Ethical statementAlpaca immunizations were supervised by Dr. Charles Shoemaker at Tufts Cummings Veterinary School Campus. All animal experiments were conducted in accordance with protocols approved by both Yale and Tufts universities’ Institutional Animal Care and Use Committees (Yale IACUC protocol no. 10584 and Tufts IACUC protocol no. G2011-08).Statistics and reproducibilityNo statistical method was used to predetermine sample size. In general, ELISA experiments used a minimum of three biological replicates and flow cytometry results shown are representative of at least two independent experiments. Recombinant VHH transfection and purification were repeated at least once. The extended high-throughput phage display panning experiment was conducted using three biological replicates per selection condition; other panning experiments used only one replicate and were not repeated. As discussed in the main text, we excluded samples where two closely-related CDR3 clonotypes, ad6f8f and 2c7c51, comprised >80% of the reads; the round 8 input sample for each of the following selections was excluded from the indicated figure and was not used to choose rVHHs for resynthesis: 1.A2, 1.C2, 1.B6, 1.C5 (Supplementary Fig. 25, OprM); 1.A4,1.C4 (Supplementary Fig. 27, OprN); 1.A3,1.B3 (Supplementary Fig. 26, OprJ). The underlying reads are included in the SRA submission and in the processed data packages deposited in Zenodo. The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment.Bacterial strains and growth conditionsBacterial strains discussed in the main text are shown in Table 1. Strains used in each phage display selection are shown in Supplementary Data 1–4. The following abbreviations are used in the Supplementary Data tables:
Strain MB5890 “∆efflux” has genotype ∆(mexAB-oprM) ∆(mexCD-oprJ) ∆mexJKL ∆(mexHI-opmD) ∆opmH36
Strain PA0397 “∆efflux” has genotype ∆(mexAB-oprM) nfxB ∆(mexCD-oprJ) ∆mexJKL ∆mexXY ∆opmH362 ∆(mexEF-oprN)37
Strains marked “::(ZTP)” carry the ZTP riboswitch reporter described by38 and have the genotype ∆exsD attB::PexoT::ZTPlacZ
If not otherwise noted in Table 1 or in the tables below, the strains mentioned in the tables below are first described in this study.
Table 1 Bacterial strains used in this studyLuria Broth (LB) agar and LB broth were used for growth of P. aeruginosa unless noted otherwise. Bacteria were streaked from glycerol stocks onto LB agar plates, then liquid cultures inoculated from well-isolated single colonies and grown overnight. Bacterial growth was at 37 °C, with liquid cultures shaking at 220 rpm.Unless noted, washes of P. aeruginosa cells were performed in PBS + MC (PBS pH 7.4 plus 0.9 mM calcium, 0.9 M magnesium). Pelleted cells were resuspended by “racking” (gently dragging a tube across a plastic tube rack) or by shaking at 1200 rpm for 30 s (for cells in V-bottom 96-well plates).Alpaca immunizationFour P. aeruginosa laboratory strains (PA14 [serotype O10], PAO1 [O5], PAK [O6] and PA103 [O11]) were grown to exponential phase in liquid LB and overnight on LB agar. Cells were collected by centrifugation or scraping plates and resuspended in 20 mM HEPES, 150 mM NaCl, 10 mM KCl, 1 mM phenylmethylsulfonyl fluoride (PMSF). Cells were lysed by french press ( >14,000 psi) and the lysate clarified by centrifugation (5000 × g for 10 min). Membrane proteins were separated from soluble proteins by centrifugation (20,000 × g for 30 min).Membrane proteins were enriched by centrifugation through sucrose density gradients; membrane proteins were collected from the interface of a 60%/25% sucrose step gradient. Additionally, PA14 and PAO1 were grown as static biofilms in Petri dishes in both LB and M9 minimal media at 30 °C for 36 h, harvested by scraping, and lysed as above. The soluble proteins, enriched membrane proteins, and biofilm extracts were pooled to generate a mixed antigen preparation and stored in single-use aliquots at −80 °C.A single adult male alpaca (Vicugna pacos) was immunized with the P. aeruginosa antigen (2 mg per injection) adjuvanted with alum. Four subcutaneous immunizations were administered at 3–4 week intervals. Serum was collected prior to each immunization and four days after the final immunization. Peripheral blood lymphocytes were harvested from the final bleed, and RNA was prepared from one aliquot of fresh peripheral blood lymphocytes (PBLs) using TRI Reagent LS (Molecular Research Center, Inc.) according to the manufacturer’s protocol, and as previously described39. RNA was column-purified using a RNeasy Mini Kit following the manufacturer’s protocol (Qiagen). The yield (87 µg/mL) was calculated in a spectrophotometer at 260 and 280 nm, and RNA was stored at −80 °C.ELISA wells were coated with the P. aeruginosa antigen preparation used for immunization (10 µg mL−1), then incubated with 2–5 fold dilutions of alpaca serum (beginning at 1:100), followed by secondary incubation with anti-llama IgG-HRP conjugate. Protein preparations from the different P. aeruginosa strains were prepared as described above, then separated by SDS-PAGE (1 µg per lane), blotted to PVDF, and incubated with alpaca pre-immunization and post-immunization serum (diluted 1:5000).Construction of a VHH phage display libraryPlasmids and primers used in this study are listed in Table 2 and Table 3, respectively. Three cDNA synthesis reactions were performed in parallel using Superscript III First-Strand Kit (Invitrogen), each with different primers: random hexamers, oligo dT, and alpaca gene-specific primers (Al.CH2, AlCH2.2) described previously39. cDNA was amplified by PCR, using AlVHH-F1 as the forward primer and either AlVHH-shR1 or AlVHH-lhR1 as the reverse primer; these primers recognize the