General information for plasmid and strain constructions and media

All cloning procedures were performed using Escherichia coli strain MC1061. Saccharomyces cerevisiae strain R1158, described in [49], served as the background for generating the doxycycline-repressible S. cerevisiae PDR1 strain (ScPDR1-DOX, details below).

All antimicrobials were prepared as 1,000× stock solutions: ampicillin (AMP), doxycycline (DOX), geneticin (G418), and nourseothricin (NAT) were dissolved in sterile water, filtered, and stored at −20 °C; posaconazole (POSA) and itraconazole (ITR) were dissolved in dimethyl sulfoxide (DMSO) and also stored at −20 °C.

For bacterial liquid cultures, E. coli was grown in LB medium (0.5% yeast extract, 1% tryptone, 1% NaCl) at 37 °C with shaking at 250 rpm. For solid cultures, cells were grown on 2YT agar plates (1% yeast extract, 1.6% tryptone, 0.5% NaCl, 0.2% glucose, 2% agar) and incubated at 37 °C. When required, AMP was added at a final concentration of 100 μg/mL for plasmid selection. S. cerevisiae was cultured in YPD medium (1% yeast extract, 2% tryptone, 2% glucose) at 30 °C with shaking at 250 rpm. For solid cultures, 2% agar was added to YPD. Yeast transformants with pRS31N plasmids were selected on YPD plates supplemented with 100 μg/mL G418 and 100 μg/mL NAT.

All PCRs were performed using KAPA High-Fidelity HotStart polymerase unless otherwise specified. Strains, reagents, plasmids, computational tools, and additional resources are listed in S2 Table. Oligos are listed in S3 Table, and PCR conditions in S4 Table.

All transformations were performed using Escherichia coli NEB 5-alpha Competent Cells (High Efficiency) following the manufacturer’s recommended heat-shock protocol. Briefly, 2–5 µL of assembly reaction (Gibson or Golden Gate; see below) was added to 50 µL of chemically competent cells, followed by incubation on ice for 30 min, a 30-s heat shock at 42 °C, and immediate cooling on ice for 5 min. Cells were then recovered in 450 µL of SOC medium (0.5% Yeast Extract, 2% Vegetable Peptone, 0.5% Yeast Extract, 10 mM NaCl, 2.5 mM KCl, 10 mM MgCl₂, 10 mM MgSO₄, and 20 mM glucose) and incubated at 37 °C for 60 min with shaking (250 rpm) before plating on selective agar plates. The complete transformation protocol is available from New England Biolabs [50].

The integrity of all assembled and mutagenized plasmids was verified either by Sanger sequencing (Plateforme de séquençage et de génotypage des génomes, Centre de recherche du centre hospitalier de Québec-Université Laval (CRCHUL), Canada) or by whole plasmid sequencing using Oxford Nanopore Technology with custom analysis and annotation (Plasmidsaurus, USA or Flow Genomics, Canada).

pRS31N-ScPDR1 WT

S. cerevisiae PDR1 (PDR1) was expressed from a yeast centromeric plasmid pRS31N-ScPDR1 that was constructed as follows. First, a plasmid containing the PDR1 sequence was created, including its promoter and terminator. The plasmid pRS31N containing AMP and NAT resistance markers was doubly digested with 20 U of both HindIII-HF and XbaI restriction enzymes. The insert, a fragment containing S. cerevisiae PDR1 native promoter (834 bp), coding sequence (3,207 bp), and terminator (345 bp), was amplified from the pMoBY-ScPDR1 plasmid with primers adding homology to the pRS31N plasmid cloning site (PCR 1). The resulting PCR product corresponding to the insert was incubated for 1h30 at 37 °C with 10 U of DpnI enzyme to remove parental DNA. Subsequently, the plasmid and insert were purified on magnetic beads and assembled by Gibson assembly to obtain pRS31N-ScPDR1.

