The use of minimal residual disease in thoracic oncology: Gaps between promises and the on‐the‐ground reality of daily practice

The assessment of minimal residual disease (MRD) from blood samples of patients with resected non‐small cell lung carcinoma (NSCLC) is promising and opens up many opportunities for the optimisation of patient care in daily practice. Notably, this includes the potential for escalation or de‐escalation of adjuvant therapies. Thus, the evaluation of MRD status can directly contribute to an increase in the overall survival of early stage NSCLC patients and/or limit therapeutic but also “financial” toxicity. Therefore, several clinical trials recently evaluated MRD in early stage NSCLC by integrating and retrospectively comparing the results of MRD assessments. In this context, there is an urgent need to close the gap between clinical research and the use of the evaluation of MRD in routine daily practice. Further action needs to be taken, particularly in evaluating the pertinence of the detection of MRD in prospective interventional clinical studies. This may be done in part by comparing different parameters, such as the techniques used, the different time points and the cutoffs of MRD assessments. This article investigates the assessment of MRD in non‐small cell lung cancers, with a special focus on the issues associated with the various assays and the limitations of using circulating free DNA analyses for MRD assessment in early stage lung cancer. Recommendations and tips for the optimisation of MRD evaluation in non‐small cell lung cancers are provided.

4][5][6] MRD evaluation is mostly based on the detection of somatic genomic alterations in cf-DNA, which can therefore be considered an indicator of MRD.A considerable amount of research effort has been devoted to clinical trials evaluating the use of this biomarker for resected NSCLC patients, and this work continues.
However, whether we are ready to use MRD to improve the therapeutic strategies in the treatment of early stage NSCLC patients in routine clinical practice remains unclear.
The purpose of this review is to describe the current approaches to MRD assessment for NSCLC and the associated challenges, focusing mainly on cf-DNA, but also including other approaches, to close the gap between clinical research and daily practice.

| IDENTIFIC ATION OF MINIMAL RE S IDUAL D IS E A S E IN THOR ACI C ON COLOGY: IMPAC T ON FUTURE CLINI C AL PR AC TI CE
The clinical utility of cf-DNA for the evaluation of MRD has been investigated in many studies, which show a significant association between circulating tumour-DNA (ct-DNA) and decreased recurrence-free survival. 7Post-radical treatment, the analysis of cf-DNA has the potential to become a highly valuable tool to guide treatment and surveillance.Thus, cf-DNA analysis can be used to stratify patients into groups with low and high risks of recurrence after treatment with curative intent, paving the way to personalised treatment and surveillance of NSCLC patients.Importantly, drawing frequent blood samples is a low-cost and risk-free approach to surveillance, in contrast to the high cost and radiation-related risks associated with frequent radiological examinations.However, studies into the evaluation of MRD use a variety of diagnostic assays and different designs, making it difficult to compare their results.Table 1 shows the key studies conducted on NSCLC patients who underwent curative-intent treatment and for whom different circulating free DNA technologies were used to detect MRD.The sensitivity in predicting disease relapse ranged from 36% to 100% when restricting the evaluation of MRD to the initial post-treatment endpoint, but improved to between 80% and 100% when considering all posttreatment follow-up time points.
In 2017, Chaudhuri and colleagues published one of the first studies on the subject. 8cf-DNA was detected at the first time point (i.e.within 4 months of treatment completion) for 17 of the 32 patients; these patients had shorter periods of freedom from progression and lower rates of disease-specific survival than those with negative cf-DNA at the same time point.All 17 positive patients died of lung cancer.Posttreatment monitoring of cf-DNA at several time points showed 100% sensitivity and 100% specificity for the prediction of recurrence. 8The prospective TRACERx study of the evolution of lung cancer by Abbosh and colleagues included 24 patients. 9All patients had resectable stage IA-IIIB tumours, half of whom received adjuvant chemotherapy following surgical treatment.Fourteen patients relapsed, and 13 of the 14 patients had detectable cf-DNA before or at the time of clinical relapse. 9Another interesting study by Chen and colleagues evaluated 25 patients during a median follow-up of more than 500 days; 9 patients developed disease recurrence, 4 of whom had detectable cf-DNA 3 days post-surgery, whereas 14 out of 16 patients who did not relapse had negative cf-DNA at this same time point. 10Patients with detectable cf-DNA either 3 or 30 days after surgery had shorter overall survival than those with negative circulating tumour DNA at the same post-operative time points.These findings suggest that post-operative detection of cf-DNA as early as 3 days after surgery can be used to stratify patients on the basis of the risk of relapse and prognosis for survival.Although these three studies used a limited panel for targeted deep sequencing to detect cf-DNA, the study by Zviran and colleagues aimed to investigate whether whole genome sequencing of cf-DNA would enable MRD detection in patients with early stage lung cancer. 11asma samples were collected before and after surgery from 22 patients with stage I-III lung cancer.Twelve of the patients had negative cf-DNA following surgery, and none relapsed.Patients with detectable cf-DNA had lower survival rates compared with patients with undetectable cf-DNA. 11This suggests that the tumour informed plasma whole genome sequencing approach could be used as a prognostic cf-DNA MRD biomarker for early stage lung cancer.