alpaca VHH cDNA at conserved sites within the FR1 domain and the short or long hinge domains respectively. Replicate PCRs were pooled and amplicons purified by silica column cleanup (Qiagen), then restriction-digested with NotI/AscI and the digest gel-purified.Table 2 Plasmids used in this studyTable 3 Primers used in this studyThe phagemid vector pD was derived from pCANTAB (GE Healthcare) via pJSC. The cloning site of this vector is in-frame with the C-terminus of phage coat protein P3; downstream from the cloning site is an E-epitope tag, separated by an amber stop codon. The vector was prepared, digested as above, and gel purified. Vector and insert were ligated using T4 ligase (NEB), cleaned up (Geneclean Turbo column, MPBio), and transformed into electrocompetent TG1 (Agilent). An aliquot of the transformation reaction was serially-diluted to estimate a titer of 4 × 106 transformants; the remaining transfomants were plated on a large area of selective media, scraped, combined, aliquotted, and stored at −80 °C.A total of 96 clones from the transformed library were randomly selected; plasmid preparation and Sanger sequencing were performed by a vendor (Beckman-Coulter). Chromatograms were manually trimmed. A multiple sequence alignment was constructed using MUSCLE and the consensus sequence was used as a reference sequence in downstream analysis.To prepare infectious phage from the library, 1 mL of frozen TG1 cells containing library phagemid transformants were diluted into 25 mL of 2YT media with carbenicillin. This culture was grown for two hours, M13KO7 added to 1 × 1010 pfu mL−1 and grown for 1 h further. This culture was added to 1 L of 2YT plus carbenicillin and kanamycin and grown overnight. Cells were removed by centrifugation; ~5 × 109 cells were collected and phagemids isolated for sequencing; the supernatant was PEG/NaCl-precipitated as described below. Aliquots of this library were considered “passage 1.” One aliquot from passage 1 was amplified as described below to ~1 L to form passage 2. All subsequent experiments were performed using aliquots of passage 2.Isolation of flagella and piliFlagella were isolated following40. 10 mL of exponential-phase cells were harvested by centrifugation, 3000 × g for 10 min at 4 °C and resuspended in 100 mL of flagella buffer (50 mM sodium phosphate [pH 7], 10 mM magnesium chloride). Flagella were sheared in a Waring blender for 30 s, then the pulse repeated. Loss of swimming was confirmed by microscopy. Cells were removed by centrifugation at 20,000 × g for 30 min at 4 °C. Flagella were collected from this supernatant by ultracentrifugation at ~105,000 × g for 1 h. Pellets were resuspended in a total volume of 3 mL Tris-NaCl (150 mM NaCl, 50 mM Tris pH 7.6).Pili were prepared from solid media and removed by vortexing41; cells were grown overnight on 2–3 150 cm2 plates, then collected using a cell scraper and resuspended in 10 mL total volume of TPM buffer (10 mM Tris HCl [pH 7.5], 1 mM KPO4 [pH 7], 8 mM MgSO4). Cells were vortexed for 3 min, then cells removed by centrifugation at 20,000 × g for 5 min at 4 °C. Supernatants were transferred to microfuge tubes, magnesium chloride added to a final concentration of 100 mM, and incubated on ice overnight. Pili were harvested by centrifugation >20,000 × g for 15 min at 4 °C. The supernatant was resuspended in a total volume of 5 mL Tris-NaCl. Large insoluble aggregates were removed by centrifugation 13,000 × g for 5 min at 4 °C.Protein preparations were kept on ice and quantified by SDS-PAGE and/or bicinchoninic acid (BCA) assay the same or next day, then diluted to single-use aliquots, snap frozen, and stored at −20 °C until ready for use.Phage display panningTerminology“Selection” refers to enrichment of a distinct population of phage-displayed VHHs via multiple rounds of panning against a particular pair of antigen conditions (e.g. counter-selection and selection bacterial cells, antigen-negative and antigen-positive protein-coated wells, etc.). An observation of that population at one point in time is a “sample.” The high-titer phage population, prior to round N of panning, is the “round N input phage”; phage eluted from the wells/cells after that round are the “round N output phage.” The round N output phage are expanded in E. coli to produce the round (N + 1) input phage, and so on. Unless otherwise noted, sequencing data is obtained for high-titer “input” phage populations, as sequencing library preparation is most consistent from these templates. Therefore in all figures showing sequencing data, round N phage refers to the round N input phage, i.e. the result of performing (N-1) rounds of panning.Common phage display methodsOmniMAX 2 T1R E. coli were used as the host strain for all phage display experiments. OmniMAX E. coli were grown in 2YT media supplemented with tetracycline 10 µg mL−1 to maintain the F episome. Kanamycin 25 µg mL−1 was added when necessary to maintain the M13KO7 helper phage. Carbenicillin 100 µg mL−1 was added when necessary to maintain the phagemid. E. coli was grown overnight in 2YT + tetracycline, then subcultured 1/50–1/100x in the same and grown to exponential phase. Before use, E. coli were routinely tested for maintenance of the F episome and for pre-infection with phagemid or helper phage by confirming an exponential phase culture could grow in 2YT plus tetracycline and could not grow in 2YT plus carbenicillin or kanamycin.For each selection, one aliquot of library at 1 × 1013 pfu mL−1 was used; library was PEG-precipitated after thawing from −80 °C. Sufficient quantity of library for a given experiment was thawed and pooled; four volumes of ice-cold PBT (PBS plus 0.9 mM calcium, 0.9 M