Sequence-activity mapping via depletion reveals striking mutational tolerance and elucidates functional motifs in Tur1a antimicrobial peptide

Abstract

Proline-rich antimicrobial peptides (PrAMPs) are attractive antibiotic candidates that target gram-negative bacteria ribosomes. We elucidated the sequence-function landscape of 43 000 variants of a recently discovered family member, Tur1a, using the validated SAMP-Dep platform that measures intracellular AMP potency in a high-throughput manner via self-depletion of the cellular host. The platform exhibited high replicate reproducibility (ρ = 0.81) and correlation between synonymous genetic variants (R² = 0.93). Only two segments within Tur1a exhibited stringent mutational requirements to sustain potency: residues ₉YLP₁₁ and ₁₉FP₂₀. This includes the aromatic residue in the hypothesized binding domain but not the PRP domain. Along with unexpected mutational tolerance of PRP, the data contrast hypothesized importance of the ₁RRIR₄ motif and arginines in general. In addition to mutational tolerance of residue segments with presumed significance, 77% of mutations are functionally neutral. Multimutant performance mainly shows compounding effects from removed combinations of prolines and arginines in addition to the two segments of residues showing individual importance. Several variants identified as active from SAMP-Dep were externally produced and maintained activity when applied to susceptible species exogenously.

Introduction

Antimicrobial peptides (AMPs) play an important role in the interactions of microbial communities and their surroundings (Maróti et al. 2011, Raffatellu 2018, Peterson et al. 2020). AMPs are naturally produced in situ and act locally to combat a broad swath of species competing for resources. AMPs also present a compelling option for a potent, selective, efficiently engineerable source of antimicrobials to address the pressing healthcare challenge of antibiotic resistance (Rea et al. 2013, Li et al. 2014, Raffatellu 2018, Heilbronner et al. 2021). To realize this technological potential, knowledge of the sequence-activity relationship is needed to identify variants with potency against the desired targets as well as the ability to maintain activity during pursuit of supplementary characteristics such as stability or expression. Sequence-activity mapping (McConnell and Hackel 2023) also provides insight on the intra- and inter-molecular interactions that drive natural function. Yet the immensity of protein sequence space and the ruggedness of sequence-activity relationships challenges such mapping, particularly for complex activities that are often not efficiently quantified in high-throughput.

We have recently developed a high-throughput platform to quantitatively map AMP sequence-activity relationships (Dejong et al. 2021). The platform, sequence-activity mapping of AMPs via depletion (SAMP-Dep), entails intracellular expression of variant AMPs within the target cell; potent AMP variants hinder growth, which is quantified via deep sequencing of the AMP vectors within the cell population before and after induction of expression. We validated SAMP-Dep with the oncocin AMP acting on E. coli (Dejong et al. 2021) and extended its breadth to a set of endolysins via periplasmic expression (Tresnak and Hackel 2023). More recently, we have expanded the study of proline-rich AMPs (PrAMPs) by implementing SAMP-Dep on oncocin homologs metalnikowin and apidaecin (Collins et al. 2024). During this pursuit, we also observed the ability of a non-homologous, mammalian, proline-rich AMP—Tur1a—to hinder growth of E. coli bacteria. Herein, we evaluate the sequence-activity landscape of Tur1a.

Tur1a is a 32-amino acid AMP expressed in dolphins. It contains the YLPRP and PRPxRR motifs homologous to oncocin but is otherwise non-homologous and mechanistically distinct (Table 1). It exhibits homology, including the PPxLP motif, with several cetacean PrAMPs (Sola et al. 2020) (Table 1). Whereas insect PrAMPs, like oncocin, mainly interact with the ribosome with hydrogen bonds (mainly with the peptide backbone) and stacking interactions (within the binding domain), interactions with mammalian PrAMPs outside of the binding domain predominantly engage the ribosome through arginine side chains (Gagnon et al. 2016, Seefeldt et al. 2016, Graf et al. 2017, Mardirossian et al. 2018, Graf and Wilson 2019). Further, Tur1a enacts more contacts within the upper polypeptide tunnel of the ribosome, where insect PrAMPs only have two C-terminal contacts in this region (Mardirossian et al. 2018, Graf and Wilson 2019). Structurally, these mammalian PrAMPs have at least seven extra residues in the C-terminus that form a short loop within the upper polypeptide tunnel, acting as an anchoring mechanism in the ribosome. This mechanistic addition is unique to this taxonomic class. Curiously, mammalian PrAMPs do not bind DnaK, conversely from insect PrAMPs (Frimodt-Møller et al. 2022). Finally, ribosomal mutations that create resistance for oncocin and apidaecin did not provide resistance to mammalian PrAMPs (Gagnon et al. 2016, Florin et al. 2017, Mardirossian et al. 2018). This suggests a significant difference in ribosomal binding mechanics and overall antibiotic activity of mammalian PrAMPs, including Tur1a.

