Targeting therapy-persistent residual disease

Abstract

Disease relapse driven by acquired drug resistance limits the effectiveness of most systemic anti-cancer agents. Targeting persistent cancer cells in residual disease prior to relapse has emerged as a potential strategy to enhance the efficacy and durability of current therapies. However, barriers remain to implementing persister-directed approaches in the clinic. This article discusses current preclinical and clinical complexities and outlines key steps toward the development of clinical strategies targeting therapy-persistent residual disease.

Molecularly targeted anti-cancer therapies often elicit dramatic and durable initial responses, however, for most patients, disease recurrence is inevitable. Incomplete eradication of cancer cells leads to therapy-persistent residual disease (RD) and eventual regrowth of a cancer that is resistant to the initial therapy. Some cells within the original tumor could harbor pre-existing resistance mutations to the selected therapy^1–3. Alternatively, some cells lacking pre-existing genetic mechanisms can survive the treatment and over time develop new resistance mechanisms^4,5. Gaining a better understanding of non-genetic sources of RD has become an important focus in cancer research, however, the development of therapeutic strategies to eliminate them is challenged by preclinical and clinical complexities.

1. Non-genetic sources of residual disease

Since the seminal work by Sharma et al.⁶, a decade of research has shown that subpopulations of tumor cells can initially survive therapy even without apparent genetic resistance mutations^4,5. This phenomenon is referred to as cancer “persistence”, with the surviving cancer cells called “persisters”. Evidence obtained from laboratory and clinical settings, across a wide range of cancer types and treatment modalities, indicates that eradicating persistent RD could improve the effectiveness of standard treatments and delay, or even prevent, disease progression.

Defining properties of cancer persistence

What characterizes cancer persisters? How do they stand apart from dormant cancer cells or cancer stem cells (CSCs)? Here, we focus on three defining properties that differentiate cancer persistence from other cancer cell states (Figure 1).

Figure 1. Non-genetic sources of therapy-persistent residual disease.

A: Inadequate elimination of cancer cells by standard-of-care therapies results in residual disease and subsequent tumor recurrence. B: A subpopulation of cancer cells, termed “persisters,” can initially survive drug treatment without harboring genetic resistance mutations, thereby serving as a reservoir for the development of diverse acquired resistance mechanisms. C: The evolution of persister cells culminating in drug resistance is complex and context dependent. A myriad of extracellular and intracellular signals promotes their emergence; once established, these cells exhibit molecular heterogeneity; and diverse mutational processes underpin their evolutionary trajectory.

i. Drug treatment:

Cancer persistence is identified within the context of a drug treatment. While cancer cells may develop certain survival-enhancing characteristics before treatment⁷, persisters emerge from within a bulk-sensitive population by surviving drug treatment. This specific criterion distinguishes persistence from dormancy and quiescence, which describe non-dividing cancer cells irrespective of drug treatment. Additionally, both persistence and stemness characterize cells capable of division, but cancer stemness is independent of drug treatment. Importantly, unlike CSCs that can be detected by certain cell surface markers, the identification of persisters relies solely on the functional state of cells surviving drug treatment. Thus, persistence only emerges with drug treatment, and depends on the selection pressure applied.

ii. Heterogeneous response:

Persistence reflects a heterogeneous survival outcome of cancer cells to treatment. Cancer persistence is most relevant for patients who experience a partial response (PR) or stable disease (SD) during treatment, as opposed to those whose cancer either resists (progressive disease or PD) or is entirely eradicated by the therapy. Persistence specifically describes the heterogeneous response of a cancer cell population to treatment in which some cells survive while other cells die.

iii. Non-genetic mechanisms:

Cancer persistence highlights that the lack of treatment response is not dictated exclusively by genomic alterations. Evidence of this is that drug-tolerant persister cells in vitro revert to a drug-sensitive state following a “drug holiday”—a period of drug treatment pause. Although reversibility tests are less applicable in patients, clinical therapy-persistent RD can progress to resistance with or without acquired genomic alterations^8–13, indicating the involvement of non-genetic mechanisms in tumor evolution. Our own labs demonstrated that tumor cells can follow distinct evolutionary paths to resistance, and persister cells serve as a reservoir from which diverse acquired resistance mechanisms can develop^4,5. How the genetic context sculpts mechanisms of persistence is an open question worthy of further investigation, as clinical data demonstrates that co-occurring mutations in tumor suppressor genes can shorten the time to acquired resistance to oncogene-targeting therapies independently of specific resistance mechanisms¹⁴. Nevertheless, persistence acknowledges the contribution of non-genetic mechanisms of cell survival to drug treatment.

