Oligo-CALL: A next-generation barcoding platform for studying resistance to targeted therapy.

INTRODUCTION
Clonal heterogeneity, a hallmark of cancer, contributes to tumor progression and the development of drug resistance (1–4). Recent advances in cellular barcoding technology have provided essential lineage-tracing tools for elucidating the evolutionary trajectories of tumor clones (5–8). These technologies facilitate the isolation of clonal populations before and after exposure to selection pressures, enabling genomic, epigenetic, and proteomic analyses and functional assessments of initial clones and their evolved progenies.
Despite these advancements, the current cellular barcoding system, such as the CaTCH (9), which uses a CRISPR transcriptional activator (CRISPRa)– linked reporter for retrospective tracking and isolation of tumor clones, exhibits limitations. Integrating the genetic barcode (BC) within the cis-element of the reporter cassette renders it incompatible with single-cell RNA sequencing (scRNA-seq). ClonMapper (10), another system designed for live clone isolation, exhibits suboptimal efficiency, achieving only about 50% efficiency with a threshold of 1000 genes per cell as per CRISPR droplet sequencing (CROP-seq) (11). More recently, CloneSelect was introduced as a lineage-tracing tool for live clone isolation, leveraging CRISPR-mediated base editing to restore the impaired start codon of a reporter gene (12). However, its design requires paired upstream and downstream barcodes within the reporter construct, posing substantial challenges for integration with scRNA-seq technique. Moreover, similar to CaTCH and ClonMapper, CloneSelect relies on additional manipulations such as cloning barcode-specific constructs and performing secondary cell transduction after barcode identification. These added steps increase technical complexity and may compromise both the accuracy and sensitivity of clonal analyses. These challenges highlight the urgent need for a more effective cellular barcoding system, ideally capable of tracing cell lineages at the single-cell level and isolating live tumor clones with high efficiency and reliability.
An elegant oligonucleotide-inducible guide RNA (gRNA) design, combined with CRISPRa (13), has garnered our attention for its potential applications in cellular barcoding technology. Using a spacer-blocking hairpin structure at the 5′ end, the activity of the gRNA can be induced by loop-specific antisense oligonucleotides (ASOs). This configuration enables the introduction of a genetic barcoding loop into gRNAs sharing a common spacer, thereby achieving barcoding-specific regulation of gRNA activity. Moreover, recent investigations into various tailing strategies of gRNA to enhance single-cell gRNA-seq have demonstrated that appending an “8A8G” tail at the 3′ end did not compromise gRNA functionality in the CRISPRa system (14). The 8A8G tailing confers compatibility with standard scRNA-seq workflows that rely on the capture of polyadenylate [poly(A)] RNA molecule (14).
Building on these advances, we developed Oligo-CALL (Oligonucleotide-inducible CRISPRa-assisted lineage labeling), which couples ASO-inducible gRNA with a poly(A)-tailed scaffold and a CRISPRa-linked reporter. By embedding the barcode in a 5′ spacer-blocking hairpin structure, Oligo-CALL enables precise ASO-triggered gRNA activation without additional genomic alterations, allowing for retrospective isolation of live BC clones. We validate Oligo-CALL in a targeted therapy resistance model involving KRASG12C inhibitor (G12Ci) in human lung cancer cells, demonstrating its potential to illuminate the clonal underpinnings of therapeutic resistance.

RESULTS

Design of Oligo-CALL system
We first engineered an ASO-inducible gRNA with a spacer-blocking hairpin structure (Fig. 1A). The gRNA contains a semirandom barcode sequence (BNSNNNVDNVNVWVM; 15 bases) flanked by the folding spacer regions. In its default state, the gRNA remains inactive because its spacer is sequestered by a complementary sequence that folds back on itself. The gRNA was expected to be activated upon exposure to BC-ASOs. These ASOs target the barcode loop, inducing cellular ribonuclease H to degrade the folding segment, thereby freeing the spacer and activating the gRNA (13). To facilitate direct capture by 10x Genomics beads for scRNA-seq, we also incorporate an 8A8G poly(A) tail at the 3′ end of the gRNA (14). Computational predictions using RNAfold (ViennaRNA Web Services) indicated that these modifications do not substantially disrupt the major secondary structures of gRNA (fig. S1A).
A single semirandom barcode within a gRNA offers a diversity of ~3.58 × 107. To further expand this range and facilitate later use of distinguishable ASOs for live clone isolation, we introduced two distinct gRNAs with different spacers. Each of them is controlled by either a human or mouse U6 promoter, achieving a theoretical diversity of up to ~1015 (fig. S1B). The Oligo-CALL system also features a reporter construct bearing three tandem repeats of target sequences for both gRNA, fused to a mini CMV (minimal cytomegalovirus) promoter driving green fluorescent protein (GFP) expression. Notably, the target sequences were designed to avoid off-target effects in both the human and mouse genomes (Fig. 1B and fig. S1, C and D). Moreover, the dCAS9-VPR [deactivated Cas9 fused to VPR (VP64, p65, and Rta)] was used as the transcriptional activator, as it was demonstrated as the most efficient transcriptional activator in such a system (9) (Fig. 1B). All constructs were cloned into lentiviral vectors to facilitate cell transduction.
In successfully transduced cells bearing Oligo-CALL, GFP expression remained silenced. Upon transfection with BC-ASOs, gRNAs will be activated and guide the dCAS9-VPR to initiate GFP reporter expression (Fig. 1C). This allowed for fluorescence-based sorting of live clones bearing the barcodes specific to the applied ASOs. In essence, the compatibility with scRNA-seq, high barcode diversity, and ASO-inducible GFP expression enable both high-resolution lineage tracing and selective retrieval of target clones from heterogeneous cell populations.

