Use of circulating tumour DNA to prospectively guide a switch from targeted to immune therapy in BRAF mutant advanced melanoma: the randomised phase II CAcTUS trial Authors: Rebecca J. Lee, Dominic G. Rothwell, Nigel Smith, Shien Chow, Juan Delgado-SanMartin, Hitesh Mistry, Yvonne Sylvestre, Shih-Chieh Chiang, Harry Clarke, Gabriela Gremel, Avinash Gupta, Kimberley Hockenhull, Noel Kelso, Rohit Kochhar, Damian Mullan, Ruth Plummer, Patricio Serra, Heather Shaw, Holly Summersgill, Samra Turajlic, Florent Mouliere, Richard Marais, Caroline Dive, Paul Lorigan Published: 2026 ABSTRACT Checkpoint inhibitor immunotherapy (CPI) for BRAF mutant advanced melanoma first-line results in a better long-term survival compared to targeted therapy (TT), however TT induction may benefit poor prognosis groups. The parallel-arm, randomised phase II, multicentre, feasibility CAcTUS trial (Clinicaltrials.gov NCT03808441) randomised 21 patients to receive standard of care investigators choice TT or CPI, switching to the alternative upon progression (n = 10), or commencing TT and switching to CPI upon an ≥80% reduction of BRAF variant allele frequency (VAF) in circulating tumour DNA (ctDNA; n = 11). The study achieved its primary endpoints with 100% (95% confidence interval [CI]: 94-100%) of critical results provided within 7 days to inform a decision to switch and 100% of patients commencing TT achieving an ≥80% reduction of BRAF VAF (95% CI: 80-100%). Secondary outcomes included progression-free survival and overall survival. No new safety signals were observed for TT/CPI. Post-hoc analysis of clinical features, circulating cytokines and chemokines at ctDNA nadir following TT induction suggested a more favourable profile prior to CPI initiation. Longitudinal ctDNA dynamics revealed ctDNA provided an early signal of CPI benefit and that rechallenge with TT following CPI progression resulted in a further ctDNA response. These data support the utility of ctDNA to guide treatment decision-making within a clinically relevant timeframe to optimise treatment scheduling strategies. Introduction The advent of targeted therapy (TT; BRAF + MEK inhibitors) and CPI (anti-PD-1 ± anti-CTLA-4) have significantly improved overall survival (OS) for patients with advanced BRAF mutant cutaneous melanoma1–3. More recently, several trials and pre-clinical studies have examined the optimal scheduling of these treatments. The DREAMSeq trial showed that CPI with standard dose ipilimumab plus nivolumab (N + I) first line and TT with dabrafenib plus trametinib as second line in patients progressing on CPI, was associated with a 20% absolute improvement in 2-year OS (72% (95% CI: 62.5 to 79.1) vs. 52% (95% CI: 41.7 to 60.4), p = 0.010) compared to the reverse schedule4. Studies have shown that when tumours are responding to TT, the tumour microenvironment can have increased T-cell infiltration, improved T-cell recognition of melanoma-associated antigens and reduced production of immunosuppressive cytokines5–7. Mechanistic pre-clinical data have shown that in melanoma with acquired resistance to TT, a decrease in CD103+ dendritic cells is observed, resulting in inferior antigen presentation and response to subsequent CPI8. Thus, there appears to be an ‘optimum point’ where patients responding to TT are also most likely to derive benefit from switching to CPI, before the tumour becomes resistant to TT and thus less likely to respond to immunotherapy. Based on this concept, the SECOMBIT study (NCT02631447) examining optimal treatment scheduling of TT and CPI in advanced melanoma, included an experimental arm where N + I was preceded by an 8-week run-in of TT with encorafenib plus binimetinib (E + B), with patients reverting to E + B if progression occurred on N + I9. SECOMBIT showed that 3-year OS for N + I first and the experimental arm were comparable, whilst patients starting on TT and switching to CPI on progression had a poorer outcome9. The EBIN study (NCT03235245) evaluated a fixed dose run-in with 12 weeks of targeted therapy E + B and elective switch to immunotherapy, compared to upfront immunotherapy with N + I10. There was no difference in progression-free survival (PFS) between N + I first and the induction TT then N + I (HR: 0.87, 95% CI: 0.67–1.12, p = 0.36) for the whole trial population10. However, in a pre-specified sub-group analysis of patients with adverse prognostic factors, there was a benefit for the induction in patients with LDH >2xULN (HR: 0.46; 95% CI: 0.21–1.03) and in those with liver metastasis (HR: 0.48; 95% CI: 0.28–0.80, p value for interaction 