Uncovering functional variants using a high-efficiency PE3-based screening platform Authors: Shisheng Huang, Wanyu Tao, Xiangyang Li, Junfan Guo, Yu Zhang, Guanglei Li, Pengfei Zhang, Guiquan Zhang, Gaoyang Li, Gaofeng Fan, Shang-Min Zhang, Xingxu Huang Published: 2026 ABSTRACT Prime editing (PE) enables precise nucleotide changes but has limited utility in high-throughput functional screening due to low editing efficiency. Here, we developed EvoPRIME, a PE3-based screening platform that integrates Csy4-mediated processing of Pol II–driven guide RNAs with a fluorescence-based enrichment strategy. EvoPRIME achieves comparable efficiency to state-of-the-art prime editing systems without requiring mismatch repair (MMR) suppression. We validated the performance of EvoPRIME through a loss-of-function (LOF) dropout screen targeting essential genes, which showed improved sensitivity and reproducibility, and a gain-of-function (GOF) saturation mutagenesis screen of the EGFR tyrosine kinase domain under osimertinib selection, which identified both known and uncharacterized resistance-conferring mutations. Leveraging EvoPRIME, we further conducted a high-throughput screen that functionally assesses both gain- and loss-of-phosphorylation (GOP and LOP) mutations. Among the hits, GAB1 S419E emerged as a phosphomimetic mutation that activates AKT signaling and confers osimertinib resistance. These studies establish EvoPRIME as a versatile platform for uncovering functional variants. INTRODUCTION CRISPR-Cas technology has revolutionized functional genomics, providing a powerful platform for genome-wide functional screens to dissect genotype-phenotype relationships (1–4). While traditional CRISPR-Cas9 knockout screens can efficiently identify gene-level functional effects, they lack the resolution to interrogate single-nucleotide variants or single amino acid changes. To overcome this limitation, base editing (BE) and prime editing (PE) extend CRISPR applications by enabling precise, programmable nucleotide-level modifications (5–7). Recently, several studies have used these tools to conduct large-scale phenotypic characterization of human nucleotide variants, enhancing our understanding of their functional impact (8–19). However, they still have certain limitations. BE is restricted by its protospacer adjacent motif (PAM) requirements and editing scope, which is limited to specific transition mutations (C → T or A → G). In addition, unintended bystander edits within the activity window can reduce editing precision (20). PE, in contrast, enables all possible nucleotide substitutions as well as small insertions and deletions, offering broad editing flexibility. Whereas its relatively low efficiency remains a major limitation, even when using optimized engineered prime editing guide RNAs (pegRNAs) in the context of the conventional PE2 system (14). The PE3 system improves efficiency by introducing an additional nicking single guide RNA (sgRNA) but increases the complexity and cost of library construction (11). Alternatively, editing efficiency can be further enhanced by inhibiting or knocking out MutL homolog 1 (MLH1) (14, 16). However, this necessitates the use of engineered cell lines, which involve additional experimental steps and may compromise genomic stability by disrupting the mismatch repair (MMR) pathway. As a result, multiple pegRNAs per target site are often needed, further increasing experimental complexity and cost and ultimately limiting the scalability of PE for high-throughput functional screening. To address these challenges, we developed EvoPRIME, a PE3-based platform that uses Csy4-mediated processing of an RNA polymerase II (Pol II)–driven transcript harboring both pegRNA and nicking sgRNA and integrates fluorescence-based enrichment strategies to enhance editing efficiency, enabling high-throughput functional screening without the need for MLH1 knockout or MMR suppression. We first benchmarked EvoPRIME using a dropout-based loss-of-function (LOF) screen targeting essential genes, which demonstrated improved sensitivity and reproducibility compared to conventional PE3 systems. We then performed a gain-of-function (GOF) saturation mutagenesis screen across the epidermal growth factor receptor (EGFR) tyrosine kinase domain under osimertinib selection to identify resistance-conferring variants. Next, we conducted a gain-of-phosphorylation (GOP) and loss-of-phosphorylation (LOP) screen targeting phosphorylation sites identified through comparative phosphoproteomic analysis between wild-type and osimertinib-resistant cells, revealing key functional phosphorylation sites. On this basis, we identified a previously uncharacterized phosphorylation site, GRB2-associated binder 1 (GAB1) S419, and found that its phosphomimetic mutation activates protein kinase B (AKT) signaling and suppresses apoptosis of PC9 cells in response to osimertinib. RESULTS Development of EvoPRIME for efficient introduction of amino acid substitutions As the relatively low efficiency of PE2 limits its utility in high-throughput applications, we focused on optimizing PE3-based designs to improve editing performance for precise amino acid substitutions. We first constructed lentiviral plasmids encoding tevo-PE3 (21), csy4-PE3 (22), and p2PE3 (23), each using PEmax (24) as the editor (Fig. 1A). These constructs represent different strategies for coexpressing pegRNA and nicking sgRNA: tevo-PE3 uses separate Pol III promoters and includes a 3′ tevopreQ1 modification to enhance pegRNA stability; csy4-PE3 expresses both guides under a Pol III promoter, separated by a Csy4 recognition site; and p2PE3 embeds both guides within the intron of a Pol II transcript for Csy4-mediated processing. To assess editing efficiency in a relevant model, we delivered these systems into PC9 cells, a non–small cell lung cancer (NSCLC) cell line harboring a common EGFR mutation (E746-A750 deletion) that is sensitive to EGFR tyrosine kinase inhibitors (TKIs). We then targeted six endogenous sites, introducing specific amino acid substitutions to assess editing outcomes. Seven days after lentiviral infection, both csy4-PE3 and p2PE3 demonstrated higher average editing efficiency (27.9% with csy4-PE3, 28.4% with p2PE3) compared to tevo-PE3 (15.0%), with p2PE3 showing a lower indels (2.3%) than csy4-PE3 (4.9%) (Fig. 1B). However, the editing efficiency at some sites remained relatively low (fig. S1A). EvoPRIME enables the efficient introduction of amino acid substitutions. (A) Schematics of the PE3 systems evaluated. All systems use PEmax with EGFP and blasticidin S deaminase (BSD) as the editor. The tevo-PE3 system uses a U6 promoter for pegRNA and an H1 promoter for nicking sgRNA expression and incorporates a tevopreQ1 modification at the 3′ end of pegRNA to enhance stability and editing efficiency. The csy4-PE3 system expresses both pegRNA and nicking sgRNA from U6 promoters separated by a Csy4 recognition site, alongside a Csy4 expression frame. The p2PE3 system links pegRNA and nicking sgRNA via Csy4 sites and embeds them within the intron of the Csy4 transcript. All constructs include mCherry and puromycin resistance (PuroR) for selection. (B) Quantification of editing efficiency and indels at six endogenous sites using the three systems. PC9 cells stably expressing PEmax were analyzed 7 days after pegRNA lentiviral transduction. (C) Schematic of the EvoPRIME system. Built on p2PE3, EvoPRIME incorporates a stop codon–containing mTagBFP2 frame, along with an additional pegRNA-nicking sgRNA pair targeting the stop codon. Successful editing restores blue fluorescence, allowing for enrichment via fluorescence-activated cell sorting (FACS). (D) Evaluation of different stop codons and corresponding pegRNA/nicking sgRNA pairs in HEK293T cells. (E) Quantification of editing efficiency and indels across six sites, comparing cells sorted by different fluorescence gates at two time points (day 7 and day 14). The percentage of total sequencing reads containing either the intended edit (red dots) or indels (gray dots) is shown. Data represent the mean ± SEM (n = 3 independent experiments). cPPT, central polypurine tract; WPRE, woodchuck hepatitis virus posttranscriptional regulatory element; nsgRNA, nicking single guide RNA; RRE, Rev