Validation of the Oligo-CALL system
To assess the efficiency of ASO-inducible GFP expression, we first generated tumor cells carrying a pair of known Oligo-CALL barcodes (BC1 to BC6; fig. S1E). H358 cells, a human NSCLC cell line harboring the KRASG12C mutation, were initially transduced with lentiviruses carrying the dCas9-VPR and GFP reporter. Cells that successfully incorporated the constructs [Cherry+/blue fluorescent protein (BFP)+)] were isolated via fluorescence-activated cell sorting (FACS) and subsequently infected with another lentivirus carrying specific gRNA barcodes. Each Oligo-CALL barcoding lentivirus can deliver two distinct gRNA subbarcodes (barcodes A and B) to the cell. GFP expression was induced specifically by ASOs, precisely matching the gRNA barcodes (Fig. 1, C to F). ASO targeting either barcode successfully triggered GFP expression in the cell, with maximal expression achieved when ASOs targeting both barcodes A and B were introduced (fig. S2).
GFP expression was dose dependent upon ASO treatment, with peak levels at ~100 nM (fig. S3, A to D). This ASO-induced GFP activation was transient. After the transfection, GFP peaked between days 2 and 3 and then returned to baseline by day 7 (fig. S3E). We further validated the system using additional Oligo-CALL barcodes (BC2, BC3, BC4, and BC5). GFP expression was induced only by matching ASOs, but not mismatched ASOs (fig. S4). These results demonstrate the high specificity and controllability of barcode-targeting ASOs in modulating gRNA activity and selective reporter expression in tumor clones using clone-specific ASOs.
To evaluate the efficacy of the Oligo-CALL system in tracing tumor clones, we performed a “spike-in” experiment (Fig. 2A). A tumor clone with a known barcode (BC1) was labeled by CellTracker dye and mixed with cells carrying different barcodes at varying concentrations (0.02 to 10%). As expected, introducing ASOs against BC1 selectively lightened the GFP+ signal within the CellTracker+ (spike-in) cells (Fig. 2A and fig. S5A). The decrease in the overall GFP+ cells correlated with the dilution of the spike-in cells. Even at the lowest spike-in concentration (0.02%), the ASO-activated GFP expression was well confined to the spike-in cells. These findings suggest that the Oligo-CALL’s efficiency in detecting barcoded cells is robust.

Application of Oligo-CALL system in the G12Ci resistance study
We next applied the Oligo-CALL system to investigate therapy resistance to G12Ci in lung cancer. H358 [CRL-5807; American Type Culture Collection (ATCC)] is a human lung cancer line bearing a heterozygous KRASG12C mutation and characterized as semisensitive to G12Ci therapy (15). Consistent with prior findings, AMG510 (sotorasib) treatment notably suppressed H358 cell proliferation by inducing G0/G1 arrest (fig. S6). However, with prolonged exposure (2 weeks), the proportion of cells in S phase rebounded, indicating the emergence of G12Ci resistance (fig. S6).
To explore clonal evolution during G12Ci resistance development, we first labeled H358 cells with dCas9-VPR and a GFP reporter (Fig. 2B). The Cherry+/BFP+ cells were sorted and subsequently infected with lentiviruses carrying a gRNA barcoding library. A multiplicity of infection (MOI) of <0.1 of lentivirus was applied to avoid multiple infections per cell. Successfully barcoded cells (Cherry+/BFP+/Thy1.1+) were then resorted into pools of various sizes and expanded for subsequent experiments.

Clonal fate assays uncover deterministic mechanisms underlying G12Ci resistance
Using Oligo-CALL–barcoded H358 cells, we initially validated barcode detection by next-generation sequencing (NGS). A clonal fate assay was conducted with a cell pool comprising ~1 × 105 clonal diversity. The barcoded cells were divided into multiple replicates, consisting of progenies from the same pool, with four treated with vehicle and four treated with AMG510 (2 μM) for ~3 weeks (Fig. 3A). Clonal populations of each replicate were sampled at multiple time points (days 0, 6, 14, and 19) and analyzed via barcode DNA amplicon sequencing. We reconstructed the clonal compositions of each replicate using identified Oligo-CALL barcodes and their abundances (Fig. 3A and fig. S7A). AMG510 treatment significantly reduced the number of surviving clones (Fig. 3B) and their Shannon diversity (fig. S7B) compared to controls, indicating drug-induced clonal selection. Notably, a high degree of similarity in clonal compositions was observed among treated replicates (Fig. 3C), suggesting deterministic rather than stochastic processes governing clonal outcomes.
Oligo-CALL barcodes further enabled analysis of barcode overlaps among treated replicates. Approximately 10% of barcodes (~1225 barcodes) were consistently identified across all AMG510-treated replicates at day 19 (Fig. 3D). This overlap significantly exceeded the expected random overlap of 2.67% (~287 barcodes), highlighting nonrandom selection. These overlapping clones comprised about 80% of the resistant cell population (Fig. 3D and fig. S8), underscoring their biological importance. Collectively, these data demonstrate that H358 cell clones harbor predetermined mechanisms, enabling persistent survival under G12Ci pressure.