As a BsaI restriction site is used in the construction of the PDR1 DMS plasmid library, we had to remove three BsaI restriction sites in pRS31N-ScPDR1. The three sites were mutated by site-directed mutagenesis (site 1) or by homologous recombination with a short double-stranded DNA fragment (gBlock) corresponding to PDR1 coding sequence between positions 2238 and 2720 without BsaI restriction sites (sites 2 and 3). The BsaI site 1 was removed by performing site-directed mutagenesis based on the QuickChange Site-Directed Mutagenesis System (Stratagene, La Jolla, CA). We amplified the pRS31N-ScPDR1 plasmid using a pair of primers containing the desired mutation at the center (PCR 2). The PCR product was then incubated for 1h30 at 37 °C with 10 U of DpnI enzyme to remove parental DNA, and pRS31N-ScPDR1 mutated plasmids (pRS31N-ScPDR1 minus site 1) were retrieved directly by transformation in bacteria. For sites 2 and 3, the plasmid pRS31N-ScPDR1 minus site 1 was first PCR amplified (PCR 3). The resulting PCR product was incubated for 1h30 at 37 °C with 10 U of DpnI enzyme to remove parental DNA and then, purified on magnetic beads. Finally, the plasmid and gBlock insert were assembled by Gibson assembly to obtain pRS31N-ScPDR1 WT. For clarity, throughout this study, pRS31N-ScPDR1 WT refers to the plasmid after the BsaI site replacements, which served as the template for constructing the whole PDR1 DMS plasmid library.

oPools

The DMS library was generated using only the PDR1 coding sequence (CDS). Due to the limited length of synthetic DNA and to enable barcode-mutation association with short-read sequencing, the 3,204 bp PDR1 gene was divided into 43 fragments of 75 bp each, except the last one which is of 54 bp. Fragment 1 goes from codon 1–25, fragment 2 goes from residue 26–50, and so on, up to the last fragment 43 (the smallest) going from residue 1,050–1,068. For each fragment, a library was built using affordable pools of degenerate oligonucleotides (oPools). oPools are composed of five parts designed as follows (Fig 1A):

1. ~40 bp of homology upstream of the PDR1 fragment of interest;
2. PDR1 sequence of the fragment containing a single NNK codon (32 codons, including a stop codon);
3. BsaI module (two BsaI sites separated by CCGAAGCT to avoid self-dimerization); and
4. 30 bp DNA barcode with six random N and two fixed sequences (12 bp each) for each codon position. This design facilitates the identification of the mutated codon while allowing multiple distinct barcodes to represent the same mutation. As a result, each unique mutation is linked to several barcodes, providing internal replicates for downstream DMS analyses. oPools currently only allow for nine degenerate positions per DNA fragment. Three are used for the codon and six for the barcode;
5. the sequencing PBS_i7 site (CTGTCTCTTATACACATCTCCGAGCCCACGAGAC), which is common to all oPools and used for rapid and efficient short-read sequencing library preparation.

To generate the barcodes, we use the following Python script: github.com/lzamparo/DNAbarcodes, which generates all possible 12-nucleotide barcode sequences. From these, 30,000 barcodes with a GC content between 40 and 60% and a hamming distance of 3 were selected. We then removed barcodes in which BsaI was present to prevent any problem in the Golden Gate assembly step of our protocol. We also removed barcodes with two identical bases at the beginning or at the end to prevent stretches of the same bases. We split our filtered barcodes based on their GC content. We generated all individual 30 bp barcodes for the oPools by combining two 12 bp barcodes with six degenerate nucleotides, NN+barcode1+NN+barcode2+NN. We made sure the GC content of the sum of the two barcodes would be ~50%.

All oPool sequences were then generated using a custom Jupyter notebook (opool_generator_20240212.ipynb). We cut the PDR1 coding sequence into 75 bp fragments (25 codons per fragment), and we identified the 40–90 bp before and the 4 bp after each fragment. For each fragment, we generated 25 variant oligonucleotides in which a single codon was replaced by an NNK degenerate codon. Each oPool contains 25 different oligonucleotide sequences, each targeting one specific codon. The final sequences were generated by combining the 40–90 bp before the fragment, the NNK sequence, the 4 bp after the fragment, the BsaI module (GGAGACCgaagCTGGTCTCGACAG), one 30 bp barcode and the PBS_i7 (CTGTCTCTTATACACATCTCCGAGCCCACGAGAC).

