2014), or into the salivary glands (Ghosh et al., 2001). Similarly, a pea aphid gut binding peptide, GBP3.1 isolated by screening the same phage display library (Bonnycastle et al., 1996) reduced the uptake of a luteovirus into the hemocoel of the pea aphid vector, thus reducing virus transmission to healthy plants (Liu et al., 2010). An additional application has been use of phage display libraries for identification of protease inhibitors with insecticidal activity against aphids (Koiwa et al., 1998; Ceci et al., 2003).

Importantly, phage display screens hold considerable potential for increasing the efficacy of bacteria-derived pesticidal proteins against pestiferous insects. While these proteins have been successfully adopted for the management of insects of both public health and agricultural importance (Bravo et al., 2011), effective pesticidal proteins are only known for some pest species, and repeated use has resulted in resistance in some instances (Fabrick et al., 2020). The phage display approach has been used to improve the efficacy of Bacillus thuringiensis-derived pesticidal proteins against common lepidopteran and hemipteran pests, by use of two distinct approaches. In the first, phage display of variants of pesticidal protein domains is used to identify domain sequences with improved toxicity against a target species, as reviewed by Deist et al. (2014). For example, variants of Cry1Aa13 with increased insecticidal activity were identified by screening for increased binding of loop 2 of domain 2 to larval gut homogenates of Anopheles gambiae (Domínguez-Flores et al., 2017). In the second approach for use of phage display to increase the efficacy of bacteria-derived pesticidal proteins, a library of peptides with random sequences is screened against a target protein or tissue. The pea aphid gut binding peptide, GBP3.1 selected from a phage display library in this manner (Liu et al., 2010) was used to enhance the binding and toxicity of the cytolytic pesticidal protein, Cyt2Aa against aphid species (Chougule et al., 2013). A similar approach was used to increase the efficacy of Cry1Ab against the brown plant hopper, Nilaparvata lugens (Hemiptera) (Shao et al., 2013; Shao et al., 2013; Shao et al., 2016). The modification of pesticidal proteins with gut binding peptides holds particular promise for targeting hemipteran pests; relatively few pesticidal proteins with activity against Hemiptera are known. In addition, the same approach can theoretically be used for overcoming resistance to transgenic crops that express bacterial pesticidal proteins.

A key question for peptide modification of bacterial pesticidal proteins relates to the specificity of peptide binding. Ideally, an insect gut binding peptide would bind well to the gut epithelium of the targeted pest and perhaps closely related species, but not to the gut of other non-target and beneficial organisms, such as the honey bee. The specificity of peptide binding is therefore important to avoid potential alteration in host range of pesticidal proteins.

In all of the examples using phage display of peptides provided above, three or more rounds of biopanning were conducted, followed by Sanger sequencing of a limited number of phage clones to identify enriched sequences of interest. With the significant advances in sequencing technologies, sequencing to deduce the DNA sequences of peptides enriched during phage display screens has moved from Sanger sequencing of DNA from selected plaques, to Illumina sequencing of all eluted phage (AC't Hoen et al., 2012). As expected, this advance increased the sensitivity of the technique and a single round of bio-panning coupled with Next generation sequencing (NGS) is reportedly sufficient to identify positive clones (AC't Hoen et al., 2012). Concomitant with increased sequencing capacity, bioinformatics tools have been developed to identify enriched peptides that are false positives (AC't Hoen et al., 2012; He et al., 2019). The enrichment of peptides during screening of a phage display library can result from (1) selection for binding to the desired target or to non-target ligands such as plastic, and (2) enhanced phage propagation if a peptide sequence confers a competitive advantage to the phage (He et al., 2019; Matochko et al., 2014; Derda et al., 2011). The peptide HAIYPRH from the PhD-7 library (New England Biolabs) has been identified in at least 13 different biopanning experiments for example (Vodnik et al., 2011; Brammer et al., 2008). Thus, biopanning output is a mixture of true binding peptides and target-unrelated peptides (TUPs). TUPs are comprised of selection-related TUPs (SrTUPs) (Menendez and Scott, 2005) and propagation-related TUPs (PrTUPs) (Thomas et al., 2010). Several experimental (He et al., 2019; Matochko et al., 2014; Brinton et al., 2016) and in silico approaches (Huang et al., 2010) have been designed to identify and exclude putative TUPs from phage display biopanning data.

