Biopanning bacterial display libraries is a proven technique for peptide affinity reagent discovery for recognition of both biotic and abiotic targets. Peptide affinity reagents can be used for similar applications to antibodies, including sensing and therapeutics, but are more robust and able to perform in more extreme environments. Specific enrichment of peptide capture agents to a protein target of interest is enhanced using semi-automated sorting methods which improve binding and wash steps and therefore decrease the occurrence of false positive binders. A semi-automated sorting method is described herein for use with a commercial automated magnetic-activated cell sorting device with an unconstrained bacterial display sorting library expressing random 15-mer peptides. With slight modifications, these methods are extendable to other automated devices, other sorting libraries, and other organisms. A primary goal of this work is to provide a comprehensive methodology and expound the thought process applied in analyzing and minimizing the resulting pool of candidates. These techniques include analysis of on-cell binding using fluorescence-activated cell sorting (FACS), to assess affinity and specificity during sorting and in comparing individual candidates, and the analysis of peptide sequences to identify trends and consensus sequences for understanding and potentially improving the affinity to and specificity for the target of interest.

Biopanning is a proven technique for affinity reagent discovery, with bacterial display libraries a convenient source for peptide affinity reagents ^{1 , 2 , 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 , 12 , 13 , 14 , 15 , 16 , 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24} . Similarly to phage display ^{25 , 26} and yeast display ^{27 , 28} , the peptides are exposed on the cell surface and are available to bind to both biotic and abiotic targets, with these interactions driven by both the peptide backbone and the unique properties of the amino acid side-chains ²⁹ . In contrast to phage display, the bacteria themselves are self-replicating and contain all genetic material required for display without elution of bound phage and reinfection of cells. In contrast to yeast display, bacterial display is faster due to the rapid (approximately 20 min ³⁰ ) doubling time of Escherichia coli . The resulting peptide affinity reagents can be produced off-cell for use in a variety of sensing platforms ^{16 , 17 , 26} and have potential for use as living materials when displayed on-cell ^{2 , 31 , 32} . As compared to monoclonal antibodies for recognition of specific targets, peptides are much smaller and more flexible, and peptide display libraries can be easily manipulated at the DNA level to avoid specific amino acids, such as cysteine ³³ . Peptides are increasingly exploited for therapeutics ^{34 , 35 , 36} and other applications where antibodies would typically be utilized due to their ease of discovery, reproducible synthetic (off-cell) production, and superior maintenance of performance in extreme environments, providing a functional reagent even after exposure to high temperature or long-term storage without refrigeration ^{37 , 38} . The ability to overcome the cold chain for storage of reagents is of particular interest to the Department of Defense, for instance, which must operate in extreme environments. Thermal stability was therefore a desirable feature for the affinity reagents recognizing the chosen targets given as examples in this manuscript.

Recently, biopanning bacterial display libraries has become even more straightforward and reliable with the use of semi-automated magnetic-activated cell sorting (MACS or MCS) methods ^{9 , 14 , 39} . These methods have been demonstrated for biological targets using several similar sorting platforms with different advantages, including a commercially available automated magnetic-activated cell sorting (autoMACS or autoMCS) instrument (see table of materials) ^{1 , 14 , 15} and more specialized devices which enclose the sample in a disposable cartridge ^{7 , 9} or chip ³⁹ , a valuable specification for use with organisms or toxins that are Biosafety Level 2 (BSL-2) or higher ⁷ . These semi-automated methods could be extended to sort for peptides able to bind to abiotic materials if the materials are magnetic or have the ability to be coated onto a magnetic bead, and for use with different organisms (such as anaerobic bacteria) if the sorting and growth conditions are altered. In addition to sorting bacterial display libraries, the commercial autoMCS instrument used here has also been used in part for the initial steps of discovery of human single-chain antibody fragments displayed on the surface of yeast, followed by further isolation using Fluorescent Activated Cell Sorting (FACS) ⁴⁰ . The primary advantages to semi-automation are decreased sorting time and effort, and more robust and reproducible elimination of unbound cells from the magnetic beads during wash steps, leading to reduced false positives, fewer required sorting rounds, and less complicated down-stream analysis ^{9 , 14} . In the case of the commercial autoMCS instrument, there are multiple pre-loaded programs which may be optimal for different applications, such as negative pre-sorting (depletion), prioritizing purity, minimizing final sample volume, and increasing or decreasing sensitivity for isolation of rare cells ⁴¹ .

