Adaptive therapy achieves long-term control of chemotherapy resistance in high grade ovarian cancer

Systemic cancer therapy traditionally aims to eradicate malignant cells using maximum tolerated doses, yet this often fails in advanced solid cancers due to selection for pre-existing or phenotypically plastic drug-resistant subclones. Evolutionary theory posits that resistance carries fitness costs that are revealed in resource-limited environments. Adaptive Therapy (AT) leverages competitive interactions between sensitive and resistant cells, seeking disease control rather than cure by maintaining a suppressive sensitive-cell population. Prior modelling and preclinical work predict benefit, and an initial clinical AT study in metastatic castration-resistant prostate cancer using PSA-guided abiraterone showed prolonged time to progression and overall survival, but lacked randomisation and mechanistic tracking of resistance. High grade serous ovarian cancer (HGSC), which exhibits relapsing-remitting responses to platinum chemotherapy and evolving resistance, provides an ideal context to test AT. The study hypothesises that platinum-resistant HGSC cells pay a fitness cost in resource-poor conditions, that chemotherapy reverses relative fitness to favour resistant cells, and that dynamically adjusting carboplatin dosing can prolong tumour control. The work also evaluates whether LiqCNA can track emergent resistant populations and relate to tumour burden, enabling more precise AT guidance.

- Evolutionary trade-offs in cancer imply resistant phenotypes expend resources that can reduce proliferation or survival in low-resource settings. - Mathematical and preclinical models suggest AT can prolong control by exploiting competition (Gatenby et al., Anderson et al., Enriquez-Navas et al.). - A non-randomised AT trial in prostate cancer using abiraterone guided by PSA showed extended time to progression and overall survival compared to continuous dosing but lacked direct resistance tracking and randomisation; larger randomised trials are ongoing. - Intermittent therapy trials in other cancers (e.g., BRAF-mutant melanoma, androgen deprivation in prostate cancer, TKIs in renal cancer) generally did not link dosing to individual tumour dynamics and had mixed results, emphasising the need for response-guided adaptation. - Platinum resistance in HGSC lacks recurrent point mutations or canonical copy number drivers; however, patient-specific CNAs emerge post-treatment. LiqCNA uses longitudinal CNA patterns to quantify emergent resistant subclones from tissue and liquid biopsies. - Prior studies in colorectal and breast cancer under hypoxia or low glucose observed competition between sensitive and resistant cells, but mechanisms for resistant decline were less defined.