Table 1.

Selected cetacean PrAMP sequences, as well as oncocin, aligned to Tur1a reference. Identical amino acids in relation to Tur1a labelled as •. Homologous amino acids (BLOSUM62 ≥ 0) are colored gray

	Site
AMP	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16	17	18	19	20	21	22	23	24	25	26	27	28	29	30	31	32
Tur1a	R	R	I	R	F	R	P	P	Y	L	P	R	P	G	R	R	P	R	F	P	P	P	F	P	I	P	R	I	P	R	I	P
Bal1	•	•	•	•	•	•	•	•	R	•	•	•	•	R	P	•	•	W	I	•	•	R	•	•	F	•	•	•	•	G	K	R
Lip1	•	•	•	•	I	•	•	•	R	•	•	•	•	R	P	•	•	W	•	•	•	R	•	•	•	•	•	•	•	G	K	R
Orc1	•	•	•	P	•	W	•	•	N	•	•	G	•	R	•	P	•	W	•	L	•	D	•	R	•	•	•	•	•	•	K	R
Del1	•	•	•	P	•	W	•	I	P	•	R	W	Q	W	P	P	•	W	•	•	•	S	•	•	•	•	•	•	S	•	K	R
oncocin				V	D	K	•	•	•	•	•	•	•	R	P	P	R	•	I	Y	N	R

Beyond the dissonant behaviors to respective insect PrAMPs, previous studies have indicated hypotheses on active sites and ribosomal interactions with the Tur1a peptide. The N-terminal region seems to be the most functionally important segment of the peptide. A truncated peptide of the first 16 residues maintained partial activity in contrast to inactivity from fragments Tur1a(8-24) and Tur1a(16-32) (Mardirossian et al. 2018). The N-terminal RRIR motif is conserved between Bac7 and Tur1a (Benincasa et al. 2004, Mardirossian et al. 2018). However, several arginines did not show noticeable electron density within this segment from resolved bound complexes of Tur1a with the ribosome, suggesting these residues could be more flexible for this peptide (Roy et al. 2015, Seefeldt et al. 2015, Gagnon et al. 2016).

To elucidate sequence-activity relationships for Tur1a, we have implemented SAMP-Dep on a systematic set of mutants varying at adjacent (n, n + 1) sites as well as next-neighbor (n, n + 2) sites. Tur1a exhibited a strikingly high tolerance for mutation. 77% of mutants perform comparably to wild-type, with essentially neutral mutational effects. Only three residues, ₉YLP₁₁, were shown to be essential for activity and a further two, ₁₉FP₂₀, were shown to have decreased mutational tolerance relative to the rest of the peptide. Additionally, our results counter the hypothesis that RRIR and PRP motifs are essential for ribosomal inhibition of Tur1a. Further, arginines that generate the majority of contacts between the ribosome during inhibition do not show importance for antimicrobial function. These results were corroborated via exogenous treatment with identified active variants from SAMP-Dep.

Results

Tur1a library construction

We constructed a genetic library of Tur1a variants with full NNK codon diversity at either adjacent (n, n + 1) or next-neighbor (n, n + 2) sites. We transformed the library, under tight transcriptional control, into E. coli T7 LysY/I^q bacteria. For library creation and preparation, transformation yielded 3.7 ± 0.4 million transformants encompassing 115 ± 15-fold coverage of the designed library. Experimental host transformation used in platform variant expression yielded 35 ± 2 million transformants (560 ± 28-fold library coverage). Sequence analysis revealed that library assembly matched design with 85% of the designed library present. The majority of the identified sequences were single and double mutants within this library, with a small proportion of additional undesigned triple and quadruple mutants (Supplementary Fig. S1). AMP expression was induced at three different concentrations of IPTG, AMP variant frequency was determined before and after induction, and the slope of frequency relative to inducer concentration is used to quantify potency (i.e. the SAMP-Dep platform (Dejong et al. 2021)). Application of SAMP-Dep to Tur1a in E. coli yielded high reproducibility, particularly for variants with high initial read depth (Fig. 1A). 61% of variants had at least 20 initial reads (Fig. 1B), which exhibited an average 0.81 correlation coefficient across replicates (Fig. 1A). Assay function was largely driven at the protein level, rather than via genetic constraints, as silent genetic mutants largely matched performance (Δamino acid variant slope: ΔDNA variant slope correlation = 0.93; Fig. 1C).