The objectives for this study, were to (1) streamline the use of phage display libraries for identification of insect gut binding peptides through development of a bioinformatics pipeline for analysis of peptide sequence data, and (2) address whether selected gut binding peptides can bind to insects from more than one order. To this end, we screened a phage display library for gut binding peptides in three insect species from different orders including two pests of agricultural importance - the Asian citrus psyllid, Diaphorina citri (Hemiptera), and beet armyworm, Spodoptera exigua (Lepidoptera), and a beneficial insect, the western honey bee, Apis mellifera (Hymenoptera). A bioinformatics pipeline was developed to remove false positive peptide sequences, and to identify optimal candidate insect gut binding peptides for further analysis.

2. Materials and methods

2.1. Feeding assays

All insects were fed the Ph.D.-C7C phage display library (New England BioLabs) using insect-appropriate feeding assays (Fig. 1). PhD-C7C contains randomized, loop-constrained heptapeptides at the N-terminus of the phage pIII coat protein.

Fig. 1.

Workflow for identification of insect gut binding peptides. Insects were fed on the Ph.D. C7C phage display library using methods appropriate to their feeding habit. Unbound phages were washed away from homogenates of whole insects (D. citri), or dissected guts (S. exigua, A. mellifera) followed by elution of bound phage. Eluted phage were amplified, ssDNA extracted and sequenced by Illumina MiSeq. Amplified phage were fed again to D. citri and A. mellifera for two and three rounds of enrichment for gut binding peptides respectively. (ssDNA gel labels: R1, round 1; R2, round 2; M, molecular mass markers; +ve control, naïve library from same lot).

2.1.1. Membrane feeding

To select phages displaying peptides that bind the gut of D. citri, 100 adults (<10 days old) were fed on 500 µl of 25% sucrose solution in TBS pH 7.5 containing 0.1% green food dye and 0.4% yellow food dye (McCormick & Co., Inc. Hunt Valley, MD) with 10¹¹ pfu/ml of PhD-C7C phage display library by membrane feeding (Fernandez-Luna et al., 2019). The adults were allowed to feed on the phage solution for 18 h at 28 °C and 60% RH. Whole psyllids rather than dissected guts were used for phage recovery due to the small size and delicate nature of the D. citri gut.

2.1.2. Droplet feeding

Thirty, second instar S. exigua larvae were raised on an artificial diet (Benzon Research Inc., Carlisle, PA) until fourth instar, after which the larvae were transferred individually to wells of a sterile six well plate and starved overnight. Each larva was then droplet fed with 10 µl of 30% sucrose solution in TBS pH 7.5 containing green food dye with 10¹¹ pfu/ml of PhD-C7C phages. After 1 hr, 25 of the larvae were dissected to recover the midgut, and the peritrophic membrane lining the gut lumen was discarded.

2.1.3. Force feeding

Newly emerged Apis mellifera (20) were hand fed with 10 µl of 30% sucrose solution (TBS, pH 7.5) containing 10¹¹ pfu/ml of PhD-C7C phages. After 1 hr, the midguts of 15 of the honey bees were dissected for recovery of bound phage.