Single-cell transcriptome analysis of Oligo-CALL–barcoded cells
A key feature of Oligo-CALL barcode is to integrate cell lineage information with transcriptome data at the single-cell level. By appending an 8A8G poly(A) tail to gRNAs, we could use the standard scRNA-seq platform, relying on poly(A)-based mRNA capture, to simultaneously capture both transcriptome and barcode sequences (fig. S9). A recognized limitation of current scRNA-seq technologies is the restricted number of cells (~1 × 104) analyzable in a single experiment. Consequently, it is crucial to manage clone diversity within experiments carefully.
Two pools of Oligo-CALL–labeled H358 cells containing approximately 500 unique clones were treated with AMG510 (2 μM) for 2 weeks. Surviving cells and corresponding treatment-naïve controls were then subjected to scRNA-seq. Because gRNA transcripts (~180 bases) are shorter than typical mRNAs, we introduced a size-selection step to separate barcode-derived cDNA from gene expression cDNAs (fig. S9). A targeted polymerase chain reaction (PCR) amplification with gRNA-specific primers was performed to enrich gRNA barcode sequences further, and a dedicated barcode library was subsequently prepared for Illumina sequencing. By integrating this extra amplification step, we achieved robust lineage assignment of single cells (Fig. 4A). Each scRNA-seq run yielded approximately 7000 cells with high-quality transcriptomes, of which more than 95% showed at least one detectable barcode. Furthermore, the median barcode (Unique molecular identifier (UMI) count per cell exceeded 100, indicating highly efficient barcode capture in scRNA-seq library preparation. This integrated approach enabled us to investigate clonal-specific features involved in G12Ci resistance, leveraging both transcriptomic and lineage information of single cells.

Clonal-specific pathways underlying G12Ci resistance identified by integrated Oligo-CALL scRNA-seq analysis
We next annotated single cells with their Oligo-CALL barcodes and systematically explored the clonal responses to G12Ci. Graph-based barcode clustering, in assistance with the exact barcode sequences, identified 10 dominant clones, each containing more than 100 cells (Fig. 4B, left, and fig. S10A). In parallel, Uniform manifold approximation and projection (UMAP) projections derived from single-cell transcriptomic profiles revealed a clear distinction between AMG510- and vehicle-treated cells, highlighting substantial transcriptional changes induced by treatment (Fig. 4B, middle). Overlaying the barcode-defined clonal identities onto the transcriptome-based UMAP provided an integrated perspective, illustrating clone-specific transcriptomic responses to G12Ci therapy (Fig. 4B, right). To elucidate the molecular mechanisms underlying these clonal responses, we conducted differential gene expression analyses and gene set enrichment analysis (GSEA), identifying distinct transcriptional signatures unique to individual clones (fig. S10B).
On the basis of their survival ratios under AMG510 treatment, clones were categorized into three distinct groups: “expanded” (class I), “neutral” (class II), and “eliminated” (class III) (Fig. 4C and fig. S10A). Expanded clones, such as BC#1, #5, and #9, exhibited substantial expansion under AMG510 treatment, with more than 80% of their cells present in treated samples, indicating a resistant phenotype. In contrast, eliminated clones, including BC#3, #7, #8, and #10, significantly declined following treatment, rarely appearing in AMG510-treated samples and suggesting sensitivity. The neutral clones (BC#2, #4, and #6) showed comparable abundance in both vehicle- and AMG510-treated samples.
GSEA within the vehicle-treated population revealed distinct molecular characteristics for each clone class. Class I (expanded) clones displayed a pronounced epithelial-to-mesenchymal transition (EMT) phenotype alongside lower activity in E2F_TARGET and the G2M CHECKPOINT pathways (Fig. 4D and fig. S11). This phenotype likely contributes to the low growth rates and subsequent survival advantage under G12Ci treatment. Class II (neutral) clones exhibited robust activation of MYC_TARGETS and proliferation-associated pathways (fig. S11), indicating potential reliance on alternative oncogenic signaling to survive KRAS inhibition.
Moreover, a global comparison of surviving AMG510-treated cells with vehicle-treated controls revealed profound transcriptional reprogramming (fig. S12). Specifically, surviving cells exhibited efficient suppression of KRAS signaling pathways (e.g., reduced KRAS_SIGNAL_UP and increased KRAS_SIGNAL_DN). Concurrently, these cells showed up-regulated gene sets associated with myogenesis, oxidative phosphorylation, DNA repair, and mTOR (mechanistic target of rapamycin) signaling, suggesting adaptive metabolic responses to treatment stress. Notably, clone-specific analyses confirmed similar pathway activations in AMG510-treated cells (fig. S13), although the degree of pathway activation varied across different clones.
Collectively, these findings underscore the substantial advantage of integrating lineage tracing with single-cell transcriptomics to uncover clone-specific transcriptional adaptations within tumor populations. These insights into the unique initial characteristics of individual tumor clones that influence their responses to G12Ci therapy can inform the development of targeted combination therapies. The Oligo-CALL platform provides a powerful tool to further investigate intratumoral heterogeneity in the development of resistance on the single-cell level.