Figure 1.

SAMP-Dep data exhibit high reproducibility. Four fully independent replicates of SAMP-Dep were performed. (A) The replicate-to-replicate correlation is plotted for all variants at or above the indicated number of clonal reads in the 0 mM sample. (B) Distribution function indicating the fraction of protein variants with at least the indicated number of clonal reads in the 0 mM sample. The dashed line at 20 reads indicates the threshold used for analysis on data. (C) The slope of frequency versus inducer concentration for each protein variant in each replicate is plotted against the slope of the corresponding gene. The dashed lined represents when the ΔAmino Acid Slope = ΔDNA Slope

Tur1a sequence-activity landscape

The analysis of mutational impact is striking as 62% of all variants screened are statistically neutral with regard to activity, which is drastically distinct from the mutational intolerance observed for the insect PrAMPs, metalnikowin and apidaecin, in a separate study (Collins et al. 2024) (Fig. 2A). Along with high reproducibility (Fig. 1), the inactivity of premature truncations up to P20 provides evidence of assay accuracy (Fig. 2B). In holding to behaviors of Class I PrAMPs (Bulet et al. 1999, Welch et al. 2020, Collins et al. 2024), the N-terminus of Tur1a has a lower mutational tolerance than the C-terminus (Fig. 2B and C). Contrary to most of the peptide, there is very low mutational tolerance for, ₉YLP₁₁. Surprisingly, the last two residues of the hypothesized ₉YLPRP₁₃ binding domain are not necessary, unlike in insect PrAMPs (Collins et al. 2024). Further, even though ₁RRIR₄ is a conserved motif across mammalian PrAMPs, many single mutations in this region exhibit comparably maintained activity to wild-type. Notably, the SAMP-Dep assay does not require peptide internalization as the AMP is expressed intracellularly. To assess the possibility of mutational robustness resulting from saturating concentrations of intracellular expression, we examined the induction dependence of growth inhibition for wild-type Tur1a. Strong dose dependence and minimal growth inhibition at low induction (Supplementary Fig. S2), along with the observation of beneficial mutations in the library evaluation (Fig. 2), counter this hypothesis and support the conclusion that Tur1a is robust to a wide array of mutations within the context of activity upon intracellular expression.

Figure 2.

AMP sequence-function maps of Tur1a. (A) Distributions of mutant performance for each PrAMP. Metalnikowin and apidaecin data reproduced from (Collins et al. 2024) for comparison. (B) Functional impact of all Tur1a single mutations. Each square represents the average slope change (clonal frequency versus inducer concentration) from the median wild-type of all variants with that column’s amino acid at that row’s site. Grey dots represent no data. Wild-type baseline in graphs is average wild-type variant slope. (C) The mutational distributions for Tur1a at each site are aggregated

The maintenance of Tur1a activity upon mutation is achieved with both conservative and non-conservative mutations (Fig. 3A; blue and green). Moderately detrimental mutations preferentially result from non-conservative mutations albeit with some exceptions (Fig. 3A; yellow). Strongly detrimental mutations more frequently result from highly non-conservative mutations (Fig. 3A; red), which is evident in a substantially reduced average conservatism (−2.5 BLOSUM62 score; Fig. 3B) relative to all mutations (−1.7).

Figure 3.