Subsequent clinical trials have assessed whether detectable
postoperative cf-DNA MRD can be used to stratify patients on the basis of risk of relapse and be prognostic for survival, and this TA B L E 1 Detection of minimal residual disease (MRD) in patients with non-small cell lung carcinoma: performance of the assays used for cf-DNA assessment.research is ongoing.The MERMAID 1 and 2 studies included early stage lung cancers that were wild-type for EGFR and ALK. 12,13MRD status was evaluated post-surgery, and stratification of the disease according to adjuvant immunotherapy was based on three parameters: (i) stage of disease, (ii) PD-L1 expression, and (iii) MRD status post-surgery.

Clinical
Systematically obtaining blood samples before surgery can provide additional insight. 14Currently, cf-DNA is not always detectable pre-treatment; however, it has been shown that the detection of cf-DNA before curative-intent treatment is associated with lower survival rates and could be a biomarker for micrometastatic disease and thus aid in the selection of neoadjuvant treatment.

| CURRENT ISSUE S WHEN US ING E VALUATION OF MINIMAL RE S IDUAL D IS E A S E IN RE AL LIFE
Despite the increasing interest in using MRD in early stage NSCLC, many studies have highlighted the challenges involved in implementing the various MRD assays in routine clinical practice. 3,8,15ese challenges are described below, and the issues are listed in Table 2.

| Techniques for MRD detection
Currently, one of the biggest challenges in this field is choosing the most suitable assay with which to detect the molecular MRD of lung cancer patients with a high sensitivity and a high specificity. 16veral assays can be used, and they are briefly described here.A fixed panel assay that tracks driver mutations and targetable gene alterations can be used with a single plasma sample.Alternatively, a personalised panel, requiring previous tumour DNA sequencing, can be used. 8,9CAPP-seq is a typical representative technology for a tumour-naïve assay; it employs a fixed panel to detect cf-DNA in more than 95% of tumours via a non-invasive method.
The "lung cancer likelihood in plasma" (Lung-CLiP) approach is based on CAPP-seq, but the detection sensitivity is enhanced by adding unique molecular identifiers, increasing the sequencing depth and refining the algorithm. 17Lung-CLiP uses an extensive panel of 255 genes.However, the baseline sensitivity for stages I, II, and III was found to be 41%, 54% and 67%, respectively, with a 98% specificity. 17Tumour-informed assays have been developed to improve sensitivity by customising personalised panels paired with tumour tissues.The Signatera test developed by Natera, which was employed in the TRACERx study, used tumour tissue samples for whole exome sequencing (WES) at a depth of 100,000×; customised, patient-specific personalised panels enabled the number of mutations to be tracked. 9However, the positive rate for early stage lung cancer obtained using this method was found to be quite low, with only 19% for lung adenocarcinoma. 9The sensitivity of cf-DNA testing was increased by adding more tracer variations, leading to a positive detection rate of 49% for early stage lung adenocarcinoma and a detection limit as low as 0.003% of mutated DNA.Another method, MRDetect, attempted to improve the sensitivity by increasing the sequencing breadth. 11This assay used a fixed panel, WGS, and single nucleotide variant/copy number alteration models to back-calculate the cf-DNA abundance and 35× sequencing depth to reach a detection limit of 10 −5 .The sensitivity for this approach was shown to be 67%, with 96% specificity, for a cohort of lung cancer patients, 78% of whom were at stage I-IIA. 11However, this method cannot be used to obtain comprehensive information on variants.Another method, called "phased variant enrichment and detection sequencing" (PhasED-Seq) was demonstrated to have a 100% positivity rate in the detection of relapse before imaging in patients with a stage III lung cancer who received radiotherapy, while CAPP-seq could not detect any cf-DNA. 18However, the application of this method needs to be validated in further prospective studies. 18Taken together, the large discrepancies in the cohorts and numbers of patients recruited in these studies made it impossible to directly compare the testing methodologies and the performance of the assays used.

