FormalPara Key Points

Custom NGS, MALDI–TOF mass spectrometry and ddPCR assays provide concordant and specific detection of ctDNA in melanoma patient blood plasma samples.

A high level of dynamic mutational complexity was identified in longitudinal blood plasma samples, including multiple BRAF mutations and concurrent BRAF and NRAS mutations in the same patient.

Undetectable ctDNA in one or more blood samples taken during the first five therapy cycles, and ctDNA level in the last measured plasma sample for each patient, had a significant association with clinical outcome.

1 Introduction

Melanoma is genomically complex [1] and progresses through complex cooperative evolutionary mechanisms [2]. It has a relatively large number of mutations relative to other tumour types [3], especially in melanoma tumours with a sun exposure aetiology. In some patients, these mutations can be detected in the blood as circulating tumour DNA (ctDNA), which can provide a convenient window into the evolution of tumour mutations over time [4]. Plasma ctDNA levels are claimed to have prognostic utility in melanoma, including a significant association between the baseline detection of ctDNA in pre-treatment plasma samples and lower progression-free patient survival [5], as well as an association with responses to both targeted therapy and immunotherapy [6,7,8,9,10]. A longitudinal reduction in plasma ctDNA levels was associated with significantly improved survival of immunotherapy-treated patients in a recent pooled analysis of five clinical trials [11]. More specifically, clearance of BRAF V600E ctDNA from plasma in the first month following T-cell transfer immunotherapy directed against BRAF V600E has been associated with an enduring treatment response [12]. Plasma ctDNA analysis can also identify tumour sub-clonality and mutations associated with drug resistance [13]. Some studies have suggested that plasma ctDNA levels can reflect therapeutic response or progression in real time [6, 14,15,16,17,18], even prior to radiological imaging [15, 19,20,21,22,23,24]. However, a recent prospective study found that elevated ctDNA levels indicated progression earlier than imaging in only 2 out of 16 patients [16]. Nevertheless, ctDNA analysis appears, at the very least, to be emerging as a useful adjunct to clinical imaging [13, 20].

Several ctDNA technologies have been developed to track tumour mutations in plasma. Two of the most commonly utilised are droplet digital polymerase chain reaction (ddPCR) [25,26,27,28,29] and next-generation sequencing (NGS) [30,31,32,33]. ddPCR is highly sensitive, specific and cost-effective, but only allows for a small number of defined mutations to be investigated simultaneously. In contrast, NGS allows a considerable number of many genomic regions to be interrogated concurrently and can detect both known and unknown mutations. Although, historically, the sensitivity of NGS was limited by factors such as polymerase error rate and low depth of coverage due to cost, advances in sequencing and bioinformatic technologies [34] including unique molecular tags/identifiers (UMT) now provide NGS with comparable or greater sensitivity to ddPCR [35]. A third methodology, matrix-assisted laser desorption/ionisation–time-of-flight (MALDI–TOF) mass spectrometry, can identify moderate numbers of specific mutations with similar sensitivity and cost to NGS [36,37,38]. Using these methodologies, it has been reported that the concentration of released ctDNA positively correlates with tumour size [39]. Assay sensitivity can, however, be impacted by delays to sample processing [40, 41] for both NGS [42] and ddPCR [43, 44]. Cell-stabilising blood collection tubes allow delayed analysis of ctDNA collected from patients living in geographically remote locations without loss of sensitivity [40].

In New Zealand (NZ), where this study is based, people from geographically remote regions [45], as well as indigenous Māori [46, 47] and Pacific people [48], suffer significant inequities in cancer care [49]. In 2020, people in NZ also suffered the highest age-adjusted mortality from metastatic melanoma in the world (5 per 100,000 person-years) [50, 51]. Although the incidence of melanoma in NZ is lower in Māori and Pacific People than in European New Zealanders, Māori and Pacific People on average experience poorer outcomes once diagnosed [52]. This is in keeping with patterns reported among Indigenous people in other parts of the world [53, 54]. The potential to either reduce or perpetuate these cancer inequities in NZ is important to consider when assessing the local use of new technologies such as ctDNA.

In this observational cohort study, we set out to identify the breadth and complexity of tumour genomic information that ctDNA analysis can reliably report. This included analysis of the consistency of information generated by different ctDNA analysis technologies (ddPCR, NGS and MALDI–TOF mass spectrometry), factors that may contribute to variability in mutation detection and the effect on assay sensitivity of altering blood collection protocols to suit patients living in geographically remote locations.

