Engineering Selective Gene Drives to Steer Tumor Evolution to Counteract Drug Resistance

As tumors evolve, targeted therapies against cancer often fail due to the evolution of drug resistance. This study demonstrates a way to manipulate tumor evolution repeatedly to design therapeutic opportunities, even in the presence of genetic heterogeneity. We developed a selective gene drive system that can be stably introduced into cancer cells, consisting of two genes (or switches), coupling an inducible adaptive advantage with a shared fitness cost. Utilizing evolutionary dynamics from stochastic models, we identified design criteria for selective gene drives and constructed prototypes that could leverage selective pressures of multiple approved tyrosine kinase inhibitors, employing various therapeutic mechanisms such as prodrug catalysis and immune activation. We proved that selective gene drives could eliminate multiple forms of genetic resistance in vitro. Finally, we showed that model-guided activation of switches can effectively target pre-existing resistance in mouse solid tumor models. These results establish selective gene drives as a powerful framework for evolution-guided cancer therapy.

For many cancers, the evolution of drug resistance is one of the biggest challenges in developing curative cancer therapies. Studies on single-cell heterogeneity have revealed that small resistant subclones often exist in tumors from the baseline, indicating that most cases will lead to treatment failure. Moreover, a recent study on hundreds of advanced lung cancer patients uncovered astonishing levels of genetic diversity at the baseline, with few clues on how to combat this profound heterogeneity. Drug treatments significantly change the evolutionary landscape of tumors, often resulting in drug-resistant and selectable tumors with fewer treatment options.

Efforts to combat resistance are hindered by the inherent unpredictability of resistance evolution. In most cases, resistant variants are too scarce at the onset of treatment to be reliably detected, so the evolutionary trajectory of the tumor cannot be predicted. Hence, traditional methods to combat resistance involve waiting for a subclone to grow large enough to be clinically detectable, then responding with the appropriate therapeutic strategy. In the case of targeted therapy, if resistance is driven by point mutations in the target gene, this strategy often means developing and responding to a new generation of inhibitors. For example, in EGFR-mutated non-small cell lung cancer (NSCLC), if the tumor acquires the T790M resistance mutation during first-line treatment with the TKI erlotinib, it suggests a next-generation tyrosine kinase inhibitor osimertinib. However, these next-generation treatments typically provide only temporary responses. Waiting for resistance to grow during first-line treatment provides the tumor enough time and selective pressure to develop secondary resistance.

We propose a “forward-engineering” treatment strategy, that is, by genetically altering cancer cells, we use small molecule drugs to reverse the evolutionary landscape of the tumor, selecting altered cancer cells instead of resistant subclones. In redesigning the tumor, we can create more therapeutic opportunities with drug selection, rather than fewer.

Inspired by CRISPR-based systems control of disease vector evolution, we call this approach the “dual-switch selective gene drive”. The gene circuit consists of two genes or “switches” introduced into cancer cells by a carrier. The basic function of this modular platform is to couple an inducible adaptive advantage (switch 1) with a shared fitness cost (switch 2). Switch 1 acts as a synthetic resistance gene, conferring a temporary resistant phenotype, expanding the frequency of engineered cells during treatment. Switch 2 is a therapeutic payload gene. Similar to standard GDEPT, it activates a diffusible therapeutic that can kill both engineered and unmodified cancer cells. This bystander effect is enhanced by hitchhiking on the switch 1 gene, maximizing the therapeutic potential of the suicide gene. This bystander activity is unaffected by any local resistant populations selected during the switch 1 treatment phase. Furthermore, since switch 2 activity is theoretically limited to the tumor environment, higher local concentrations of the active agent may be achieved than with systemic administration. By promoting more robust therapeutic behavior, the risk of cross-resistance is minimized, and the promise of combination therapy is fully realized.

In this paper, we design, construct and evaluate dual-switch selective gene drives for cancer therapy using model-guided design. By designing controllable analogs of drug targets, we demonstrate switch 1 activity in multiple biological contexts. In addition, we established the therapeutic functionality and bystander killing effect of GDEPT and immune version switch 2 genes. Our complete dual-switch circuit demonstrated the ability to eliminate pre-existing resistance, including complex genetic library resistance variants within drug targets and across the whole genome. Finally, model-guided switch activation in vivo demonstrated strong therapeutic efficacy, emphasizing the benefits of leveraging evolutionary principles rather than opposing them. Overall, our findings support the concept of selective gene drives rooted in evolutionary theory as a framework for redesigning tumors and targeting diverse native genetic heterogeneity.