Complexities of cancer persistence

Attacking persistence mechanisms is most relevant before acquired resistance mechanisms are present. Three such strategies are to prevent the emergence of persisters, inhibit the maintenance of their survival, or block their evolution into acquired resistance. What is the “optimal” timing for eliminating persisters?

i. The emergence of persisters:

Cancer persistence has two primary, non-mutually exclusive, origins: cell autonomous and non-autonomous. Cell-autonomous persistence is due to inherent heterogeneity in cancer cells, influenced by factors such as cell cycle position at the time of treatment¹⁵, stochastic expression of survival or apoptotic signaling proteins^16–18, or spontaneous entry into a slow-cycling state¹⁹. Non-autonomous persistence stems from environmental interactions. Extracellular signals can alter intracellular states which subsequently affect life-or-death decisions upon drug treatment²⁰, and the tumor environment with gradients of nutrients, cytokines, and oxygen levels diversifies drug responses. Treatment-induced changes, such as therapy-stimulated tumor secretome proteins or ATP release from dying cells, may support survival of emerging persister cells^21,22. Although completely preventing persisters from emerging is challenging, a reasonable hypothesis is that even partial prevention will enhance therapeutic response, extending progression-free or even overall survival of patients.

ii. The maintenance of persisters:

Upon drug treatment, cancer cells display features distinct from their pretreatment states, from epigenetics and gene transcription, epitranscriptomics and mRNA translation, to post-translational modification and metabolism (reviewed in refs^23–27). Persister composition can vary both within and among tumors^13,28, amplifying the complexity of addressing cancer persistence. A key question is, which features of persisters are vital for their survival? Omics profiling may provide insights into persister characteristics and associated vulnerabilities. In certain cases, the correlation between persister features and their biological function is clear, such as the link between persistence and enhanced cytochrome P450 drug detoxification function²⁹. However, identifying causal mechanisms driving persister cell survival is often more challenging, especially with the growing volume of data depicting features of persister cells down to single cell resolution. As with any complex biological system, experimental perturbations are critical for determining the functional importance of these features and revealing potential vulnerabilities.

iii. The evolution of persisters:

Persister cancer cells left unchecked can evolve and develop acquired resistance. They can adopt genetic alterations, in either the drug target or in proteins that bypass it^4,5, and transition from initial non-genetic persistence to more stable genetic resistance mechanisms. Recent studies have begun to reveal the molecular processes underlying this transition. For example, lung cancer targeted therapies can induce the expression of APOBEC3A, a cytidine deaminase, which leads to double-strand DNA breaks and genomic instability, thus promoting mutagenesis and the evolution of persister cells during therapy³⁰. In colorectal cancer, targeted therapies have been shown to decrease the expression of mismatch repair and homologous recombination DNA-repair genes, while increasing error-prone polymerases in persister cells, resulting in elevated DNA damage, mutability, and genomic instability³¹. Additionally, persisters can pursue non-genetic evolution paths to develop acquired resistance, as evidenced in studies on lung cancer and melanoma by our and other groups^5,32. These observations raise several critical questions: What determines whether persisters follow a genetic or non-genetic path of evolution? How do genetic and non-genetic mechanisms interact? What dictates the specificity and redundancy of DNA-editing enzymes and DNA repair pathways in shaping persister evolution? Furthermore, what is the role of RNA-editing enzymes in the transition of persisters to non-genetic acquired resistance? The evolution of persisters is an area of active research, with many avenues yet to be explored.