In vitro studies: - Cell lines: Human HGSC lines OVCAR4 and Cov318 cultured in DMEM with 10% FBS and antibiotics. Platinum-resistant derivatives generated by stepwise exposure to cisplatin or carboplatin to achieve 2–10x IC50 increases: Ov4Cis (cisplatin-resistant OVCAR4), Ov4Carbo (carboplatin-resistant OVCAR4), CovCis (cisplatin-resistant Cov318). OVCAR4 and Cov318 transduced with GFP; resistant OVCAR4 derivatives transduced with RFP for lineage tracing. - Co-cultures: Sensitive and resistant cells seeded at defined ratios and maintained either in high-resource conditions (10% FBS) or low-resource conditions (0.5% FBS with daily media exchange). Total cell counts obtained (Countess IIR). Proportions of GFP/RFP measured by flow cytometry (BD LSRFortessa) and analysed in FlowJo. Growth rates and carrying capacities inferred using Mathematica v11 and PopDynamics. Linear models fitted to log(sensitive:resistant) ratios to estimate growth-rate differences (g = g_s − g_r) and assess independence from initial seeding ratios. - Mechanism assays under low-resource conditions: Cell cycle profiling by PI staining and RNaseA with flow cytometry; apoptosis by Annexin V (BV605) with etoposide-treated positive controls; conditioned media experiments exchanging media pre-conditioned by same/opposite cell line or 50:50 co-culture, assessing effects on growth; senescence markers p16 and p21 quantified by qRT-PCR (normalized to β-actin) with doxorubicin-treated controls. - Chemotherapy perturbation: 50:50 OVCAR4-GFP:Ov4Cis co-cultures in low-resource conditions treated with a single cisplatin/carboplatin dose (0.1–1 µM) on day 6, drug washed off after 24 hours. Time-course FACS quantification of sensitive/resistant fractions post-treatment. In vivo studies: - Mouse models: Female CD1 nu/nu mice injected subcutaneously (both flanks) or intraperitoneally with mixtures of sensitive OVCAR4 (GFP or non-fluorescent) and resistant Ov4Cis-RFP at varying ratios (e.g., 100:0, 80:20, 50:50, 10:90). Tumours measured by callipers; endpoint criteria predefined. - Quantification of lineage proportions: qPCR for GFP and RFP DNA normalized to human GAPDH from harvested tumours; time-course assessments at weeks 4, 8, 12 for subcutaneous tumours seeded at 50:50 or 80:20. - Histology and IHC: Tumour sections stained for p53 (tumour burden), GFP (sensitive cells), cleaved caspase-3 (apoptosis). Pixel quantification of sensitive vs resistant regions, noting spatial distribution. Adaptive therapy (AT) vs standard therapy in vivo: - Tumour-bearing mice (flanks injected with 100% sensitive OVCAR4, 80:20 OVCAR4:Ov4Carbo-RFP, or 100% Ov4Carbo-RFP) randomized when a target tumour reached 300 mm3. Groups: vehicle (PBS IP every 4 days ×3), standard carboplatin (60 mg/kg IP every 4 days ×3), or AT. - AT dosing algorithm (weekly dosing): start/max 60 mg/kg. If target tumour volume change ≤20%, no dose change; >20% increase, increase by 50% (cap 60 mg/kg); >20% decrease, reduce by 50%; if target tumour <60 mm3 or <20% of initial, omit dose. Calliper measurements weekly; cumulative dose tracked; toxicity monitored. Patient sample analysis and LiqCNA: - Patients with stage III/IV HGSOC provided tissue biopsies and serial blood samples under ethics approval. Plasma processed for cfDNA; leucocyte pellets stored as germline controls. - DNA processing: Tissue DNA (including LCM for small tumour foci) and cfDNA libraries prepared (NEBNext kits, QIAseq cfDNA kit). Shallow whole-genome sequencing: mean depth ≈0.5× for tissue, 1.9× for cfDNA, 0.3× for leucocytes. Reads aligned to hg19; CN profiles generated with QDNAseq. - LiqCNA pipeline applied to longitudinal samples to infer emergent resistant (ER) subclone proportions by identifying resistant-specific CNAs and estimating subclonal ratios. Bootstrapping: 150 runs on random 75% segment subsamples to derive 95% CIs per estimate. - Clinical correlation: ER growth rates compared with CA125 trajectories; Pearson correlations computed. Cases with only two timepoints flagged due to potential confounding by pervasive CN instability or measurement bias.

- Fitness costs in resistant cells under low resources: - In high-resource (10% FBS) co-culture, resistant populations increased as a fraction over time across multiple sensitive–resistant pairs, with independent exponential growth and no competition (e.g., growth rates ≈0.31±0.05 doublings/day for sensitive vs 0.43±0.04 for resistant in one setting). - In low-resource (0.5% FBS with daily exchange), co-cultures exhibited logistic growth with competition; resistant populations declined relative to sensitive, independent of initial ratios. Estimated growth-rate difference g = g_s − g_r < 0 for resistant cells, indicating a fitness cost penalised by limited resources. - Mechanism of resistant decline: - Cell cycle profiles for sensitive cells unchanged by co-culture; resistant cells showed increased sub-G0 and significantly higher Annexin V positivity that scaled with larger sensitive fractions (e.g., Day 8 and Day 11 in 10% resistant co-culture vs 100% resistant monoculture, p=0.0001), indicating apoptosis drives resistant decline. - Conditioned media had no effect on growth rates/carrying capacities, arguing against secreted-factor mediation; no induction of p16 or p21 under low-resource co-culture, suggesting senescence is not the primary mechanism. - In vivo validation of fitness cost: - In both subcutaneous and intraperitoneal co-injection models, the sensitive population expanded beyond the initial ratio while resistant declined at 12 weeks, mirroring in vitro low-resource dynamics. - IHC showed discrete GFP-negative (resistant) islands embedded in GFP-positive (sensitive) tumour, with cleaved caspase-3 positivity confined to resistant regions and increasing over time (weeks 4 to 12), consistent with apoptosis-mediated decline. - Chemotherapy reverses relative fitness and enables AT: - A single carboplatin exposure in co-culture led to dose-dependent transient reductions in the sensitive fraction with subsequent recovery and outgrowth as drug effects waned, demonstrating dynamic sensitive–resistant fluctuations through therapy. - In vivo AT vs standard carboplatin: - 100% sensitive tumours: AT yielded undefined median survival (too few deaths) vs 18.75 weeks with standard dosing (p=0.0082); durable control to study endpoint in most mice. - 80:20 sensitive:resistant tumours: AT achieved durable control with undefined median survival vs 11.25 weeks with standard therapy (p=0.574; small n). - 100% resistant tumours: AT increased survival vs standard carboplatin (p=0.0389), though all progressed by week 8; standard dosing reduced median survival to 5.75 weeks vs 15.25 weeks with vehicle (p=0.0896), consistent with competitive release. - AT cohorts received higher cumulative carboplatin without observed excess toxicity; in sensitive-dominant tumours, doses were often reduced or omitted as tumour control was maintained. - Tracking emergent resistance in patients: - LiqCNA detected resistant-specific CNAs and quantified ER subclone proportions in sequential tissue and cfDNA from five HGSC patients, with patient-specific CNA patterns. - Strong correlation between ER growth rate and CA125 levels (R=0.94, p=0.00015), supporting CA125 as a practical guide for AT and LiqCNA as a potential biomarker to direct dosing. - Concordance between tissue and blood ER estimates varied; in one case with closely timed samples, estimates were similar; in another, cfDNA suggested higher ER than tissue, highlighting sampling and shedding biases.