magnesium, 0.5% w/v bovine serum albumin [BSA], 0.05% v/v Tween-20, filter sterilized) and one volume ice cold sterile PEG-NaCl solution (20% w/v PEG-8000, 2.5 M NaCl) were added to one volume library. Phage were PEG-precipitated on ice for 20 min, harvested by centrifugation at 20,000 × g for 20 min; the phage pellet was resuspended in PBT and stored on ice. Immediately before use, phage solutions were spun as before for 5 min to remove insoluble material.Preparation of bait, washes, and elution are described below. After elution, phage were amplified by infecting E. coli. One volume of neutralized phage eluate was added to nine volumes (solid-phase and small-scale cell-based panning) or two volumes (high-throughput panning) of exponential-phase E. coli (OD 0.3–0.6). Infected cultures were grown for 45 min; helper phage M13KO7 (NEB) was added to 1010 pfu mL−1 final concentration and cultures were grown for 1 h. Finally, one volume culture was added to five volumes (solid-phase and small-scale) or one volume (high-throughput) 2YT, supplemented with carbenicillin and kanamycin to the proper final concentration. Cultures were grown for at least 16 h.After amplification, cells were removed by centrifugation at 3000 × g for 10 min. The supernatant was transferred to a new tube and one volume ice cold sterile PEG-NaCl solution was added to five volumes culture. Phage were PEG-precipitated on ice for 20 min (solid-phase and small-scale) or 1 h (high-throughput), then harvested at 20,000 × g for 20 min (solid-phase and small-scale) or 5500 × g for 1 h (high-throughput). Phage pellets were resuspended and spun before use as described above.Phage titers were measured by preparing serial 10-fold dilutions of phage particles in PT buffer (PBS plus calcium, magnesium, and 0.05% v/v Tween-20, filter sterilized), then transferring one volume of phage to 10 volumes of exponential phase E. coli. Infected cells were incubated shaking for 15–30 min, then 10 µL of each dilution was spotted onto LB plus carbenicillin and LB plus kanamycin plates. Colonies were enumerated using Ilastik 1.4.0 in object density mode42. The greatest dilution with more than 10 visible colonies was used to calculate the titer.To assay individual clones, the phagemid-infected E. coli culture was plated on LB agar plus carbenicillin. Individual colonies were picked to 1 mL of 2YT plus carbenicillin plus M13KO7 at 1 × 1010 pfu mL−1 and grown shaking overnight.Solid-phase panningSolid phase panning protocol was adapted from10. Briefly, for each antigen, four wells of a high-binding ELISA plate (Nunc MaxiSorp, Invitrogen) were coated with 0.5 µg of purified protein (or mock purification) diluted in sodium carbonate buffer (50 mM, pH 9.6) overnight at 4 °C. Wells were decanted and blocked for 90 min in blocking buffer (PBS plus 0.5% w/v BSA). 100 µL of input library at 1013 pfu ml−1 was added to each counter-selection well and incubated for 1 h at room temperature nutating. Phage were transferred from counter-selection to selection wells and incubated for 2 h. Phage were decanted and wells washed 10 times with PT buffer. Phage were eluted by adding 100 µL per well HCl (100 mM), incubating 5 min, then neutralizing with 1/8 volume (50 µL) of Tris 1 M, pH 11. Eluted phage were rescued and amplified as described above.Cell-based panningP. aeruginosa cells were grown overnight, then subcultured 1/50x into 5 mL of LB and grown to mid-exponential phase (OD 0.4–0.6); subculture, washing, and blocking of selection cells was staggered one hour later than counter-selection cells. Cells were harvested by centrifugation at 3000 × g for 5 min, then washed twice by resuspending in 1 mL PBS + MC and centrifuging at 2700 × g in a microcentrifuge. 1 × 107 cells were transferred to a 2 mL microcentrifuge tube, pelleted, resuspended in 2 mL blocking buffer, and incubated rocking for 1 h at RT. Counter-selection cells were collected by centrifugation, resuspended in 1 mL of re-precipitated, cleared library at 1 × 1013 pfu ml−1, and incubated for 1 h rocking at room temperature. Counter-selection cells were pelleted and selection cells resuspended in the supernatant; selection cells were incubated with the phage library for 2 h. Selection and counter-selection cells were pelleted, then washed four times in 1 mL PT buffer; cells were pelleted and supernatant decanted thoroughly. Phage were eluted by addition of 800 µl of 0.1 N HCl to cells and incubation for 5 min; cells were removed by centrifugation at 18,000 × g at 4 °C for 5 min; the supernatant was neutralized in 100 µL 1 M Tris-HCl, pH 11. Residual phage attached to cells were eluted by addition of 640 µL per well 0.1 M triethylamine (TEA) and incubation for 5 min. Cells were removed as before and supernatant neutralized in 260 µL per well of Tris pH 6.8. The neutralized supernatants from the acid elution and the base elution steps were combined to form the final eluate. 