response element; LTR, long terminal repeat. To enhance the editing efficiency of PE in functional screens, previous studies have used the knockout or inhibition of MLH1 (14, 16), a crucial component of the MMR pathway (24). While this approach can enhance editing efficiency, it may compromise genomic stability (25) and imposes a requirement for extensive cell engineering before screening, thereby limiting its applicability in certain contexts such as primary cells. Therefore, we sought to adopt a more universally applicable strategy to enhance PE efficiency without altering the native cellular environment. Inspired by previous work using fluorescence-based reporter systems to enrich PE-edited cells (26, 27) and motivated by the broad editing scope and combinatorial editing capability of p2PE3 (23), we integrated a stop codon–containing fluorescence reporter system that restores blue fluorescence upon successful editing (Fig. 1, C and D), hereafter referred as EvoPRIME (evolutionary screening platform for the reliable introduction of mutations in endogenous contexts). Next, we evaluated the ability of this system to enhance editing efficiency at the same sites. Seven days postinfection, cells were sorted on the basis of mCherry fluorescence to identify successfully transduced populations. Within the mCherry-positive cells, two subpopulations were further distinguished: mCherry-positive/blue fluorescent protein (BFP)-negative cells (indicating no editing of the reporter system) and mCherry-positive/BFP-positive cells (indicating successful editing of the reporter system). Editing efficiency was then assessed in each group. The double-positive population exhibited the highest editing efficiency (12.0 to 82.1%, average efficiency of 42.6%), which was 1.7-fold higher than the mCherry-positive cells without reporter enrichment (42.6% versus 24.5%) and 4.0-fold higher than the mCherry-positive/BFP-negative cells (42.6% versus 10.5%) (Fig. 1E and fig. S1B). These results demonstrate the effectiveness of our fluorescence-based enrichment strategy. We further extended the infection period to 14 days, which resulted in a further increase in editing efficiency (18.6 to 85.3%, average efficiency of 51.6%) (Fig. 1E and fig. S1C). During the revision of our manuscript, another group reported a puromycin-based activity reporter for BE that enables coselection of highly edited cells, thereby improving the resolution of mutational scanning screens, consistent with the general principle underlying EvoPRIME (28). Together, these findings indicate that EvoPRIME enables efficient introduction of targeted mutations, establishing a robust foundation for downstream functional screening. Evaluation of EvoPRIME across editing systems and cellular contexts Recent advances in PE have yielded next-generation systems such as PE7 (29) and PE-SB (30), which respectively enhance editing by La protein binding to the poly-U tract of U6-driven pegRNAs to prevent degradation and by an artificial intelligence–generated MLH1 small binder that inhibits MMR with limited toxicity. To determine whether EvoPRIME could be combined with or perform comparably to these updated designs, we compared tevo-PE3, p2PE3, and EvoPRIME, each tested alone or in combination with PE7, PE-SB (PE5), or PE7 + PE5. Consistent with previous reports, PE7 alone substantially boosted the editing efficiency of tevo-PE3 by stabilizing U6-transcribed pegRNAs (average efficiency 33.3% versus 19.9%, 1.7-fold), but it did not improve editing efficiency in p2PE3 (32.3% versus 31.4%) or EvoPRIME (48.6% versus 50.1%), since these systems use Pol II rather than U6 promoters (Fig. 2A and fig. S2, A and B). p2PE3 achieves an efficiency comparable to tevo-PE7. Considering that Csy4 is also an RNA binding protein, thus after Csy4 cleaves and releases the nicking sgRNA, we assume that it will remain bound to the Csy4 recognition motif at the 3′ end of the pegRNA, thereby protecting the pegRNA from degradation. Characterization of EvoPRIME across various editing systems and different cellular contexts. (A) Quantification of editing efficiency and indels across six endogenous sites using different PE systems. PC9 cells stably expressing PEmax (PE3), PEmax-La (PE7), PEmax-SB2 (PE5), or PE7-SB2 (PE5 + 7) were transduced with lentiviral pegRNA constructs derived from the tevo-PE, p2PE, and EvoPRIME systems and analyzed 14 days posttransduction. The percentage of total sequencing reads containing either the intended edit (red dots) or indels (gray dots) is shown. Data represent the mean ± SEM (n = 3 independent experiments). (B) Deep sequencing of three on-target loci and their five off-target sites predicted by Cas-OFFinder. Editing frequencies are shown as percentages of modified alleles for untreated, tevo-PE3, the BFP-negative fraction, and the BFP-positive fraction (EvoPRIME). Sequence mismatches relative to the on-target are highlighted in red lowercase, and PBS regions are underlined. (C) Specificity scores were calculated for each off-target as 1 − (off-target editing/on-target editing), yielding values between 0 and 1. Each dot represents one off-target measurement. Horizontal bars show the median with 95% confidence intervals (n = 3 independent experiments). Significance levels of specificity scores were calculated with the nonparametric Friedman test. Multiple comparisons were corrected using a Dunn’s test. (D) Chromatin state characterization of six endogenous loci targeted in PC9, HeLa, and K562 cells. The H3K79me2 score was used as a marker to distinguish loci with open versus closed chromatin configurations. (E) Quantification of editing efficiency and indel formation at six endogenous loci across three cell lines. The left shows results for three open chromatin loci, and the right shows results for three closed chromatin loci. Data represent mean ± SEM (n = 3 independent experiments). PE5 alone also increased editing efficiency for both tevo-PE3 (37.6% versus 19.9%, 1.9-fold) and p2PE3 (45.0% versus 31.4%, 1.4-fold) by reducing MMR-mediated correction of prime edits. Combining PE5 with PE7 further enhanced editing, bringing tevo-PE3 and p2PE3 to their highest efficiencies (40.3 and 45.0%, respectively), although these efficiencies remained below that of EvoPRIME alone (50.1%). It is worth noting that combining PE5 with EvoPRIME resulted in a further increase in editing efficiency (55.2% versus 50.1%, 1.1-fold) (Fig. 2A and fig. S2, C and D). While these results confirm that PE5 can be used in conjunction with EvoPRIME, we believe that suppressing the MMR pathway alters the intrinsic DNA repair environment of the cell, which may not be desirable in large-scale or unbiased high-throughput screens, as it may introduce confounding effects on cell physiology and downstream phenotypes. Therefore, we conclude that EvoPRIME alone provides the optimal balance of efficiency and physiological relevance for pooled functional screening. Given its strong enhancement of on-target editing, we next asked whether EvoPRIME might also elevate off-target activity. Targeted deep sequencing of three highly active loci and their top predicted off-target sites revealed that EvoPRIME substantially improved on-target efficiency without increasing off-target modification (Fig. 2B). We calculated a specificity score (31) defined as 1 − (off-target editing/on-target editing) for each site, which showed significantly higher values in the BFP-positive fraction compared with the BFP-negative or tevo-PE3 samples (Fig. 2C). Together, these findings demonstrate that EvoPRIME enhances editing efficiency without promoting off-target activity. The strong performance of EvoPRIME in PC9 cells prompted us to extend our evaluation to additional cellular and chromatin contexts, especially given that a recent study highlighted the profound influence of chromatin environment—particularly H3K79me2 levels—on PE outcomes (32). To this end, we expanded our analysis beyond PC9 cells to include HeLa and K562 cells. Because the CBH promoter can be silenced in hematopoietic cells such as K562, we uniformly replaced the promoters in this comparative experiment, using the elongation factor 1α (EF1α) promoter to drive PEmax and the human phosphoglycerate kinase (hPGK) promoter to drive the Pol II