The focus of this work is to demonstrate the methodology for biopanning a bacterial display library using the commercially available autoMCS instrument, and to expound the thought process applied in analyzing and minimizing the resulting pool of candidates in order to facilitate extension to other applications. The display library used here (see table of materials) ^{12 , 13} contains approximately 10 ⁹ – 10 ¹¹ unique 15-mer peptides displayed via a modified version of the bacterial outer membrane protein OmpX in an unconstrained nature at the cell surface, which is expected to be representative of the free peptide in solution if produced synthetically. This protein scaffold additionally contains a positive control P2X peptide at the C-terminus for monitoring expression via binding to YPet Mona. However, a purchased or novel library with suitable diversity and other characteristics necessary for the specific application desired should work similarly with this biopanning procedure, though some minor changes may be required. A schematic of this biopanning protocol is illustrated in Figure 1 . Additionally, the protocol could be adapted to other automated magnetic sorting platforms and other sorting strategies. Manual magnetic sorting with a benchtop magnet has been successfully demonstrated, albeit with more complicated and time consuming downstream analysis required due to increased false positives ¹⁴ , and nonetheless provides an alternative option to the autoMCS steps if no automated magnetic cell sorting device is available.

Biopanning bacterial display libraries has been a successful approach to the isolation of affinity reagents, and peptides offer a useful alternative to antibodies for recognition when specific criteria are required, such as stability in extreme environments. Biopanning of these libraries has been streamlined using the autoMCS methods described here, and a commercially available autoMCS platform extends this technology to a much wider audience. In comparison to traditional alternating MCS/FACS sorting methods for bacterial display libraries, autoMCS methods are less expensive because they do not require investment in a more expensive FACS instrument capable of sorting and isolating the bound cells, rather than the type of flow cytometer that is used here, which is only capable of analysis ¹⁴ . The critical steps for successful biopanning by autoMCS include separation program selection, negative sorting against potential cross-reactants to reduce false positives, monitoring library enrichment using FACS to determine when to stop biopanning, and smart analysis of the candidate pool to avoid cumbersome screening of hundreds of candidates.

The examples described herein demonstrate successful biopanning of the chosen bacterial display library for two separate targets, PA and abrax, although with slightly different analysis strategies due to differences in the peptide sequences for each pool of candidates after four rounds of biopanning. PA was the more straightforward case, with sequence analysis demonstrating clear trends from the peptide sequences that repeated at least once in 144 colonies. This alone was enough to determine a consensus sequence for the PA target, and those colonies that contained the consensus or a similar sequence were the best candidates in terms of the ratio of binding to PA versus binding to the streptavidin negative control. However, this will not always be the case. For the abrax target, sequencing 100 colonies led to five repeating sequences but the trend in these sequences was less clear, other than a tendency to contain aromatic amino acid residues and aspartic acid or asparagine. The predicted "consensus" from the repeating sequences only could have been produced and tested for affinity to abrax, but since no parent sequences contained that exact consensus sequence, we returned to the list of sequenced colonies and observed that other candidates that had more trends in common with each other. In fact, two colonies contained a similar sequence to the "consensus" from the repeats alone, (FWDTWF instead of FWAWF), which had the same trend of tryptophan, phenylalanine, and aspartic acid residues but with different spacing. One of these peptides was a repeating sequence, one was not. These two sequences, along with a third sequence containing a similar variant, were among the top 5 binders tested in terms of affinity for abrax. The top binder overall, AX-A05, also displayed similar trends. The results for abrax demonstrate that an approach which alternates several times between sequence analysis and affinity and specificity analysis will sometimes be required, depending on the chosen target. The results for the abrax target also demonstrate the level of specificity that can be achieved over a very similar protein (rivax ^{1 , 15} ).

The autoMCS programs selected for our studies were Posselds and Posseld, which are both for positive selection of labeled target cells with low frequency, less than 5% of the initial population. In both cases, cells pass through two magnetic columns. In the case of the Posselds program, however, cells pass more slowly through the first column to increase exposure time ⁴¹ . In addition to the program descriptions given by the manufacturer of the commercial autoMCS device (see Materials Table ) ⁴¹ , these programs were chosen after an initial comparison of several potential programs. Each program's ability to avoid pulling out false positives from a homogenous culture of negative control cells, and to successfully isolate a known binder to a target of interest from a mixed culture containing negative control cells spiked with a decreasing percentage of the known binder, were compared. If significantly deviating from the applications described here, a similar comparison could be helpful in program selection for the new application. However, previously published results comparing these two programs (Posseld and Posselds) for isolation of peptide affinity reagents for PA showed that using either of these programs for all four rounds of biopanning led to isolation of peptides with similar affinity and specificity for PA and determination of the WXCFTC consensus. The main differences were that Posseld, with its faster flow rate, sorted more rapidly and binding affinity for the target was therefore higher after only two rounds of biopanning, while using Posselds led to more unique sequences after four rounds of biopanning ¹⁴ . Learning from this, the strategy was changed slightly when biopanning for abrax binding peptides to incorporate both benefits: Posselds was used for Round 1 to allow more exposure time during this critical step where diversity is highest, but then the program was switched to Posseld for faster isolation of the highest affinity binders during Rounds 2-4 ^{1 , 15} . This appeared to be ideal for abrax, especially since the nMFI and percent binding were both higher for abrax sorting round 4 as compared to PA sorting round 4 with either program ( Figure 2 and Figure 3 and Sarkes et al. 2015 ¹⁴ ), but a direct comparison using one strategy versus the other with the same target would be required for a more conclusive comparison. Which program is best for a particular application may depend on the individual target, the sorting library used, and the level of peptide expression that is achieved with any change to the conditions described here.