Activity is maintained across mutations with a broad level of conservatism whereas mutational detriment largely results from non-conservative mutations. (A) For each level of conservatism (via BLOSUM62 score), the frequency of mutants within a particular activity tier (based on the indicated SAMP-Dep slope) is normalized relative to total mutants of any activity. The log₂ of this frequency ratio (or ‘enrichment’) is plotted for each activity tier. (B) The weighted average conservatism (via BLOSUM62 score) across all mutants is plotted for each activity tier. The average conservatism across all mutants is plotted as a dotted line

We next sought to evaluate intramolecular residue interactions via epistatic analysis. Most single- and double-mutants effected minimal change in activity thereby precluding an opportunity for epistasis. However, several residue pairs are observed to have significant and consistent interactions, most consistently at proline and arginine residues, which are hypothesized to be the most important residues within the peptide. Numerous double mutations from two wild-type prolines or two wild-type arginines have highly negative epistatic effects (Fig. 4, red and yellow, respectively). Double mutations from one wild-type arginine and one wild-type proline are highly epistatic with either negative or synergistic effects observed (Fig. 4, orange). The most striking epistatic effects are detrimental to variant activity. The strongest synergies in more potent variants occurs around the binding domain (₉YLP₁₁) and the location immediately N-terminal to permitted truncations (Fig. 3). Simultaneous mutation of L10 and P11 yields several strong synergies with improved potency (Fig. 4, blue). F19 variants (M or G) synergize with multiple P21 variants (Fig. 4, green). While the frequent synergistic detrimental mutations could be consistent with a scenario in which wild-type Tur1a is expressed beyond its inhibitory threshold—thereby yielding non-linear inhibition from this saturation—the induction titration data (Supplementary Fig. S2) demonstrate this is not the case.

Figure 4.

Residue interactions within Tur1a. Double-mutant epistasis was calculated as fold change between double mutant slope change compared to multiplied performance of each single mutation component and plotted versus the double-mutant slope change relative to wild-type Tur1a. Notable residue interactions are highlighted

Exogenous exposure on susceptible species

We next sought to evaluate peptide variants identified within the intracellular platform for their antimicrobial activity upon exogenous administration. Five peptides were selected for chemical synthesis and applied to cultures of potentially susceptible species: the three statistically most active variants (G14R, I28S, and P29L/I31R), an active truncation mutant (P22A/P24*), and a variant inactive in SAMP-Dep (P8R/L10P) (Fig. 5A and B). The intracellularly-active variants exhibited minimum inhibitory concentrations (MICs) of 2.2–20 μg/ml, which ranges from improvement to mild hindrance of potency relative to wild-type (5 μg/ml against E. coli BW25113 as measured in Mueller-Hinton broth with overnight growth (Mardirossian et al. 2018) in contrast to lysogeny broth with 9-hour growth in the current work) (Fig. 5C). The most potent variant, Tur1a I28S, decreased the MIC by a factor of at least three compared to all other variants against all strains besides T7 LysY E. coli. The truncated peptide retained its activity without the seemingly unnecessary extended C-terminal segment and with removal of five prolines (Fig. 5A and C).

Figure 5.

Variants identified for external exposure in susceptible species from SAMP-Dep Assay. (A) Sequences of the wild-type peptide, the most inactive variant, the three most active variants, and an active truncated mutant identified from their performance within the SAMP-Dep platform, shown in (B). (C) MIC (μg/ml) against six strains of two species mirroring activity assays shown in Lai et al. (2019)

Strikingly, the peptide that was inactive within the SAMP-Dep platform (internal production and activity) displayed high potency when externally applied to all tested strains (Fig. 5C). The SAMP-Dep inactivity could be due to low expressibility, although multiple DNA variants encoding this same peptide are inactive within SAMP-Dep (Fig. 5B). Overall, the evidence of no internal activity contrasted with high exogenous potency suggests Tur1a targets more than just the ribosome in its antimicrobial activity. Further work will be needed to evaluate this hypothesis.