0.008)10. Both studies used a pre-defined TT induction period in the experimental arm, though the optimal length of the induction period, if any, remains unclear. We and others have shown that circulating tumour DNA (ctDNA) can be used to accurately follow tumour burden over time and that ctDNA monitoring is acceptable to patients11–16. Critically, by analysing tumour-specific mutations in the blood, we were able to identify treatment response and disease progression at an early stage13,15. We therefore hypothesised that ctDNA could be used to tailor the duration of TT and identify the optimum time to switch to CPI, when the tumour was responding to TT, and therefore most likely to respond to CPI therapy. Informed by pre-clinical data (Supplementary Fig. 1a) and the variance criteria for the test, we determined that even those patients with high disease burden were likely to have an ≥80% reduction of BRAF variant allele frequency (VAF) early on TT, although the response might be short-lived. Not all patients completely cleared their ctDNA at maximal response. Defining complete ctDNA clearance as the time point to switch would prevent some patients from switching to immunotherapy early in the course of their treatment, risking the emergence of targeted therapy-resistant clones, which would be less likely to respond to CPI therapy. We therefore designed the CAcTUS trial to test whether personalising a treatment switch from TT to CPI guided by an ≥80% reduction of BRAF VAF would enable us to better define that ‘optimum point’ between response and resistance to TT and so individualise the decision to change treatment electively. Results Real-time clinical decision-making based on circulating tumour DNA level is deliverable To enable individualised monitoring of response to treatment we developed and validated (to good clinical practice regulations, see ‘Validation of droplet digital PCR (ddPCR) BRAF V600 assay for the CAcTUS trial’ report Supplementary Information) droplet digital assays (ddPCR) for the detection of BRAF V600E, K and R mutations. We used these assays to detect an 80% BRAF VAF drop in ctDNA when blood was longitudinally sampled. Assay specificity was demonstrated previously for all ddPCR BRAF assays using healthy volunteer circulating free DNA (cfDNA) (HNV), where all samples were found to be negative11. We confirmed assay specificity using cell line DNA with known BRAF status, where a VAF of 0% was observed in all negative controls (n = 3 HNV in triplicate, n = 2 experiments, Supplementary Data, negative in all cell lines without the mutation present, Supplementary Table 1). Assay reproducibility (%VAF difference between wells of the same experiment) and repeatability (between 3 different experiments) was demonstrated to be within ±5% of the determined %VAF (Supplementary Tables 2 and 3) for all assays. To demonstrate the VAF lower limit of detection (LLOD) and lower limit of quantitation (LLOQ), mutant positive cell-line DNA was mixed with wild-type cell-line DNA in serial dilution and a standard curve produced (Supplementary Table 3, Supplementary Fig. 1b, c). All wildtype controls had 0.0% VAF. Using this, we determined that the lowest input for the assay to detect ≥3 mutant positive droplets (LLOD) was 0.40 ng and 2.12 ng to detect ≥10 mutant positive droplets (LLOQ), and therefore we planned to have a minimum of 5 ng input cfDNA to reliably determine an 80% BRAF VAF drop with treatment. Having demonstrated the reliability of the assay, between 2nd May 2019 and 30th May 2022, we randomised 21 patients to the trial (Consort diagram, Fig. 1). The control arm (n = 10) consisted of either investigators’ choice first-line immunotherapy with nivolumab 1 mg/kg and ipilimumab 3 mg/kg given 3-weekly for 4 cycles then nivolumab maintenance 3 mg/kg 4-weekly, (N + I) followed on progression by second-line TT (dabrafenib 150 mg bd + trametinib 2 mg od), (D + T), or the reverse schedule (Supplementary Fig. 1d). In the experimental arm of ctDNA guided therapy (n = 11), patients were commenced on D + T, had 2 weekly ctDNA assessments for the first 4 weeks then 4-weekly based on standard of care visit schedules and switched to N + I when an ≥80% decrease in BRAF VAF from the baseline pre-treatment sample was observed. If the ≥80% decrease in BRAF VAF was not achieved, patients would continue on TT. To reduce the incidence of ‘crossover’ toxicity when switching treatments, we stipulated a wash-out period based on drug half-lives. Trametinib was stopped 7 days prior to first cycle of N + I, and dabrafenib stopped the day before. If