| The low sensitivity of current assays
One major challenge in implementing the technology required for MRD assessment, particularly in routine practice, is the low sensitivity of most of the currently available assays, which give false-negative results.In general, the sensitivity of the different assay technologies seems to be insufficient for use in daily practice.Moreover, the great majority of these tests are not commercially available and/or require the use of a centralised platform, meaning that the sample analysis process must be outsourced.

| The risk of false positivity
One challenge in the assessment of residual molecular disease includes the risk of occasional false-positive results due to the clonal haematopoiesis background in certain deep sequencing approaches.
As well as ageing, certain diseases that are often observed in lung cancer patients, such as COPD, are associated with clonal haematopoiesis. 19,20It is noteworthy that total cf-DNA increases progressively with patient age, independent of the presence of cancer, and this should be considered during the assessment of MRD assays. 21

| Study design and missing training and validation sets
Different study designs have been used in the clinical trials of MRD, most notably in terms of the technology used for sequencing and the time points at which the blood samples are obtained.Comparative evaluation of their efficiency is therefore not possible.

| The cutoff for circulating tumour DNA
Due to the different technologies used, as well as variations in the time point after surgery at which blood samples are obtained, the level of cf-DNA that may be associated with tumour progression, and thus the prescription of an adjuvant therapy (or even treatment escalation or de-escalation) has not yet been determined.Multicenter and prospective studies must be performed in the near future to identify cf-DNA cutoffs.

| The detection time: when to take blood samples
There is no standard testing time for postoperative MRD in NSCLC.
Depending on the assay, blood samples are obtained at different time points.It is therefore impossible to compare the sensitivity of these assays at the same time point.An elevated level of cf-DNA due to surgical trauma dilutes the cf-DNA in the plasma, reducing the sensitivity of the assay and leading to false-negative results.cf-DNA levels increase within 1 week following surgery; 22 thus, absolute MRD levels will be insufficient if assessed immediately after surgery.To enable early clinical decision-making, MRD should be assessed 2 weeks post-surgery.In general, longitudinal cf-DNA levels should be dynamically monitored preoperatively, postoperatively and after adjuvant therapy to improve sensitivity and specificity.

| Mastering the pre-analytical phase
Considering the sophisticated biotechnologies used for MRD assessment, most notably the deep sequencing approaches, mastering the different steps of the pre-analytical phase is a pivotal challenge for the future use of this biomarker in daily practice. 23,24e-analytical factors strongly impact the accuracy of the cf-DNA analysis.While the buffy-coat layer of blood is rich in genomic DNA from peripheral mononuclear blood cells, blood plasma con- How can the best technology among the currently available assays be selected?
How can the sensitivity of the currently available blood assays be increased?
How can false-positive results be identified?
How can the different MRD studies be compared, and how can the data be standardised and controlled for using validation sets?
How can the cutoff values of cf-DNA be established, and how can these cutoffs be compared for each assay?
How can the different blood sampling time points be established?
How can the pre-analytical phase be mastered using established quality control parameters?
What is the best strategy for a given institution-outsourcing or in-house MRD testing?
How can the cost of MRD monitoring be integrated into the health care budget system?
How can MRD testing be developed within the public health care reimbursement program?
How can the implementation of the in vitro diagnostic regulations be best anticipated for the different MRD assays?
TA B L E 3 Solutions for closing the gap between clinical trials and daily clinical use of minimal residual disease (MRD) for non-small cell lung carcinoma.
Increase the sensitivity of future assays with the development of new technologies