Discussion

Tur1a has a striking internal activity sequence-function relationship highly divergent from other PrAMPs and contradictory to previously held functional hypotheses. Tur1a follows similar rules to insect-derived class I PrAMPs but has few residues that are necessary for function. The C-terminus exhibits multiple options for mutational benefit. Hotspots of mutational intolerance include segments ₉YLP₁₁ and ₁₉FP₂₀. ₉YLP₁₁ entails a portion of the PPxLP motif shared across Tur1a, oncocin, and cetacean PrAMPs (Mardirossian et al. 2020). These hot spots also include consistent epistatic interaction with surrounding residues, as seen in Fig. 4. The ₉YLP₁₁ segment resides within the hypothesized binding domain; however, the peptide still requires residues through P20 to maintain activity on target species even though residues 12 through 18 can include vast variability. The function of these residues past the binding domain is not yet known but could range from providing steric hindrance blocking diffusion of the PrAMP back out of the ribosome to preventing the peptide from being pushed out of the A-site binding pocket from the incoming aa-tRNA (Mardirossian et al. 2018, Graf and Wilson 2019). Given this noteworthy sequence-function profile for Tur1a, there were prominent variants with increased potency for the 70S ribosome identified within the SAMP-Dep platform. Several identified variants with high potency were further tested through environmental exposure, requiring cellular internalization to target the ribosome. However, the most statistically inactive Tur1a variant maintained high levels of activity comparable to the previously reported wild-type activity level. Within this variant (P8R/L10P), P8 and L10 that surround the Y9 within the hypothesized ribosomal binding domain are both mutated to inhibiting individual substitutions (Fig. 2B). Given this segment is highly conserved and important within other PrAMPs that target the 70S ribosome (Collins et al. 2024) and all DNA variants encoding this peptide showed no internal activity, these results highly suggest other targets present most likely on the surface of these bacteria are affected by the presence of Tur1a.

These results provide direct commentary on previously held functional hypotheses of Tur1a and mammalian PrAMPs in general. The RRIR motif beginning the peptide has been shown to be highly important to its functional capabilities (Benincasa et al. 2004, Mardirossian et al. 2018) with R2 creating a pi stacking interaction with the ribosomal components in the A-site binding pocket (Seefeldt et al. 2015, Seefeldt et al. 2016, Mardirossian et al. 2018, Graf and Wilson 2019). Indeed, in the SAMP-Dep assay, R2 exhibits mild mutational intolerance (Fig. 3B and C). Proline and arginine residues are also hypothesized to be critical beyond their numerous inclusions within the peptides. Prolines are thought to prevent alpha helical formation for ribosomal insertion, and arginines make up most of the physical contacts through hydrogen binding with the ribosome during inhibition (Graf et al. 2017, Graf and Wilson 2019). R2 and R6 are sensitive to mutation (median Δslope >0), and R1 and R12 are mildly sensitive (median Δslope ~0) (Fig. 3B and C). Conversely, R4, R15, R16, R27, and R30 exhibit mutational tolerance. For prolines, substantial sensitivity is observed at P11 and P20 along with mild sensitivity at P8 and P32. Other proline sites (7,13, 17, 21, 22, 24, 26, and 29) permit a broad array of mutations without loss of activity. Notably, it appears several of these residues interact with each other, where multiple mutations of either two arginines, two prolines, or one arginine and one proline can have compounding deleterious effects (Fig. 5). It is important to note that these mutations were not tested directly within an environmental exposure context, meaning it is unknown if they are important for the internalization mechanism. Moreover, the exit tunnel anchor that forms within the ribosome is not necessary for potent inhibition as the extra residues extending the C-terminus of mammalian PrAMPs are not needed to maintain or increase wild-type function. Lastly, the overall hypothesized mechanism Tur1a uses to assert its antimicrobial activity has also been called into question. The consistent disparate results between internal vs. external potency for the P8R/L10P variant shows internal/ribosomal inhibition may not be the only or most important mode of action when interacting with susceptible species. Notably, Tur1a has more positive charge than other proline-rich AMPs, yet it is not known how this impacts cellular engagement, internalization, or ribosomal binding. It is surmised that this characteristic allows greater negatively-charged cellular membrane interaction like cationic AMPs but remains non-lytic in contrast to this class of AMPs (Mardirossian et al. 2018, Graf and Wilson 2019, Welch et al. 2020).

