The PDB database was queried for the keyword ‘plants’, and proteins with the exact same sequences were removed. This resulted in a set of ~2000 proteins (see list.plants.txt in Dataset 1). These proteins were analyzed using DSSP ³⁵ to identify the AHs, and AHs with the same sequence were removed. This resulted in ~6000 AHs (see ALPHAHELICES.zip in Dataset 1). PAGAL was applied to this set of AHs (see RawDataHelix.txt in Dataset 1). This data was refined to obtain peptides with different characteristics. We also computed the set of all pairs of AHs that are connected with a short (less than five residues) loop (see HTH in Dataset 1). This set is used to extract a pair of AHs, such that one of them is cationic with a large hydrophobic moment, while the other comprises mostly of hydrophobic residues. The PAGAL algorithm has been detailed previously ²⁰ . Briefly, the Edmundson wheel is computed by considering a wheel with centre (0,0), radius 5, first residue coordinate (0,5) and advancing each subsequent residue by 100 degrees on the circle, as 3.6 turns of the helix makes one full circle. We compute the hydrophobic moment by connecting the center to the coordinate of the residue and give it a magnitude obtained from the hydrophobic scale (in our case, this scale is obtained from Jones et al. ¹⁹ ). These vectors are then added to obtain the final hydrophobic moment. The color coding for the Edmundson wheel is as follows: all hydrophobic residues are colored red, while hydrophilic residues are colored in blue: dark blue for positively charged residues, medium blue for negatively charged residues and light blue for amides. All protein structures were rendered by PyMol ( http://www.pymol.org/). The sequence alignment was done using ClustalW ³⁶ . The alignment images were generated using Seaview ³⁷ . Protein structures have been superimposed using MUSTANG ³⁸ .

Synthesized chemical peptides were obtained from GenScript USA, Inc. The protein molecular weight was calculated per peptide then diluted to 2000 µM or 3000 µM stock solutions with phosphate buffered saline. Stock solutions were stored in -20°C and thawed on ice before use.

Using the stock solutions, we made dilute solutions of 300 µM, 250 µM, 200 µM, 150 µM, 100 µM, 75 µM, 50 µM, 30 µM, 25 µM, and 10 µM to a final volume of 100 µl of phosphate buffered saline. Dilute peptide solutions were stored in -20°C and thawed on ice before use.

Xylella fastidosa 3A2 (PD3) ³⁹ , Xanthomonas arboricola 417 (TYS) ⁴⁰ , and Liberibacter crescens BT-1 (BM7) ⁴¹ media were prepared and autoclaved at 121°C for 15–30 minutes, then cooled and poured into 100 × 15mm sterile petri dishes. Kanamycin (50 µg/ml) was added to PD3.

Bacteria were inoculated and allowed to grow in liquid medium at 28°C: Xf (5 days), Xa (3 days), and Lc (3 days) to reach the exponential phase. The inoculum was diluted to a working OD of 0.5 (1×10 ⁷ cells/ml). 10 µl of the OD 0.5 was plated with 90 µl of liquid media and spread on the pre-made agar plates to create a confluent lawn of bacteria. The bacteria were given an hour to set at room temperature. 10 µl of each peptide concentration was spotted onto a plate of agar preseeded with a layer of bacterium. After spotting the plates were incubated at 28°C for 2 to 10 days till zones of clearance were clearly visible and the plates were scored for the minimum inhibitory concentration (MIC) as that beyond which no visible clearance was observed. Data presented is in triplicate, and were identical.

Cecropin B (CECB) was used as a positive control, as it is known to target membrane surfaces and creates pores in the bacterial outer membrane ^{30, 31} . CECB consists of an cationic amphipathic N-Terminal with a large hydrophobic moment ( Figure 1a), and a C-Terminal comprising mostly of hydrophobic residues, which consequently has a low hydrophobic moment, ( Figure 1b) joined by a short loop. Another positive control was a linear AH-AMP consisting of the residues 2-22 of the N-Terminal in CECB (CBNT21) ( Figure 1a). The sequences of these are shown in Table 1.

Linear AH-AMPs . In order to choose a peptide mimicking CBNT21 (cationic, amphipathic, large hydrophobic moment), we directed our search to ‘locate a small peptide with a large hydrophobic moment and a high proportion of positively charged residues’ on the raw data computed using PAGAL (See RawDataHelix.txt in Dataset 1). A small peptide is essential for quick and cost effective iterations. Table 2 shows the best matching AHs. Next, we used the sequence of these AHs to search the grapevine and citrus genomes, choosing only those that are present in both genomes. This allowed us to locate an AH from phosphoenolpyruvate carboxylase from sunflower, a key enzyme in the C4-photosynthetic carbon cycle which enhances solar conversion efficiency (PDBid:3ZGBA. α11) ³³ . Figure 2a shows the specific AH located within the protein structure, marked in green and blue. Although DSSP marks the whole peptide stretch as one AH, we chose the AH in blue due to the presence of a small π helix preceding that. We named this peptide PPC20 ( Figure 2, Table 1). This peptide is fully conserved (100% identity in the 20 residues) in both grapevine (Accession id:XP_002285441) and citrus (Accession id:AGS12489.1). Figure 2b,c shows the Pymol rendered AH surfaces of PPC20. The Asp259 stands out as a negative residue in an otherwise positive surface ( Figure 2c). Since previous studies have noted dramatic transitions with a single mutation on the polar face, it would be interesting to find the effect of mutating Asp259 to a cationic residue ⁴² .