Set up a common algorithm for MRD assessment according to international guidelines
Standardise the pre-analytical steps for good practices (from venule puncture to sequencing approach) Establish external quality assessment and accreditation processes (ISO 15189/ISO 17020 norms) Combine multiple circulating blood biomarkers (cfDNA and circulating tumour cells and/or exosomes and/or plasma proteins) Use different body fluids (cerebrospinal fluid, pleural effusion, etc.) in association with blood samples for MRD detection according to the clinical presentation Integrate MRD detection for neoadjuvant and/or adjuvant treatments

Disseminate the usefulness of MRD detection to different stakeholders
Develop in-house testing as an alternative to outsourced testing according to the organisation of the institution and budget downstream DNA more difficult and the quantitation of the variant allele frequency (VAF) much less accurate.

| RECOMMENDATI ON S TO CLOS E THE G AP B E T WEEN THE US E OF MINIMAL RE S IDUAL D IS E A S E IN CLINI C AL RE S E ARCH AND DAILY ROUTINE APPLI C ATI ON
Various approaches may help to integrate the knowledge acquired from clinical research and trials into MRD assessment in routine clinical practice in the near future.Several potential solutions are listed in Table 3.

| Improve test sensitivity
Currently, the most pressing need in MRD detection is to improve the sensitivity of the assays.Sensitivity can be improved by increasing the number of mutations and signal abundance.tions and methylation, 25,26 (ii) the fragmentation approach, which detects the size of the cf-DNA, 27 (iii) an approach that combines variant calling with analysis of cf-DNA fragment lengths, 28 and (iv) an approach that combines analyses of cf-DNA mutations and protein biomarkers, such as those used in the CancerSEEK method. 29,30

| Establish a common algorithm in daily practice
One of the challenges in incorporating MRD into routine clinical practice is the establishment of a common algorithm for cf-DNA MRD evaluation that is independent of the therapeutic strategy being used.Preoperative cf-DNA testing must be systematically performed.According to Henriksen and colleagues, detection of MRD should be performed at 2 weeks and 5 weeks after surgery and should be followed up every 3 months to monitor relapse dynamically. 31Of course, elevated cf-DNA levels alone cannot identify recurrence or metastasis, and simultaneous imaging must be performed.It may seem easy, in theory, to proceed to adjuvant therapy only in patients who are positive for MRD.However, setting up a common algorithm to homogenise best practices in daily practice is challenging-most studies have their own protocols and time points for blood sampling, and there are a large number of neoadjuvant and/or adjuvant therapies in the pipeline.So, since blood MRD is a very good biomarker for overall survival, it should be used for escalation or de-escalation treatments in early stage lung cancer.Thus, a switch to a more aggressive therapeutic approach in cases of a positive cf-DNA MRD could lead to an improvement in the overall survival of early stage lung cancer patients in the near future.However, the effects of adjuvant therapy on cf-DNA and cf-DNA release into the blood are unclear and may vary according to the type of therapy.

| Increase the number of prospective interventional clinical trials
Comparing the results of clinical trials that included the evaluation of MRD retrospectively, as an ancillary test or a secondary objective, is quite difficult.Interventional clinical trials should be conducted at different centres to standardise the different assays and to confirm the usefulness of MRD detection to improve the overall survival of resected lung cancer patients. 32