All oPools are flanked by homology sequences at both the 5′ end (40 bp upstream of the target fragment within the PDR1 sequence) and the 3′ end (PBS_i7 sequence, facilitating rapid sequencing library preparation), for a total of 186–258 bp per oPool. The complete list of oPool sequences used in this study is provided in S1 Table.

Because cloning applications require double-stranded DNA inserts, oPools were amplified using two PCR reactions. The first PCR consisted of a single cycle using only the reverse primer (PCR4). Amplicons were purified using magnetic beads at a bead:DNA ratio of 1.8 to enrich for full-length oligonucleotides and to minimize the incorporation of truncated sequences (linked to oPool incomplete synthesis) into the final product. Then, 2 µL of the purified product was used as template for a second PCR, involving both forward and reverse primers and limited to five cycles to reduce PCR-related errors (PCR5). Each oPool amplification was performed in four replicates, which were pooled after the second PCR to maximize the diversity of variants carried into downstream cloning steps. This approach is intended to improve the overall quality of the oPools, ensuring sufficient insert quantity and diversity for subsequent steps of DMS library construction.

Cloning strategy of the DMS plasmid library (Fig 1)

The cloning strategy was divided into two steps: Gibson assembly followed by Golden Gate assembly.

In the first step, the plasmid backbone was amplified from pRS31N-ScPDR1 WT (PCR 6). The amplified region includes homology arms corresponding to the sequence upstream of the targeted fragment and extends to the plasmid region downstream of the terminator (i7 region). The resulting linear PCR product was incubated for 1h30 at 37 °C with 10 U of DpnI enzyme to remove parental plasmid DNA and then, purified using magnetic beads. The oPools are amplified as described in the oPools section. Backbone and insert fragments were assembled using Gibson assembly protocol [51]. To reach a target of ~25,000 transformants per fragment, necessary to ensure good barcode diversity and mutation coverage (Fig 1), two parallel Gibson reactions (10 μL each) followed by ~5 transformations (3 μL of Gibson reaction per transformation) using NEB 5-alpha Competent E. coli cells were performed for each fragment. Following transformations, 5 mL of 2YT medium was added directly onto each agar plate, and colonies were scraped using a sterile glass spreader. All transformants corresponding to a given fragment were pooled together. Aliquots were prepared, each corresponding to an optical density (OD₆₀₀) of 20 per fragment, and stored at −80 °C without (cell pellet) or with (glycerol stock) 15% glycerol. Plasmid libraries were extracted and purified from one cell pellet aliquot per fragment. These purified plasmids served as backbones for the subsequent cloning step.

In the second step, the PDR1 missing 3′ end region, including the 3′ end coding sequence, terminator, and PBS_i5 (Fig 1A), was amplified (PCR 7). BsaI recognition sites were added to both ends of the amplicon to generate non-palindromic overhangs compatible with those in the Gibson-assembled plasmids. The resulting PCR product was incubated for 1h30 at 37 °C with 10 U of DpnI enzyme to remove parental DNA and then, purified on magnetic beads. The previously prepared plasmid libraries served as vectors, and the PDR1 missing 3′ region acted as inserts for Golden Gate assembly, a one-pot cloning method alternating between BsaI digestion and ligation (PCR 8). To ensure sufficient barcode diversity and adequate mutation coverage (Fig 1), approximately 100,000 transformants per fragment were needed. To achieve this, two parallel Golden Gate reactions (20 μL each) and ~10 transformations (2.5 μL of Golden Gate reaction per transformation) were performed for each fragment using NEB 5-alpha Competent E. coli cells, according to the manufacturer’s instructions. After transformation, 5 mL of 2YT medium was added directly to each agar plate, and colonies were scraped using a sterile glass spreader. All transformants corresponding to a given fragment were pooled together. Aliquots were prepared, each corresponding to an optical density (OD₆₀₀) of 80 per fragment, and stored at −80 °C without (cell pellet) or with (glycerol stock) 15% glycerol. Plasmid libraries were extracted and purified from one cell pellet aliquot per fragment. The resulting plasmid preparations corresponded to the final PDR1 DMS plasmid libraries, with a distinct library for each PDR1 fragment.