2.2. Biopanning for selection of insect gut binding peptides

In vivo biopanning for phage selection was performed according to Liu et al. (2010), with a few modifications. In brief, the midguts of larvae (S. exigua), adults (A. mellifera), or whole insects (D. citri) were collected and suspended in 1 ml TBS pH 7.5 containing 1% BSA. Samples were homogenized using a polytron (Kinematica PT2500E) or hand-held homogenizer on ice for 5 min and centrifuged at 1000 g in a bench top centrifuge for 5 min at 4 °C. The supernatant was removed, and the pellet was resuspended in 1 ml TBST (TBS with 0.1% Tween-20) to wash off unbound phages. The sample was centrifuged as before, and the washing step repeated thrice. After the third wash, the supernatant was discarded and bound phages eluted from the pellet by adding 220 µl of elution buffer (50 mM glycine-HCl, pH 2.2, 1 mg/ml BSA), followed by incubation with gentle rotation in a Labquake™ shaker (Barnstead/Thermolyne, Duduque, IA) for 10 to 15 min at room temperature. The suspension was neutralized by adding 22 µl of 1 M Tris pH 9.1, followed by centrifugation at 1000 g for 5 min at 4 °C. The eluted phages were transferred to a 1.5 ml microcentrifuge tube. The number of recovered phages was determined by titering according to the phage display library manual, and the eluted phages immediately amplified in E. coli (ER2738) for sequencing and / or for a second round of enrichment. Amplified phages were titrated and used for additional rounds of bio-panning with two rounds conducted for D. citri and three for A. mellifera. A single round of phage enrichment was conducted for S. exigua.

For A. mellifera only, 40 to 50 individual clones were randomly selected from each round of eluted and amplified phage for Sanger sequencing by Genewiz (South Plainfield, NJ).

2.3. Amplification of eluted phage and DNA isolation (for NGS)

A 250 ml Erlenmeyer flask with 20 ml of LB was inoculated with 20 µl of eluted phage and 200 µl of an overnight culture of E. coli ER2738. The flask was shaken at 250 rpm, 37 °C for 4.5 hr. The culture was then centrifuged at 12,000 g for 10 min, 4 °C. The supernatant was transferred to a clean, sterile tube and centrifuged once more. The top 16 ml of supernatant was recovered for DNA isolation. Single-stranded DNA (ssDNA) was extracted from phage eluates from rounds 1 (R1), 2 (R2) and 3 (R3) for A. mellifera, R1 and R2 for D. citri, and a single round (R1) for S. exigua, following the EZNA M13 mini kit protocol (Omega BioTek, Norcross, GA, USA). The ssDNA samples from each round were run on a 1% agarose gel with a naïve PhD-C7C library from the same lot used as control.

2.4. PhD-C7C naïve reference library

Twenty µL of a 1:1000 dilution of the PhD-C7C library was used to create a reference “naïve” library. This aliquot was taken from the same lot number as the library used for in vivo bio-panning for each insect to avoid bias due to potential variation between lot numbers (Brinton et al., 2016). Phages in the aliquot were amplified, and ssDNA extracted for deep sequencing as described above for eluates from in vivo biopanning experiments. NGS was performed by Genewiz.

2.5. TargetGxOne amplicon sequencing

TargetGxOne amplicon sequencing libraries were prepared using a 2-step PCR workflow. First, PCR was used to amplify the template using primers to the region of interest with overhang Illumina adapters. The following primers were designed based on the M13 phage genome: Forward primer: 5′ CCGATACAATTAAAGGCTCC 3′; Reverse primer: 5′ GTATGGGATTTTGCTAAACAAC 3′. The amplicon (212 bp) was purified away from free primers and primer dimers using magnetic beads. Indices and Illumina sequencing adapters were then added by limit cycle PCR (first round, 9 cycles; second round, 6 cycles). The sequencing libraries were validated on the Agilent TapeStation (Agilent Technologies, Palo Alto, CA, USA), and quantified using a Qubit 2.0 Fluorometer (Invitrogen, Carlsbad, CA), as well as by quantitative PCR (KAPA Biosystems, Wilmington, MA, USA). DNA libraries were multiplexed in equal molar mass and loaded on an Illumina MiSeq instrument according to the manufacturer's instructions (Illumina, San Diego, CA, USA). Sequencing was performed using a 2 × 150 paired end (PE) configuration; image analysis and base calling were conducted by the MiSeq Control Software (MCS) on the MiSeq instrument.