Isolation of live tumor clones using the Oligo-CALL platform
Next, we isolated surviving clones from the resistant populations and retrieved their barcode-matched counterparts from the Oligo-CALL–labeled, cryopreserved treatment-naïve pool. By comparing each matched pair, we could determine whether resistance traits were preexisting or acquired during G12Ci treatment. Notably, the complete sequences of Oligo-CALL barcodes were retrievable from both DNA amplicon sequencing and scRNA-seq, enabling the design of specific ASOs for targeted clone isolation. We isolated clones retrospectively from preserved cell pools generated during the earlier clonal fate assay. These pools represent extensive clonal diversity (~1 × 105 unique clones), thereby providing an optimal environment to evaluate the efficiency of Oligo-CALL technology. In addition, these barcodes had been sequenced using DNA amplicon sequencing, which enabled reliable pairing of the Oligo-CALL barcodes. The purity of isolated clones was facilitated by two sequential rounds of GFP+ cell sorting using ASOs targeting two barcodes in one Oligo-CALL construct (Fig. 5A).
We prioritized the analysis of overlapping clones from Day 19_AMG510–treated replicates, as these clones potentially harbored preexisting resistance traits enabling survival under treatment conditions (fig. S8C). Approximately 62% (764 of 1225) of Oligo-CALL barcodes detected in G12Ci-resistant clones were also present in the treatment-naïve pools. Five relatively abundant clones in the resistant pool were selected, and ASOs targeting their Oligo-CALL barcodes were synthesized (fig. S14A) for targeted clone isolation. We successfully isolated three pairs of H358 clones from both treatment-naïve and resistant cell populations (Clone#1, #2, and #3). Subsequent barcode amplicon sequencing of expanded isolated clones confirmed that more than 95% of cells within these populations had the intended barcode identities (Fig. 5B).

Paired biological assay with isolated treatment-naïve and resistant tumor clones
The clones isolated from the treatment-naïve pool were considered sister cells, derived from the same ancestral cells that gave rise to the resistant clones with the identical Oligo-CALL barcodes. We next investigated whether these treatment-naïve clones had innate drug-resistant traits. The naïve clones were treated with serial doses of AMG510 and compared their responses to those of the original bulk population (naïve-pool). Notably, none of these clones exhibited immediate resistance; rather, they showed sensitivity to AMG510 comparable to the bulk naïve population (fig. S14B). These observations suggest that G12Ci resistance traits likely emerged after drug exposure rather than preexisting intrinsic characteristics.
During the ~4-week period required for clone isolation and expansion, cells were maintained in drug-free medium. This “drug holiday” provided an opportunity to assess whether resistant phenotypes persisted in the absence of continuous selective pressure. To evaluate the stability of resistance, we rechallenged the resistant clones with AMG510 using pooled resistant cells and their treatment-naïve counterparts as controls. Pooled resistant cells exhibited no significant differences in AMG510 sensitivity compared to the pooled treatment-naïve control cells (Fig. 5C). However, individual resistant clones demonstrated distinct response patterns. Notably, resistant Clone#3 retained a pronounced shift toward higher viability upon AMG510 reexposure, indicative of stable and heritable resistance (Fig. 5C). These findings highlight the latent and clonally specific characteristics of G12Ci resistance, revealing that distinct clones may use mechanistically diverse pathways to achieve resistance. Some clones acquire stable, heritable resistance, whereas others lose their resistance phenotype once the selective drug pressure is removed.
Furthermore, we conducted transcriptomic profiling of isolated clonal pairs to investigate the molecular basis underlying clonal G12Ci resistance. Three resistant clones, their treatment-naïve counterparts, and the parental pool were treated with AMG510 (2 μM) or vehicle for 3 days and subjected to bulk RNA-seq. Principal components analysis (PCA) revealed clear transcriptomic separation between AMG510- and vehicle-treated samples, indicating a robust treatment effect (Fig. 5D). Notably, distinct clustering of individual clones was also observed, underscoring substantial transcriptomic heterogeneity across clonal populations.
To dissect the baseline transcriptional differences among the clones, we performed differential gene expression analysis and GSEA. Clone#1 was enriched for MYC_TARGETS gene sets, suggesting activation of alternative oncogenic pathways independent of KRAS signaling (fig. S15A). Clone #1 also showed elevated activity in cell cycle–associated pathways, including G2M_CHECKPOINT, E2F_TARGETS, and MITOTIC_SPINDLE. In contrast, Clone#2 displayed relatively lower activation across these proliferative and oncogenic pathways. Clone#3 demonstrated strong activation of the EMT program, accompanied by reduced proliferative signaling. These results highlight the phenotypic and transcriptional heterogeneity among tumor clones. Notably, these clone-specific features, such as MYC signaling activation in Clone#1, diminished pathway activity in Clone#2, and, EMT phenotype in Clone#3, were largely retained in their resistant counterparts (fig. S15B), further supporting a model in which preexisting clonal heterogeneity contributes to G12Ci resistance.
The paired treatment-naïve and resistant clones also enabled assessment of acute transcriptional responses to G12Ci reexposure. Comparison of AMG510- versus vehicle-treated cells within the same clone revealed consistent inhibition of KRAS signaling and cell proliferation–related pathways, along with up-regulation of xenobiotic metabolism, fatty acid metabolism, and oxidative phosphorylation pathways across clones (fig. S16A). Notably, similar transcriptional responses were observed in resistant clones (fig. S16B), indicating a largely conserved cellular reaction to G12Ci treatment. Notably, persistent suppression of KRAS signaling in resistant clones suggests that the KRASG12C mutation remains targetable in these cells and that resistance may arise through engagement of alternative or downstream survival pathways rather than KRAS reactivation.