transcript harboring the Csy4 guide cassette, thereby ensuring robust expression across all cell types. We then selected three loci in open chromatin regions and three in closed chromatin regions (Fig. 2D) to directly compare EvoPRIME with p2PE3. EvoPRIME consistently achieved approximately 1.4-fold higher editing efficiency than p2PE3 at open-chromatin loci (16.2% versus 8.9% in PC9, 22.4% versus 15.6% in HeLa, 11.5% versus 8.3% in K562), demonstrating broad compatibility with different cell types and Pol II promoters. However, at closed-chromatin loci, both EvoPRIME and p2PE3 exhibited uniformly low editing efficiencies (all below 1%), indicating that EvoPRIME does not overcome the intrinsic barrier posed by compact chromatin. Nevertheless, even in these challenging regions, EvoPRIME still provided a relative improvement compared with p2PE3 (0.71% versus 0.44% in PC9, 0.22% versus 0.12% in HeLa, 0.30% versus 0.13% in K562), highlighting the utility of the enrichment strategy (Fig. 2E and fig. S3). Last, because EvoPRIME uses a Csy4 processing module, we evaluated whether this could induce cytotoxicity. We compared the growth rates of PC9, HeLa, and K562 cells stably expressing PEmax and transduced with lentiviral constructs encoding either tevo-PE3 (without Csy4) or EvoPRIME (with Csy4). Across all three cell types, EvoPRIME-expressing cells displayed no detectable proliferation defects relative to tevo-PE3, indicating that stable Csy4 expression does not impair cell growth (fig. S4A). We next examined the transcriptomic effects of Csy4 expression using RNA sequencing in PC9 cells. Hierarchical clustering revealed highly similar transcriptome profiles between EvoPRIME and tevo-PE3, indicating that Csy4 expression does not cause broad transcriptomic perturbation (fig. S4B). Differential expression analysis identified only nine significantly altered genes, and manual inspection confirmed that the down-regulated transcripts lacked Csy4 recognition motifs (fig. S4C). Since Csy4-mediated cytotoxicity has previously been reported only under conditions of extremely high plasmid expression (33), we reason that lentiviral integration at near-single-copy levels may avoid this issue. Benchmarking EvoPRIME for LOF screen in essential genes Having demonstrated that EvoPRIME enables efficient introduction of targeted mutations at endogenous loci, we next sought to evaluate its feasibility in pooled format. To this end, we adapted a published PE library designed to introduce premature stop codons into essential genes (14). Four thousand five hundred eighty pairs of spacer- and codon-matched pegRNAs targeting 998 essential genes, with three to five induced stop codon (iSTOP) edits per gene, were selected and appended corresponding nicking sgRNAs to enable PE3-based editing. In addition, we included 650 nontargeting pegRNAs as negative controls (fig. S5A, table S4, and Materials and Methods). To specifically assess the impact of reporter-based enrichment on screen sensitivity, we performed a side-by-side comparison of EvoPRIME (which incorporates a BFP restoration reporter) and a nonenrichment PE3 system. PC9 cells stably expressing PEmax were transduced with the pegRNA library and selected with puromycin. On day 7, mCherry-positive cells were collected and split into two groups: BFP-positive cells were sorted to represent the BFP-enriched condition, while the remaining mCherry-positive cells were maintained as nonenriched condition. Both groups were cultured until day 14, at which point genomic DNA was extracted for sequencing analysis (Fig. 3A). This parallel design allowed us to evaluate screening performance with and without reporter-based enrichment. EvoPRIME enables improved sensitivity and reproducibility in LOF pooled screening. (A) Schematic overview of the essential gene dropout screening workflow. PC9 cells stably expressing PEmax were infected with the pegRNA library and selected with puromycin. On day 7, cells were sorted by flow cytometry into mCherry+ and BFP+ populations, which were then cultured separately until day 14. Genomic DNA was collected for target site amplification and deep sequencing. (B and C) MAGeCK analysis comparing day 14 to day 7 samples in the nonenriched screen (B) and the BFP-enriched screen (C). Each point represents a pegRNA, plotted by its z-score and −log10 RRA score. pegRNAs introducing nonsense mutations are shown in red, synonymous mutations in pink, and nontargeting controls in gray. (D) Gene-level summary of significant hits in the dropout screen. Genes were grouped according to the number of significant iSTOP pegRNAs detected in the nonenriched screen (red) or the BFP-enriched screen (blue), using thresholds of z-score < −2 and RRA score < 0.05. (E) Receiver operating characteristic curves evaluating the ability of the nonenriched screen and the BFP-enriched screen to distinguish nonsense from synonymous pegRNAs. Area under the curve (AUC) is shown for each screen. The resulting data were quantified and analyzed using MAGeCK (34) (Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout) to identify pegRNAs depleted at days 14 compared to day 7. The BFP-enriched screen exhibited greater reproducibility, as evidenced by higher correlation between biological replicates compared to nonenriched screen (fig. S5, B to D). Log2 fold change for each pegRNA was normalized to nontargeting controls and converted into z-scores (table S5 and Materials and Methods). Using thresholds of z-score < −2 and robust rank aggregation (RRA) score < 0.05, we found that 6.7% (288 of 4307) of pegRNAs introducing premature stop codons caused significant negative growth phenotypes in the nonenriched screen (Fig. 3B), which increased to 12.7% (548 of 4307) in the BFP-enriched screen (Fig. 3C). We compared the log2 fold change distributions of iSTOP pegRNAs that were identified as significant only in the BFP-enriched screen, only in the nonenriched screen, or in both (fig. S5E). The results showed that pegRNAs identified as significant only in the nonenriched screen displayed weak depletion signals, with most log2 fold change (LFC) values greater than −1. By contrast, the BFP-enriched screen uniquely identified pegRNAs with stronger depletion signals, while shared hits showed the most robust effects (fig. S5F). At the gene level, this corresponded to 247 of 998 genes (24.7%) detected in the nonenriched screen and 366 of 998 genes (36.7%) in the BFP-enriched screen. To further illustrate the robustness of these findings, we categorized genes based on the number of significant iSTOP pegRNAs they contained. The results showed that BFP sorting not only increased the total number of identified genes but also yielded more genes supported by multiple independent pegRNAs targeting the same gene, demonstrating improved reliability for gene-level dropout screening (Fig. 3D). Meanwhile, we noted that 1.9% (83 of 4307) of synonymous edits also be defined as significant in the BFP-enriched screen. A likely explanation is indel formation inherent to the PE3 system. Because PE3 uses a nicking sgRNA to enhance editing efficiency, there is a possibility of small indels occurring at the target locus. If such an indel introduces a frameshift in an essential gene, a synonymous pegRNA could mimic a true LOF phenotype, resulting in apparent dropout. To further quantify the sensitivity and specificity of these two platforms, we constructed receiver operating characteristic curves, defining true positives as pegRNAs introducing stop codons into essential genes and true negatives as pegRNAs introducing synonymous mutations into the same essential genes. The BFP-enriched screen showed superior performance, achieving an area under the curve of 0.655 compared to 0.571 in the nonenriched screen (Fig. 3E). Collectively, these results demonstrate that EvoPRIME provides enhanced sensitivity and reliability for pooled LOF screening compared to the nonreporter-enriched system. GOF screen of EGFR variants under osimertinib selection using EvoPRIME Having demonstrated the utility of EvoPRIME in LOF pooled screening, we next aimed to evaluate its performance in identifying GOF mutations. We focused on EGFR, a clinically important oncogene frequently mutated in NSCLC (35, 36). Osimertinib, a