Plasmid library quality control by sequencing

At each cloning step, library quality was assessed by high-throughput sequencing, focusing on parameters such as mutation coverage and barcode diversity per mutation (Figs 1C, 1D, and S1–S4).

Sequencing library preparation

To perform sequencing-based quality control after Gibson assembly, we started with 20 ng of plasmid library obtained from bacteria minipreps. Sequencing libraries were prepared via three PCRs. The first two PCRs (PCR9 and PCR10) amplified the fragment of interest along with the downstream DNA barcode, while simultaneously adding the PBS_i5 adapter sequence at the 5′ end. The forward primer included the PBS_i5 and 20 bp upstream of the targeted region, while the reverse primer corresponded only to the PBS_i7. The amplicons were purified, diluted 1:1,000 and used as a template for a third PCR (PCR11), adding the appropriate Illumina i5 and i7 sample indexes for multiplexed sequencing. For each sample, the last PCR was performed in triplicate to maintain the diversity of the sequenced sample. All amplicons from the same sample were pooled and purified using magnetic beads. All samples were sent for high-throughput sequencing (sample description in S5 Table). Although this approach ensures robust amplification of the library and sufficient sequencing depth, it does not directly account for potential amplicon sequencing biases. Such biases can occur when random DNA molecules are over-amplified compared to others, which can cause significant issues for downstream quantitative analyses. Incorporating Unique Molecular Identifiers (UMIs) during PCR when preparing sequencing libraries would help mitigate the possible occurrence of such biases.

Following Golden Gate assembly, only the DNA barcode region was sequenced, in order to verify that barcode diversity and mutation coverage were maintained. Because the PBS_i5 and PBS_i7 were already present on each side of the DNA barcode, sequencing libraries after Golden Gate assembly were generated in a single PCR (PCR12), which added the appropriate Illumina i5 and i7 sample indexes. Like after Gibson assembly, this PCR was also performed in triplicate per sample to prevent sampling bottlenecks. All replicate amplicons were pooled and purified on magnetic beads prior to sequencing. All samples were then sent for high-throughput sequencing (sample description in S6 Table).

Sequencing platform

High-throughput sequencing was performed in paired-end 100 bp or 150 bp at the Centre de recherche du CHU de Québec—Université Laval sequencing platform (CHUL, Canada) using an Illumina NovaSeq 6000 S4 system. Libraries were sequenced at an average depth of ~250 reads per expected variant to assess barcode diversity and mutation coverage.

Sequencing read processing

To analyze quality control after Gibson assembly, first, low-quality reads were removed with fastp using default parameters -q 15 (the minimum phred score quality value that a base is qualified equal to 15) and -u 40 (40% of bases allowed to be unqualified). Reads were then merged with bbmerge from bbmap. All subsequent processing was done using custom Python scripts available at https://github.com/Landrylab/Jann_et_al_2025. To remove potential sequencing errors, the following reads were discarded: reads with incorrect length (reads including insertions or deletions), reads with more than one mutated codon within the PDR1 coding region (75 bp), reads with unexpected mutations within barcode sequences, and reads with missing nucleotides within the PDR1 or barcode sequence. After filtering, more than 60% and 80% of all reads were retained for F13 and F43, respectively.

Mutations with coverage of less than two reads were discarded. To measure the impact of combining results from multiple transformations (16 reactions, each with 5,000 (F13) or 7,500 (F43) transformants), we calculated the cumulative % of different metrics across all transformations sorted in a randomized manner. Means and confidence intervals were obtained for 100 random draws of transformation reactions. Metrics included: (i) informative barcodes, (ii) mutation coverage, (iii) barcode diversity: #barcodes >4 per mutation, and (iv) barcode diversity: #barcodes >9 per mutation. To standardize data, all metrics have been converted to the number of transformants/bp.

