ctDNA & MRD in Immunotherapy Trials: Early Response Signals That Hold Up

ctDNA & MRD in Immunotherapy Trials: Early Response Signals That Hold Up

Immune checkpoint inhibitors have always had a measurement problem. Imaging response is slow and the first scan usually comes eight to twelve weeks in. Some patients appear to progress and then respond. And in the adjuvant setting there is nothing on a scan to measure at all. Circulating tumour DNA (ctDNA) offers a faster, quantitative read on tumour burden from a blood draw, and MRD testing (detecting molecular residual disease after curative-intent treatment) applies the same idea to patients who look disease-free.

In 2026 the field crossed a regulatory line. On 15 May 2026 the FDA approved adjuvant atezolizumab for muscle-invasive bladder cancer in patients with ctDNA molecular residual disease after cystectomy, with Natera’s Signatera CDx as the companion diagnostic. A blood-based MRD test now decides who receives a checkpoint inhibitor.

This article is for translational, clinical development and diagnostics teams. It covers how MRD assays differ, which ctDNA response signals on immunotherapy have held up in prospective and pooled data, where pseudoprogression fits, how to design the sampling plan, and what regulators and pharma partners now expect.

Key takeaways

• Tumour-informed MRD assays track variants found in each patient’s tumour and generally give higher specificity. Tumour-agnostic assays use a fixed panel or genome-wide signals and turn around faster, but not every patient is trackable.
• Early ctDNA decline or clearance on PD-1/PD-L1 therapy is associated with longer overall survival in prospective studies (INSPIRE, BR.36) and in pooled data from 918 patients across four randomised NSCLC trials.
• IMvigor011 showed that serial ctDNA surveillance can select patients who benefit from adjuvant atezolizumab, and it led to the first FDA approval of a drug with a blood-based MRD companion diagnostic.
• FDA’s final guidance (November 2024) supports ctDNA for patient selection, enrichment and signal finding in curative-intent solid tumour trials, but says ctDNA is not yet validated as an early endpoint to support approval.
• Sampling time points, molecular response thresholds, limit of detection and clonal haematopoiesis filtering all change the answer. They need to be prespecified, not chosen after the data arrive.

MRD testing, ctDNA and molecular response: getting the terms straight

ctDNA is the fraction of cell-free DNA in plasma that comes from the tumour. FDA describes it as tumour-derived fragmented DNA shed into the bloodstream that is not associated with cells.

Minimal residual disease is the long-standing haematology term. In solid tumours, FDA and most sponsors now say molecular residual disease: ctDNA detected after surgery, radiotherapy or (neo)adjuvant therapy, in a patient with no radiographic evidence of disease. MRD testing is therefore a prognostic and selection question: is cancer still there?

Molecular response is the on-treatment question, mostly in the metastatic setting: does ctDNA fall, and how far, after the first cycles of therapy? Definitions vary (a 50% fall, a 90% fall, complete clearance), and that variability is one of the main obstacles to using ctDNA as an endpoint.

Tumour-informed vs tumour-agnostic MRD assays

FDA’s guidance on ctDNA in curative-intent solid tumour drug development sets out the trade-offs between the two main assay designs. The summary below follows that framework.

FDA also flags that tumour-informed panels are more reliable when built from tumours with higher mutation burden and well-characterised variants, a point that matters for low-TMB indications. Our explainer on tumour mutational burden covers that biology.

ctDNA molecular response on checkpoint inhibitors: the evidence

INSPIRE: personalised ctDNA on pembrolizumab

The phase 2 INSPIRE study (Bratman et al., Nature Cancer, 2020) applied bespoke, tumour-informed ctDNA assays to 316 plasma samples from 94 patients across five pembrolizumab cohorts, sampled at baseline and every three cycles. Baseline ctDNA correlated with progression-free survival, overall survival and response, and the association became stronger when on-treatment kinetics were added. All 12 patients whose ctDNA cleared during treatment were alive at a median follow-up of 25 months.