450 µL, approximately 1/4 of this volume, was expanded as above to generate the input library for subsequent rounds.High-throughput cell-based panningP. aeruginosa cells were arranged and stored in glycerol in 96-well format; 96-well microplates containing 150 µl per well of sterile LB agar were prepared. Two days prior to the experiment, 10 µL per well of sterile LB broth was added to each well and a multichannel pipette was used to inoculate this semi-solid master plate from the arrayed glycerol stocks. For bacteria panned after growth on solid media, a flame-sterilized loop was used to densely streak from this solid master plate 1/8th of a 100 mm petri dish containing LB agar; these plates were incubated overnight. For bacteria panned in liquid, a multichannel pipette was used to inoculate duplicate liquid cultures of 1.2 mL per well from the solid media microplate. These cultures were incubated overnight. For bacteria panned in exponential phase, the stationary phase cultures were diluted 1/50 in appropriate media and incubated for 3 h. Solid-media cultures were scraped into 1.2 mL per well of LB; solid-media and stationary-phase cultures were transferred to a single microplate along with exponential-phase cultures. Duplicate plates were combined and optical density was measured but not standardized. Subculture, wash, and blocking of the selection cells was staggered by 2 h after the counter-selection cells. Cells were harvested by centrifugation at 3000 × g for 10 min, washed twice with 1 mL per well PBT, and resuspended by shaking 750 rpm × 5 min in 100 µl per well of PBT. 250 µL of re-precipitated, cleared library at 1 × 1013 pfu ml−1 was added to counter-selection cells and incubated at RT shaking 750 rpm for 1 h. Counter-selection cells were removed by centrifugation 3000 × g for 15 min at RT; the supernatant was transferred to the selection cells and resuspended by gently pipetting up and down. Selection cells were incubated with phage for 1 h shaking, then 1 h nutating. Selection cells were pelleted, supernatant decanted, then selection and counter-selection cells were washed three times as above. Phage were eluted from selection and counter-selection cells by both acid and base as above, with the following modifications: add 200 µL per well 0.1 N HCl; shake 750 rpm for 10 min; add 25 µL 1 M Tris-HCl, pH 11 to neutralize; spin 3000 × g for 10 min; the supernatant was passed through a 0.22 µm PVDF filter (Millipore MSGVS2210). We found this filter was particularly effective at removing residual bacteria while allowing phage to pass through without clogging. The pellet was eluted again with base: add 200 µl per well of 0.1 M TEA; incubate shaking 750 rpm for 10 min; add 80 µL per well Tris pH 6.8 to neutralize; spin 3000 × g for 10 min; pass supernatant through a 0.22 µm filter. Filtered acid- and base-eluted eluates were combined, titered, and amplified.Phage were expanded as described above; 200 µl per well of eluate phage was added to 400 µl per well of exponential-phase E. coli in 96-well format, incubated, M13KO7 was added to a final concentration of 1 × 1010 pfu ml−1, incubated, and 2YT plus kanamycin and carbenicillin were added to a final volume of 1.2 mL per well. This culture was grown overnight. Cells were removed by centrifugation 6000 × g for 10 min and the supernatant transferred to a clean microplate on ice. PEG-precipitation was performed in this microplate; the PEG-precipitated phage were resuspended by shaking at 750 rpm for 10 min, then stored on ice overnight until the next round of panning was to be conducted.Extended cell-based panningPrior to subsequent rounds of panning, the round 5 input phage from the high-throughput panning were expanded as follows: 200 µL of phage at ~1 × 1013 pfu ml−1 was added to a 4.5 mL culture of OmniMAX E. coli cells in 2YT plus tetracycline grown to OD 0.3–0.6; cultures were incubated shaking at 37 °C for 30 min, then M13KO7 added to 1 × 1010 pfu ml−1 and incubated for 45 min. The entire volume of culture was added to a final volume of 30 mL 2YT plus carbencillin and kanamycin and incubated overnight. Phage particles were harvested by PEG-precipitation as described earlier for the small-scale panning experiments.Subsequent rounds of phage display panning were performed as above, except that three separate sets of counter-selection cells were prepared identically. Phage were incubated with the first set of counter-selection cells for 30 min, then cells removed and the supernatant transferred to the second set of counter-selection cells. This process was repeated once more for a total of three counter-selection phases. After applying phage to selection cells, the number of washes was increased from three (rounds 1–4) to five (round 5), seven (round 6), and twelve (round 7).Library preparation for high-throughput sequencingFor initial high-throughput sequencing of the input library, phagemid DNA was purified, digested using NotI/AscI, and gel purified. The sequencing library was prepared by end-repair and blunt-end ligation of sequencing adapters using a commercial kit (Illumina Nextera). Paired-end 300 bp sequencing was performed on an Illumina MiSeq.Subsequent HTS libraries were prepared by polymerase chain reaction (PCR). 2 µL of amplified phage particles at ~1 × 1013 pfu ml−1, or 10 µL of phage eluate were used as template for an initial PCR (Phusion, NEB) with primers CDR123-seq-F/CDR123-seq-R which added Illumina R1 and R2 primer binding sites. Silica column cleanup was performed and an additional PCR added combinatorial dual indexes. Yield and size of all products was verified by gel electrophoresis and individual reactions repeated as necessary. Amplicons were pooled using Just-a-Plate 96 PCR Normalization (Charm Biotech). The pooled library was column- and then gel purified. Library purity and concentration was verified by automated electrophoresis (e.g. Fragment Analyzer, Agilent) and qPCR. Paired-end 150 bp sequencing was performed in an Illumina NovaSeq S4 to a target depth of 100,000 reads per round per selection.High-throughput sequencing data analysisPrimer sequences were removed with Cutadapt43, and reads lacking a primer sequence were discarded. Reads were deduplicated and denoised using DADA233, with each sample processed independently. Distinct read sequences were aligned to the reference sequence described above using Bowtie 244. For paired-end 150 nt sequencing experiments, the reads were not expected to overlap in