2. Experimental models for cancer persistence studies.

What are relevant models for studying cancer persistence? As in most biological and disease contexts, experimental models balance tradeoffs between assay scalability and physiological relevance.

i. Cancer cell culture models:

Among all available models, cancer cell cultures remain mainstay tools for studying persisters. Many pivotal discoveries in the field have been made using simple monolayer cell cultures. These include the initial identification of cancer persistence⁶, the evolution from persistence to acquired resistance^4,5, treatment-induced epigenetic alterations³³, adaptive mutability^30,31, and metabolic reprogramming³⁴. Additionally, three-dimensional cultures of tumor cells offer insights into persister biology^35,36, although tumor spatial arrangements captured by these models are not necessarily relevant in every study. Although cancer cell lines do not capture interactions between tumor cells and non-tumor cells within the tumor microenvironment, they are indispensable tools for investigating autonomous persistence of cancer cells. Their affordability, scalability, and the wide range of cell lines reflecting diverse clinicopathological features and treatment histories, empower investigations into both general and context-specific mechanisms and susceptibilities of persister cancer cells.

ii. Tumor-stroma culture models:

Integrating cancer cell cultures with elements of the tumor environment provides reductionist methods to identify signals that non-autonomously modulate persistence. These environmental “add-ons” can include soluble factors, such as cytokines and metabolites, or co-cultures of stroma cells, such as cancer-associated fibroblasts (CAFs) and immune cells. For example, our previous research utilized lung cancer cell models supplemented with tumor secretome factors, leading to the discovery of a novel persistence-modulating signal through the IFNγ/STAT1/type I PRMT axis²⁰. In another study, using co-culture models derived from patients with lung cancer, we identified three subtypes of CAFs, each with a distinct role in modulating sensitivity to targeted therapies, characterized by their expression of HGF, FGF7, and p-SMAD2³⁷. These findings demonstrate the utility of such models in pinpointing causal signals and cellular interactions that affect persister survival. Although tumors encounter combinations of environmental signals too vast to test exhaustively, these scalable models allow exploration in a broad cell signaling space to identify persister-specific mechanisms. The study of cancer behavior in relation to its environment is particularly important in persister studies, given the non-genetic, cell plasticity roots of cancer persistence.

iii. In vivo models:

Animal models hosting human or murine tumors serve as tools for in vivo study of cancer persistence. Human cancer cell line- or patient-derived xenograft models, though lacking interaction with a human host environment, are useful for studying cancer cell autonomous persistence. This is well illustrated by a study that identified neural crest stem cell transition in melanoma persisters following RAF/MEK inhibition²⁸. Syngeneic murine tumor models do not always recapitulate the complexity of human disease, however, they retain host immunity, allowing the study of non-autonomous persistence with therapies targeting both tumor and immune cells^38,39. In the context of basal cell carcinoma, where other experimental models are less accessible, syngeneic models have been especially useful, enabling the discovery of a cell identity switch in persister cells following the inhibition of Hedgehog pathway⁴⁰. Humanized models of patient-derived xenografts, preserving elements of tumor-host interaction, may be valuable for studying immunotherapy persistence. Despite their substantial costs and limited scalability, animal models enable exploration of persister biology that is unattainable in vitro, and they offer a complementary system to validate persister findings derived from in vitro studies.

By design, every experimental model has strengths and limitations, therefore employing orthogonal and diverse models is essential to validate persister discoveries and avoid model-specific artifacts.