As done following the Gibson assembly, raw reads obtained after Golden Gate assembly product sequencing were filtered using fastp (default parameters), and overlapping paired-end reads were merged using bbmerge. Reads with incorrect length, unexpected barcode sequences, or missing nucleotides were removed, along with single-count reads. Barcode-mutation associations were performed by merging barcode reads from the Golden Gate assembly with the reference association generated after the Gibson assembly. To evaluate the library quality after the Golden Gate step, we re-calculated key metrics: informative barcodes, mutation coverage, and barcode diversity.

Quality control of final plasmid libraries

To assess the accuracy of the cloning strategy and evaluate potential PCR-induced mutations, we performed Whole Plasmid Sequencing of six plasmids derived from independent fragments assembly by Golden Gate. No unintended mutations were present in the PDR1 promoter, coding sequence, or terminator regions. This quality-control step validates the reliability of our cloning approach. Although pre-sequencing of all PCR-amplified fragments prior to Golden Gate assembly could further ensure accuracy, it was not performed here due to cost and scalability considerations.

Yeast strain construction (S12A Fig)

The ScPDR1-DOX strain was generated based on the Yeast Tet-Promoters Hughes Collection (yTHC) [52]. The endogenous promoter of PDR1 (spanning 50 bp upstream of the ATG codon; chrVIII:121683-121733) was replaced with a tetracycline-regulatable promoter. The KANMX-tetO7 cassette was amplified from plasmid pKB33 (PCR13) and introduced into competent S. cerevisiae R1158 cells using an adapted lithium acetate transformation protocol [53]. Transformants were selected on YPD medium containing 200 μg/mL G418.

Validation of genomic PDR1 expression control in ScPDR1-DOX (S12B Fig)

To validate the control of genomic PDR1 expression in the ScPDR1-DOX strain, PDR1 expression levels were quantified in the presence and absence of DOX in the culture medium. For this purpose, the gene coding for a monomeric enhanced Green Fluorescent Protein (mEGFP) was fused to the PDR1 coding sequence in the genome of ScPDR1-DOX strain. The mEGFP sequence along with a hygromycin resistance module was amplified from plasmid pFA-mEGFP-HPHNT1 (see construction below) using primers introducing homology arms targeting the 3′ end of the PDR1 CDS and its terminator (PCR 14). Amplicon was treated with 20 U of DpnI for 1 h at 37 °C, purified using magnetic beads, and used to transform ScPDR1-DOX competent cells using the standard lithium acetate transformation protocol [53].

Yeast strains (ScPDR1-GFP-DOX and ScPDR1-DOX) were cultured overnight at 30 °C with shaking in synthetic complete (SC) medium in 24 deep-well plates. The SC medium contained 0.17% yeast nitrogen base without amino acids and ammonium sulfate, 0.1% monosodium glutamate, 2% glucose, and amino acids drop-out without tryptophan. Saturated cultures were diluted to an OD₆₀₀ of 0.15 in fresh medium with or without DOX (10 µg/mL), and grown again under agitation at 30 °C until reaching an OD₆₀₀ of 0.5. Cells were then diluted to ~500 cells/µL in 0.2 µm filtered distilled water for flow cytometry analysis. GFP fluorescence was measured using a Guava easyCyte flow cytometer (MilliporeSigma) with excitation at 642 nm (GRN-B channel, 525/30 filter). A preliminary gating step (FSC-H (Forward Scatter Height) <15,000 and FSC-A (Forward Scatter Area) <15,000) was applied to select the main population of morphologically normal cells and to exclude non-representative events. A total of 5,000 events were recorded per replicate. Three independent biological replicates were analyzed per condition. Fluorescence signals were processed using a custom Python script and all the data (including control lacking GFP) are shown in S12B Fig.