third-generation EGFR TKI, has shown substantial clinical efficacy (37), yet the emergence of additional EGFR mutations remains a major barrier to sustained therapeutic response (38, 39), highlighting the need for a comprehensive functional evaluation of EGFR variants under osimertinib treatment. The EGFR tyrosine kinase domain is encoded by exons 18 to 24, but clinically relevant mutations are concentrated in exons 18 to 21, which are located near the adenosine triphosphate (ATP)–binding pocket (40, 41). Therefore, we designed and synthesized a saturation mutagenesis library on exons 18 to 21 to maximize biological and clinical relevance while maintaining manageable library size (Fig. 4A). Considering that PE can induce small deletions, we expanded our library by including deletions of one to five amino acids in addition to the amino acid substitutions. As controls, we incorporated synonymous mutations and mutations targeting the AAVS1 locus (fig. S6A, table S6, and Materials and Methods). Saturation mutagenesis screen for osimertinib resistance mutations in EGFR using EvoPRIME. (A) Schematic of the EGFR gene structure, with annotated functional domains. Exons 18 to 21, encoding the tyrosine kinase domain, were selected for a saturation mutagenesis screen to identify mutations conferring resistance to osimertinib. (B) Experimental workflow. PC9 cells stably expressing PEmax were transduced with the EvoPRIME saturation library on day 0. Following puromycin selection and recovery of BFP expression, BFP-positive cells were sorted by FACS on day 7 and cultured until day 14. Cells were then split into two groups: one treated with 300 nM osimertinib (+osimertinib) and one untreated (−osimertinib) and cultured until day 28. Genomic DNA was isolated from three conditions (day 14 untreated, day 28 treated, and day 28 untreated), followed by polymerase chain reaction (PCR) amplification, deep sequencing, and data analysis. (C) z-Score distribution of mutations in the osimertinib-treated population compared to day 14 untreated controls while excluding pegRNAs also enriched in the day 28 untreated group. For each mutation, the pegRNA with the highest absolute z-score was used to represent the variant. Each point represents a mutation, with color indicating RRA score. Data were analyzed using MAGeCK. (D) Structural mapping of top-ranking resistance mutations onto the EGFR tyrosine kinase domain (Protein Data Bank ID: 6JXT), visualized in PyMOL. Mutated residues are shown in magenta; osimertinib is shown in gold. The right highlights key mutations near the ATP-binding site. The library was packaged into lentiviruses and transduced into PC9 cells stably expressing PEmax. On day 7 postinfection, puromycin-selected cells were sorted for BFP-positive enrichment and maintained in culture for an additional 7 days. To ensure the identification of only the most critical resistance mutations, we selected 300 nM osimertinib based on dose-response experiments performed in PC9 cells (fig. S6B), where this concentration achieved maximal drug effect (Emax). On day 14, cells were split into two groups: one treated with 300 nM osimertinib and the other left untreated. Both groups were cultured until day 28. Samples were collected at three time points: day 14, day 28 with osimertinib treatment, and day 28 without treatment (Fig. 4B). High replicate correlation across groups confirmed dataset robustness and suitability for downstream analysis (fig. S6C). We first compared the osimertinib-untreated day 28 group with the day 14 control group and observed both LOF and GOF effects (fig. S6, D and E). Among missense variants, 13.9% (1130 of 8157) were significantly depleted and 3.6% (291 of 8157) enriched; deletions showed a similar pattern with 7.8% (129 of 1661) depleted and 4.2% (69 of 1661) enriched. Nonsense variants largely conferred a disadvantage, with 15.4% (65 of 423) depleted, although a small subset was enriched (1.9%, 8 of 423). Synonymous and control variants showed only modest shifts (26 of 423 and 3 of 475 depleted; 16 of 423 and 13 of 475 enriched, respectively), which may be attributable to PE3-induced indels and statistical noise (table S7). Together, these results indicate that EvoPRIME can capture both enrichment and depletion signals in the absence of drug pressure, supporting its broader utility for saturation mutagenesis screens. To prioritize EGFR variants enriched in the presence of osimertinib, we next compared the osimertinib-treated day 28 group with the day 14 control group and focused on the positive selection results (fig. S6, F and G). Using thresholds of z-score > 2 and RRA score < 0.05, 2.1% (220 of 10,241) of mutation edits, 2.4% (10 of 423) of synonymous edits, and 3.4% (16 of 475) of AAVS1 controls were significant (table S7). The enrichment of synonymous mutations likely results from PE3-induced indels that alter the drug-binding region and mimic resistance. We then compared the osimertinib-treated D28 group with the untreated D28 control group to evaluate osimertinib response relative to control growth (fig. S6, H and I). Across these comparisons, we observe that growth-associated effects are present for a subset of variants, whereas the top-ranked candidate variants exhibit consistent enrichment patterns (table S7). Accordingly, for downstream functional validation, pegRNAs from the osimertinib-treated group that were also enriched in either paired synonymous or parallel untreated samples were excluded. Top-ranking hits included the known resistance-associated residues C797 and L792 (42), as well as additional enriched positions including D800, S720, T847, P733, I780, and L828, along with an in-frame deletion Δ733-736 (Fig. 4C). Structural mapping revealed that C797 and L792 are located within the ATP-binding site of the EGFR kinase domain, consistent with their established roles in drug resistance. D800 is also positioned within this site and may affect drug binding (Fig. 4D). Other enriched residues are not located within known drug-binding regions, highlighting the possibility of noncanonical mechanisms contributing to resistance. To validate the reliability of these enriched variants, we performed two-color competition assays (9) in which cells expressing individual pegRNAs were tracked over time in the presence or absence of osimertinib (Materials and Methods). As expected, mutations at C797 and L792 conferred strong resistance, with rapid outgrowth under drug selection. Other enriched variants also exhibited a growth advantage in the presence of osimertinib, although to a lesser extent (fig. S7). These results demonstrate that EvoPRIME-based saturation mutagenesis enables sensitive detection of both strong and moderate resistance-conferring mutations, establishing it as a robust platform for GOF screening. High-throughput GOP and LOP screen for functional assessment of phosphorylation sites Encouraged by these results, we expanded our screening approach to assessing the function of phosphorylation sites associated with osimertinib resistance in NSCLC. Acquired resistance to osimertinib frequently involves the reactivation of downstream signaling pathways and bypass mechanisms (43, 44). Previous studies have shown that mutations in specific proteins, particularly kinases and phosphatases closely linked to phosphorylation, can reactivate critical signaling pathways and confer drug insensitivity (45, 46). Therefore, understanding the functional roles of phosphorylation sites is critical for elucidating resistance mechanisms and developing effective strategies for targeted therapy and combination treatments. Current functional studies typically use site-directed mutagenesis to mimic phosphorylated or dephosphorylated states, followed by phenotypic analysis (47–49). While effective, these approaches are labor intensive and low throughput. Recently, our group and others have applied BE screens for LOP studies (17–19). However, high-throughput functional studies of GOP remain unexplored. To address this gap, we applied EvoPRIME to perform a high-throughput GOP and LOP screen by editing endogenous serine, threonine, or tyrosine residues to alanine or glutamic acid, thereby mimicking constitutive loss or gain of phosphorylation, respectively. This approach enabled systematic