2.6. Sequence data analysis

Raw data and total unique nucleotide counts were provided by Genewiz. In brief, raw sequence data were demultiplexed using bcl2fastq version 2.17.1.14. Read pairs were trimmed for adapter sequences and low quality base calls using Trimmomatic version 0.36 (Bolger et al., 2014); reads less than 30 bases long were discarded. The target sequence between conserved flanking primers was extracted from each merged pair: When forward and reverse reads overlapped, the target overlapping region was identified by the reformat function within bbmap and each read pair was merged using the bbmerge tool (Bushnell et al., 2017) (bbtools software toolkit). Reads that could not be merged were discarded from further analysis. For each sample, an Excel file listing unique nucleotide sequences and their counts was generated by Genewiz using bash and perl scripts. Further analyses for identification of consensus sequences was conducted: Nucleotide sequences were converted to amino acids (aa) and unique 7 aa sequences extracted with the number of sequences counted with in-house Python3 scripts (Supplementary Files). Before normalization all environmental phage sequences (i.e. wild type phage without C7C peptide inserts) were removed from the naïve library, R1, R2 and R3 as applicable. The peptide counts were normalized to the total number of peptides: Percent normalization = (Total count of a particular aa seq/ Total count of all aa seqs) x 100. The log₂ fold change of R1 and R2 eluate peptides relative to naïve peptides was calculated using R script (Tidyverse_1.3.0). Peptides with log₂ fold change ≥5 between naïve and R1, and naïve and R2 were selected for further analysis.

The SAROTUP version 3 TUPScan tool (Huang et al., 2010; He et al., 2016; Huang et al., 2012) was used to predict whether sequenced peptides had a target-unrelated peptide (TUP) motif. The PSBinder tool in the SAROTUP suite was used to predict whether the identified peptides are polystyrene surface-binding peptides. The MimoBlast tool in the SAROTUP suite was also used to compare the gut binding peptides against those in the MimoDb (Huang et al., 2012) database containing TUP sequences. Highly similar peptides obtained with different binding targets might also be TUPs (Thomas et al., 2010; He et al., 2016; Huang et al., 2012). Instability indices and GRAVY scores were calculated for peptides that passed the SAROTUP suite filter test using the Peptides version 2.4.2 (R package). Peptides were classified into four groups based on their instability index and GRAVY scores: stable and hydrophilic, stable and hydrophobic, unstable and hydrophilic, and lastly unstable and hydrophobic. Peptides with an instability index of < 40 and GRAVY score < 0 were selected for clustering analysis.

2.7. Clustering analyses of candidate gut binding peptides

Gibbs clustering (Andreatta et al., 2013) analyses (GibbsCluster-2.0 Server) were performed on all peptide sequences that had passed the SAROTUP suite filter test, were stable and hydrophilic, and had a log2 fold change ≥5. This analysis identifies motifs based on aa sequence identity and biochemical properties. The Gibbs clustering algorithm groups peptide sequences into clusters and determines the optimum local sequence alignment for each cluster.

3. Results

3.1. Screening of PhD-C7C for isolation of insect gut binding peptides

To select for gut binding peptides, D. citri, S. exigua and A. mellifera were fed with the PhD-C7C phage display library using insect-appropriate methods. Fig. 1 provides the workflow from feeding of insects on the phage display library to Illumina sequencing of extracted single stranded DNA of phage eluates. Following NGS of amplified, eluted phage. The numbers of unique peptides common between species in different rounds of biopanning, and common between rounds for a given species were determined (Fig. 2A and B). A total of 1158 peptides were common to all three rounds of biopanning for A. mellifera and 951 peptides were common to the first and second rounds for D. citri (Fig. 2B). The numbers of peptides exclusive to rounds 1, 2 and 3 for A. mellifera were 52,756, 8840 and 2584 respectively highlighting the diminishing number of additional unique peptides from successive rounds of phage enrichment. Finally, 13,427 unique peptides sequences from D. citri, 89,561 from S. exigua, and 69,053 from A. mellifera were identified from all rounds of in vivo biopanning (Fig. 2C). None of these peptides were common to all three species. However, 173, 23 and 3 peptides were common to D. citri and S. exigua, S. exigua and A. mellifera, and D. citri and A. mellifera, respectively (Fig. 2C).