3. How do we detect and assess persistent residual disease in the clinic?

The development of clinical strategies to target persistence has been slowed by an incomplete understanding of therapy-persistent RD in patients. Limitations in tumor sampling during the course of treatment are a fundamental obstacle. Successful translation of experimental findings into effective therapies will require characterizing clinical RD and should focus on two major efforts: First, a more comprehensive assessment of the landscape of cell states and interactions in residual tumors is needed to prioritize therapeutic targets for specific clinical contexts. Second, the development of novel technologies for non-invasive detection of RD will enable identification of biomarkers necessary for deployment of persister-directed therapies.

i. Assessing the biology of clinical persistence:

Investigation of persistence mechanisms in patients in the clinic has largely focused on studying tumor cells, either by analyzing resected tumors after neoadjuvant therapy or through “on-treatment” biopsies, or by isolation of circulating tumor cells from blood samples. Deep molecular profiling using techniques such as single cell RNA-seq have begun to illuminate phenotypic heterogeneity of cancer and non-cancer cells associated with therapy-persistent RD^13,41–43. Spatial transcriptomic and proteomic analysis can be applied to reveal the physical arrangement and interactions of persister cells within the tumor microenvironment. Spatial approaches are particularly valuable, given that persistence is inherently a result of heterogeneous drug response within the bulk tumor, and the persister population likely exhibits molecular heterogeneity. The integration of imaging and omics technologies with spatial information may offer unique insights into the mechanisms of clinical persistence. We have recently developed such an integrative method, combining spatial analysis of multi-modal data, which led to the discovery of biologically meaningful tissue subpopulations that were not identifiable through individual modalities alone⁴⁴. Leveraging opportunities for tissue sampling during therapy will continue to be essential for better understanding the biology of clinical therapy-persistent RD, and whenever possible, should be incorporated into clinical trials. However, because these efforts are invasive and resource-intensive, they are challenging to scale to larger clinical cohorts. Therefore, the development of less invasive methods will be critical to expand detection and analysis of clinical RD.

ii. Non-invasively detecting clinical persistence:

Clinical monitoring of therapeutic response, RD and relapse has traditionally been accomplished by quantifying overall disease burden, either by use of radiologic imaging (e.g., CT, MRI, PET) or tumor markers in the plasma (e.g., PSA, CEA). Identifying secretome markers specific to tumor persistence (e.g., glycosylated plasma proteins) could be particularly valuable, as these markers might provide insights into the persistence biology and help select persister-guided treatment. In the past several years, plasma circulating tumor DNA (ctDNA) has become increasingly useful for disease monitoring, for instance in the detection of RD after local therapy⁴⁵, and can predict prognosis after initial response to targeted therapy^46–48. Because ctDNA provides information about specific genomic alterations, it can reveal genomic resistance mechanisms that serve as biomarkers to guide subsequent therapy⁴⁹. Importantly, blood-based “liquid biopsies” can easily be integrated into routine clinical care. Although current ctDNA assays are of more limited utility in evaluating non-genetic tumor persistence, liquid-biopsy approaches currently under development hold promise for assessing non-genetic mechanisms. For instance, methods that can detect tumor specific gene expression in plasma cell free RNA (cfRNA)^50,51 or by cell free DNA fragmentation patterns⁵² may enable monitoring of transcriptional states associated with drug tolerance. Complementary methods that infer tumor-specific epigenetic states based on analysis of DNA methylation patterns^53,54, chromatin modifications⁵⁵ or integrative approaches⁵⁶ may be well suited to non-invasively detect lineage plasticity. Recently, we along with other researchers reported a method for comprehensive epigenomic profiling of cancer from patient plasma. This approach may serve as a robust proxy for transcriptional activity, enabling inference of the expression levels of diagnostic markers and drug targets, measurement of the activity of therapeutically targetable transcription factors, and detection of epigenetic mechanisms of resistance⁵⁷. This proof-of-concept study demonstrates the significant potential of plasma epigenomic profiling in providing clinically actionable information that, until now, could only be accessed through direct tissue sampling.

With liquid biopsies becoming potentially useful in detecting non-genomic tumor features, an important question is whether they will also have sufficient sensitivity to assess persisters in low-burden RD. This is particularly relevant for therapies that induce dramatic responses, as highlighted by studies showing that ctDNA becomes undetectable in most patients treated with lung cancer targeted therapies^46,47. Significant progress must be made to optimize sensitivity and specificity of these assays, yet there is reason to be optimistic. Similar challenges exist in the context of early cancer detection and optimizing treatment for early-stage disease^58,59, and applying lessons learned in these settings will accelerate progress in the RD setting.