all cases, as the full-length sequenced amplicon was greater than 300 nt; to ensure that a given read pair captured the full length of the variable portion of the VHH gene, the forward read was required to contain the entirety of CDR1 and CDR2, plus at least 3 nt, while the reverse read was required to contain the entirety of CDR3 plus at least 3 nt. Reads which did not meet this criteria were discarded. Overlapping read pairs were stitched together, with any disagreements being resolved in favor of the forward read. For read pairs which passed the filter but did not overlap, the gap between the two reads was filled with corresponding bases from the reference sequence. The merged read pairs were translated to amino acid sequences (amber stop codons UAG were translated as Gln). Amino acid sequences were grouped into zero edit-distance clusters (i.e. two or more reads with 100% identity and overlap beyond a threshold length were grouped together) using the linclust routine45 from the MMseqs2 package46. Amino acid sequences were then aligned to a reference sequence in order to identify the boundaries of CDR1–3. Reads were filtered according to minimum lengths for CDR1 (≥4 aa), CDR2 (≥6 aa), CDR3 (≥3 aa), and FR4 (≥2 aa) and the full-length amino acid sequences (≥69 aa). Two feature tables (contingency tables of sample vs. feature) were constructed in the biom format47, one where the columns (features) were distinct amino acid sequences (e.g. all nucleic acid sequences with the same translation were summed) and one where the columns were distinct CDR3 sequences. Feature tables from multiple sequencing runs, if applicable, were summed. Relative abundance was calculated by dividing counts by the sum for each sample (row). Enrichment was calculated for each feature for each selection by dividing the ending abundance by the starting abundance. The starting abundance was defined as either: the relative abundance of that feature in the input library or, if undefined, the relative abundance of that feature in the first round of selection where the feature was observed. The ending abundance was the relative abundance of that feature in the last round of selection.Diversity analysisThe number of distinct full-length VHH amino acid sequences (encompassing the region beginning with CDR1 and ending after CDR3) in the input library was estimated using the breakaway method from the breakaway R package48.VHH selection metricsLet \({X}_{i,\, j,r}\) be the relative abundance of VHH \(j\) in selection \(i\) at round \(r\). Define \({E}_{i,j}={X}_{i,\, j,R}/{X}_{i,\, j,1}\) as the enrichment of VHH \(j\) in selection \(i\) during a selection campaign of \(R\) rounds.We defined the enrichment probability, \(P\big({E}_{i,\, j} \, > \, e{{{\rm{|}}}}{X}_{i,\, j,0}=x\big)\), as the probability of observing an enrichment \(e\), given a starting abundance of \(x\), due only to experimental noise unrelated to the selective pressures of phage display panning. To estimate this quantity, we repeatedly re-sequenced two passages of the input library. This sample captures both variation due to library preparation and due to changes in phage proportion during expansion in E. coli, precipitation of the phage, etc. We obtained \(K=12\) samples of the library and calculated a table of enrichments and initial abundances: \({\hat{E}}_{i,j}={X}_{k1,j}/{X}_{k2,\, j}{{{\rm{;}}}}\, {\hat{X}}_{i,j,0}={X}_{k1,\, j},\, \forall k1,\,k2\in \left(1…K\right)\times \left(1…K\right),\, \, j\in \left(1…J\right),\, i\in \big(0…{K}^{2}\big)\). For all further calculations, we took the logarithm of both abundance and enrichment. We visually confirmed this joint distribution was unimodal. We then performed a Gaussian kernel density estimate on this table to estimate a joint probability mass function \(f\left(e,\, x\right)=P\big({\hat{E}}_{i,\, j}=e\cap {\hat{X}}_{i,\, j,0}=x \big)\). We evaluated this function on a 2D grid of abundances and enrichments, then summed over the first axis to calculate the marginal probability of abundance (e.g. \(P\big({\hat{X}}_{i,j,0}=x\big)={\sum }_{e}\, f\left(e,\, x\right)\)). We divided the discretized joint PMF by the marginal distribution of abundance to create a conditional PMF. We took the cumulative sum of these quantities over the first axis to calculate a conditional cumulative distribution function. Then we fitted a 2D spline to this surface in order to estimate the enrichment probabilities for unknown values.To calculate the binary enrichment probability, \(P\left({N}_{k}^{+} < \, {N}_{k}^{-}\right)\), we used simulations. For a given antigen, assume there are \(k\) antigen-positive selections, and a particular VHH is significantly enriched in \({N}_{k}^{+}\) of those \(k\) selections. Among \(k\) random antigen-negative selections, let \({N}_{k}^{-}\) be the number of antigen-negative selections where the VHH is significantly enriched. \({N}_{k}^{-}\) is a random variable with a support of \(\left(0…k\right)\). We estimated this probability by repeatedly choosing \(k\) antigen-negative selections, counting \({\hat{N}}_{k}^{-}\) for those selections, then counting the fraction of these simulations where \({N}_{k}^{+} \, < \, {\hat{N}}_{k}^{-}\). We only performed a finite number of simulations, so if \({N}_{k}^{+} \, < \, {\hat{N}}_{k}^{-}\) in all simulations, we say \(P\left({N}_{k}^{+} \, < \, {N}_{k}^{-}\right)\, < \,1/q\), where \(q\) was the number of simulations.To calculate the normalized rank: Let \({E}_{i,j}\) be the enrichment of enrichment of VHH \(j\) in selection \(i\); let \({E}_{i,\left(0\right)} \, < \, {E}_{i,\left(j\right)} \, < \, {E}_{i,\left(J\right)}\) be the ranks of VHHs in selection \(i\) where \(J\) VHHs were