2 Materials and Methods

2.1 Ethical Approval, Study Design, Patient Recruitment and Clinical Follow-up

Twenty-nine patients with stage IV metastatic cutaneous melanoma commencing pembrolizumab or nivolumab treatment in Auckland, NZ were enrolled on the study between October 2017 and December 2019 under the NZ Health and Disability Ethics Committee ethical approval 16/NTA/180. A further 16 patients with stage III/IV melanoma that underwent full surgical resection were recruited as above, but data for only 2 were included in this manuscript. Study numbers were dictated by the number of patients who could be recruited during the study time period. Eligible patients were either patients with stage III melanoma , or patients with stage IV melanoma who met the funded immunotherapy treatment criteria with radiologically measurable disease. All donors were at least 18 years of age and provided written informed consent. Patients with an unknown primary melanoma were included, but patients who had been diagnosed with a second malignancy in the previous 5 years were excluded. Clinical characteristics of the patients with stage IV melanoma undergoing immunotherapy are presented in Supplementary Table 1. Computed tomography (CT) imaging was conducted at approximately 3-month intervals, in keeping with local clinical practice. Clinical outcomes were analysed retrospectively and included duration of treatment and overall survival. Using RECIST version 1.1 criteria [55], patients were classified according to best tumour response achieved during the course of therapy, and their disease status at the time of final blood sample and overall survival. Presence or absence of durable clinical benefit (DCB) was defined as partial or stable response lasting > or ≤ 6 months, respectively, as previously described [56].

2.2 Sample Collection and Processing

Peripheral venous blood samples were collected from pre-surgical and subsequent clinical appointments for each surgical patient, and prior to each cycle of pembrolizumab/nivolumab therapy (cycles 1–5) and then 3-monthly thereafter for immunotherapy patients. Blood samples were collected into either: (i) three 10 mL K2-EDTA vacutainer tubes, with plasma prepared within 4 h of venesection, or (ii) three 8 mL Cell-Free DNA Collection Tubes (Roche) left at room temperature for 7 days before plasma preparation. In addition, one 10 mL clot activator tube was collected for each timepoint and processed within 4 h of venesection for serum. To isolate plasma and serum, blood tubes were centrifuged at 1500g for 10 min at room temperature. For plasma, supernatant from all three K2-EDTA or Cell-Free DNA tubes were pooled, re-centrifuged at 4000g for 10 min and the supernatant plasma was stored at −80 °C in 2 mL aliquots. A total of 0.25 mL serum was used for lactate dehydrogenase (LDH) analysis conducted by the Auckland City Hospital LabPLUS clinical laboratory.

2.3 Preparation of Germline Genomic DNA and Cell-Free DNA

Cell-free DNA (cfDNA) was isolated from 5 mL plasma using the QIAamp Circulating Nucleic Acid kit (Qiagen), as per the manufacturer’s protocol. Leucocyte buffy coats collected after the 1500g K2-EDTA vacutainer centrifugation were used to prepare germline genomic DNA (gDNA) using the QIAamp DNA Blood Mini Kit (Qiagen) as per the manufacturer’s instructions. Fresh frozen tumour tissue samples were available from four patients, from which tumour gDNA was isolated using a NucleoSpin Tissue Kit (Machery Nagel). Both gDNA and cfDNA were eluted in ultrapure DNase/RNase-free water (Invitrogen). DNA concentration was determined using a Qubit Fluorometer (dsDNA HS or dsDNA BR Assay Kits, ThermoFisher Scientific, MA, USA).