Non-linear AH-AMPs consisting of two AHs. Next, we located two AHs within chitinase from Nicotiana tobaccum (PDBid:3ALGA. α4 and 3ALGA. α5) ³⁴ connected by a short random coil such that one of the AHs is cationic and hydrophobic, while the other AH is comprised mostly of hydrophobic, uncharged residues (CHITI25, Figure 3a, Table 1). This peptide mimics the complete CECB protein ( Figure 3b). While the properties of the AHs in CHITI25 is reversed from that of CECB, the order in which these AHs occur is not important for functionality. The multiple sequence alignment of CHITI25 from grapevine, citrus and tobacco is shown in Figure 3c. CHITI25 from tobacco is the most cationic (five), followed by citrus (four) and grapevine (three). Thus, it is possible that the anti-microbial properties of CHITI25 from grapevine would be lower than CHITI25 from tobacco. These peptides can be subjected to mutations to enhance their natural anti-microbial properties in such a scenario ⁴³ .

Negative control - an anionic AH-AMP. We also located an anionic AH-AMP using a similar strategy - a 13 residue peptide present within the structure of isoprene synthase from gray poplar (PDBid:3N0FA. α18) ⁴⁴ . We also used phosphate buffered saline as a negative control. We have extended this helix on both terminals by including one adjacent residue from both terminals to obtain ISS15 ( Table 1).

We have validated our peptides using plating assays ( Table 3, Figure 4). CECB, the well-established AH-AMP, is the most potent among all the peptides tested, having minimum inhibitory concentrations of between 25 µM (for Xa) to 100 µM (for Xf and Lc). This shows the variations in susceptibilities of different organisms. Understanding this differential susceptibility would require a deeper understanding of the underlying mechanism by which these AH-AMPs work ⁴⁵ , as well as the difference in the membrane composition of these gram-negative pathogens ⁴⁶ . Mostly, CBNT21 has a slightly lower potency, indicating a role for the C-terminal AH in CECB, which comprises of mostly hydrophobic residues for Xf and Lc. This results corroborates a plausible mechanism suggested by others in which the anionic membranes of bacteria is targeted by the cationic N-terminal, and followed by the insertion of the C-terminal AH into the hydrophobic membrane creating a pore. PPC20 and CHITI25 have comparable potencies with CECB and CBNT21, although Lc appears to be resistant to CHITI25. Finally, the anionic peptide used as a negative control shows no effect on these pathogens.

Alternate computational methods for finding such new AMPs based on known AMPs could be of two kinds, although neither method is as effective in obtaining our results. Firstly, a sequence search using BLAST can be done to find a corresponding peptide in the genome, say for cecropin B. However, a BLAST of the cecropin sequence does not give any significant matches in the grapevine or citrus genomes, and is a dead end. In principle, what we need is a peptide with cecropin B like properties - and that information is not encoded in the linear sequence, but in the Edmundson wheel of the AH. The second method for such a search is to find structural homology in the PDB database through a tool like DALILITE ⁴⁹ . However, AHs are almost indistinguishable structurally, and the results will give rise to many redundancies. Thus, there are no existing methods tailored to incorporate the quantifiable properties of AHs in the search. We, for the first time, have proposed such a method in SCALPEL.

Computer-assisted design strategies have also been applied in designing de novo AMPs ^{50, 51} . Other hand curated comprehensive databases for ‘for storing, classifying, searching, predicting, and designing potent peptides against pathogenic bacteria, viruses, fungi, parasites, and cancer cells’ ⁵² do not enjoy the automation and vastness of available data elucidated in the SCALPEL methodology.

There are several caveats to our study. We are yet to ascertain the hemolytic nature of the identified peptides, and will be performing these experiments in the near future. In fact, the selective cytotoxicity against human cancer cells, might be used as a substitute therapy in place of conventional chemotherapy ^{53, 54} . It must be noted that the development of a selective peptide with anti-cancer cell properties has been a challenge ⁵⁵ . Although, we have not measured the lipid permeabilizing abilities of our peptides, a recent study has found that potency in permeabilizing bacteria-like lipid vesicles does not correlate with significant improvements in antimicrobial activity, rendering such measurements redundant ⁵⁶ . The electrostatic context of an peptide is known to have a significant bearing on its propensity to adopt an AH structure. The ability to predict the folding of peptides requires significant computational power and modelling expertise ⁵⁷ . Peptides often remain in random coil conformations, and achieve helical structures only by interacting with anionic membrane models ⁵⁸ . It is also possible to measure peptide helicity through circular dichroism spectroscopy ⁵⁹ . However, our results have been all positive based on selected choices of peptides arising from our search results, and suggest a high likelihood of getting anti-microbial activity from these peptides. Additionally, we may have to resort to other innovative techniques that have been previously adopted to overcome thermodynamic instability or proteolytic susceptibility ^{60– 63} .