observed. The normalized rank of VHH \(j\) in selection \(i\) is the rank divided by the number of VHHs in a sample, i.e. \({\mbox{NR}}\left(i,\, j\right)={E}_{i,\left(j\right)}/J\). The normalized rank sum for VHH \(j\) in a group of selections \(S=\{{s}_{0},\, {s}_{1},\ldots,\, {s}_{N}\}\), is the sum of the normalized ranks, \({\mbox{NR}}\left(S,\, j\right)={\sum }_{i\in s}{{{\rm{NR}}}}\left(i,j\right)\).OrdinationRelative abundances at the final round of selection were normalized for sequencing depth with the scran package49,50. truncated single value decomposition (TSVD) was performed using the scikit-learn package51 with the top 100 components. For CCA, ordination was performed on the log-transformed enrichment matrix using the scikit-bio package52.Machine learning and antigen predictionsClassifiers were gradient-boosted trees trained using XGBoost via the scikit-learn package51. Models were evaluated with repeated stratified \(k\)-fold cross-validation, with \(k=5\), and \(n=15\) repeats, and scored by mean AUROC over all test folds. The effect of learning rate \(\eta\) was evaluated for several antigens and optimal values (the fastest learning rate that still achieved maximum mean AUROC) found to be ~0.01–0.05. For each antigen, models were trained using three datasets: enrichment matrix only, enrichment plus final round abundance, or enrichment plus abundance for all rounds; the dataset with highest performance was chosen per-antigen. Additional model hyperparameters (e.g. min_child_weight, subsample, colsample_bytree, gamma, reg_alpha) were chosen by Bayesian optimization using the BayesSearchCV class from the scikit-optimize package53. Using this final set of hyperparameters, models were trained and evaluated using the cross-validation procedure described above; receiver-operator characteristic curves (ROC) for each training fold, plus a mean over all folds, are shown in Fig. 6d–g. \(p\)-value for mean accuracy determined by one-sided Student’s \(t\)-test. \(p\)-value for mean AUROC >0.5 was determined by Mann-Whitney U test.Expression of recombinant VHHs in bacteriaPhagemid dsDNA was purified from overnight cultures of phagemid-infected E. coli, digested with NotI/AscI, and the gel-purified insert was ligated with gel-purified, NotI/AscI-digested pJEG3, then transformed into DH5α. Clones were verified by Sanger sequencing, then transformed into BL21(DE3) for expression. BL21(DE3) cells were grown overnight in 2YT plus carbenicillin plus 2% w/v glucose, subcultured 1:20 in 100 mL of the same, then media removed and cells transferred to 200 mL 2YT plus carbenicillin and 1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). Cells were grown 19 h shaking at 27 °C. Cells were harvested at 17,650 × g in JA-10 rotor (10,000 rpm) and resuspended in a lysis buffer (50 mM sodium phosphate [pH 8.0], 300 mM NaCl, 10 mM imidazole, 1x cOmplete™, EDTA-free Protease Inhibitor Cocktail [Sigma], 1 mM PMSF). Cells were spheroplasted for 30 min at RT by addition of lysozyme to ~1 mg ml−1 and DNaseI to 5 µg ml−1, then lysed by three freeze-thaw cycles. Lysates were cleared by centrifugation 20,000 × g at 4 °C for 20 min. Immobilized metal affinity chromatography (IMAC) was performed; Ni-NTA resin (2.5–5 mg protein/1 mL slurry) were diluted 1/5x in sample and lysis buffer. Slurry was packed on a gravity filtration column and washed with eight bed volumes of wash buffer (50 mM sodium phosphate [pH 8.0], 300 mM NaCl, 20 mM imidazole) and 1/2 bed volume of 100 mM imidazole. Protein was eluted with 2 bed volumes of 250 mM imidazole, and buffer exchanged to PBS + MC using PD-10 desalting columns (Cytiva). Samples were flash-frozen and stored at −20 °C. Yield was measured by BCA assay and purity monitored by SDS-PAGE.Expression and purification of recombinant VHHs in human cellsNucleic acid sequences were synthesized as gene fragments (Twist Biosciences), reconstituted, and amplified by PCR. Unpurified amplicons were cloned using NEBuilder HiFi DNA Assembly system into the pCER243 mammalian expression vector, which had previously been linearized with XhoI and gel purified. pCER243 is modified from pD2610-v12 (ATUM Bio), adding an upstream H7 leader sequence and downstream an in-frame (GGGGS)3 linker and human IgG1 Fc with the N297A mutation54. Assembly products were transformed into DH5-alpha competent E. coli (NEB), plated on solid selective media, scraped, and grown overnight in liquid; plasmids were purified. For each VHH clone, presence of the insert was verified by NotI/XbaI digestion. rVHH-Fcs were expressed by transient transfection of the pCER243-rVHH plasmid using the Expi293F Transfection Kit (Gibco), following the manufacturer’s instructions. Briefly, Expi293F cells were grown in Expi293 media in shaking at 125 rpm in non-baffled, vent-cap polystyrene Erlenmeyer flasks at 37 °C, 8% CO2 to a density of 3–5 × 106 cells ml−1. Cells were seeded to 2 × 106 cells ml−1 and grown shaking overnight. In a 2 mL, V-bottom 96-well plate, 0.5 µg plasmid DNA was diluted in Opti-MEM media (Gibco) to 25 µL final volume. A master mix of 1.35 µl per well Expifectamine and 23.65 µl per well Opti-MEM media was prepared, then 25 µL per well of this mixture was added to each well and incubated at room temperature for 20 min to form DNA-Expifectamine complexes. Expi293F cells were diluted to 2 × 106 cells ml−1 and 425 µl cells were added dropwise to this mixture. Cells were grown shaking at 1200 rpm in conditions described above. After at least 20 h of growth, 2.5 µL of transfection enhancer 1 and 25 µL of transfection enhancer 2 were added to each well, and cells were returned to grow for 4 days further (5 days total).rVHHs were purified from clarified supernatants using protein A magnetic beads. Cultures were clarified by centrifugation at 600 × g for 10 min at room temperature. 