functional interrogation of phosphorylation sites potentially involved in mediating osimertinib resistance. To this end, we first used quantitative phosphoproteomics to identify phosphorylation sites involved in the osimertinib resistance process. Osimertinib-resistant cell lines were established in PC9 and HCC827 cells using a stepwise drug escalation method (fig. S8, A and B). Both parental cell lines harbor the EGFR E746-A750 deletion mutation and are sensitive to osimertinib. These resistant cell lines were then compared to their parental cell lines to evaluate changes in their phosphoproteomic profiles using PTM-SEA (50) (posttranslational modification signature enrichment analysis). Although an osimertinib signature was not represented in posttranslational modification signature database (PTMsigDB), we identified positive enrichment of signatures associated with the first-generation small-molecule TKIs erlotinib and gefitinib, as well as the mitogen-activated protein kinase (MAPK) kinase (MEK) inhibitor U0126. In addition, we observed negative enrichment of signatures related to the EGFR pathway and extracellular signal-regulated kinase 1/2 (ERK1/2) upon osimertinib treatment, supporting the reliability of our analysis. We also found negative enrichment in the p38α/β MAPK inhibitor losmapimod and the c-Jun N-terminal kinase inhibitor SP600125 (Fig. 5A), suggesting that activation of these two pathways may contribute to the resistance process, consistent with previous clinical reports (51, 52). High-throughput functional phosphorylation site screen associated with osimertinib resistance. (A) PTM-SEA analysis of phosphoproteomic data comparing osimertinib-resistant (OR) and parental cells in PC9 (top) and HCC827 (bottom) cells. Normalized enrichment scores (NES) and −log10(FDR) (false discovery rate) are plotted for each signature set. Dot size reflects the percentage of scored phosphorylation sites within each set. Key inhibitor- or kinase-associated signatures are annotated. (B) Identification of phosphorylation sites associated with osimertinib resistance. Each point represents a phosphorylation site. The higher z-score from either the to E or to A edit was used, with values from to A edits multiplied by −1 so that GOP sites are positive and LOP sites are negative. (C) Western blot analysis of parental PC9 and monoclonal S419E mutant lines treated with 300 nM osimertinib or dimethyl sulfoxide (DMSO) for 24 hours. Phosphorylation of FOXO3 and AKT was examined. p-FOXO3, phosphorylated FOXO3; p-AKT, phosphorylated AKT; GAPDH, glyceraldehyde phosphate dehydrogenase. (D) Apoptosis assay of PC9 and S419E cells following 24-hour treatment with 300 nM osimertinib or DMSO. Early and late apoptosis were measured by annexin V–fluorescein isothiocyanate (FITC)/propidium iodide (PI) staining. Data represent mean ± SEM (n = 3 biologically independent replicates). **P < 0.01 and ***P < 0.001 (two-sided t test). (E) Dose-response curves of PC9 and S419E cells treated with osimertinib alone or in combination with 500 nM ipatasertib for 48 hours. Cell viability was assessed using the Cell Counting Kit-8 (CCK-8) assay. On the basis of our quantitative phosphoproteomic data (table S8), we designed a library targeting 2010 phosphorylation sites across 982 genes (fig. S8, C to F). For each site, we designed one to three sets of pegRNAs depending on the number of distinct available spacers. Each set included three variants: glutamic acid (to mimic GOP), alanine (to mimic LOP), and a synonymous mutation. In addition, mutations targeting the AAVS1 locus were included as controls (fig. S9A, table S9, and Materials and Methods). We applied the same screening and analysis strategy as in the EGFR screen to functionally evaluate these phosphorylation sites (Fig. 4B and fig. S9B). Among all pegRNAs, 2.7% (91 of 3353) GOP edits, 2.6% (87 of 3353) LOP edits, 2.8% (93 of 3353) synonymous edits, and 2.2% (13 of 585) AAVS1-targeting controls were identified as significant (table S10). The enrichment of synonymous mutations likely reflects PE3-induced indels that generate gene LOF events conferring resistance to osimertinib. Consistent with the EGFR screen, the same filtering criteria were applied in downstream analyses (fig. S9C). Ultimately, we identified 69 GOP sites and 71 LOP sites associated with osimertinib resistance (Fig. 5B). Among these, seven sites were shared between the two groups. This overlap may reflect that, in certain cases, substitution of a phosphorylation site with alanine or glutamate fails to recapitulate the regulatory effects of phosphorylation, leading both GOP and LOP mutations at the same site to exhibit resistance (53). We then performed pathway and GO enrichment analyses on GOP- and LOP-related genes separately. The pathway analysis showed that GOP-related genes were significantly enriched in pathways related to mRNA processing, cell survival, proliferation, and EGFR-TKI resistance (fig. S9D). GO enrichment analysis revealed biological processes such as RNA splicing, cytoskeleton organization and cell adhesion, and membrane protein localization and signal transduction (fig. S9E). In contrast, LOP-related genes did not reach statistical significance at the biological process or pathway level after multiple hypothesis testing correction. This suggests that GOP mutations tend to converge on a few core pathways that drive cell growth and survival, whereas LOP mutations may affect more dispersed regulatory nodes and adaptor proteins. To assess their prevalence and potential clinical relevance, we cross-referenced the phosphorylation sites identified in our screen against the missense mutations cataloged in the COSMIC database. Our analysis revealed that 37 phosphorylation sites harbored at least one reported missense mutation, including 20 GOP sites and 17 LOP sites (table S11). Notably, among the 20 GOP sites, none of the reported mutations were phospho-mimicking substitutions (glutamic acid or aspartic acid). This finding suggests that phosphorylation represents a distinct and dynamic regulatory mechanism independent of genomic mutations, highlighting the importance of systematically investigating posttranslational modifications in diseases. Furthermore, we performed motif-centric analysis using the Kinase Library (54) to identify candidate upstream kinases (fig. S9F). GOP-associated phosphorylation sites were enriched for serine/arginine-rich protein-specific kinase 3 (SRPK3), transforming growth factor beta receptor 2 (TGFBR2), and CDC-like kinase 1 (CLK1) motifs, which have been implicated in promoting cancer cell growth and survival (55–57), while LOP-associated sites were enriched for p53-related protein kinase (PRPK), STE20-like kinase (SLK), and TAO kinase 3 (TAO3) motifs, which are linked to DNA damage response signaling (58–60). These results provide insights into the regulatory network underlying resistance and highlight potential targets for further investigation. Last, we validated the top three phosphorylation mutations in both PC9 and HCC827 cell lines and confirmed that all three were found to confer resistance to osimertinib (fig. S10). Among these, GAB1 S419E emerged as a particularly interesting hit. GAB1 is a multisite docking protein that contains binding domains for various MAPK- and PI3K-modulating proteins, positioning it as a critical hub in intracellular signaling pathways (61, 62). Previous studies have implicated GAB1 in EGFR TKI resistance in NSCLC (63). Although the functions of several GAB1 phosphorylation sites have been examined in the context of drug resistance (62, 64), the function of the S419 site, whose phosphomimetic mutation was highly enriched in our osimertinib resistance screening, remains unexplored. To investigate the contribution of this site to drug resistance, we generated PC9 cell lines harboring the phosphomimetic mutation GAB1 S419E (fig. S11). Western blot analysis showed comparable GAB1 protein levels in parental and mutant clones, with or without osimertinib treatment. Osimertinib induced up-regulation of GAB1 protein in both parental and mutant cells. Notably, the S419E mutation led to increased phosphorylation of AKT, with minimal effects on mechanistic target of rapamycin (mTOR) or ERK. In contrast, phosphorylation