Targeting DNA repair pathway to overcome G12Ci resistance
Transcriptomic profiling of Oligo-CALL–barcoded cells revealed that tumor heterogeneity is a major obstacle to sustained G12Ci responses. While mapping clone-specific resistance mechanisms for tailored combinations is one approach, it requires extensive profiling of many more clones. As an alternative, we examined pathways consistently altered across three clones to identify shared vulnerabilities amenable to broadly effective G12Ci combinations.
The hallmark DNA repair pathway was differentially regulated in AMG510-treated cells from all clones (fig. S16), with pronounced suppression of homologous recombination (HR) genes, including BRCA1, BRCA2, CHEK1, CHEK2, RAD51, and BLM (Fig. 6A). GSEA using the reactome HDR homologous recombination gene set [Molecular Signatures Database (MSigDB)] consistently confirmed this suppression in both scRNA-seq–defined clone clusters and bulk RNA-seq of isolated clones (fig. S17). This convergence across independent datasets indicates that KRAS pathway inhibition attenuates HR repair capacity, exposing a potential vulnerability. Targeting this defect with HR-directed strategies, such as poly(adenosine 5′-diphosphate–ribose) polymerase (PARP) inhibition, may represent a practical approach to constrain the emergence of G12Ci resistance.
To assess the potential of PARP inhibition in G12Ci-resistant tumors, we established an AMG510-resistant H358 line (H358R) by continuous exposure to AMG510 (2 μM) for 2 weeks. These H358R cells showed markedly reduced sensitivity to AMG510 reexposure (Fig. 6B), but increased sensitivity to the PARP inhibitor [talazoparib (Tal)], compared to their parental H358 cells (Fig. 6C). Combination treatment with AMG510 and Tal produced a significant synergistic effect in H358R cells (Fig. 6, D and E).
The promising in vitro findings prompted us to evaluate the therapeutic potential of combining PARP inhibition with G12Ci in an immunocompetent mouse model. We first established a mouse KrasG12C (mG12C) lung tumor line from primary lung tumors induced in a KrasLSL-G12C+/−; p53fl/fl mouse via intratracheal Cc10-Cre adenovirus. In vitro cytotoxicity assays confirmed the sensitivity of mG12C cells to AMG510, and more importantly, a similar synergistic effect of AMG510 and Tal in mG12C cells (fig. S18, A and B). Orthotopic lung tumors were generated with mG12C cells through tail vein injection. Tumor-bearing mice were treated with AMG510, Tal, or the combination. AMG510 significantly inhibited tumor growth, whereas Tal alone showed minimal effect (Fig. 6F). AMG510-treated tumors eventually developed resistance, regrowing after two rounds of treatment. In contrast, recurrent tumor growth was suppressed in 50% mice receiving the combination therapy (Fig. 6F), which was also associated with significantly prolonged overall survival (fig. S19). These results demonstrate that PARP inhibition can impair G12Ci-resistant tumor regrowth and support the efficacy of AMG510/Tal combination therapy in vivo.