12.5 µL protein A magnetic beads (Lytic Solutions) were added to supernatants and supernatants shaken at 750 rpm for 2 h at 4 °C. Beads were harvested, washed three times in PBS, then rVHHs eluted in 100 µL glycine (100 mM, pH 3.0–3.2). Eluates were neutralized with 10 µL tris (1 M, pH 8) and 100 µL PBS. Concentration was measured using the Pierce BCA assay kit (Thermo Scientific). rVHHs with yields > 100 ng/µL in final eluate were tested for activity.Standard ELISAsMaxiSorp plates were coated with antigen and blocked as described above for solid-phase phage display. Each condition was performed in triplicate. Individual phage clones were grown as above; cells removed by centrifugation 3000 × g for 10 min, then supernatants diluted 1/3x in PBT. Rabbit antisera was used as a positive control; YU573 is rabbit polyclonal antisera raised against PAK flagella55; YU586 is polyclonal antisera raised against PAK pilin56. Antisera was diluted 1/10,000 in PBT. 100 µL per well of diluted phage or antisera was applied to coated wells and incubated 1 h, nutating. Liquid was decanted and wells washed 4 times with 100 µL per well PT buffer. Secondary antibodies (Goat anti-rabbit::HRP [Bio-Rad Cat #170-6515, RRID:AB_11125142], Mouse anti-M13::HRP [Sino Biological Cat #11973-MM05T, RRID:AB_2857926]) were diluted 1/3000x in PBT; 100 µl per well was applied and incubated 45 min at RT. Wells were washed 5 times, then plates blotted dry on paper towel. 3,3’,5,5’-Tetramethylbenzidine (TMB), 100 µL per well was added and incubated at RT until the control wells containing no primary antibody began to turn blue. 100 µL per well 1 N H2SO4 was used to quench the reaction, and absorbance read at 450 nm.Cell-based ELISAsPlates were coated with cells and fixed with methanol following the method of57, with modifications. Overnight cultures of P. aeruginosa were subcultured 1/50x and grown to mid-exponential phase, harvested by centrifugation 3000 × g × 10 min, resuspended and washed twice in PBS + MC. Cells were diluted to OD 0.3 and 100 µL per well (~3 × 107) were added to a microplate. Cells were incubated at 37 °C for 2 h. 100 µL per well of ice cold methanol was added and cells were incubated rocking at room temperature for 10 min. Methanol was removed by gentle aspiration from the wall of the well, and plates were dried 10 min uncovered in the fume hood. PEG-precipitated phage clones at ~1 × 1013 pfu ml−1 were diluted 1/100 in PBT; antisera were diluted 1/20,000x in PBT; 100 µL per well of primary antibody was added and incubated 2 hours with fixed cells. Wells were washed five times with PT, then stained with secondary antibody and developed as above.Cell-based ELISAs with bacterial rVHHs were performed as above, except purified rVHHs at 2 mg ml−1 were diluted 1/100 and used as primary antibodies. For brVHHs, rabbit anti-E-tag (Novus Biologicals Cat# NB600-527, RRID:AB_10001463), 1 mg ml−1 was used at 1/500x dilution as the secondary antibody; cells were incubated 30 minutes and washed three times. Goat anti-rabbit::HRP (Bio-Rad Cat #170-6515, RRID:AB_11125142) was used as the tertiary antibody for detection.Live cell-based ELISAsFor cell-based ELISAs performed on live cells, overnight cultures were subcultured 1/50x and grown to mid-exponential phase, then harvested by centrifugation 3000 × g for 5 min. Cells were washed once in PBS, then ~1 × 109 cells were resuspended in PBS + 0.5% BSA + 0.05% Tween-20, added to a microcentrifuge tube, and incubated rocking for 1 h. Cells were pelleted, then resuspended in 1 mL of PBS + 0.5% BSA + primary antibody (either antiserum at 1/100x dilution or brVHH @ ~40 µg ml−1 final concentration) + SYTO9 at 1/250x and incubated rocking for 1 h protected from light. Cells were pelleted and washed once; brVHH samples were resuspended in secondary antibody rabbit anti-E-tag (Novus Biologicals Cat# NB600-527, RRID:AB_10001463), 1 mg ml−1 at 1/500x dilution in 1 mL PBS + 0.5% BSA + 0.05% Tween-20; antisera samples were resuspended in buffer; samples were incubated 1 h, then pelleted and washed once. Cells were resuspended in Goat anti-rabbit::HRP (Bio-Rad Cat #170-6515, RRID:AB_11125142) tertiary antibody at 1/1000x and incubated 1 h, then pelleted and washed once. Cells were resuspended in 100 µL of PBS, then diluted 1/2x, 1/4x, or 1/8x; each dilution of cells was added to the plate. OD600, SYTO9 fluorescence, and A450 were measured. TMB and H2SO4 were added as with standard ELISA. A450 signal saturated detection in some wells, so samples were diluted 1/2x in PBS and read.Bacterial dot blotsP. aeruginosa cells in stationary phase were harvested, washed once in PBS, and diluted to 2 × 1010 cells ml−1 (nominal OD = 20). Serial 5-fold dilutions were performed in PBS. 3 µL of cells were spotted at each dilution on a 2.5 × 2.5 cm nitrocellulose membrane. Cells were allowed to dry 30 min, then stained 5 min in 0.1% w/v Ponceau S stain in 1% glacial acetic acid and destained 2 min in water. Blots were blocked and fully destained in 1x TBST (Tris HCl pH 8, 100 mM; NaCl, 1.5 M; Tween-20, 0.05% v/v) plus 5% non-fat milk. Blots were washed once with TBST and probed with primary antibody. 