of the ribosomal protein S6 was reduced in S419E mutant cells following osimertinib treatment, potentially reflecting altered signaling feedback (fig. S12). AKT, a key downstream effector of the PI3K pathway, is known to regulate apoptosis, cell growth, and metabolism and has been widely implicated in drug resistance (65). Transcriptomic analysis further revealed activation of DNA repair pathways in S419E mutant cells compared to parental controls (fig. S13). The interplay between AKT signaling and DNA repair is well established in cancer therapy (66). As AKT activation often makes cellular responses toward DNA repair rather than apoptosis (66, 67), we next examined the phosphorylation status of FOXO3, a downstream proapoptotic transcription factor. AKT-mediated phosphorylation of FOXO3 drives its nuclear export and inhibits its transcriptional activity (68). Consistent with AKT activation, S419E mutant cells exhibited increased FOXO3 phosphorylation (Fig. 5C) and reduced early apoptosis upon osimertinib treatment (Fig. 5D). When we reexamined the PTM-SEA results, we found significant activation of the PI3K/AKT pathway in HCC827-OR but not in PC9-OR cells (fig. S14, A and B). This likely reflects model-specific resistance mechanisms, as AKT activation is not a universal feature across resistant NSCLC lines. On the basis of these findings, we assessed whether targeting the PI3K/AKT pathway could restore osimertinib sensitivity. Combination treatment using the PI3Kα inhibitor alpelisib or the AKT inhibitors capivasertib/ipatasertib markedly enhanced the cytotoxic effects of osimertinib in both mutant and wild-type cells (Fig. 5E and fig. S14, C and D). These results suggest that the GAB1 S419E mutation promotes antiapoptotic responses via AKT activation, thereby contributing to osimertinib resistance, and that cotargeting the PI3K/AKT axis may offer a therapeutic strategy to overcome resistance. DISCUSSION PE offers precise and versatile genome modification capabilities, but its application to high-throughput screening has been limited by suboptimal efficiency and the complexity of library construction. In this study, we developed EvoPRIME, a PE3-based platform that integrates Csy4-mediated processing of Pol II–driven guide RNAs (gRNAs) with a fluorescence-based enrichment system. A key technical advantage of EvoPRIME is its ability to achieve high-efficiency editing without requiring MLH1 knockout or MMR suppression. This obviates the need for extensive cell line engineering and broadens the applicability of PE in diverse cellular contexts. In addition, Pol II–driven expression bypasses the sequence constraints of Pol III–based systems (23), while Csy4-mediated coexpression reduces library complexity and cost. EvoPRIME supports both GOF and LOF screening in a pooled format, as demonstrated across distinct applications. First, EvoPRIME enabled a dropout-based screen by introducing stop codons in essential genes, with improved sensitivity and reproducibility compared with the conventional PE3 system. Second, in a saturation mutagenesis screen targeting the EGFR tyrosine kinase domain, EvoPRIME successfully identified resistance-associated variants under osimertinib selection, capturing both known and uncharacterized mutations. Taking advantages of EvoPRIME, we simultaneously performed GOP and LOP functional screening across more than 2000 phosphorylation sites, highlighting associated signaling pathways and kinases that contribute to osimertinib resistance. Notably, we identified and validated GAB1 S419E as a previously uncharacterized phosphomimetic mutation that promotes osimertinib resistance. Mechanistically, the S419E mutation activates the AKT pathway and suppresses apoptosis, leading to drug resistance. Cotreatment with PI3K/AKT pathway inhibitors restored sensitivity in resistant cells, highlighting the translational potential of targeting downstream survival signaling. This study underscores the power of EvoPRIME not only for systematic discovery but also for gaining mechanistic insights and guiding rational combination therapy design. While EvoPRIME enables high-throughput functional mapping, we must explicitly acknowledge the limitations of the current screening architecture. First, indel formation is an intrinsic characteristic of PE3-based editing strategies in general, reflecting the necessary trade-off between achieving efficient editing and maintaining precision. Second, the observation of nonzero hit rates among synonymous and AAVS1-targeting controls suggests a baseline of stochastic noise in large-scale PE screens. To rigorously account for these factors, we used a conservative statistical framework: Enrichment was quantified via control-relative z-scores, and we implemented stringent filtering to exclude pegRNAs enriched in either paired synonymous or parallel untreated samples. These measures ensure that the reported hits represent robust functional signals rather than background artifacts. Looking forward, the resolution of EvoPRIME could be further enhanced by adopting a dual-library design that encodes reference and alternative alleles in matched pegRNA libraries (11). Such an approach would allow variant effects to be inferred from parallel relative comparisons, providing an even more refined lens for variant interpretation. While this study focused on drug resistance in NSCLC, the broader potential of EvoPRIME remains to be explored. PE offers exceptional flexibility, including base substitutions, small insertions, and deletions. Although these capabilities were not directly evaluated here, EvoPRIME inherits this potential and could be applied to perturb diverse genetic variants at endogenous loci in future studies. Additional validation in other biological systems will be of interest to expand its utility in functional genomics research. In conclusion, EvoPRIME provides a robust and flexible platform for high-throughput functional screening at single amino acid resolution. Its demonstrated utility in LOF, GOF, and posttranslational modification-mimetic screening establishes a foundation for broader efforts to map genotype-phenotype relationships and guide the development of mechanism-based therapeutic strategies. MATERIALS AND METHODS Plasmid construction The fragments of PEmax, PEmax-La, PE7-SB2, tevo-PE3, csy4-PE3, p2PE3, and EvoPRIME were gene synthesized at GenScript Biotech Corporation. Then, these fragments were inserted into lentiCRISPR v2 (Addgene, 52961) using the ClonExpress Ultra One Step Cloning Kit V2 (Vazyme). To construct individual target vectors, synthesized oligonucleotides were annealed and cloned into the BsmBI site of the lentiviral backbone using the Golden Gate assembly. Plasmids, pegRNAs, and nicking sgRNAs used in this study are listed in tables S1 and S2. Cell culture Human embryonic kidney (HEK) 293T and Hela cells were cultured in Dulbecco’s modified Eagle’s medium (Gibco), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin (Gibco). K562, PC9, and HCC827 cells were cultured in 1640 medium (Gibco), supplemented with 10% FBS and 1% penicillin-streptomycin (Gibco). All cell models were confirmed to be mycoplasma free and authenticated through short tandem repeat profiling while being maintained in 37°C incubators with 5% CO2. Lentivirus packaging HEK293T cells were seeded in a 10-cm dish and transfected at ∼80% confluence using Lipofectamine 3000 (Thermo Fisher Scientific). A total of 10 μg of lentiviral plasmid, 5 μg of pMD2.G plasmid, and 7.5 μg of psPAX2 plasmid were transfected into cells per dish. The transfection process was carried out according to the manufacturer’s protocol. After 48 hours, the supernatant was collected and centrifuged to remove cell debris and filtered through a 0.45-μm membrane. The virus was concentrated by ultracentrifugation, resuspended in 100 μl of serum-free medium, aliquoted, and stored at −80°C. Target deep sequencing We amplified the target sites from extracted genomic DNA using Phanta Max Super-Fidelity DNA Polymerase (Vazyme). These polymerase chain reaction (PCR) products were taken to Novogene for deep sequencing using NovaSeq X Plus Series PE150. Sequencing reads were demultiplexed using AdapterRemoval (69) (v.2.2.2), and the pair-end reads with 11 bp or more alignments