| Standardisation of good practices in mastering the pre-analytical steps
To efficiently control the pre-analytical phase and obtain highquality molecular results, it would be beneficial to standardise the manner in which common procedures such as venipuncture, centrifugation, plasma collection, nucleic acid extraction and sample storage are conducted. 33One major issue is the turnaround time from venipuncture to centrifugation, which is critical for blood preservation and DNA release from the circulating haematological cells.As such, a growing number of reagents and blood collection tubes are available for the post-collection preservation of cf-DNA profiles of whole blood. 33The pre-analytical workflow for the analysis of cf-DNA includes several steps, from venipuncture to the preparation of the isolated cf-DNA for downstream analysis.Blood collection and preservation, sample storage and transport, the time that elapses between specimen collection and processing for plasma preparation, plasma storage and/or transport conditions, cf-DNA extraction and storage must all be taken into account.Each step of the preanalytical workflow can affect the analytical outcome, and different approaches can lead to contradictory results.In particular, the dilution of ct-DNA with genomic DNA released from apoptotic and/or lysed leukocytes after blood sampling constitutes one of the biggest challenges in the analysis of cf-DNA.In the event that cf-DNA stabilisation collection tubes are not available, EDTA tubes are preferred.
Relative to citrate and heparin, the use of as an anticoagulant results in a more moderate release of genomic DNA within the first 24 h after blood collection.Most researchers recommend processing the plasma from EDTA-anticoagulated blood within 1-6 h of collection.The consensus is that plasma is preferred over serum for the purification of cf-DNA because of the ex vivo release of DNA during the clotting process.Where logistics do not allow for strict control over the storage time or conditions between venipuncture and plasma preparation, the use of a dedicated blood collection tube with a standard cf-DNA reagent, such as Cell-Free DNA BCT (Streck Tube) is essential. 33

| External quality assessment and accreditation
One major issue in the use of MRD assays in daily practice relates to compliance with legislation and anticipation of the future of the in vitro diagnostic regulations (IVDR).In this context, it is highly recommended that external quality assessments are conducted for projects aiming to use cf-DNA in routine clinical practice. 34Such assays would need to be performed according to specific procedures established in an accredited laboratory.

| Integration of multiple circulating blood biomarkers into MRD assessment
[40][41][42] However, as there are multiple approaches to CTC detection and the risk of false-negative and false-positive results varies depending on the method used, it is currently difficult to integrate this biomarker into daily practice. 39,43Additionally, other circulating blood components, such as exosomes and microRNA, have been evaluated as potential biomarkers of MRD in early NSCLC. 35,36ch of these circulating biomarkers has advantages and limitations and has the potential to improve MRD detection performance in NSCLC. 44,45

| Usefulness of non-blood biofluids
Body fluids other than blood have been evaluated for their potential usefulness in MRD assessment in NSCLC. 46The sensitivity of MRD detected in the plasma of patients with brain metastases only is quite low due to the presence of the blood-brain barrier, and cerebrospinal fluid (CSF) may provide a good opportunity for MRD detection in such cases.However, MRD testing with CSF is not routinely performed for NSCLC patients, and specific protocols need to be established.

| Neoadjuvant therapies and MRD
In addition to evaluating MRD to assess the efficacy of adjuvant therapy, MRD assays can also be used to determine the effectiveness of neoadjuvant therapy (with or without associated adjuvant therapy). 47

| Establishment of policies for cost effectiveness and reimbursement
The development of a non-invasive blood test to monitor patients with NSCLC after surgical resection will require multiple analyses at different time points, including blood sampling before and after surgery, and repeated sampling during follow-up, regardless of whether a patient is receiving adjuvant therapy.The rapid increase in the number of molecular tests being performed in laboratories will require budgeting for reagents, sequencing and salaries.The use of different molecular tests for personalised medicine, particularly NGS tests, raises the question of the associated costs and the need for reimbursement to maintain the sustainability of MRD detection-especially with regards to achieving equality between different countries-and to provide all institutions with access to these tests. 48It is crucial that policies are adopted at a national level to enable the widespread adoption of these tests for the benefit of all patients.