Fig. 2.

Total unique and selected candidate gut binding peptides resulting from bioinformatics pipeline across three species. The total number of unique peptides identified from each phage display library screen, including the number of peptides common to two species in a given round (A), and between rounds of a given species (B) are indicated. No peptide sequences were common to all three phage library screens (C). The numbers of candidate peptides selected for further investigation following SAROTUP suite, stability and hydrophilicity analyses of sequences from each screen (bioinformatics pipeline) are shown (C). Importantly, none of the candidate peptides of interest are common to more than one species.

3.2. Bioinformatics pipeline to identify candidate gut binding peptides

To analyze the thousands of peptides identified from eluted phage, we designed a bioinformatics pipeline to select the best candidate gut binding peptides (Fig. 3). The first part of the pipeline identifies the enriched gut binding peptides (true binders) and reduces the PrTUPs (false positives) from further experimental analysis. First, peptide sequences were determined along with the number of times each peptide sequence appeared in the biopanning experiment (peptide count). Each peptide's count was then normalized by calculating the percentage contribution to the total peptide count in the sample (Fig. 3). This step removes discrepancies generated due to differences in number of reads from independent screens when samples are run on different lanes. The increase in peptide count (fold-change) of each peptide was normalized relative to the naïve (control) library (Fig. 3). This normalization removes bias that could originate from differences in bacterial amplification or in unequal distribution of phage clones within the naïve library (Matochko et al., 2014; Brinton et al., 2016). Highly enriched peptides with log₂ fold change ≥ 5 were selected. The second part of the pipeline predicts and eliminates SrTUPs and PrTUPs from NGS biopanning datasets with log₂ fold change ≥ 5 using the SAROTUP suite (Fernandez-Luna et al., 2019; He et al., 2016; Huang et al., 2012) (Fig. 3). SAROTUP contains three categories of tools: 1) motif based (TUPScan), 2) machine learning based (PhD7Faster, SABinder and PSBinder) and 3) database search based (MimoSearch, MimoBlast and MimoScan). The Peptides R package was used to predict the instability index and grand average of hydropathy (GRAVY).

Fig. 3.

Bioinformatics pipeline for identification of optimal gut binding peptides. First, peptides that were highly enriched during biopanning are identified. Second, the SAROTUP suite of tests is used to exclude peptides resulting from propagational bias and selection-related false positives.

Analysis with our bioinformatics pipeline led to the identification of 728, 638, and 7281 peptides with log₂ fold change ≥ 5 in D. citri, S. exigua, and A. mellifera, respectively (Table 1). Further analysis of these peptides with the SAROTUP suite to predict PrTUP or SrTUP sequences led to the removal of 66%, 58.3% and 54.8% of peptides from the D. citri, S. exigua, and A. mellifera datasets, respectively (Table 1). For A. mellifera, 27% of the R2 peptides (391) were also present in R1, 35% R3 peptides (236) were present in R1, and 86% R3 peptides (577) were present in R2.

Table 1.

Numbers of enriched peptides from phage display library screens that passed SAROTUP analyses. Numbers of peptides derived from each round of each screen are indicated.

Insect	Total unique peptides per round a			Unique peptides with fold change ≥5 a				Passed SAROTUP suite analysis (% b)
	R1	R2	R3	R1	R2	R3	Total
D. citri	10,343	4035	NA	667	63	NA	728	247 (34)
S. exigua	89,561	NA	NA	638	NA	NA	638	266 (41.7)
A. mellifera	55,608	13,489	5987	6206	1466	669	7281	3291 (45.2)

^aSome peptides common between rounds.

^bPercentage was calculated with respect to number of peptides with ≥5 fold change relative to naïve library; NA, not applicable.