were combined into a single consensus read. For PE analysis, collapsed reads were aligned to the reference sequence with CRISPResso2 (70) in HDR (homology-directed repair) mode using the parameters “-q 30 --discard_indel_reads -a -e -g -qwc.” The quantification window was positioned to include the entire sequence between pegRNA- and nicking sgRNA-directed Cas9 cut sites, as well as an additional 10 nt beyond both cut sites. All PE frequencies were quantified as: (number of HDR-aligned reads)/(number of reference-aligned reads). All indel frequencies were quantified as: (number of indel-containing reads)/(number of reference-aligned reads). For off-target analysis, collapsed reads were aligned to the reference sequence with CRISPResso2 using the parameters “-q 30 -w 10 -a -g.” Editing frequencies were quantified as percentages of modified alleles. Primers used in this study are listed in table S3. Cell line construction Osimertinib-resistant cell lines were established in PC9 and HCC827 cells using a stepwise drug escalation method. The drug concentration was incrementally escalated until cells could grow and proliferate in the 300 nM osimertinib. Each cell line (PC9 and HCC827) underwent three independent biological replicates of this procedure. To generate the PC9 GAB1 S419E mutant cell line, PC9 cells were transfected with the EvoPRIME system for target site editing using FuGENE HD Transfection Reagent (Promega) according to the manufacturer’s protocol. After transfection, cells were subjected to flow cytometry for single-cell sorting into a 96-well plate. Once colonies formed, individual clones were carefully picked and expanded. Genomic DNA was extracted from each clone, and PCR was performed to verify the targeted mutation. This process resulted in the successful isolation of three independent clones carrying the GAB1 S419E mutation, which were further expanded and characterized for downstream applications. Cell growth curve analysis PC9, HeLa, and K562 cells were first stably engineered to express PEmax and then transduced with lentiviral constructs encoding tevo-PE3 (without Csy4) or EvoPRIME (with Csy4). Forty-eight hours postinfection, cells were selected with puromycin (1 μg/ml) for 5 days and subsequently sorted by fluorescence-activated cell sorting (FACS), mCherry+ cells were isolated for tevo-PE3, and mCherry+/BFP+ double-positive cells were isolated for EvoPRIME. On day 14 postinfection, a total of 2 × 105 cells from each condition were counted to determine the starting cell number. The cells were then continuously passaged every 2 days, and at each passage, the actual split ratio was recorded. Theoretical total cell numbers at each subsequent time point (days 16, 18, and 20) were calculated by multiplying the day 14 baseline count by the cumulative passage ratio. RNA analysis RNA was immediately extracted from harvested cells using Total RNA Extraction Reagent (Vazyme) and sequenced at the Novogene using NovaSeq X Plus Series PE150. Transcript abundance was quantified using Salmon (71) (v0.14.0) using annotation from GENCODE version 30. Differential gene expression analysis was performed using DESeq2 (72) (v1.30.0). Gene Ontology, Pathway enrichment, and gene set enrichment analysis (GSEA) were performed using clusterProfiler (73) (v4.12.5) with wikipathways data v20241010. Chromatin environment analysis H3K79me2 chromatin immunoprecipitation sequencing datasets were downloaded from the ENCODE portal, including PC9 (ENCSR705SZP), K562 (ENCSR000APD), and HeLa (ENCSR000AOG). The calculation of H3K79me2 scores followed a previously reported method (32). Specifically, the R package Genomation (74) was used to extract signal values from the bigWig files within nonoverlapping 1-kb windows across the genome. For each dataset, window scores were calculated as the natural logarithm of the summed signal values within each window after adding a small offset corresponding to the minimum nonzero signal value detected across all genomic windows. The resulting H3K79me2 scores were used to compare chromatin environments between open- and closed-chromatin loci in PC9, HeLa, and K562 cells. Phosphoproteomic analysis PC9, PC9-OR, HCC827, and HCC827-OR cells in biological triplicates were harvested. Two hundred microliters of lysis buffer (8 M urea dissolved in 50 mM ammonium bicarbonate; 1 phosphatase and protease inhibitor tablet per 10 ml of buffer) was added to each sample for ultrasonication-based lysis. The lysates were centrifuged, and the supernatants were collected for protein quantification using the bicinchoninic acid (BCA) assay (Thermo Fisher Scientific). Equal amounts of protein from each sample were taken for subsequent experiments. The proteins were reduced with dithiothreitol and alkylated with iodoacetamide before being digested overnight with trypsin at 37°C. The following day, digestion was terminated using trifluoroacetic acid, and the supernatants were collected by centrifugation and desalted. Samples were labeled using the TMTpro 16-plex Label Reagent Set (Thermo Fisher Scientific) according to the manufacturer’s protocol. After labeling, a small aliquot from each sample was pooled to evaluate the labeling efficiency via liquid chromatography– tandem mass spectrometry, ensuring a labeling efficiency of >95%. The labeled peptides were fractionated to enhance proteome coverage. Subsequently, phosphorylated peptides were enriched using the TitanSphere Phos-Tio Kit (GL Sciences) according to the manufacturer’s instructions. The enriched phosphopeptides were then desalted. Mass spectrometry analysis was performed using the Q Exactive HF-X instrument (Thermo Fisher Scientific). The raw data were first converted to mzML files using ThermoRawFileParser (75). The resulting mzML files were then searched against the GENCODE version 30 reference proteome using MSGFPlus (76) with the parameters “-tda 0 -protocol 1 -minNumPeaks 5 -addFeatures 1.” Reporter ion information was extracted using MASIC (77). The spectral-level false discovery rate (FDR) was controlled to ≤1% using a target-decoy searching strategy. Peptide identification was filtered by SpecEvalue < 1 × 10−9, with a mass error (DelM_ppm) between −10 and +10 parts per million and a PepQvalue < 0.01 to maintain a global FDR of ∼1%. Phosphorylation site localization was performed using AScorePro (78), requiring a minimum score of 13 for confident site assignment. Differential analysis of peptide intensities was conducted using DEqMS (79). Library design and construction For the essential gene screen, pegRNAs were selected from the StopPR library (14). Specifically, only the iSTOP pegRNA with the strongest depletion effect for each iSTOP edit in the original screen was retained, resulting in 7267 pairs. We then removed pegRNAs that targeted multiple genomic loci and added matched nicking sgRNAs to fit the PE3 strategy used in our study, yielding 6641 pairs. Next, we filtered out genes with fewer than three pairs and limited each gene to a maximum of five top-performing pairs. The final library contained 4580 pairs of spacer- and codon-matched pegRNAs targeting 998 essential genes, with three to five iSTOP edits per gene. In addition, we included 650 nontargeting pegRNAs as negative controls. For EGFR and phosphorylation screens, we extracted 200-bp sequences upstream and downstream of the mutation sites to be used as input files for the PRIDICT (80) and DeepPrime (81) software. On the basis of the output files from tools, we first filtered out pegRNAs and nicking sgRNAs that contain BsmBI and BsaI recognition sites. Then, for each mutation, we selected the top one to three pegRNAs with different spacers based on the average ranking of scores from both tools. For nicking sgRNA, preference was given to those with the highest scores at distances between 40 and 90 bp from the pegRNA-mediated nick. If no suitable nicking sgRNA could be designed within this range, the selection extended to distances between 20 and 120 bp. If no appropriate nicking sgRNA was found, the corresponding pegRNA was no longer considered. The library oligos were synthesized by Twist Bioscience based on the following template: 5′-[primer]-[BsaI site]-[pegRNA spacer]-[BsmBI site]-[BsmBI site]-[pegRNA extension]-[Csy4 site]-[nicking