DISCUSSION
Intratumoral heterogeneity necessitates robust technologies capable of both single-cell lineage tracing and retrospective isolation of live clones. In this study, we present Oligo-CALL, a next-generation cellular barcoding platform designed to address these needs. Oligo-CALL overcomes two key limitations of existing technologies, such as CaTCH (9). First, by embedding the barcode sequence within the spacer-blocking hairpin of an ASO-inducible gRNA, Oligo-CALL enables efficient and selective isolation of live tumor clones without introducing additional genomic modifications, thereby preserving the genetic state of the targeted cells. Second, using a transcribable gRNA barcode rather than cis-regulatory elements, Oligo-CALL allows concurrent capture of lineage information and transcriptomic profiles at the single-cell level. The inclusion of a distinctive 8A8G 3′ tail (14) ensures compatibility with standard poly(A)-based scRNA-seq workflows, which enhances the resolution and utility of lineage-tracing applications.
Using Oligo-CALL, we investigated mechanisms of resistance to G12Ci in lung cancer cells. Clonal fate mapping under drug treatment revealed a “predestined” rather than purely “stochastic” mode of resistance, with specific clones reproducibly selected under drug pressure across independent replicates. This pattern is reminiscent of observations in other targeted therapies, such as epidermal growth factor receptor inhibitors (16, 17), where certain clones harboring intrinsic features determine relapse trajectories. These results underscore the concept that drug resistance is not uniformly stochastic but shaped by intrinsic clonal predispositions. To explore the clonal features, we exploited Oligo-CALL’s ability to isolate paired clones from cryopreserved treatment-naïve and drug-resistant pools. As these pairs share the same barcode and originate from a common ancestor, direct phenotypic comparisons are possible. Notably, treatment-naïve clones responded to G12Ci similarly to the parental population, indicating that resistance traits were not constitutively present but instead acquired upon drug exposure. This adaptive mode of resistance aligns with clinical observations in lung cancer patients, where resistance-associated mutations or pathway alterations, including KRAS secondary mutations, bypass signaling, and histological transformation, often emerge only after therapy (18–21).
Furthermore, paired-clone analysis revealed that resistance can be dynamic: Some resistant clones reverted to sensitivity after drug withdrawal, while others retained stable resistance, underscoring the complexity and plasticity of resistance mechanisms. These findings suggest that clone-specific adaptive patterns may contribute to the divergent clinical and experimental observations of resistance. In some cases, such resistance presents as transient and reversible drug-tolerant persister states, whereas in others, it becomes stable and irreversible, likely due to fixed genetic alterations or enduring transcriptional reprogramming (18, 22, 23). This clonal heterogeneity highlights the need for experimental systems, such as Oligo-CALL, which can distinguish between reversible and persistent resistance trajectories at the single-clone level.
Integrated transcriptome analyses, combining bulk RNA-seq of isolated clones and lineage-resolved scRNA-seq, revealed that the heterogeneous parental population comprises multiple clone-specific transcriptional programs. Despite this diversity, several pathway alterations consistently appeared after treatment with AMG510, particularly in DNA repair, mTOR signaling, metabolic pathways, and oxidative phosphorylation. Among these changes, the DNA repair pathway stood out: homologous recombination genes, including BRCA1 and BRCA2, were down-regulated across resistant clones. Loss of BRCA function makes cells especially vulnerable to PARP inhibition through synthetic lethality—a strategy already effective in breast and ovarian cancers (24, 25). Preclinical studies also suggest that suppressing mitogen-activated protein kinase signaling can decrease BRCA2 expression and increase tumor sensitivity to PARP inhibitors (26). In line with these observations, the PARP inhibitor Tal showed synergy with AMG510 in both in vitro and in vivo KRASG12C models. Collectively, these findings support combining G12Ci with PARP inhibition as a promising strategy to overcome or prevent resistance.
While these convergent vulnerabilities offer therapeutic opportunities, the breadth of intratumoral heterogeneity indicates that single-pathway targeting will be insufficient to eliminate resistance. Moving forward, integrating high-resolution transcriptomic profiling with functional validation of resistant clones will be critical to delineating distinct resistance programs and informing rational combination therapies that target multiple survival mechanisms. Oligo-CALL extends beyond discovery science: Its unique ability to couple lineage tracing with live-clone recovery provides a framework for patient-tailored experimental pipelines. For example, an individual’s tumor biopsy could, in principle, be isolated, expanded, and labeled with Oligo-CALL. The emerging resistant clones can be functionally profiled in comparison to treatment-naïve clones bearing the identical barcodes to design individualized therapy regimens, bridging clonal biology with precision oncology. By enabling nondisruptive live-clone isolation and efficiently linking lineage identity to transcriptomic state, Oligo-CALL provides a robust platform for elucidating the interplay between clonal evolution and drug resistance. Extending this approach to additional tumor models and therapeutic settings could accelerate the identification of exploitable vulnerabilities and inform strategies to overcome treatment resistance in cancer.

MATERIALS AND METHODS

Cloning of Oligo-CALL constructs
Lentiviral backbone vectors—Dual-sgRNA (plasmid #154194), CaTCH_empty (plasmid #157746), and dCas9-VPR_P2A_mCherry (plasmid #154193), developed initially in A. Obenauf’s laboratory (9)—were obtained from Addgene. Oligonucleotides (table S1), including Poly(A) tail template, gRNA target sequences, and PCR primers for amplification, were synthesized from Integrated DNA Technologies (IDT Inc.).
The inducible gRNA-barcoding library (Inducible gRNA_BC) was constructed in two steps: (i) Addition of the 8A8G Poly(A) tail to the Dual-sgRNA: We selected an 8A8G design, wherein mismatched “G” residues stabilize the gRNA without affecting its interaction with Cas9 (14). Two Poly(A) tail fragments were amplified by PCR with synthetic templates and primers (table S1) and inserted downstream of the gRNA scaffold (at XhoI and EcoRI sites, respectively) in the Dual-sgRNA using NEBuilder HiFi DNA Assembly Cloning Kit (New England Biolabs Inc). (ii) Constructing Dual-gRNA barcoding library: Semirandom barcodes for gRNA-BC1 and gRNA-BC2 (table S1) were synthesized as polyacrylamide gel electrophoresis–purified Ultramer oligonucleotides. Limited-cycle PCR (5 to 7 cycles) was conducted to amplify these gRNA-barcoding oligos, minimizing overamplification and bias. The resulting PCR products were purified and inserted into the poly(A)-tailed Dual-gRNA construct at BsmbI and BbsI sites. To preserve library diversity, we performed 20 separate ligations (10 μl each). All ligation products (200 μl) were electroporated into competent Escherichia coli (ElectroMAX DH10B Cells, Thermo Fisher Scientific). The transformants were plated on 20 20-cm agar plates for library expansion.
For the GFP reporter construct, a template oligonucleotide containing three tandem repeats of the targeting sequences for gRNA-BC1 and gRNA-BC2 (see table S1) was synthesized and PCR amplified. The resulting DNA fragment was inserted upstream of the miniCMV promoter (BsmBI site) in the CaTCH_empty vector. The dCas9-VPR_P2A_mCherry plasmid expressed the dCas9-VPR fusion protein, functioning as the CRISPR-based transcriptional activator.