Findings provide direct empirical evidence that platinum-resistant HGSC cells bear a fitness cost that is exposed in resource-limited environments typical of tumours. This cost manifests as apoptosis-driven decline of resistant cells when competing with sensitive cells, and chemotherapy transiently reverses relative fitness to favour resistant populations. These dynamics enable AT to cyclically modulate drug pressure, preserving a sensitive-cell population that suppresses resistant clones and delays progression, as demonstrated by significant survival benefits in mouse models, particularly when tumours are sensitive-dominant. The observation that standard high-dose carboplatin can worsen outcomes in fully resistant tumours underscores the risk of competitive release and the evolutionary rationale for AT. While AT cohorts accrued higher cumulative doses, durable control with dose reductions/omissions suggests benefits arise from dosing dynamics rather than dose intensity alone, consistent with prior preclinical reports. The study further shows that emergent resistant populations can be inferred from longitudinal CNAs using LiqCNA, and that ER dynamics correlate strongly with CA125, supporting the use of CA125 in current clinical AT protocols and LiqCNA as a future precision biomarker to guide dosing based on resistance evolution. Spatial structure, microenvironmental constraints, and tumour ecology likely modulate clonal competition in vivo; the in vivo results, despite immunodeficient models, align with ecological theory and bolster the translational premise for AT in HGSC.

This study demonstrates that platinum-resistant HGSC cells incur a measurable fitness cost in low-resource settings, leading to apoptosis-driven decline when competing with sensitive cells. Chemotherapy dynamically alters relative fitness, producing predictable oscillations in sensitive and resistant populations that can be leveraged by adaptive therapy. In mouse models, carboplatin AT significantly prolonged tumour control and survival compared to standard dosing without added toxicity, while avoiding competitive release. In patients, LiqCNA quantified emergent resistant subclones from serial tissue and liquid biopsies and correlated closely with CA125, supporting its potential to guide future AT. These insights have led to the multicentre, randomised ACTOv phase 2 trial evaluating carboplatin AT in relapsed, platinum-sensitive HGSC. Future work should integrate resistance-tracking biomarkers like LiqCNA into AT algorithms, dissect microenvironmental and spatial effects on competition, and determine how dosing dynamics vs cumulative exposure contribute to clinical benefit.

- Preclinical in vivo studies used immunodeficient mice, limiting assessment of immune contributions to tumour ecology and therapy response. - AT-treated mice received higher cumulative carboplatin doses than standard therapy, so survival benefits cannot be fully disentangled from total dose effects in this experiment. - Small cohort sizes in some in vivo groups limit statistical power (e.g., 80:20 tumours). - LiqCNA estimates from only two timepoints may be confounded by pervasive CN instability and measurement bias, potentially overestimating ER fractions in some cases. - Tissue vs cfDNA ER concordance can be affected by sampling and ctDNA shedding biases, which may not represent whole tumour heterogeneity. - Lack of recurrent, easily trackable genomic drivers of platinum resistance in HGSC constrains universal biomarker development; LiqCNA relies on patient-specific CNAs. - Generalisability to human clinical outcomes awaits results from the ACTOv randomised trial.