| CON CLUS ION
Patients with a localised NSCLC may be considered for upfront surgical treatment.However, the overall 5-year survival rate is around 60%.To improve survival rates, the use of adjuvant chemotherapy (ACT) has been explored in depth, but analyses have shown an overall improvement in survival at 5 years of less than 7%.The evaluation of the risk of recurrence and subsequent need for ACT is mainly based on the tumour stage (according to the TNM classification system), but more than 25% of patients with stage IA/B tumours will relapse.Recently, adjuvant targeted therapy has been approved for EGFR mutated resected NSCLC and trials are being carried out to evaluate other targeted therapies and immunotherapies in adjuvant settings. 1However, due to the high costs, prolonged treatment duration, emergence of resistant clones and significant side effects, there is an urgent need to better select patients for this treatment approach.In this context, it is essential to establish and validate prognostic and theranostic markers that enable better stratification of patients, to identify those who are most likely to benefit from adjuvant therapies.Molecular MRD in blood samples from these patients is a promising biomarker that can be used to monitor and evaluate neoadjuvant and adjuvant therapies. 1,5,49,50e gaps discussed here currently act as "bottlenecks" in the movement towards the everyday use of MRD as a tool for the escalation or de-escalation of systemic therapies in the early stages of NSCLC; however, future advances should help to fill these gaps (Figure 1).As described above, the detection of cfDNA, the most studied biomarker for MRD, remains challenging in NSCLC.
The sensitivity of the available assays is low, resulting in an undetermined status when the post-surgery cfDNA is negative.True MRD-negative patients have longer progression-free survival and overall survival, and patients with longitudinal undetected MRD at 18 months may be cured. 51 relatively low quantities of extracellular DNA.Cf-DNA arises from dying cells from the entire body, including healthy cells and tumour cells dying from apoptosis, necrosis and immune cytotoxicity.Thus, only a small fraction of the cf-DNA is comprised of tumour-derived ct-DNA.Moreover, the cf-DNA in blood may be damaged during sample collection, transport, temperature variations, delay before centrifugation, centrifugation and storage, resulting in modified nucleosides that are incorrectly recognised by DNA polymerases during polymerase chain reaction (PCR) amplification.This leads to amplicon DNA sequences with variants that may be interpreted as cancer-specific mutations or oxidation, the two most commonly observed types of DNA damage.In general, contamination of cf-DNA is linked to genomic DNA from leukocytes.Contamination of cf-DNA with leukocyte genomic DNA dilutes the fraction of cf-DNA that contains useful information (corresponding to ct-DNA), rendering the mutation analysis of TA B L E 2 Questions and potential issues for the use of blood samples for molecular/minimal residual disease assessment (MRD) in daily practice for non-small cell lung carcinoma.
Fixed panels that do not need to be personalised are much easier to extend into clinical practice.Molecular biology forms the foundation of several measurement technologies, and an increase in throughput has resulted in the development of omic analyses.Huge advances in the development of parallel sequencing technologies have enabled access to fundamental molecular data and revealed genomic and transcriptomic signatures.Therefore, multi-omic technology can be integrated by adding signal-rich biological dimensions.New technologies based largely on ultra-deep sequencing should be able to improve the sensitivity of MRD detection in the near future, especially for the quantitation of very rare somatic mutations.Different approaches are being developed, including (i) the detection of combinations of cf-DNA muta- Thus, MRD offers great opportunities for new therapeutic strategies for early stage NSCLC, helping to avoid unnecessary toxicity as well as unnecessary costly treatments.MRD testing could allow patients with a high risk of recurrence to be selected and may therefore improve overall survival, and it should enable the quantification of cf-DNA and the identification of somatic alterations for targeted therapy.In the near future, the major strength of MRD detection in blood samples would be the ability to adapt the therapeutics according to the presence and level of cf-DNA, leading to the escalation or de-escalation of different therapeutic strategies.Studies of MRD detection in NSCLC have so far been modest in size and mostly retrospective.As such, a major weakness in the application of LB for MRD assessment is the lack of prospective clinical validation with large cohorts of patients.Thus, MRD is not yet ready for use as a predictive biomarker to guide treatment strategies in the routine care of localised-stage NSCLC patients who are not involved in clinical trials.Whether treatment personalisation according to cf-DNA MRD detection status will translate into improved patient outcomes remains to be seen.More studies with larger patient populations and multiple early post-treatment time points are urgently needed.The real impact of the detection of MRD on patients' overall survival in a prospective setting still needs to be demonstrated.