4. How will we run clinical trials and treat patients?

As our understanding of clinical RD matures and leads to persister-targeting drugs, clinical trial frameworks will need to evolve to test RD-targeting strategies (Figure 2). For example, a drug might be given before initiation of standard-of-care (SoC) therapy to eliminate cells in a state primed to become persisters, or as an upfront combination with SoC to prevent establishment of RD. Alternatively, a drug targeting induced mechanisms of persistence could be added after an initial period of SoC therapy. Ideally, trials should incorporate biomarkers that inform patient and drug selection as well as treatment timing. Given that many SoC targeted therapies now achieve deep and durable responses lasting years, early surrogate endpoints beyond the traditional metrics of progression free survival or overall survival will be necessary to efficiently evaluate the added clinical benefit of persister-directed therapies. There is unlikely to be a universal trial design that is suitable for all persister-directed therapies. Clinical trials will need to be tailored to optimize patient selection, treatment timing and measures of clinical efficacy.

Figure 2. Framework for clinical evaluation of persister-directed therapies (PTx).

A: A PTx (orange) can be combined upfront with standard-of-care therapy (SoC; black) to prevent therapy-persistent residual disease (RD). The selection of patients would be guided by prognostic and/or predictive markers present prior to initiation of therapy. B: Upon achieving a stable residual disease state with an initial period of SoC therapy, a PTx can be introduced in combination with the SoC. Patients could be selected based on the burden of residual disease. C: The initial phase of SoC therapy allows for monitoring of predictive biomarkers induced in persister cells. These biomarkers would guide selection of specific PTx matched to the biology of the residual tumor. Right: Early indicators of response, such as clearance of ctDNA or decrease in tumor volumetric measurements from RD “baseline”, could increase efficiency in evaluating the efficacy of a PTx.

i. Patient selection:

With the increasing effectiveness of many SoC therapies, incorporating prognostic biomarkers to identify sub-groups of patients predicted to have shorter duration of response to SoC may enable clinical trials to enroll those most likely to benefit from addition of a persister-targeted drug. For example, recent studies have shown that the level of ctDNA at the time of diagnosis or shortly after initiation of lung cancer targeted therapies inversely correlates with duration of response^46,47. Focusing clinical trials on “high-risk” populations who have a higher burden of RD or other poor-prognostic features would have practical benefits of decreasing cohort size while reducing the time to trial readout. Predictive biomarkers that reveal specific biology of persistent disease and guide the selection of persister-directed therapies will also be essential, especially in disease contexts for which multiple possible persister states might arise. Clinical trial designs that incorporate prognostic and predictive biomarkers will need to consider the optimal timing of biomarker assessment, whether initially present, or dynamically emerging after initiation of treatment.

ii. Treatment timing:

Many persister-targeting drugs will be used in conjunction with SoC therapy and have little efficacy when used in isolation. Clinical trials will need to be designed to deploy the persister-targeted drug in the optimal window of activity. This will necessitate understanding the dynamics of the persister state, informed by biomarkers that can define clinical timescales relative to initiation of SoC. In particular, strategies targeting persisters after initial response to SoC will require a clear clinical definition of RD. On-going trials testing local consolidative therapy^60,61 for residual oligo-metastatic non-small cell lung cancers after initial response to EGFR and ALK targeted therapies (NCT03410043, NCT03707938) may provide an instructive roadmap for the timing of persister-targeting drugs.

iii. Measures of success:

With SoC therapies that now elicit responses lasting years (for example, the median PFS of ALK fusion-positive lung cancers to the third-generation ALK TKI lorlatinib is greater than 3 years⁶²), the time and cost required to demonstrate improvement in traditional outcomes such as PFS or OS present barriers to rapid clinical testing of persister-directed therapies. Surrogate endpoints that assess efficacy earlier could enable acceleration of the cycle of clinical testing and promote better engagement of patients, oncologists, and industry stakeholders. For example, if a persister-directed drug is introduced at the point of stable RD after initial tumor regression, detecting a second radiologic response or clearance of ctDNA could provide an early indication of efficacy and justify continuing the persister-targeted therapy until PFS endpoints can be reached. Additionally, early indication of efficacy could support a decision to continue therapy despite a modest increase in toxicity.