The results show that introducing and selecting a genetic structure will introduce more heterogeneity into the tumor population and purposely expand the group of genetically engineered cancer cells. To assess the evolutionary risk of this counterintuitive therapeutic method, we developed a stochastic tumor evolution mechanism model. Such a model can foresee and investigate evolutionary risks associated with the selective gene drive system. Moreover, understanding the expected evolutionary dynamics under selection helps to formulate key system design standards. These standards cover the hypothetical gene transfer efficiency required to achieve evolutionary control and the adaptiveness of gene-driven cells during the switch 1 treatment phase.

The model considers an initially sensitive small cancer cell population that expands until a portion of the tumor cells are transformed into gene drive cells, at which point treatment begins. Treatment with switch 1 is maintained until the gene drive cells become the dominant population, and then switch 2 treatment begins. Throughout the simulation process, mutation events generate subclone models, simulating potential system failure points. These mutations include acquiring resistance to targeted therapy, resistance to the therapeutic activity of the switch 2 gene, and loss of switch 2 activity in gene drive cells.

Furthermore, we wanted to assess potential spatial risks with the selective gene drive system. In particular, the bystander effect of the therapeutic switch 2 gene requires proximity to unmodified cells to clear them. Therefore, we anticipated that the spatial distribution of gene-drive cells and the range of bystander activity would be important determinants of therapeutic success. To expect spatial failure points, we constructed a spatial surrogate model of the selective gene drive system. The model considered a mixed population of sensitive cells, resistant cells, and gene drive cells. However, gene drive cells were seeded according to spatial diffusion parameters.

In this study, we propose using model-tested design criteria to determine the feasibility of the gene drive system in a broader context. Compared to clinical demonstrations, results indicate that evolutionary control is achievable even with low gene transfer rates. Moreover, even if gene drive cells are less fit relative to natural resistant populations, simulation results suggest the possibility of achieving evolutionary control. This is because, even with low gene transfer rates, gene drive cells are expected to number several orders of magnitude more than resistant subclones at the onset of treatment.

Overall, these model results explore many conceivable failure modes with physiologically reasonable parameters and predict the likelihood of successful forward-engineering of tumor populations. However, selective gene drive therapy is expected to prolong progression-free survival in all cases and enhance eradication potential in most conditions. Moreover, model studies suggest partial independence in simulation results with various dispersal and bystander parameters, benefiting the selective gene drive system when bystander activity is dispersed.

Theoretical compartmental and spatial surrogate models suggest that selective gene drives are a viable path toward achieving evolutionary control, and so we designed and assembled genetic structures based on theoretical outcomes. Manufacturing gene structures applicable to existing standard of care treatments and capable of using evolutionary principles to eradicate heterogeneity in pre-existing forms is feasible.

Our original prototype gene constructs were controllable, using molecular parts that have been proven safe in humans, and their gene drive behavior matched quantitative models. Furthermore, we demonstrated the modularity of selective gene drive patterns, with different switch designs proving inducible adaptive advantages in various drugs and tumor types. Alternative switch 2 systems, including immune-mediated anti-cancer mechanisms, displayed strong bystander activity. To demonstrate the evolutionary robustness of our approach in the face of genetic diversity and spatial failure modes, we showed that selective gene drives could eradicate a wide range of genetic heterogeneity within drug targets and throughout the genome. Finally, using evolutionary models to optimize in vivo switch dynamics, we applied these findings to in vivo drug kinetics experiments to demonstrate the effectiveness of dual-switch functionality.

All experiments we conducted in vitro and in vivo took place within a large, pre-existing resistant cell population at super-physiological levels. Thus, our model-driven design is expected to achieve evolutionary control in tumor cell populations that could rapidly fail treatment.

Selective gene drive technology is built on the advance of the emerging field of evolution-guided therapy. A primary consideration for selective gene drive therapeutic approaches is delivery. Theoretically, tumor cells can be modified in vivo to express the genetic circuit. Local delivery with non-resectable late-stage tumors provides a low threshold to begin transformation.

We have demonstrated the conceptual framework of selective gene drives, which can redirect the evolution of tumors to create more therapeutic states. Our initial selective gene drive designs, validated with stochastic evolutionary models, exhibited predictable domains and considerable robustness against genetic and spatial failure modes. Although selective gene drives carry risks, the tricky problems of treating advanced tumors and the vast genetic diversity of tumor baselines require bold new approaches. By leveraging evolutionary models, we can design tumors to reliably and effectively target their own heterogeneity.