BR.36: molecular response as a trial decision point

The Canadian Cancer Trials Group BR.36 trial (Anagnostou et al., Nature Medicine, 2023) went further and built ctDNA into the design. Stage 1 treated 50 patients with advanced NSCLC with pembrolizumab and used a tumour-agnostic plasma panel, with matched white blood cell sequencing to remove clonal haematopoiesis variants. Molecular response was defined as clearance of the maximal mutant allele fraction by the third cycle.

• Of 45 patients evaluable for ctDNA, 10 had no detectable ctDNA at baseline, leaving 35 evaluable for response; 15 (43%) had a molecular response.
• Against RECIST response, sensitivity was 82% and specificity 75%.
• Median time to molecular response was 2.1 months, ahead of the 12-week imaging assessment.
• Overall survival was markedly longer with molecular response (hazard ratio 0.16, 95% CI 0.05 to 0.50).

Stage 2 uses molecular response to randomise patients without it to treatment intensification or continued therapy, which is exactly the adaptive use case sponsors want to test.

Pooled data across trials

Single studies use different assays and thresholds. The Friends of Cancer Research ctMoniTR project pools patient-level data to test whether the signal survives harmonisation. Its first analysis of 200 patients from five NSCLC checkpoint inhibitor studies (Vega et al., JCO Precision Oncology, 2022) found that on-treatment ctDNA reductions were associated with better overall and progression-free survival. A larger analysis of 918 patients from four randomised trials (Journal for ImmunoTherapy of Cancer, 2025) tested three thresholds (a fall of at least 50%, at least 90%, and complete clearance) at two windows: up to seven weeks, and seven to 13 weeks. In the anti-PD(L)1 group, reductions at both windows were associated with improved survival at every threshold. Patients with molecular response at both windows had the strongest association, and the later window was marginally stronger.

The pattern holds, but these are patient-level associations. Showing that a treatment’s effect on ctDNA predicts its effect on survival at trial level, the test a surrogate endpoint must pass, still needs randomised data.

Adjuvant immunotherapy: MRD testing becomes a selection biomarker

The bladder cancer programme with atezolizumab is the clearest example of MRD testing moving from prognosis to selection. IMvigor010 did not meet its endpoint in the intention-to-treat population. But a tumour-informed ctDNA analysis of 581 patients (Powles et al., Nature, 2021) found 214 (37%) ctDNA-positive at the start of therapy. These patients had a poor prognosis and did better with atezolizumab than observation (disease-free survival HR 0.58; overall survival HR 0.59), with no difference between arms in ctDNA-negative patients. ctDNA clearance at week 6 was 18% with atezolizumab against 4% with observation.

The prospective test was IMvigor011 (Powles et al., NEJM, 2025). Of 761 patients enrolled, those who turned ctDNA-positive during up to a year of serial surveillance were randomised 2:1 to atezolizumab or placebo (n=250). Median disease-free survival was 9.9 vs 4.8 months (HR 0.64) and median overall survival 32.8 vs 21.1 months (HR 0.59). Among 357 patients who stayed ctDNA-negative, disease-free survival was 95% at one year and 88% at two years without adjuvant treatment. The FDA label tells clinicians to keep testing ctDNA-negative patients until a positive result or the end of the 12-month window.

Two design lessons follow. First, serial surveillance captures patients a single landmark test misses: in TRACERx (Abbosh et al., Nature, 2023), postoperative landmark ctDNA was detected in 49% of patients who went on to relapse, and three- to six-monthly surveillance flagged impending relapse in a further 20% of landmark-negative patients. Second, a negative MRD result is a probability, not a cure. De-escalation designs need high negative predictive value and long follow-up.

Pseudoprogression: where ctDNA helps most

Pseudoprogression (apparent growth on imaging followed by response) is uncommon but costly when it causes effective therapy to be stopped. In a cohort of 125 patients with metastatic melanoma on anti-PD-1 therapy (Lee et al., JAMA Oncology, 2018), nine of the 29 with RECIST progression had pseudoprogression. A favourable ctDNA profile (undetectable at baseline, or a more than 10-fold fall on treatment) predicted pseudoprogression with 90% sensitivity and 100% specificity.