3.3. Properties of candidate peptides

Instability index and GRAVY scores were determined for the subset of peptides that passed the SAROTUP filtering step (Fig. 4). The majority (58 to 68%) of candidate gut binding peptides from all three species were hydrophilic. The majority of candidate peptides were stable for S. exigua (59%) and A. mellifera (58%), but not for D. citri (41%). The “stable and hydrophilic” category indicating stability in solution, contained 21, 40 and 43% of peptides analyzed in D. citri, S. exigua and A. mellifera, respectively. This category comprised of 53 peptides for D. citri, and 107 for S. exigua (Fig. 2C) is of primary interest for use in pest control.

Fig. 4.

Predicted properties of candidate gut binding peptides. Predicted instability and GRAVY scores of peptides that passed SAROTUP suite tests as determined by use of the Peptides R package (version 2.4.2). Image generated using ggplot2 version 3.3.2.

Gibbs cluster analysis was used to form groups of candidate peptides based on their biochemical properties and sequences. Candidate gut binding peptides from D. citri, S. exigua, and A. mellifera, were clustered into 3, 5, and 3 groups respectively with the majority of amino acids have a binding affinity for metals (Yamashita et al., 1990) (data not shown).

3.4. Analysis of final set of D. citri gut binding peptides of interest

Analysis of the fold change of peptides selected for D. citri gut binding that passed SAROTUP analysis, and were stable and hydrophilic is shown in Fig. 5. In Fig. 5, Peptides 1 to 3 were highly enriched from round 1 to round 2 of biopanning, while peptides 4 to 13 had higher fold enrichment in round 1 of biopanning (Fig. 5). Peptides that were present in both rounds, passed the bioinformatic screening process, were stable and hydrophilic, and belonged to different Gibbs cluster groups were prioritized for further analysis.

Fig. 5.

Enrichment of D. citri gut binding peptides by round of biopanning. Peptides that passed SAROTUP analysis and were stable and hydrophilic from rounds 1 (R1) and 2 (R2) of the D. citri screen are shown. The more red the color, the higher the log₂ fold change. Peptides 1–3 were highly enriched in R2, while peptides 4–13 were highly enriched in R1 and to a lesser extent in R2. Image generated using ggplot2 version 3.3.2.

3.5. Selected gut binding peptides were unique to each insect

None of the selected peptides were common across the three species examined (Fig. 2). Of the peptides common among pairs of species (Fig. 2) 2 (67%), 99 (56.6%), and 21 (87.5%) were predicted to be TUPs (false positives) between D. citri and A. mellifera, D. citri and S. exigua, and A. mellifera and S. exigua respectively. The remainder of the peptides common among pairs of species were poorly enriched (i.e. fold change ≤ 5).

3.6. Comparison of classical and NGS approaches for phage sequencing

For A. mellifera, the classical approach of using Sanger sequencing for a limited number of phages was also used for comparison with NGS results. Table 2 outlines the number of phages identified after each round of biopanning, and the proportion that were also detected by NGS from the same screen. A total of 67 unique peptide sequences were identified from the three rounds of biopanning. Highlighting the sensitivity of use of NGS for peptide sequencing, of the 67 unique peptides identified from Sanger sequencing of selected phage clones from the A. mellifera screen, 26 were predicted to be true binders. Of these, five were predicted to be stable and hydrophilic, compared to 1423 from the NGS data (Fig. 2). Interestingly, some R1 and R2 peptides identified by Sanger sequencing were not identified by NGS. The use of sequencing with greater depth (HiSeq or NextSeq) is expected to result in identification of more peptide sequences than the use of MiSeq.

Table 2.

Comparison of A. mellifera gut binding peptides identified from Sanger sequencing of a limited number of phage clones, and by NGS (MiSeq) of all eluted phage.