This internal sequence-function landscape for Tur1a is highly notable given the likelihood of protein variants that are functional. These landscapes tend to be highly rugged and mainly barren (Romero and Arnold 2009). From previous data collected from many studies, trends show roughly 30–50% of single mutations remove most or all activity, another 50–70% inhibit activity slightly or are fully neutral, and lastly 0.1–1% mutations allow for increased activity (Supplementary Fig. S3A) (Rennell et al. 1991, Moore et al. 1997, Guo et al. 2004, Drummond et al. 2005, Romero and Arnold 2009). Neutral and beneficial mutations become more rare with increased simultaneous residue changes (Tokuriki and Tawfik 2009). For Tur1a single mutants, 3% are deleterious (less than 20% wild-type activity), 18% partially inhibit activity, 77% are statistically neutral, and 2.5% are beneficial (Supplementary Fig. S3B). This pattern is surprisingly maintained with multiple simultaneous mutations: 6% are deleterious, 24% partially inhibit activity, 69% are statistically neutral, and 2% improve activity. These results are contrasted by insect PrAMPs within the same platform: for single mutations, 33%–48% are deleterious, 29%–30% decrease activity, 20%–38% are statistically neutral, and 0.1%–0.4% are beneficial; even more strikingly for multimutants, 63%–78% are deleterious, 14%–19% inhibit activity, 18%–20% are statistically neutral, and 0.05%–0.15% are beneficial (Supplementary Fig. S3C) (Collins et al. 2024). Such a rich activity landscape is highly unexpected and speaks to the potential mechanism of the peptide itself. These observations may be present due to the increased structural flexibility of the Tur1a peptide when bound to the ribosome compared to Bac7 (Mardirossian et al. 2018, Graf and Wilson 2019) or a feature of the assay bypassing the internalization mechanism. Overall, more study is required to determine all targets of mammalian PrAMPs, or Tur1a individually, that would provide more context to the results exhibited here.

Materials/methods

Gene construction

Protocols were modified from Collins et al. (Collins et al. 2024). Overlapping oligonucleotide primers were used to individually generate the n/n + 1 and n/n + 2 Tur1a variant libraries via PCR. Each sub-library contained NNK degenerate codons at each selected site. NheI and BamHI restriction enzymes were used to digest the pET vector with added calf intestinal phosphatase. Amplicons and vector digests were purified through gel electrophoresis and miniprep extraction (Epoch Life Sciences). Sub-library amplicons were mixed together in equimolar amounts. NEBuilder HiFi Assembly using 0.5 pmol total DNA consisting of amplicon:vector in a 5:1 ratio was used to assemble the full plasmid for the entire library. MinElute purification columns (Qiagen) were used to purify and concentrate final HiFi assemblies into ultrapure water. The concentrated library sample was transformed into electrocompetent 10β E. coli cells and grown overnight on lysogeny broth (LB) plates with 50 μg/mL kanamycin (kan) at 37°C. After plating, a total of 3.3 million and 4 million transformants were present for n/n + 1 and n/n + 2 Tur1a libraries, respectively (~100–130× library oversampling). All colonies were collected, and DNA was harvested using miniprep extraction. Extracted DNA stock concentrations used for SAMP-Dep exceeded 450 ng/μl.

SAMP-dep transformation and growth

Dejong et al. (Dejong et al. 2021) introduced the SAMP-Dep protocol, modified in Collins et al. (Collins et al. 2024), used in this study. 5 μl of concentrated library DNA were mixed together in equimolar quantities and transformed into 50 μl of competent T7Express LysY/I^qE. coli cells (NEB). 1 ml SOC recovery media was used after transformation, incubating for 1 h at 37°C. To calculate transformation efficiency of each replicate, 10 μl was then taken from the SOC mixture and 10× dilutions were plated on LB-kan agar. The 990 μl of leftover SOC mixture was centrifuged at 6000 RCF, resuspended in LB with 50 μg/ml kan, and incubated for 1 h at 37°C with shaking. 250 μl of transformant mixtures were added to three 9.75 ml of LB with 50 μg/ml kan containing IPTG at 0, 0.15, and 0.50 mM concentrations and placed in a dynamic incubator at 37°C. The remaining 250 μl was harvested via miniprep and stored for the pre-induction sequence distribution. After incubating for 7.2 h, each IPTG concentration supernatant was pelleted. DNA was harvested and stored. Four replicates were performed on different days.

Next-generation sequencing

The same forward and reverse primers were used to amplify Tur1a variant sequence for each replicate and condition sample. Primers with a unique barcode were used to subsequently amplify each sample. Amplicons from the second PCR were purified through gel separation and miniprep purification. Equimolar replicate and condition amplicons were mixed and sequenced at the University of Minnesota Genomics Center via Illumina NovaSeq 6000 on an SP Flow Cell. USearch was used to filter, accept reads presenting less than a 0.5% expected error rate, and merge each read direction.