The numbers are small and the assay was a mutation-specific ddPCR test for BRAF and NRAS. Still, the logic carries over: when imaging and ctDNA disagree early on immunotherapy, a falling ctDNA level supports continuing treatment until confirmatory imaging. Protocols that permit treatment beyond progression can prespecify ctDNA as supporting data.

Designing ctDNA into an immunotherapy trial

Sampling time points

Collect a pre-treatment baseline in every patient. FDA asks for it, and it is the only way to know whether a later negative result reflects clearance or a tumour that never shed. Then sample early on treatment (around cycle 2 to cycle 3, roughly three to seven weeks) and again at 7 to 13 weeks; the ctMoniTR data suggest two on-treatment points are better than one. In the adjuvant setting, set a landmark at least several weeks after surgery (IMvigor011 started at least six weeks after cystectomy) and then sample serially. Timing must be identical across arms.

Thresholds

Choose the molecular response definition before unblinding. Complete clearance is easy to communicate but depends heavily on assay sensitivity. Percentage falls (50% or 90%) are more robust across assays but need a quantitative readout and a detectable baseline. Prespecify sensitivity analyses for alternative thresholds and windows.

Sensitivity and limit of detection

Ask for LoD in terms of tumour fraction, at the plasma volume and cfDNA input you will actually collect, and confirmed in clinical specimens where possible, as FDA recommends. In the metastatic setting, plan for patients with undetectable baseline ctDNA: in BR.36, 10 of 45 were not evaluable for response on that basis. Low-shedding histologies need either a more sensitive assay or a larger sample size.

Clonal haematopoiesis filtering

Clonal haematopoiesis is the biggest source of false positives in plasma. In a high-depth sequencing study (Razavi et al., Nature Medicine, 2019), most cfDNA mutations, 53.2% in patients with cancer and 81.6% in controls, had features consistent with clonal haematopoiesis. Matched white blood cell sequencing is the standard fix. Tumour-informed designs largely avoid the problem by tracking only tumour-confirmed variants; tumour-agnostic panels need explicit filtering and should report how many variants were removed.

Pre-analytics and logistics

Standardise blood collection tubes, processing time, plasma volume and storage across sites. FDA expects validation to cover the whole system from the collection tube onward. For tumour-informed assays, build tissue acquisition and panel-design turnaround into the screening timeline.

The wider principles of fit-for-purpose validation are set out in our guide on validating an IO biomarker in a clinical trial.

Regulatory status: what FDA and EMA say about ctDNA as an endpoint

FDA’s position is set out in the final guidance Use of Circulating Tumor DNA for Curative-Intent Solid Tumor Drug Development (November 2024; the May 2022 draft used “early-stage” in the title). It describes four uses:

• Patient selection by molecular alteration, with LoD studies and tumour confirmation of negative results where feasible.
• Enrichment using ctDNA MRD status, including escalation designs for MRD-positive patients and de-escalation for MRD-negative patients. FDA asks for randomised trials with DFS, EFS or OS as the primary endpoint, and assays with high sensitivity and NPV for de-escalation, or high specificity and PPV for escalation.
• Measure of response in early-phase trials for signal finding and dose optimisation, alongside imaging.
• Early endpoint: “not currently validated for use”. FDA says prospective randomised data and a trial-level meta-analysis (randomised trials only, patient-level data, prespecified timing and windows) are needed before ctDNA can serve as an endpoint reasonably likely to predict long-term outcome.

The guidance does not cover the metastatic setting, although FDA notes some principles may apply, and it points to its separate December 2020 guidance on MRD in haematological malignancies. An investigational ctDNA test used for enrolment or treatment decisions falls under investigational device exemption rules, so a study risk determination should be part of early planning. The May 2026 atezolizumab approval shows the selection route working in practice: DFS and OS supported approval, and the MRD test is the companion diagnostic.

In Europe, we have not found EMA guidance specific to ctDNA as an endpoint. ctDNA endpoint questions currently go through scientific advice, and an MRD test used to select patients for an EU-authorised medicine is a companion diagnostic under the IVDR, with the notified body and EMA consultation that brings. Our IVDR companion diagnostics guide covers that pathway, and the FDA companion diagnostic approval guide covers the US one.