patients subsequently progressed on CPI, they could then recommence D + T, under the premise that resistance to D + T had not developed, as they were initially switched when still responding (Supplementary Fig. 1d).Fig. 1Consort diagram.End of study reasons: patient withdrawal (n = 1), died (n = 8), TE; Alive at end of trial follow-up (n = 12). Immune: nivolumab plus ipilimumab. Targeted: dabrafenib plus trametinib. Consort diagram. End of study reasons: patient withdrawal (n = 1), died (n = 8), TE; Alive at end of trial follow-up (n = 12). Immune: nivolumab plus ipilimumab. Targeted: dabrafenib plus trametinib. We initially screened 10 patients with a BRAF VAF of ≥5% as the inclusion criteria, based on previous literature17 that this would identify patients with poorer prognosis disease with conventional treatment who might therefore benefit from alternative treatment strategies. Screening was performed analysing duplicate experiments using cfDNA pooled from 2 separate tubes taken at the same time, with a concordance of 0.99. We found that a cut-off of ≥5% BRAF VAF resulted in exclusion of some patients who still had poor prognostic disease (high LDH, multiple disease sites; Fig. 2a). Using the screening data collected (Supplementary Data), we reassessed the LLOD and LLOQ and determined that the inclusion VAF could be reduced to ≥1.5% without compromising the assay performance. Overall, analysis of the 44 screening samples showed that a threshold of ≥5% would lead to 45% of patients being included, whereas a ≥ 1.5% threshold would enable inclusion of 65% of patients (Fig. 2a). Despite lowering the threshold to ≥1.5% BRAF VAF, we still selected for a cohort which had features associated with a poorer prognosis compared to contemporary trials4,9,18 with the majority of patients having M1c disease, raised LDH and a median number of ≥3 disease sites (Table 1). Patients who were excluded due to a VAF of ≤1.5% were associated with lower baseline LDH (median 324.5 IU/L [interquartile range: 297.3–537.0] CAcTUS included patients (n = 21) vs. 218.0 [207.7–255.2] (n = 10) for <1.5% VAF excluded patients p = 0.0068 (Wilcoxon test)) with 76% (16/21) having >3 sites of disease for CAcTUS included patients vs. 30% (3/10) in those excluded due to <1.5% VAF.Fig. 2Real-time clinical decision-making based on circulating tumour DNA is deliverable.a Comparison of different screening inclusion criteria ctDNA VAF cut-offs. Distribution of screening ctDNA samples (average of 2 separate ctDNA tests) over time and eligibility of patients (n = 44, n = 16 not eligible, n = 21 eligible and n = 7 not eligible for other reasons) to enter the study pre and post protocol amendment, depending on inclusion criteria cut-off of 5% and 1.5% VAF. Dotted line relates to the original cut-off of 5%, and the dashed line the amended cut-off value of 1.5%, green dotted line represents the date of protocol amendment. b Distribution of days to return sample reports for critical samples for n = 44 patients. Time to report critical samples defined as time required for a clinical decision to be made. c Initial percentage change from baseline in ctDNA variant allele frequency (VAF) for each patient (n = 21), depending on initial treatment commenced (targeted therapy [TT, teal] n = 17 or immune therapy [CPI, salmon] n = 4), with 80% threshold shown (dashed line). TT visits were 2 weekly, and CPI were 3 weekly according to treatment schedules. Source data are provided as a Source data file. d Swimmer plot for n = 21 patients (10 control, 11 experimental arm), showing treatment administered (immune therapy [CPI, pink] vs. targeted therapy [TT, blue]), response type (CR complete response = green, PR partial response = orange, SD stable disease = blue, PD progressive disease = red, NE not evaluated = grey) and events (alive = triangle, COVID-19 related death = square and melanoma death = circle).Table 1Patients’ characteristics at baselineVariableLevelControl arm (N = 10)Experimental arm (N = 11)Biological sex (N)Male66Female45Age (median [range])62 [34–80]67 [39–81]ECOG (N)067144M classification (N)Ia22Ib02Ic77Id10Point Mutation (N)V600E99V600K11V600R01No of disease sites (median [range])4 [2–6]3 [2–9]RECIST sum of diameters in mm (median [range])95 [49–241]106 [51–158]LDH (IU/L)Normal12≥x1ULN66≥x2ULN33Albumin (g/L) (median [range])44 [34–48]45 [42–50]CRP (mg/L) (median [range])a20 [1–130]5 [2–217]ANC (109/L) (median [range])7.5 [1.1–10.9]5.2 [3.4–12.5]Lymphocyte (109/L) (median [range])1.2 [1–2.1]1.4 [0.5–2.0]ANC/Lymphocyte ratio (median [range])5.5 [1.1–10.9]3.4 [2.0–21.0]Platelet (109/L) (median [range])339 [160–497]253 [205–362]ctDNA (N)<1.5%001.5 to ≤10%45>10 to ≤20%30>20%361st line treatment choice in control arm (N)Dabrafenib + Trametinib6-Nivolumab + Ipilimumab4-ULN upper limit of normal.aOne patient did not have baseline/screening for CRP. Real-time clinical decision-making based on circulating tumour DNA is deliverable. a Comparison of different screening inclusion criteria ctDNA VAF cut-offs. Distribution of screening ctDNA samples (average of 2 separate ctDNA tests) over time and eligibility of patients (n = 44, n = 16 not eligible, n = 21 eligible and n = 7 not eligible for other reasons) to enter the study pre and post protocol amendment, depending on inclusion criteria cut-off of 5% and 1.5% VAF. Dotted line relates to the original cut-off of 5%, and the dashed line the amended cut-off value of 1.5%, green dotted line represents the date of protocol amendment. b Distribution of days to return sample reports for critical samples for n = 44 patients. Time to report critical samples defined as time required for a clinical decision to be made. c Initial percentage change from baseline in ctDNA variant allele frequency (VAF) for each patient (n = 21), depending on initial treatment commenced (targeted therapy [TT, teal] n = 17 or immune therapy [CPI, salmon] n = 4), with 80% threshold shown (dashed line). TT visits were 2 weekly, and CPI were 3 weekly according to treatment schedules. Source data are provided as a Source data file. d Swimmer plot for n = 21 patients (10 control, 11 experimental arm), showing treatment administered (immune therapy [CPI, pink] vs. targeted therapy [TT, blue]), response type (CR complete response = green, PR partial response = orange, SD stable disease = blue, PD progressive disease = red, NE not evaluated = grey) and events (alive = triangle, COVID-19 related death = square and melanoma death = circle). Table 1Patients’ characteristics at baselineVariableLevelControl arm (N = 10)Experimental arm (N = 11)Biological sex (N)Male66Female45Age (median [range])62 [34–80]67 [39–81]ECOG (N)067144M classification (N)Ia22Ib02Ic77Id10Point Mutation (N)V600E99V600K11V600R01No of disease sites (median [range])4 [2–6]3 [2–9]RECIST sum of diameters in mm (median [range])95 [49–241]106 [51–158]LDH (IU/L)Normal12≥x1ULN66≥x2ULN33Albumin (g/L) (median [range])44 [34–48]45 [42–50]CRP (mg/L) (median [range])a20 [1–130]5 [2–217]ANC (109/L) (median [range])7.5 [1.1–10.9]5.2 [3.4–12.5]Lymphocyte (109/L) (median [range])1.2 [1–2.1]1.4 [0.5–2.0]ANC/Lymphocyte ratio (median [range])5.5 [1.1–10.9]3.4 [2.0–21.0]Platelet (109/L) (median [range])339 [160–497]253 [205–362]ctDNA (N)<1.5%001.5 to ≤10%45>10 to ≤20%30>20%361st line treatment choice in control arm (N)Dabrafenib + Trametinib6-Nivolumab + Ipilimumab4-ULN upper limit of normal.aOne patient did not have baseline/screening for CRP. Patients’ characteristics at baseline ULN upper limit of normal. aOne patient did not have baseline/screening for CRP. Of note, the median % BRAF VAF was higher in the experimental arm (20.34% [range: 2.36–34.65%]) than control (13.4% [range: 2.53–63.48%]). In addition, in the control arm, the baseline BRAF VAF for those patients commencing CPI first was lower than those commencing TT (median 3.46% [range: 2.53–17.67%] vs. 16.91% [5.67–63.48%], respectively, p = 0.067), likely reflecting emerging evidence guidelines that CPI was associated with a better outcome for the majority of patients and first line targeted therapy should be reserved for those patients with symptomatic or poor prognosis disease requiring a rapid response to treatment. Essential to the delivery of the study was the ability to report ctDNA results quickly. A key primary endpoint for feasibility was therefore to assess whether ≥95% of samples could be processed and reported following quality assurance approval within 7 working days. Of the 344 samples collected, 61 were designated as critical, i.e. would inform clinical decision-making. For these 61 samples, the mean processing time was 5.4 ± 1 working days (with median of 6 and range of 3–7 working days), with 100% (95% CI: 94–100%) critical results reported within 7 days (Fig. 2b). The primary endpoint of CAcTUS was to assess whether we would observe an ≥80% decrease in mutant BRAF VAF. All the patients who initially received TT across both arms (control arm; n = 6, experimental arm; n = 11), were observed to have an ≥80% reduction in % BRAF VAF from baseline (95% CI: 80–100%), taking a median of 15 days (range: 13–54 days) to reach the 80% reduction threshold (Fig. 2c). In addition, all patients (n = 4) who initially received CPI, were observed to have an ≥80% reduction of % BRAF VAF with a median of 22 days to reach ≥80% reduction; range: 21–49 days (Fig. 2c). Of note, 3 of these 4 patients had a durable PR until trial end, suggesting the ctDNA reduction was associated with good outcomes to CPI. Of note, the progressing patient 18 had a rise in ctDNA before scan progression. Switch at time of ctDNA drop is associated with features of immunotherapy response The treatments received and time on treatment are shown in the consort diagram (Fig. 1) and swimmers plot (Fig. 2d). Of the 6 patients who started targeted therapy first and who were treated to resistance on the control arm, 3 progressed on TT and received second line immunotherapy but all progressed quickly and died, 1 had progressive disease (PD) on TT and was not fit to receive CPI then died, and 2 patients responded until trial end. For those on the control arm who had CPI first, 3 had durable partial responses, and 1 progressed and switched to targeted therapy but subsequently further progressed. Of those on the experimental arm receiving initial treatment with D + T, 1 patient died due to disease progression before they could be switched, and 1 patient withdrew consent and switched to E + B due to toxicity and remained on this. The rest (n = 9/11) were switched from TT to CPI based on an ≥80% BRAF VAF drop. Of these 9 patients, 1 died from COVID-19, although was responding to treatment, 2 had a durable response to immunotherapy, and 6 subsequently progressed on CPI and switched back to targeted therapy. Two of these 6 patients died subsequently due to disease progression, and 4 were alive at trial end, with 2 continuing to respond to TT. Progression-free (including both TT and CPI treatment lines) and overall survival were not significantly different between both arms (median overall survival control arm 605 days [95% CI: 377–NR] vs. experimental arm 511 days [95% CI: 197–NR], p = 0.707 [Supplementary Fig. 1e, f]). Treatment-related toxicity was higher in the experimental arm than control, with 4 patients in the control arm and 9 in the experimental arm experiencing CTCAE v5 grade 3/4 toxicity. These were in line with the known toxicities for these regimens (Supplementary Tables 4–6). To determine whether the targeted therapy run-in and switch at time of ctDNA drop could potentially affect response to immunotherapy, we examined features known to be associated with CPI response in patients on the experimental arm, from baseline to the start of immunotherapy. We observed a significant decrease in lactate dehydrogenase (LDH), neutrophils and neutrophil:lymphocyte ratio (Wilcoxon rank-sum test p < 0.05 for all), a non-significant decrease in platelets, and increase in C-reactive protein (CRP) as well as heterogeneous changes in lymphocytes (Fig. 3a–g). Furthermore, analysis of circulating cytokines/chemokines in patients on the experimental arm from baseline to cycle 1 day 1 of immunotherapy revealed an increase in a number of inflammatory mediators commonly associated with CPI response such as IFNg, TNFa, IL12, CXCL9, ICAM-1 and a reduction in those mediators associated with poorer outcomes such as IL6 and IL8 (Fig. 3h). We also observed a decrease in the melanoma-associated protein S100B, which correlated with the ctDNA decrease (Fig. 3h). When we compared those patients on the control arm commencing CPI first line with those patients on the experimental arm commencing CPI following the targeted therapy run-in, we observed lower LDH, neutrophils, platelets and neutrophil:lymphocyte ratio, with higher CRP (Supplementary Fig. 2a–f; statistical analysis not performed due to inadequate power) at the time of commencing CPI treatment in the experimental arm patients. Furthermore, we observed higher levels of circulating IFNg, CXCL10 and lower IL6 and IL8 in patients on the experimental arm vs. control arm prior to CPI (Supplementary Fig. 2g–r). To assess if the depth of radiological target lesion response reflected ctDNA response in the experimental arm, we examined the waterfall plots of best radiological response based on % change in sum of target lesion diameters (SLD) and the maximum ctDNA % change in VAF, as well as their best overall response determined by RECIST v1.1 evaluation (colour-coded bars). Of note, an additional scan was performed at the time of switch to CPI in the experimental arm. This revealed that the magnitude of ctDNA response (best % change in VAF) did not correlate well with the radiological response (best % change in SLD) due in part to the