	Phage clones sequenced	No. unique peptides	No. present in NGS data	% present in NGS data
R1	40	28	6	21.4
R2	50	30	23	76.7
R3	50	9	9	100

4. Discussion

Our goal was to optimize methodology for identification of insect gut binding peptides, and to address whether gut binding peptides of interest generated by our bioinformatics pipeline could bind to insects from more than one order. We focused on use of the Ph.D.-C7C phage display library, which has provided better results in our hands than the Ph.D.−12 library (New England BioLabs), which displays random 12-mer peptides. Our bioinformatics pipeline differs from prior workflows (AC't Hoen et al., 2012; Brinton et al., 2016), as (1) it is based on peptide enrichment (rather than peptide counts) and (2) incorporates SAROTUP, stability and hydrophilicity analyses. This pipeline allows for elimination of peptides that are poorly enriched or false positives, and for identification of stable, hydrophilic gut binding peptide candidates for further analysis. As false positive peptides are eliminated prior to validation experiments, the bioinformatics pipeline avoids the herculean task of experimentally validating thousands of candidate peptides.

While screening of phage display libraries for binding to insect-derived brush border membrane vesicles (BBMV) (Wolfersberger et al., 1987), or to recombinant gut surface proteins is also feasible, our preference is to conduct in vivo screens followed by gut dissection whenever possible. Although enriched in gut surface proteins, BBMV also contain intracellular proteins (Bayyareddy et al., 2012), which result in the selection of an additional category of false positives. Analysis of peptide binding partners and elimination of peptides that bind intracellular proteins would be required. The selection of peptides that bind to specific, abundant gut surface proteins requires appropriate expression (e.g. baculovirus expression) and purification of the recombinant protein prior to the screen, which is considerably slower than conducting an in vivo screen for gut binding peptides. In vivo screens with three rounds of biopanning can be completed within a two-week period.

None of the peptides in the final set were common to more than one of the tested species. This specificity has important implications for the use and regulation of peptide-modified pesticidal proteins for crop protection and other field applications. Based on the binding properties of GBP3.1, which binds aminopeptidase N (APN) of the pea aphid and other aphid species (Linz et al., 2015), peptides selected for binding to the gut of a given species are predicted to bind to the guts of related species, but less strongly. The ability of the peptide to bind and the strength of binding will relate to homology between the binding partner protein (e.g. APN) in the gut epithelium of the target insect compared to that of other insect species. As gut protein sequences are less conserved between orders than within orders, the lack of common peptides in the final pools of candidate peptides across the three species tested is perhaps not surprising (Likitvivatanavong et al., 2011; Shao et al., 2018). Identification of gut binding peptides from the honey bee will provide a useful reference dataset, although empirical testing for peptide-modified pesticidal proteins in the honey bee will be required to ensure safety.

The use of a single round of biopanning has been proposed, to avoid propagation-related bias in subsequent rounds of phage enrichment. Use of a single round screen can be an effective strategy to reduce false positives (Olson et al., 2012; Rentero Rebollo et al., 2014). However, Brinton et al., found that screens conducted on some ex vivo tissue samples required an extra round of biopanning due to a high degree of non-specific binding in round 1 (Brinton et al., 2016). In our case, a single round of biopanning for S. exigua, resulted in a greater number of candidate gut binding peptides than for the other two species (Fig. 2). This result could be attributed to the feeding behavior of lepidopteran larvae, with the relatively large gut functioning to process large amounts of plant material (Grant, 2006). Greater complexity of the lepidopteran gut surface proteome relative to that of hemipteran or hymenopteran insects, could contribute to the increased number of gut binding peptides observed. Alternatively, non-specific binding could contribute to the high number of peptide candidates seen in this screen. It is notable that despite the higher number of peptides identified from the S. exigua R1, there were 92.5% fewer peptides in the final pool than for A. mellifera (Fig. 2).

For A. mellifera in the present study, 86% of the peptides with log₂fold change ≥ 5 enriched in R2 were also enriched in R3 (data not shown). However only 27% of the peptides with log₂fold change ≥ 5 in R1 were also enriched in R2. Similarly, ‘tHoen et al., (AC't Hoen et al., 2012) found an 80% correspondence in the top 1000 most abundant peptides in R3 and R4. This suggests that, with the depth of NGS, two rounds of biopanning is sufficient for peptide identification, with relatively few additional peptides appearing in subsequent rounds. As illustrated in Fig. 5, D. citri peptides 1 to 3 would not have been identified had the screen been limited to a single round of biopanning.