What MRD and ctDNA vendors need to show pharma

Sponsors now choose MRD partners against a regulatory benchmark, not just a sensitivity claim. Expect to be asked for:

• Analytical validation that matches the use: LoD, precision at the cut-off, and accuracy in clinical specimens, plus a strategy for validating personalised panels across variant types and numbers.
• A prespecified, locked cut-off and evidence that positive calls near the detection limit are real.
• Clinical validity data in the right setting: the indication, line of therapy and drug class the sponsor cares about, ideally from randomised trials with long-term outcomes.
• Quantitative output, not just positive or negative, so molecular response can be defined at several thresholds.
• CHIP handling and trackability rates: the share of patients with a usable baseline or a successfully built panel.
• Operational proof: tissue-to-panel turnaround, global sample logistics, failure rates and data delivery fast enough for adaptive decisions.
• A regulatory path: IDE and Q-Submission experience in the US, IVDR readiness in Europe, and willingness to co-develop a companion diagnostic. See our piece on structuring pharma and diagnostics co-development.
• Contribution to shared evidence, such as pooled ctMoniTR-style analyses and reference materials, which FDA explicitly encourages.

Liquid biopsy & MRD is a San Diego theme

These questions sit across both streams at ImmunoMark Summit San Diego, the immuno-oncology and biomarker conference in San Diego, on 10–11 February 2027. 300+ senior pharma, biotech and diagnostics leaders meet in two parallel streams: Next-Gen Immuno-Oncology, and Biomarkers & Companion Diagnostics. The format includes keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception. Sessions on the published agenda include:

• The Growing Role of Liquid Biopsy and Minimal Residual Disease Monitoring in Personalized Cancer Care
• Overcoming Resistance Mechanisms to Improve Long-Term Responses to Immunotherapy
• Clinical Utility of Predictive Biomarkers in Personalised Oncology
• Global Regulatory Strategies for Companion Diagnostics and Precision Medicine

For the wider trends shaping this space, see our analysis of immuno-oncology biomarker trends.

Register for ImmunoMark San Diego 2027

If your company wants to meet this audience directly, whether you run an MRD or liquid biopsy platform, a central laboratory or a companion diagnostics business, see the San Diego 2027 sponsorship opportunities. You can also read why the summit works as a liquid biopsy and biomarker technology conference for sponsors, and how it serves as a companion diagnostics conference in San Diego.

FAQs

What is MRD testing in solid tumours?

MRD testing detects molecular residual disease: tumour DNA in the blood after surgery, radiotherapy or (neo)adjuvant therapy, in a patient with no visible disease on imaging. A positive result signals a high risk of relapse. In immunotherapy trials it is used to enrich for high-risk patients, to stratify randomisation, and since May 2026 to select patients for adjuvant atezolizumab in muscle-invasive bladder cancer.

What is the difference between tumour-informed and tumour-agnostic MRD assays?

Tumour-informed assays sequence the patient’s tumour and track a personalised set of variants in plasma, which generally gives higher specificity but needs tissue and time to build. Tumour-agnostic assays use a fixed panel or genome-wide signals such as methylation, so they are faster and need no tissue, but some patients are not trackable and clonal haematopoiesis must be filtered out.

Does early ctDNA response predict outcome on checkpoint inhibitors?

Patient-level evidence says yes. In the BR.36 trial, ctDNA clearance by the third cycle of pembrolizumab was associated with much longer overall survival in advanced NSCLC, and a pooled analysis of 918 patients from four randomised trials found ctDNA reductions at up to 7 weeks and at 7 to 13 weeks associated with overall survival across thresholds. Trial-level validation as a surrogate endpoint is still needed.

Can ctDNA be used as a primary endpoint for FDA approval?

Not yet. FDA’s November 2024 guidance says ctDNA is not currently validated as an early endpoint to support approval and calls for prospective randomised trials and a trial-level meta-analysis. FDA does support ctDNA for patient selection, MRD-based enrichment and signal finding, and ctDNA-guided trials such as IMvigor011 have supported approval using disease-free and overall survival.