increased frequency of ctDNA assessments compared to scan assessments particularly in those patients with aggressive/poor prognosis disease (Supplementary Fig. 2s–v). To assess the kinetics of response to different treatment types in patients in the experimental arm, we assessed the best radiological response from baseline to the scan at switch (TT component) and from the scan at switch until PD or end of trial (CPI component). Excluding the 2 patients that progressed on immunotherapy, both components of the treatment strategy contributed to the response in the remaining patients (Fig. 3i, j). Finally, to assess whether retreatment with TT following progression on immunotherapy would restore disease control in the experimental arm, we assessed both the best radiological response (SLD) and the ctDNA response in patients restarting TT. We observed that all patients re-challenged with TT responded both radiologically and in ctDNA (Fig. 3k, l, Supplementary Table 7).Fig. 3Switch at time of ctDNA drop is associated with features of immunotherapy response.a–g Difference in prognostic parameters between patients at C1D1 IO following the initial TT run-in vs. baseline in patients on the experimental arm. a ctDNA variant allele frequency (VAF). b Lactate dehydrogenase (LDH) c. Absolute neutrophil count (ANC) d ANC/lymphocyte ratio. e Lymphocyte count. f C-reactive protein (CRP). g Platelet count. Wilcoxon two-sided signed-rank test p values shown, n = 13 patients with 1 technical replicate, whisker plots represent the mean ± Standard Error of the Mean. h Heatmap showing difference in plasma pg/ml of circulating proteins between patients (numbered on each row, n = 7) at C1D1 IO following the initial TT run-in vs. baseline on the experimental arm. Red = increased and blue=decreased compared to baseline in patients on the experimental arm. Waterfall plots showing best % change in sum of lesion diameters (SLD) from i baseline to the switch (TT run-in, n = 9 patients) and j from the switch to end of trial or progression (CPI component, n = 8 patients) on the experimental arm. Waterfall plots showing best % change in k sum of lesion diameters (SLD, n = 5 patients) and l ctDNA VAF from C1D1 of restart of TT (following CPI progression, n = 5 patients) to end of trial or progression. Source data are provided as a Source data file. Switch at time of ctDNA drop is associated with features of immunotherapy response. a–g Difference in prognostic parameters between patients at C1D1 IO following the initial TT run-in vs. baseline in patients on the experimental arm. a ctDNA variant allele frequency (VAF). b Lactate dehydrogenase (LDH) c. Absolute neutrophil count (ANC) d ANC/lymphocyte ratio. e Lymphocyte count. f C-reactive protein (CRP). g Platelet count. Wilcoxon two-sided signed-rank test p values shown, n = 13 patients with 1 technical replicate, whisker plots represent the mean ± Standard Error of the Mean. h Heatmap showing difference in plasma pg/ml of circulating proteins between patients (numbered on each row, n = 7) at C1D1 IO following the initial TT run-in vs. baseline on the experimental arm. Red = increased and blue=decreased compared to baseline in patients on the experimental arm. Waterfall plots showing best % change in sum of lesion diameters (SLD) from i baseline to the switch (TT run-in, n = 9 patients) and j from the switch to end of trial or progression (CPI component, n = 8 patients) on the experimental arm. Waterfall plots showing best % change in k sum of lesion diameters (SLD, n = 5 patients) and l ctDNA VAF from C1D1 of restart of TT (following CPI progression, n = 5 patients) to end of trial or progression. Source data are provided as a Source data file. Longitudinal ctDNA monitoring through multiple treatment lines provides early signals of treatment efficacy To further assess the dynamics of ctDNA during treatment on each arm, we plotted the ctDNA level alongside the LDH and SLD (control arm Supplementary Fig. 3a, experimental arm Fig. 4). In both arms, ctDNA decrease was associated with response in SLD (e.g. patient 12, 15, 16 and 17). The ctDNA response sometimes preceded a CR in SLD (e.g. patient 10) and also differentiated true response from pseudoprogression as previously reported19. Patient 12 had an increase in size of a ureteric nodule, but the ctDNA remained 10 mg daily prednisone equivalent) or other immunosuppressive medications within 14 days of study drug administration. Patients with brain metastases and leptomeningeal metastases were excluded unless asymptomatic and untreated at presentation, or if symptomatic, the