Gut binding peptides predicted to be both stable and hydrophilic are of the greatest interest for practical application in insects, including modification of bacteria-derived pesticidal proteins and blocking the interaction of microbes with their target gut receptors. In both cases, peptides bind to gut surface proteins. Hydrophilicity reduces the likelihood of the peptide crossing the hydrophobic lipid bilayer of the plasma membrane (Prochiantz, 1996), with hydrophilic substrates of >700 Da unable to diffuse through protein channels (Etz et al., 2001). While the stability and hydrophilicity profiles of candidate peptides identified from S. exigua and A. mellifera were similar, the proportion of stable, hydrophilic peptides isolated from D. citri was about half that of the other species. Whether this results from use of the whole insect rather than dissected gut for the phage display screen, remains an open question.

Having identified a pool of candidate gut binding peptides for a species of interest, the next steps would be to confirm binding by for example, pull down assay with gut-derived BBMV, and identification of the primary binding partner (Chougule et al., 2013). The binding partner of GBP3.1, APN is highly abundant in the pea aphid gut (Cristofoletti et al., 2006). Indeed, we expect the most enriched peptides to bind the most abundant gut surface proteins, or the proteins that are most abundant in the anterior midgut in species with highly efficient protein digestion systems. Pathogens of insects, or pathogens that use insects as vectors are highly likely to exploit the most abundant proteins for binding or entry into the host or vector. In support of this, some of the most abundant gut surface proteins such as APN, alkaline phosphatase (Gómez et al., 2007), and ABC transporters (Sato et al., 2019; Banerjee et al., 2017) are receptors for Bacillus thuringiensis – derived pesticidal proteins, and APN serves as a receptor for Pea enation mosaic virus (PEMV) in the pea aphid vector (Liu et al., 2010; Linz et al., 2015). As demonstrated previously (Liu et al., 2010; Chougule et al., 2013), the same insect gut binding peptide can serve to block microbe-insect interaction as well as enhance the efficacy of bacterial pesticidal proteins.

5. Conclusion

We have developed an efficient protocol that couples phage display and next generation sequencing with a bioinformatics pipeline to identify candidate insect gut binding peptides with desirable properties. Such peptides have application as pathogen transmission blocking agents, and for increased binding of bacterial pesticidal proteins to enhance toxic action.

Availability of data

Complete peptide sequence datasets are provided for the S. exigua phage display library screen, while patentable sequences have been removed from the D. citri and A. mellifera datasets. Data and codes reported in this manuscript are available at the following sites:

Codes: https://github.com/ruchirjd/CRIS_Phage_display_2021

Raw data, S. exigua: https://doi.org/10.6084/m9.figshare.13667639.v2

Log fold change data for S. exigua: https://doi.org/10.6084/m9.figshare.13728304.v2

Datasets for all three species: https://doi.org/10.6084/m9.figshare.13667630.v1

Supplementary Files

Exctractaaseq.py: Python code to extract nucleotide sequence and convert to amino acid sequence.

Percentcount.py: Python code to calculate count and percentage normalization.

Processaaseq.py: Code to extract 7 amino acid sequences.

FoldChange.R: Tidyverse code to calculate log₂ FoldChange.

Credit author statement

Ruchir Mishra: Conceptualization, Methodology, Software, Formal analysis, Data curation, Writing – original draft, Writing – review and editing, Visualization Ya Guo: Formal analysis, Investigation, Data curation, Writing – original draft, Visualization Pavan Kumar: Investigation

Pablo Emiliano Cantón: Investigation  Clebson Dos Santos: Investigation, Writing – original draft  Rahul Banerjee: Investigation, Writing – original draft, Visualization  Suyog Kuwar: Conceptualization, Investigation  Bryony Bonning: Conceptualization, Resources, Writing – review and editing, Visualization, Supervision, Project administration, Funding acquisition

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.