Cell culture
H358 and 293T cells (originally obtained from ATCC) were expanded, cryopreserved, and used between passages 5 and 10 for all experiments. H358 cells were maintained in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS), 1% l-glutamine, and 1% penicillin-streptomycin (PS). 293T cells were cultured in Dulbecco’s modified Eagle’s medium with 10% FBS, 1% l-glutamine, and 1% PS. The mG12C cell is a cell line established from primary lung tumors induced by Adeno-CC10-Cre in KrasLSL-G12C+/−; p53fl/fl mice. G12C mutation and deletion of W53 were verified using strategies suggested by the Jackson Laboratory. Cells were labeled with Lenti-CherryLuc to facilitate in vivo bioluminescent imaging. mG12C cells were maintained in RPMI medium supplemented with 10% FBS, 1% l-glutamine, and antibiotics. Mycoplasma test was routinely conducted using the Universal Mycoplasma Detection Kit (ATCC, catalog no. 30-1012K) to ensure the absence of contamination.

KrasG12C lung cancer mouse model
To establish a lung cancer model, mG12C cells (5 × 105 cells in 100 μl of phosphate-buffered saline) were intravenously injected via tail vein into C57BL/6 mice. Tumor-bearing mice were randomized to receive vehicle, AMG510 (100 mg/kg, orally, daily), talazoparib (0.33 mg/kg, p.o. daily), or AMG510 + talazoparib combination therapy for 2 cycles of 10-day treatment (n = 10 mice per group). Tumor progression was monitored by bioluminescent imaging every 5 days. Morbidity and survival were recorded throughout the study, and Kaplan-Meier survival analysis was performed.

Lentivirus packaging and cell transduction
Oligo-CALL components (Inducible-gRNA-BC, GFP reporter, and dCas9-VPR) were packaged into lentiviruses by cotransfecting 293T cells with each Oligo-CALL plasmid, VSVG, and psPAX using the Calcium Phosphate Transfection Kit (Thermo Fisher Scientific). After 24 hours, the medium was replaced, and the supernatant containing lentiviral particles was collected at 48 and 72 hours posttransfection. To estimate viral titer, target cells were transduced with serial dilutions of the collected supernatant; the fraction of infected cells (f) was determined by flow cytometry at 3 days postinfection. Viral titer (in units per milliliter) was calculated using titer (U/mL) = −ln(1 – f) × N/V, where f is the fraction of infected cells, N is the total number of target cells, and V is the volume (in milliliters) of lentiviral supernatant used.
For cell transduction, target cells were infected with the lentiviral supernatant in growth medium supplemented with Polybrene (8 μg/ml; Santa Cruz Biotechnology). To minimize the likelihood of multiple infections per cell in the barcoding step, an MOI of 0.1 was used when delivering the Inducible-gRNA-BC lentiviruses.

ASO transfection
ASOs were designed as reverse complementary sequences targeting the barcode loop regions within gRNAs. For barcodes 1 and 2, oligonucleotides with the motifs GTC-N15-GAC and CGA-N15-TCG, respectively, were synthesized (IDT Inc., USA). Cells at ~60% confluence were seeded in 12-well plates. Transfection was performed using X-tremeGENE 360 (Roche) according to the manufacturer’s protocol. Briefly, ASOs (100 μM stock) were diluted in Opti-MEM I (Gibco) to a final concentration of 0 to 200 nM per well. X-tremeGENE 360 (4 μl per well) was added to the diluted ASO and incubated for 20 min at room temperature before being added to cells in 1 ml of fresh growth medium. Medium was replaced 24 hours posttransfection, and GFP induction was assessed by fluorescence microscopy.

Flow cytometry and cell sorting
Each Oligo-CALL construct includes a unique marker expression cassette to facilitate the sorting of successfully transduced cells by flow cytometry. Single-cell suspensions were generated by trypsinization, neutralized with growth medium containing 10% FBS, and filtered through a 40-μm cell strainer (BD Biosciences). When required, cells were stained with an anti-Thy1.1 antibody (clone OX-7, BioLegend) using a standard immunostaining protocol.
Flow cytometric analyses were performed on a BD LSRFortessa Flow Cytometer and visualized with FlowJo v10 software (FlowJo LLC). Endogenous fluorescence from BFP+, GFP+, and red fluorescent protein (RFP)+ cells was assessed alongside isotype antibody controls and nonfluorescent cells to determine appropriate gating, voltage settings, and compensation parameters. For FACS, an Aria III cell sorter (BD Biosciences) equipped with FACSDiva software was used. All cell preparations and sorting steps were conducted under sterile conditions, and postsort samples were reanalyzed to verify the purity of the isolated subpopulations.

Oligo-CALL barcode amplicon sequencing and analysis
Genomic DNA was isolated from Oligo-CALL–labeled cells using the DNeasy Blood & Tissue Kit (QIAGEN). For each sample, five parallel PCR reactions were carried out, each with 1 μg of genomic DNA to ensure sufficient coverage of barcoded cells. Amplification was performed with BC primers (BC-for and BC-rev; table S2) using Q5 Master Mix (New England Biolabs) in the presence of 1× SybrGreen. The PCR protocol included an annealing temperature of 65°C for 45 s. Reactions were monitored in real time, and amplification was halted during the exponential phase to prevent overamplification and related biases. Products from the five PCR reactions were then pooled, purified using a 0.5 to 0.65× bead-based size-selection protocol (Hieff NGS DNA Selection Beads, Yeasea Inc). Subsequently, the barcode amplicon sequencing libraries were prepared using the NEBNext Ultra II DNA Library Prep Kit for Illumina. Pair-ended reads were performed on Illumina NovaSeq X Plus sequencer at our Genomics Core Facility.
We extracted Oligo-CALL barcode sequences from raw NGS data by requiring a perfect match to the respective 5′ and 3′ flanking sequences (“GTTGTAGCTG” and “GTGCTAGTAC” for BC1; “TCAGCTGTAT” and “TAACAGGAGA” for BC2). The R packages ShortRead, Biostrings, and GenomicRanges were used to parse and count barcode reads. Clonal compositions for each sample were visualized using ggplot2 and ggforce.