lesions had to have been definitively treated with surgery or stereotactic surgery and did not require steroids for control of symptoms. See protocol (Supplemental information) for other exclusion criteria. Patients were enrolled between 2nd May 2019 and 30th May 2022, and the trial completed when the last patient finished their follow-up. Trial design and endpoints CAcTUS was designed as a parallel arm randomised phase II, multicentre, feasibility trial. It was conducted across 12 UK centres. Initially, 40 patients were planned to be randomised 1:1 to either control arm (investigator choice) or experimental arm (ctDNA-guided switch). The primary feasibility endpoints of the study were i. ≥95% return of critical (red) ctDNA results returned within 7 working days of samples being received in the laboratory ii. ≥80% of patients with a decrease in ctDNA level of mutant BRAF ≥ 80% on targeted therapy. The number of patients was later reduced to 21, as in discussion with the trial steering committee, we observed that all patients commencing TT were reaching the primary endpoint of 80% VAF drop, and therefore the trial was closed to recruitment. The original study sample size target was 40 patients (20 in each arm) based on true feasibility sample size parameters of a i. 95% (or greater) return rate of critical clinical decision dependent blood samples; ii. 80% (or greater) success rate of participants on targeted therapy achieving ≥80% BRAF VAF decrease in ctDNA levels enabling estimation of i. the return rate with a two-sided 95% confidence interval of width no more than 12% points (i.e. 87–99%) assuming at least 60 critical red blood samples (one sample from each of the 20 control arm participants and an average of at least two samples from each of the 20 the experimental arm participants); ii. the success rate of participants on targeted therapy achieving ≥80% BRAF VAF decrease in ctDNA levels with a two-sided 95% confidence interval of width equal to 30% points (i.e. 62%, 92%), assuming that 25% of the participants in the control arm will receive D + T as first line treatment (i.e. 25 participants across arms; 5 in the control arm and 20 in the experimental arm). On interim review of the return rate, there was good evidence of a high return rate, and hence a 95% confidence interval of similar width (14%) could be obtained from a sample of 21 or more participants (resulting in at least 32 critical clinical decision dependent samples, assuming one sample from control arm participants and at least two samples from experimental arm participants). Moreover, 32 samples provided 80% confidence that the lower confidence limit was at least 90% if the true return rate was at least 98%. Likewise, with a sample size of 17 participants with ctDNA reduction data, we could be confident that at least 80% of participants would achieve a reduction in ctDNA levels of at least 80% from baseline, assuming that the true ‘success rate’ percentage was 95% (or greater). Randomisation was performed centrally via the King’s College London independent randomisation service. Patients were equally randomised to either treatment arm using minimisation with a random element controlling for LDH (10 to ≤20%, >20%) and M stage (stage III unresectable plus M1a, M1b, M1c and M1d). In the control arm, the choice was between either dabrafenib 150 mg twice daily plus trametinib 2 mg once daily or nivolumab 1 mg/kg plus ipilimumab 3 mg/kg Q3wk followed by 480 mg nivolumab Q4wk first line, then switching on progression to the other treatment second line until disease progression or unacceptable toxicity. The experimental arm commenced dabrafenib 150 mg twice daily with trametinib 2 mg once daily. Following decrease of ctDNA BRAF VAF by ≥80%, trametinib was stopped for 7 days with dabrafenib continuing until the day before switching treatment. A CT scan was performed to correlate radiological imaging with ctDNA response. The patient was then switched to nivolumab 1 mg/kg plus ipilimumab 3 mg/kg Q3wks followed by 480 mg nivolumab Q4wk. Then, if progression was observed, they were switched back to dabrafenib 150 mg twice daily plus trametinib 2 mg once daily until disease progression or unacceptable toxicity. The co-primary endpoints were the number of critical (defined as required for a clinical decision) ctDNA results returned within 7 working days of samples being received in the laboratory and the number of patients commencing TT first in whom we observed an ≥80% decrease in BRAF VAF. Assessments Response and disease progression were assessed by radiologists at