Not discussed here are potential results from biopanning with abiotic materials, but we have also used the same bacterial display library for biopanning against bulk aluminum ² . This study was not performed using autoMCS, but similar studies could be accomplished using autoMCS if the material is available on, or could be coated on or conjugated to, a magnetic bead. In the bulk aluminum study, individual amino acid analysis and secondary structure modeling was helpful in understanding what was driving the binding affinity of the isolated peptides, since no consensus sequence was determined ² . Even with biological targets, this may be required to understand what is driving the binding affinity for some targets, which is why the spreadsheet generated from the "eCPX_Sequencing" macro includes a tab with individual residue frequency analysis. If no repeating sequences are seen in addition to the lack of a consensus, and residue frequency shows no pattern, however, other types of analysis may be required. Additionally, further sorting rounds may be necessary to avoid screening a much larger number of candidates for down-selection of those with sufficient affinity to the desired target.

With the increasing availability of next generation sequencing, a more thorough study could be performed to determine the most promising candidates before testing affinity and to determine the extent of enrichment after each round of biopanning. However, sequences moving on to the binding analysis step may need to be re-created by cloning, if they are not of high enough frequency to easily isolate with routine colony sequencing of tens or hundreds of colonies. As this technology advances, becoming less expensive and more routine, and especially with an option for colony isolation, it will likely be the best way to analyze biopanning data. It may lead to elucidation of the way individual sequences, and the entire library, evolve. Additionally, the bacteria in the sorting library may mutate over time, which could bias sorting towards cells with faster growth rates or bacteria that overexpress genomic proteins able to bind a material even without display scaffold and peptide expression, for instance. For that reason, once promising candidates emerge, it is worth isolating the plasmid DNA, retransforming fresh E. coli, and confirming peptide binding via FACS. If planning to use the peptide in a cell-free format, affinity should also be assessed for the free peptide. For example, peptide affinity reagents discovered through bacterial display for the target SEB were synthetically produced off-cell and analyzed by more traditional methods such as enzyme-linked immunosorbant assay (ELISA), and on-cell (via FACS) and off-cell (via ELISA) affinity were compared using the K _d s determined by both methods ⁷ .

Although the strength of the biotin-streptavidin interaction makes it an ideal strategy for capture of protein target and bound peptides, this procedure can be modified for other capture strategies. Direct coupling of protein to magnetic beads, such as epoxy and carboxylic acid beads, has been successful and removes a binding step. However, direct attachment may hinder potential binding sites in a specific or non-specific manner, depending on attachment strategy. An additional advantage to using streptavidin beads is that less stable protein targets can be freshly biotinylated in 30 min or stored frozen, if needed, while the beads themselves tend to require longer incubation with the protein for cross-linking and are generally stored at 2 – 8°C. The suggested starting concentration of biotinylated protein target here, 600 nM, has been successful for multiple targets, but it may be possible to isolate peptide capture reagents with increased affinity if this concentration is reduced. Additionally, this method can be extended to and modified for other bacterial display libraries and other organisms. For example, these steps can be performed in the absence of oxygen for use with anaerobes, or in other environments for use with extremophiles to isolate peptides with unique properties. The peptides and/or consensus determined for a particular target of interest can be potentially used on-cell as a living material, or produced synthetically off-cell. Synthetically produced peptides can be further matured for the development of even higher affinity and more robust Protein Catalyzed Capture (PCC) agents for use in sensors, or for other applications where antibodies would typically be used for binding or recognition ^{38 , 55} . Molecular modeling can also be used to help determine binding location of the peptide with its target, as was recently performed for the abrax binding peptides and consensus listed in Figure 5C ¹ . Overall, biopanning bacterial display libraries for peptide affinity reagents is a fast, straightforward, and powerful alternative to the production of antibodies for recognition and detection of protein targets, and this semi-automated biopanning approach has produced reliable results with far reaching applications.

The authors would like to acknowledge the support of the U.S. Army Research Laboratory (ARL) Postdoctoral Fellowship Program, administered by the Oak Ridge Associated Universities through a contract with ARL, through the appointment of Dr. Justin P. Jahnke. Remaining funding was provided by the Sensors and Electron Devices Directorate at ARL. The authors would also like to thank the Daugherty Lab at UCSB for sharing the bacterial display library and information for cloning the YPet Mona reagent, Dr. Bryn Adams for support in cloning the YPet Mona reagent at ARL, Dr. Joshua Kogot for his initial work on and training in biopanning of bacterial display libraries, Alena Calm of Edgewood Chemical and Biological Center for sharing plasmids used to express and purify abrax and rivax proteins, Brandi Dorsey for technical support for isolation of PA binding peptides, Qin Guo for technical support of preliminary experiments for isolation of abrax binding peptides, and Dr. Jessica Terrell for helpful discussions regarding this manuscript.