Bulk RNA-seq analysis
Total RNA was extracted from cells (~1 × 106 cells each sample) using the RNeasy Plus Kit (QIAGEN). RNA-Seq libraries were prepared and sequenced according to standard Illumina protocols at the Genomics and Epigenetics Core Facility (Weill Cornell Medicine). Quality control (QC) and alignment were carried out with customized Partek Flow software (Partek Inc.). Briefly, reads were aligned to the human transcriptome (hg38) using STAR, and gene expression was quantified with the PartekE/M annotation model and then normalized to counts per million. Differential gene expression was assessed using the DESeq2 algorithm, which calculates fold changes and P values for each gene in pairwise comparisons. Volcano plots were generated to visualize significantly altered genes (fold change > 2 and P < 0.01). Biological interpretations of differentially expressed genes, such as GSEA, were integrated into the Partek Flow platform. Gene sets of interest were obtained from the MSigDB (www.gsea-msigdb.org/gsea/msigdb).

scRNA-seq analysis
Two pools of Oligo-CALL–labeled H358 cells (each comprising ~1000 clones) were seeded in 10-cm plates (2 × 106 cells per plate). Cells were treated with AMG510 (2 μM) or vehicle (dimethyl sulfoxide) for 2 weeks. During the treatment, medium was replenished every 3 days, and cells were subcultured at 80% confluence to maintain exponential growth. For scRNA-seq, single-cell suspensions were prepared and assessed for viability (>90%) using trypan blue staining by TC20 Automated Cell Counter (BioRad Inc.). Viable cells were processed at the Genomics and Epigenetics Core Facility (Weill Cornell Medicine) with a modified protocol of Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (10x Genomics Inc.) (incorporating Feature Barcoding technology for CRISPR).
Briefly, gel beads–in–emulsion (GEMs) were created to capture ~8000 cells per channel. After GEM reverse transcription and cDNA amplification, a 0.6× size selection was used to collect larger cDNA fragments for gene expression library construction. In contrast, a 0.6 to 1.2× fraction containing shorter cDNAs was dedicated to generating gRNA barcode libraries. Targeted PCR amplification further enriched gRNA barcode transcripts with primers specific to gRNA sequences (table S2). To ensure sufficient coverage and minimize amplification bias, each cDNA sample underwent three parallel amplification reactions for both BC1 and BC2. Libraries were indexed using the Dual Index Kit TT Set A (10x Genomics Inc.) and sequenced on an Illumina NovaSeq X Plus system at the Genomics Core Facility.
scRNA-seq and barcode data were processed using both 10x Genomics Cloud Analysis and Partek Flow (Partek Inc.). Barcode identification was performed by using 14-base reference sequences—“GACGTGCTAGTACG” for BC1 and “TCGATACAGCTGAC” for BC2—located immediately downstream of the barcode. The raw transcriptomic and barcode reads were aligned and normalized against the human reference genome (hg38) via 10x Cloud Analysis, generating single-cell count matrices. Subsequent analyses and data visualization were carried out in Partek Flow. QC at the single-cell level involved filtering cells based on total UMI counts, number of detected features, and the fraction of reads mapping to mitochondrial genes, with thresholds established per sample. The top 20 principal components were then used for UMAP dimensionality reduction and visualization. Differential expression and GSEA were conducted directly within Partek Flow; gene sets were obtained from the MSigDB.

Cytotoxic assays
To evaluate the sensitivity of H358 and mG12C cells to AMG510 and the PARP inhibitor Tal, 1 × 103 cells were seeded per well in 96-well black-walled adherent plates and treated for 72 hours with serial dilutions of AMG510 (0 to 104 nM) and talazoparib (0 to 100 μM). Cell viability was assessed using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Synergistic drug interactions were analyzed using SynergyFinder 3.0 (27), with the expected combination responses calculated according to the ZIP reference model. Positive and negative deviations from the expected response indicate synergy and antagonism, respectively.

Experiment rigor and statistics
Ensuring accurate lineage tracing in barcoded tumor cells requires rigorous control of lentiviral transduction parameters. A low MOI (<0.1) was used for the barcoding viruses to minimize the likelihood of multiple infections per cell. Flow cytometry confirmed that fewer than 10% of cells were infected, aligning with theoretical predictions (MOI of 0.1 corresponds to ~9.5% infection, with ~95.2% estimated single-infection events). A further consideration is the efficiency of cell sorting, which directly influences clonal purity. Following each sort, postsort purity was evaluated and maintained at >98% to ensure a high-fidelity barcoded population.
In the clonal fate assay, to balance the number of clones tested and ensure reliable detection of clonal survival differences, we used an Oligo-CALL–labeled pool of 1 × 105 H358 clones. We seeded 2 × 106 cells from this pool—achieving a 20× coverage of clones—into four biological replicates per treatment group (vehicle or drug). Cytotoxic assays were performed in triplicate for each treatment condition, and data were analyzed using GraphPad Prism. Statistical significance was defined at P < 0.05, and error bars represent the SEM unless otherwise noted.