Library sequence-function analysis

Growth rate was calculated as outlined in Dejong et al. 2021 for all sequences passing the USearch filter and subsequently merged. The standard g_p,0mM of 12.5 for an uninduced culture calculated from Dejong et al. (2021). Sequences with a stop codon within the first three codons (no potential activity) were designated as baseline negative controls with an unaffected growth rate of 0 M⁻¹ min⁻¹. The average of slopes from all four experimental replicates was calculated for all sequences. A threshold of 20 reads for all replicates within the 0 mM IPTG-full incubation sample was used to filter resulting sequences for confidence in calculated slope. The final measurement of potency, the slope change from wild-type, was calculated based on number of codon differences from the original DNA baseline construct. The median wild-type slope was calculated for 0, 1, and 2 silent codon changes and chosen for comparison to all sequences based on their number of codon differences from the wild-type starting DNA. Variants with 1 codon change were compared to the median wild-type amino acid sequence variants with 1 silent codon change. Variants with ≥2 codon changes were compared to wild-type amino acid sequence variants with 2 silent codon mutations. Significance was calculated through an independent T-test between each variant’s four replicate delta slopes to the median wild-type sequence’s four replicate slopes. All python codes and deep sequencing counts are available at: https://github.com/HackelLab-UMN

Calculation for Epistatic analysis

The same equation in Collins et al. (Collins et al. 2024) was used to determine interactions between residues. Interactions were defined as the change in multimutant performance unaccounted by additive properties of each individual simultaneous mutation. This equation taken from Collins et al. (2024) is shown below:

The fold change of each individual amino acid substitution and multiple substitution variant was calculated in reference to the median wild-type performance with the same number of codon changes. The dual mutant relative performance is shown as FC_1,2, with each single mutant represented by FC₁ and FC₂.

Minimal inhibitory concentration (MIC) determination

In order to determine potency of each Tur1a variant, the selected peptide requires external construction and purification to perform exogenous exposure on susceptible species. Six species selected for this testing were taken from Lai et al. (Lai et al. 2019) and Collins et al. (Collins et al. 2024) in E. coli T7 Express LysY, FVEC 638, JW0013, and BW25113 strains in addition to Salmonella enterica serovar Typhimurium ATCC SL1344 and Enteritidis MH91989 strains [Source: Keio Collection, GenoBase, Minneapolis VA Hospital, and University of Minnesota].

Five Tur1a variants were identified as candidates for exogenous exposure and were constructed by ProteoGenix via solid-phase synthesis to >80% purity. These variants include the most statistically inactive variant, three most statistically active variants, and an active variant without most of the C-terminal residues as a special case. Statistical significance was determined by an independent t-test of all replicate slopes of each peptide variant against all wild-type DNA variant slopes. Production of the wild-type Tur1a peptide was attempted but could not reach necessary purity.

Susceptible species were diluted and plated on LB-agar for clonal isolation. Single colonies were plucked from LB-agar plates and inoculated in 3 ml of LB media for overnight culture at 37°C while shaking. Two hours prior to conducting the assay, 10 μl was transferred to 3 ml of fresh LB media and grown at 37°C to reach exponential phase. While growing, 75 μl of 3× serial dilutions of each Tur1a variant in LB was added to 96-well plates at 120, 60, 20, and 6.7 μg/ml concentrations. After completed outgrowth 75 μL of each species was dilution to 5 × 10⁵ CFU/ml and added to the Tur1a variant dilution samples (60, 20, 6.7, and 2.2 μg/ml at final concentration for each Tur1a variant). A BioTek Synergy H1 microplate reader was used to take OD₆₀₀ readings every 9 h. 96-well plates were incubated at 37°C at 250 rpm between measurements. Two replicates were conducted on separate days.

Media only samples for each species were done as a negative control for each plate. The minimum concentration of Tur1a variant resulting in a mean OD₆₀₀ measurement below two standard deviations of the negative control at the 9 h timepoint was reported as the minimum inhibitory concentration.

Edited by: Timothy Whitehead

Author contributions

Jonathan Collins (Conceptualization [supporting], Data curation [lead], Formal analysis [lead], Investigation [lead], Writing—original draft [equal], Writing—review & editing [equal], and Benjamin Hackel (Conceptualization [lead], Formal analysis [supporting], Funding acquisition [lead], Investigation [supporting], Project administration [lead], Supervision [lead], Writing—original draft [equal], Writing—review & editing [equal]).

Funding

This work was supported by the National Institutes of Health (R01 GM121777). We thank the University of Minnesota Genomics Center for assistance in deep sequencing.