In order to generate ScPDR1-GFP-DOX yeast strain, we first had to construct pFA-mEGFP-HPHNT1 plasmid. This was done by inserting mEGFP gene (insert 1) and ENO1 terminator fragment (insert 2) into pFA-hphNT1 (vector) using Gibson assembly. All three fragments were PCR amplified (PCR15, PCR16, and PCR17, respectively) and treated with 20 units of DpnI for 1 h at 37 °C before being purified on magnetic beads.

Yeast strain validation by complementation (S12C Fig)

In order to validate whether PDR1 expressed from the plasmid complements its genomic counterpart, we selected a condition where PDR1 expression is beneficial (in presence of antifungal) and assessed the growth of ScPDR1-DOX strain transformed either with an empty plasmid or with one containing PDR1 WT sequence. After adjusting cell density of precultures to an OD₆₀₀ of 1, three serial 1/5 dilutions were prepared in 200 μL of water (i.e., 40 μL of cells in 160 μL of water). Then, 5 μL of each dilution were spotted on YPD + NAT agar plates supplemented with either 0.1% DMSO (control), 1 μg/mL ITR, 2 μg/mL ITR, or 4 μg/mL ITR, in the presence or not of DOX (10 μg/mL). Plates were incubated at 30 °C for 48 h before imaging.

Library in yeast

The DMS plasmid library was introduced into ScPDR1-DOX yeast cells via lithium acetate transformation, performed fragment by fragment using a standard protocol [53]. For each fragment, approximately 100,000 colonies were recovered to ensure sufficient library coverage and barcode diversity. To collect transformants, 5 mL of YPD medium was added directly to each plate, and colonies were scraped using a sterile glass spreader. All transformants corresponding to a given fragment were pooled together. Aliquots were prepared, each corresponding to an optical density (OD₆₀₀) of 80 per fragment, and stored at −80 °C in 25% glycerol.

Plasmid recovery

Plasmid recovery from yeast was performed using the Zymoprep Yeast Plasmid Miniprep II kit, with the following modified protocol based on [5,24] to improve yield. Frozen cell pellets were resuspended in 200 μL of Solution 1, followed by the addition of 30 U of Zymolyase 20T. Cells were gently vortexed and incubated at 37 °C for 2 h. Following incubation, tubes were briefly vortexed and frozen at −80 °C for at least 20 min (up to overnight). After thawing at 37 °C for 3 min, 200 μL of Solution 2 was added and mixed, followed by the addition of 400 μL of Solution 3. Tubes were centrifuged at maximum speed for 3 min, and the supernatant (~800 μL) was transferred to a Zymo-Spin I column. Columns were washed with 550 μL of ethanol-containing wash buffer and centrifuged for 1.5 min at maximum speed. After removing the flow-through, columns were spun for an additional min to dry. Plasmids were eluted by adding 10 μL of Tris buffer (10 mM, pH 8.0) and centrifuging for 1 min at maximum speed. Eluted DNA was stored at −20 °C until further processing.

High-throughput sequencing

To validate whether DNA-barcode inference mutations corresponded accurately to the actual mutations present in the F13 sequence of PDR1, two types of library were generated: one containing the F13 region of PDR1 and one consisting of the barcodes.

Sequencing from yeast minipreps was performed using a protocol similar to that described in the Plasmid library quality control by sequencing section. Starting from 3 μL of yeast miniprep, an initial PCR was carried out to amplify only PDR1 from the plasmid, using Phusion High-Fidelity DNA Polymerase and M13 primers flanking the PDR1 insert on pRS31N (PCR18). This step avoids sequencing genomic PDR1 and ensures that the same sequencing material could be used for both F13 region and barcode-based analyses, allowing a direct and unbiased comparison between the two datasets.

In a standard barcode-based DMS experiment, DNA-barcode could be directly amplified using the PBS_i5 and PBS_i7 flaking sites, which would simplify library preparation and improve sequencing data quality.