each site according to RECIST guidelines (version 1.1). Toxicity was assessed according to Common Terminology Criteria for Adverse Events guidelines (CTCAE version 5). Collection of blood samples and plasma separation Blood samples were collected at pre-specified timepoints according to the protocol schedule (Supplemental information). Four tubes of 10 ml of venous blood were collected into Streck Cell-Free DNA BCT tubes and transported at ambient temperature. For ctDNA assays, double spun (DS) plasma was separated within 96 h of blood draw by centrifugation at 2000 × g for 10 min, followed by a second centrifugation at 2000 × g for 10 min, then stored at −80 °C until required. For Luminex assays, double spun plasma from EDTA tubes was separated within 1 h by centrifugation at 2000 × g for 10 min, followed by a second centrifugation at 2000 × g for 10 min and stored at −80 °C until required. Droplet digital PCR Cell-free DNA was isolated from the entire DS plasma using the QIAsymphony DSP DNA Kit (Qiagen, Cat No: 1091063) as per the manufacturer’s instructions. cfDNA was isolated from 6 ml batches of DS plasma, each eluted in 60 μl of kit elution buffer (Qiagen). These were pooled and stored at −20 °C until required. ddPCR was performed using the QX200-AutoDG Droplet Digital PCR System (Bio-Rad) as per the manufacturer’s instructions. All preparation and reaction set-up was performed in a dedicated pre-PCR room and PCR cabinet. One of three BRAF 20x primer-probe mutation detection assays (Bio-Rad) was used to measure %VAF; BRAF p.V600E c.1799T>A (Assay ID: dHsaMDV2010027), BRAF p.V600K c.1798_1799GT>AA (Assay ID: dHsaMDV2010035), and BRAF p.V600R c.1798_1799GT>AG (Assay ID: dHsaMDV2010037). To improve quantification of the 80% drop in %VAF from screening/baseline (i.e. clinical decision making), critical timepoints (D + T week 2 and D + T week 4) were run in 8 wells. Other timepoints were run in 4 wells. Each 22 μL ddPCR reaction contained 11 μL of 2x ddPCR SuperMix for probes (no dUTP) (Bio-Rad), 9.9 μL of template DNA, and 1.1 μL of 20x primer-Probe mutation detection assays (Bio-Rad). Reactions were prepared in Bio-Rad ddPCR 96-well plates and droplets generated using the AutoDG. Plates were sealed with a pierceable foil heat seal (Bio-Rad) and PCR performed on a C1000 Touch thermal cycler (Bio-Rad) as per the manufacturer’s instructions. Amplifications were performed using the following cycling conditions: 1 cycle of 95 °C (2.5 C/s ramp) for 10 min, 40 cycles of 94 °C (2.5 C/s ramp) for 30 s and 55 °C for 1 min, followed by 1 cycle of 98 °C (2.5 C/s ramp) for 10 min. The sample was held at 4 °C until further processing. After PCR, the plates were read using a QX200 droplet reader (Bio-Rad). Positive and negative controls were used to manually set thresholds, which were applied to all samples in the assay. QuantaSoft™ software (version 1.7.4) was used to assign positive/negative droplets. Counts were then converted to copies/ml and %VAF. The %VAF was quantified if the sample contained more than 10 mutant droplets and recorded as detected is it contained between 3 and 9 mutant positive droplets. Samples containing <3 mutant positive droplets were below the limit of detection. The entire sample analysis workflow was validated and performed to ICH GCP E6, and a quality-assured %VAF result was reported. Luminex analysis Analysis was performed according to manufacturer’s instructions using the human Luminex Discovery assay panels on an xMAP INTELLIFLEX® machine. Analysis was performed using Quantist v1.0.1. Further analysis/heatmap visualisation was performed using Prism v10.2.2. Statistical analysis, visualisations, and data engineering Data engineering was performed using R packages tableone (v0.13.2), tidyverse (v2.0.0), and lubridate (v1.9.2). Data visualisations were generated using ggplot2 (v3.4.2). For statistical analyses, differences in central tendency between groups or timepoints were assessed using Wilcoxon rank-sum test or a Wilcoxon signed rank test when values could be paired. All biomarker and exploratory efficacy analyses were hypothesis-generating. Unadjusted p values are reported descriptively alongside effect sizes and confidence intervals; no formal adjustment for multiple testing was applied. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Information Supplementary Data 1 and 2 Reporting Summary Transparent Peer Review File Supplementary Information Description of Additional Supplementary Information Supplementary Data 1 and 2 Reporting Summary Transparent Peer Review File Source data Source data Source data