sgRNA spacer]-[BsaI site]-[primer]-3′. The construction of the gRNA library involves two steps. First, the library oligos were amplified using NEB Q5 high-fidelity DNA polymerase. After BsaI digestion, they were ligated with T4 ligase onto the BsmBI-linearized Lenti vector backbone with corresponding sticky ends. The resulting plasmid library from the first step was obtained through electroporation and plasmid extraction. Subsequently, the first-step plasmid library was subjected to BsmBI digestion. The sgRNA scaffold with the corresponding sticky ends was ligated using T4 ligase into the region between the pegRNA spacer and pegRNA extension. The final gRNA library was obtained through a second round of electroporation and plasmid extraction after ligation. Screening workflow For pooled screening, cells stably expressing PEmax were transduced with the EvoPRIME lentiviral library at a low multiplicity of infection (<0.5) to favor single-copy integration while maintaining ≥500× library coverage at all steps. Forty-eight hours postinfection, cells were selected with puromycin (1 μg/ml) for 5 days and subsequently sorted by FACS. For p2PE3 experiment in dropout screen, mCherry+ cells were isolated; for EvoPRIME experiments, mCherry+/BFP+ double-positive cells were isolated. At the sorting step, the positive population was evenly split into two independent replicate cultures, which were maintained separately for all subsequent procedures. Screen timelines were as follows. For the dropout screen, baseline samples were collected at day 7 postinfection and endpoint samples at day 14. For EGFR and phosphorylation screens, cells were sorted at day 7, expanded to day 14, and then cultured ±osimertinib (300 nM; Emax determined in PC9 dose-response) until day 28, at which point samples were harvested from both arms. After harvesting the cells, cell lysis buffer containing ribonuclease A (100 μg/ml; TIANGEN) was added to the cell pellet, and the mixture was incubated in a 37°C shaker for 3 hours. Proteinase K (100 μg/ml; Yeasen) was then added, and the mixture was incubated at 55°C overnight. An equal volume of DNA extraction reagent (ACMEC) was added, and the mixture was vortexed vigorously before centrifugation. The upper aqueous phase was transferred to a new tube, mixed with an equal volume of isopropanol, and centrifuged again. After discarding the supernatant, the pellet was washed with 75% ethanol, centrifuged, and air-dried at room temperature until transparent. Last, prewarmed double-distilled water was added to dissolve the genomic DNA, and the DNA concentration was measured. The extracted genomic DNA was used as a PCR template to amplify and enrich the pegRNA and nicking sgRNA region using NEB Q5 high-fidelity DNA polymerase. Two rounds of PCR amplification were performed. In the first round, multiple 50-μl reactions, each containing 5 μg of genomic DNA, were set up to achieve 500× coverage per sample. The PCR cycling conditions were as follows: 3 min at 95°C, followed by 10 s at 98°C, 20 s at 64°C, and 30 s at 72°C for 24 cycles. The PCR products from the same sample were then pooled, and 10 μl of the mixture was used as the template for a second 100 μl of PCR reaction, using primers containing P5 and P7 Illumina adapters. The second PCR cycling conditions were as follows: 3 min at 95°C, followed by 10 s at 98°C, 20 s at 64°C, and 30 s at 72°C for 17 cycles. The purified PCR products were sent to Novogene for sequencing using NovaSeq X Plus Series PE150. Screening data analysis Sequencing reads were mapped to pegRNA and nicking sgRNA reference libraries using a custom Python script (table S12) requiring exact sequence matches. Read counts were normalized to 10,000,000 total reads per sample. For the essential gene screen, pegRNAs with fewer than 200 reads in the day 7 sample were filtered out, and only pegRNAs with matched stop and synonymous variants were retained for analysis. For the EGFR and phosphosite screens, pegRNAs with fewer than 20 reads in the day 14 sample were excluded. Processed read count tables were analyzed using MAGeCK (34) with the parameters “--paired --norm-method none --gene-lfc-method mean.” As described in a prior study (14), log2 fold changes from each replicate were first adjusted by subtracting the median log2 fold change of control pegRNAs, then averaged across replicates, and lastly converted to z-scores by dividing by the SD of control pegRNAs. For the essential gene screen, pegRNAs were defined as depleted if z-score < −2 and RRA score < 0.05. For the EGFR and phosphosite screens, pegRNAs were considered as enriched if z-score > 2 and RRA score < 0.05, after further filtering to exclude pegRNAs also enriched in paired synonymous controls or in the untreated day 28 group. For each mutation, the pegRNA with the highest absolute z-score was selected to represent the variant in downstream analyses. Two-color competition assays PC9 cells stably expressing PEmax were infected with pegRNAs targeting either the gene of interest (mCherry fluorescence) or AAVS1 (RFP670 fluorescence). Seven days postinfection, mCherry/BFP-positive and RFP670/BFP-positive cells were sorted. The mCherry and RFP670-expressing cells were mixed in a 1:1 ratio and seeded into a 24-well plate for continued culture. After an additional 7 days, the number of mCherry-positive and RFP670-positive cells was recorded by FACS and considered as day 0. At this point, cells were treated with 300 nM osimertinib or dimethyl sulfoxide (DMSO). Fluorescence counts were analyzed every 3 days using FACS. Data are presented as the mCherry/RFP670 ratio, normalized to the ratio on day 0 for each experimental time point. Each experiment was performed in biological triplicates. Western blot Cells were lysed using radioimmunoprecipitation assay lysis buffer, including phenylmethylsulfonyl fluoride and protease and phosphatase inhibitors. After centrifugation at 4°C for 10 min, the supernatant was collected. Protein concentrations were measured using BCA assay (Thermo Fisher Scientific). After fully denaturing the proteins, these samples were separated by SDS–polyacrylamide gel electrophoresis and transferred onto a nitrocellulose membrane. The membrane was blocked with 5% BSA (room temperature, 1 hour) and overnight incubation at 4°C with primary antibody. The next day, the membrane was washed with TBST (Tris-buffered saline with Tween 20) and incubated with a secondary antibody (room temperature, 1 hour). After washing with TBST again, the membrane was imaged. Long-term colony formation assay Wild-type PC9 cells and three independent PC9 S419E clones (clones 1 to 3) were seeded in 24-well plates at a density of 3000 cells per well and cultured for 24 hours to allow cell attachment. Cells were then treated with osimertinib at concentrations of 0, 1, 10, 50, 100, and 300 nM and maintained for 7 days, with medium replaced every 2 days to ensure stable drug exposure. At the end of the treatment period, cells were washed with phosphate-buffered saline (PBS), fixed with 4% paraformaldehyde for 15 min, and stained with 0.1% crystal violet for 30 min at room temperature. The plates were rinsed with PBS and air-dried. Colony growth was quantified using ImageJ and expressed as the stained colony area (%) per well. Apoptosis and CCK-8 assay For apoptosis analysis, cells were seeded in 12-well plates and treated the following day with DMSO or 300 nM osimertinib for 24 hours. Apoptotic cells were stained using the Annexin V–FITC/PI (fluorescein isothiocyanate/propidium iodide) Apoptosis Detection Kit (Vazyme) according to the manufacturer’s protocol. Both early apoptotic (annexin V+/PI−) and late apoptotic (annexin V+/PI+) populations were quantified using flow cytometry and analyzed with FlowJo software. Cell viability was assessed using the Cell Counting Kit-8 (CCK-8, Topscience). Cells were seeded at 5000 cells per well in 96-well plates and allowed to adhere for 24 hours. They were then treated with a range of osimertinib concentrations (1, 5, 10, 50, 100, 500 nM, and 1 μM) alone or in combination with 500 nM alpelisib, capivasertib, or ipatasertib. After 48 hours of treatment, cell viability was measured using a Spark microplate reader (Tecan). All conditions were tested in three biological replicates.