3 µg of purified rVHH was diluted in 100 µL of TBST, spotted into a clean 6-well plate, and the membrane placed face down in this puddle. Plates were sealed with tape and incubated overnight at 4 °C. Blots were washed three times for 5 min with 2 mL TBST. Secondary antibodies (Goat anti-Human IgG::HRP, Secondary Antibody, Invitrogen Cat# SA5-10283, RRID:AB_2868331 @ 1 mg ml−1) were diluted 1/5000x in 500 µL TBST and applied to blots; blots were incubated 30 min at RT and washed three times. Enhanced chemiluminescence (ECL) substrate was prepared (100 mM Tris pH 8.5, 1.25 mM luminol, 225 µM coumaric acid, 0.00001% H2O2); 1 mL ECL substrate was added to each blot, then blots were imaged for 3 min using a Chemidoc MP (Bio-Rad).Live cell dot blotsStationary phase cells were washed once in PBS and diluted to 1.5 × 108 cells ml−1, then 7.5 × 106 cells (50 µL) per well were added to a 2 mL V-bottom 96-well plate. SYTO9 was diluted to a final concentration of 1/250x and added along with 300 ng per well of rVHH to cells for in a total volume of 100 µl per well. Pre-cleared polyclonal antisera YU573 and YU586 were used at a final concentration of 1/100x. Cells were incubated, protected from light, for 40 min, then washed twice by addition of 500 µL per well PBS, centrifugation 4000 × g for 10 min @ 10 °C, and decanting. Cells were resuspended in 50 µL PBS by pipetting, then 5 µL per well was spotted onto a nitrocellulose membrane and allowed to dry for 30 min protected from light. The membrane was blocked in 5% nonfat milk + TBST at 4 °C overnight, then rinsed and stained with secondary as above. Before addition of the enhanced chemiluminescenc (ECL) substrate, blot was exposed for fluorescence in the AlexaFluor 488 channel (Ex = 460 − 490 nm, Em = 518–546 nm, 0.05 s). Blot was developed and imaged for ECL (10 min exposure).Bacterial flow cytometryAll solutions used for FCM were passed through a 0.22 µm filter on the day of the experiment.Bacterial cells in stationary phase (16 h of growth) were harvested in round-bottom culture flasks by centrifugation 3500 × g for 10 min at room temperature. Supernatants were decanted, the pellet disrupted by dragging the tube across a rack (“racking”), cells resuspended in 1 mL FACS buffer (PBS supplemented with 0.05% v/v Tween-20 and 0.5% w/v BSA, filter sterilized), and suspension transferred to 1.7 mL microcentrifuge tubes. Cells were centrifuged at 3500 × g, room temperature for 5 min, decanted, pellet disrupted by racking, then resuspended in 1 mL FACS buffer by flicking and inverting for 1–5 min. We found gentle handling (disrupting pellet by racking, resuspension by flicking and inverting rather than vortexing, low speed centrifugation) was critical to achieving reproducible staining of P. aeruginosa. Cells were resuspended to an OD of 0.2 and 50 µL of cells (1 × 107 cells) per condition added to a 2 mL 96-well V-bottom plate (Corning). The primary antibody (or VHH) was diluted in FACS buffer to 2x final concentration, then 50 µL primary antibody added to cells. Cells were incubated with primary antibody at room temperature on a microplate shaker (750 rpm) for 30 min. Soluble IgG1-Fc served as a negative control. To wash, 500 µL per well of FACS buffer was added; bacteria were collected by centrifugation at 4000 × g, 10 °C for 10 min and buffer was decanted. The wash was repeated once. A mixture of secondary antibody (PE Goat polyclonal anti-Human IgG Fc, eBioscience, [Thermo Fisher Scientific Cat# 12-4998-82, RRID:AB_465926], 1 µg per well; or PE Donkey anti-rabbit IgG [BioLegend Cat# 406421, RRID:AB_2563484], 0.125 µg per well) and SYTO9 nuclear stain at 1/500x dilution was prepared in FACS buffer; 50 µL per well of this mixture was added to equal volume of cells and mixed by gently pipetting up and down 10 times with a multichannel pipette. Cells were incubated with the secondary antibody as above, for 15 min. The cells were washed once as above.Cells were fixed in 1% paraformaldehyde (PFA) to prevent efflux of the nuclear stain SYTO9. Briefly, paraformaldehyde (Electron Microscopy Solutions) 16% w/v, was diluted to 4% in filter-sterilized PBS and stored at −20 °C with minimal headroom (to prevent oxidation). Single-use aliquots of 4% w/v PFA were thawed and diluted to 1% final concentration in sterile PBS. Cells were incubated with 1% PFA at room temperature as above for 20 min. Excess aldehydes were quenched by addition of 0.75 M Tris pH 8 and cells incubated for 10 min further. Cells were washed by addition of 500 µL FACS buffer, centrifuged, and decanted as above. Cells were resuspended in 250 µL FACS buffer (5x dilution compared to their original density), then 200 µL cells were passed through a 40 µm filter mesh. Cells were analyzed on a CytoFLEX LX flow cytometer (Beckman-Coulter) with the following settings: flow rate, 15 µL/min; event rate-setting, high; threshold, side-scatter height (SSC-H) > 10,000; voltages, FSC = 165, SSC = 400, B525-FITC = 150, Y585-PE = 500–1000. Data was collected for at least 20 s or until >50,000 events were recorded in the gate identifying bacteria.Data were gated to identify single bacterial cells (see Supplementary Fig. 30). To distinguish intact bacteria from similarly-sized debris, events with SYTO9 fluorescence greater than that of unstained cells were marked as bacteria. Analysis was performed using FlowJo.For each rVHH, we drew an “rVHH +” gate starting at the 99th percentile of fluorescence for the antigen-negative cells stained with that rVHH; we then calculated the fraction of antigen-positive cells that were rVHH-positive, as well as the mean fluorescence intensity for both antigen-negative and antigen-positive cells. We considered an rVHH successful if the %rVHH+ for antigen-positive cells was at least three times higher than the %rVHH+ for antigen-negative cells. This threshold was established by performing the same gating procedure on the isotype control and observing that the %rVHH+ cells for the antigen-positive population was never greater than three times that for the antigen-negative cells.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.