To generate the F13 sequencing libraries, two rounds of PCR were then performed. Starting with 2 μL of M13-purified amplicons, a first PCR made with Phusion High-Fidelity DNA Polymerase (PCR19) was used to amplify the F13 region while introducing PBS_i5 and PBS_i7 sequences at the 5′ and 3′ ends, respectively. In addition, a 0–3 nucleotides spacer (N) was added to the primers to increase sequencing diversity. A second PCR (PCR20) was then used to add the appropriate Illumina i5 and i7 sample indexes, as described in the Plasmid library quality control by sequencing section.

To generate the DNA barcode sequencing libraries, 2 μL of the M13-purified amplicons were used as a template for a single PCR reaction to add the appropriate Illumina i5 and i7 indexes (PCR21), as described in the Plasmid library quality control by sequencing section.

Both the F13 and DNA barcode sequencing libraries were sent for high-throughput sequencing in paired-end 150 bp at the Centre de recherche du CHU de Québec - Université Laval sequencing platform (CHUL, Canada) using an Illumina NovaSeq 6000 S4 flow cell (sample description in S10 Table). All samples were sequenced to achieve a minimum of 500 reads per expected variant (i.e., ~ 0.4 million reads per sample for the F13 libraries and ~4 million reads per sample for the DNA barcode libraries). Only samples from timepoints TP0 and TP2 were sequenced. TP1 was excluded, as the limited frequency changes observed after just four generations hindered robust genotype–phenotype associations.

Selection coefficient and variant characterization

Selection coefficients were obtained with gyōza, a Snakemake-based workflow for the analysis of deep mutational scanning (DMS) data [54]. The pipeline includes adapter trimming with Cutadapt, read merging using PANDAseq, and clustering of identical reads via VSEARCH. Singleton reads (i.e., sequences observed only once) were excluded to reduce noise from sequencing errors. A custom script then aggregated read statistics across all processing steps.

The F13 samples were processed using the ‘codon’ mode of gyōza v1.1.8 [54], with constant 20 bp sequences on either side of the mutated locus specified for trimming. For the barcoded library, the “barcode” mode of gyōza v1.1.8 was used instead, by providing the dataframe of barcode-variant associations. Briefly, (i) all singletons (read count of 1) were discarded, (ii) log₂ fold changes of allele frequencies were calculated between TP0 and TP2, (iv) selection coefficients were calculated by normalizing with the number of mitotic generations (to account for different growth rates across cultures and make sure functional impact scores reflect the per-generation selective advantage), and (v) selection coefficients were further normalized by subtracting the median log₂ fold-change of silent variants (excluding the WT nucleotide sequence).

Variants were assigned confidence scores (as defined by gyōza) based on their initial read count at TP0 across replicates. A minimum threshold of five reads per nucleotide sequence was used to define high-confidence variants, which were retained for downstream analysis. Selection coefficients were then aggregated by amino acid position, averaging across synonymous codons (S5 and S6 Data). The resulting tables were used to classify mutations from resistant to sensitive (0.6 to −0.6) based on their deviation from the WT. Finally, heatmaps were generated using the selection coefficient for each variant to enable clear visualization and comparison between F13 region and DNA barcode sequencing (Figs 2B and S14).

To assess the quality of the barcoded library, we compared the final selection coefficients per mutation obtained from sequencing of either mutated F13 region or DNA-barcode inference. To evaluate how barcode diversity influences the reliability of these coefficients, we progressively increased the number of barcodes required per mutation (from 1 to 10) and, at each level, performed 100 random samplings. To ensure consistent comparisons across variants with different barcode counts, we implemented a standardized subsampling procedure during correlation analysis. If a variant had fewer barcodes than the target number (n), some barcodes were randomly resampled multiple times to reach n. Conversely, if a variant had more than n barcodes, a random subset of n barcodes was selected. This approach allowed us to normalize barcode representation across variants and to accurately assess correlation metrics based on 100 independent subsamplings. For each sampling, we calculated the correlation between selection coefficients inferred from DNA barcodes and those obtained from the F13 region (Fig 2C). An analysis of the correlation stratified by the number of barcodes per variant confirms the robustness of our strategy, even for variants associated with a lower barcode count (S14 Fig).