Clinical trial frameworks for testing persister-targeted strategies

In the simplest case, a persister-directed therapy could be initiated simultaneously with the SoC therapy (Figure 2A). Conceptually, this approach would be best suited to prevent the establishment of persister states. Patient selection would be guided by biomarkers in the tumor or blood before treatment. Clinical efficacy would be determined by an increase in PFS compared to SoC alone, with an increased frequency of ctDNA clearance providing an early surrogate readout of clinical benefit.

Alternatively, the persister-directed therapy can be added after an initial period of SoC treatment once a stable RD state is achieved (Figure 2B). Sequential therapies may be optimal for targeting adaptive mechanisms of persistence induced in response to the SoC. This design would allow assessment of initial response to select patients with suboptimal response to SoC, who may be most likely to benefit from escalation of therapy. Additionally, analogous to an N-of-1 clinical trial⁶³, once a patient has achieved stable RD on SoC, a new baseline can be established to evaluate additional tumor shrinkage as an early metric of response to the persister-directed therapy. This period would also provide an opportunity to monitor predictive biomarkers that emerge and evolve during treatment to guide treatment selection (Figure 2C). Finally, a sequential design could be extended to test multiple persister-directed therapies in a single patient.

To foster the paradigm shift that will be required to enable clinical development of persister-targeted therapies, we envision a stepwise process. First, it is imperative to establish proof-of-concept rationale for adaptive escalation of therapy at the time of RD. The most straightforward approach is to design adaptive clinical trials that deploy established therapies, which are guided by validated prognostic biomarkers. A prime example is an on-going study testing the feasibility and utility of adding chemotherapy treatment in patients with EGFR-mutant lung cancer who have persistent ctDNA after initial treatment with osimertinib (NCT04410796). Such trials will provide opportunities to refine methods for patient stratification, biomarker monitoring and surrogate efficacy endpoints. Additionally, trials should integrate translational programs to enable research efforts to develop and benchmark novel liquid-biopsy technologies, better understand the biology of clinical RD, and identify predictive biomarkers that can facilitate the development of persister-targeting drugs. Once the utility of an adaptive trial framework is established, individual advances in each of these areas can be assembled to construct innovative clinical trials that incorporate prognostic and predictive biomarkers and test novel persister-targeted drugs matched to the biology of individual patients.

5. Summary and future outlook

In summary, we contend that effective targeting of therapy-persistent RD before acquired resistance would represent a major advance in the treatment of cancer and should be a priority within the oncology drug development community. Achieving this goal will require continued and sustained preclinical and clinical research efforts to illuminate the biology of RD, identify drug targets, and bring novel agents into clinical testing. Specifically, we propose action in the following five key areas: i. Expansion of experimental models. We need to continue developing and refining models that faithfully represent, or at least capture key aspects of, RD in cancer patients. ii. Access to on-treatment patient samples. Obtaining samples from patients undergoing treatment is crucial for a better understanding of persister biology and evolution of RD. iii. Development of biomarkers. Identifying robust prognostic and predictive biomarkers is essential for detecting clinical persistence and stratifying patients who are most likely to benefit from persister-directed therapies, particularly after an initial response to SoC treatments. iv. New clinical trial frameworks. A paradigm shift in clinical trial design is necessary. Persister-targeting therapies must be optimized in the context of the corresponding SoC. While opportunities for monotherapies may exist under certain specific conditions, the development of combination therapies will require re-adjusting expectations and commitment to innovating clinical trial designs⁶⁴. v. Development of new therapeutic modalities. Novel therapeutic approaches must be developed for targets currently lacking effective methods of intervention. It is our genuine belief that oncologists, academic researchers, industry stakeholders, and regulatory agencies share a mutual interest in advancing RD-targeting treatments. With a united effort, the cancer community can move closer to transforming tumors into manageable conditions, and possibly even achieving cures.