2.4 NGS Analysis

NGS analysis was performed to identify gene mutations present in the cfDNA isolated from plasma, using an Ion AmpliSeq HD custom melanoma panel that contained 115 amplicons, covering regions of 41 genes commonly mutated in melanoma (Supplementary Table 2). Two pools with up to 10 ng of cfDNA or tumour gDNA in each pool were used to generate sequencing libraries. Amplicon primers containing UMTs were attached to each DNA molecule as the sample underwent an initial three cycles of PCR, and any remaining UMTs were enzymatically digested with the proprietary enzyme SUPA (ThermoFisher Scientific) which digests and removes any unincorporated single-stranded primers. A second universal PCR amplification (20 cycles) was conducted to generate the completed sequencing library. Proofreading polymerase was used for both PCR amplification steps of the library generation (ThermoFisher Scientific). Libraries were quantified using an Ion Library TaqMan Quantitation kit (ThermoFisher Scientific) and with an HS-D1000 TapeStation (Agilent, CA, USA). Libraries were templated on an Ion Chef instrument before three cfDNA libraries, or up to nine gDNA libraries, were pooled to a final concentration of 15–20 pM and sequenced using an Ion 318 v2 semi-conductor sequencing chip on a Personal Genome Machine (ThermoFisher Scientific) using standard protocols. Data was processed using Ion Reporter software version 5.18 (ThermoFisher Scientific). Following bioinformatic de-duplication, each mutation was reported as variant molecular tags (VMTs), and the mutational frequency defined as variant allele fraction (VAF) where the VMTs were calculated as a proportion of all unique molecular tags (UMTs). The sequence of the UMTs and the method of de-duplication is proprietary to ThermoFisher Scientific, but the estimated unique sequencing depth per cfDNA target hotspot, and how this varied across the gDNA libraries, is presented in Supplementary Table 3. Following NGS analysis, all variants were visually verified at the level of individual reads in BAM files using the Integrated Genome Viewer [57].

2.5 MALDI–TOF Mass Spectrometry Analysis

For a subset of samples, MassARRAY UltraSEEK MALDI–TOF mass spectrometry analysis was performed to further orthogonally validate the detection of specific ctDNAs. For each sample assessed, PCR amplification of up to 10 ng of cfDNA sample was analysed with the UltraSEEK MassARRAY melanoma V2 panel (Agena Biosciences [58]), following the manufacturer’s instructions. Amplified products were treated with shrimp alkaline phosphatase, followed by single-base PCR extension with biotinylated chain terminator nucleotides specific to the mutant allele. Streptavidin-coated magnetic beads were used to capture the single-base extended oligonucleotides and the products eluted by the addition of biotin competition solution. Eluted products were conditioned with 2 μL (2 mg) of anion exchange resin slurry. A MassARRAY RS1000 Nano-dispenser was used to transfer the analyte onto the SpectroCHIP, and the target spot size was 12 nL, with a range of 5–18 nL. The SpectroCHIP was transferred into the matrix-assisted laser desorption/ionisation time-of-flight mass spectrometer, and data was acquired using the MassARRAY analyser. For the UltraSEEK analysis, further data analysis was performed using the MassARRAY Typer software version 4.0.26.74 (Agena Biosciences). The normalised intensity of each spectrum was calculated by normalising the signal intensity of the mutant allele against the capture control peaks generated within each reaction. The normalised mutant allele values were reported with a Z-score. The recommended Z-score cut-off to report a mutation was set to 10.

2.6 ddPCR Analysis

Custom ddPCR assays were performed to orthogonally validate and quantify mutations identified by NGS using six custom ddPCR assays designed to detect DNA mutations encoding: BRAF V600E, BRAF V600K, BRAF K601E, NRAS Q61K, NRAS Q61R and KIT L576P (Supplementary Table 4). Additional ddPCR assays to detect TERT C228T and TERT C250T were used as previously described in Ref. [59]. ddPCR reactions were carried out with a Bio-Rad QX200 system, in a 20 μL reaction with 1× ddPCR Supermix for Probes (no dUTP) (Bio-Rad Laboratories), 8 μL cfDNA template, 209 nM HEX (wild-type) probe and 209 nM FAM (mutant) probe, with 626 nM each of 10 μM target specific forward and reverse primers (Integrated DNA Technologies). Annealing temperature optimisation was carried out for all six custom-designed ddPCR assays (two examples are shown in Supplementary Fig. 1), and the following optimal amplification conditions were identified and used in subsequent samples: one denaturation cycle at 95 °C for 10 min, 40 cycles of 94 °C for 30 s and 61.4 °C for 1 min (ramp rate 2 °C per second), followed by a final 98 °C for 10 min. Samples with > 10,000 droplets generated were analysed using QuantaSoft version 1.0.596 software. Droplet counts are presented as the number of positive droplets per mL of input plasma and a threshold of ≥ 2.5 droplets per mL of input plasma was used for reporting the presence of ctDNA in a sample. Specificity and sensitivity of the custom ddPCR assays were assessed with gBlock fragment containing the mutation of interest (Supplementary Figs. 2–5 and Supplementary Table 5).

2.7 Statistical Analysis

All statistical analyses were undertaken using the R programming and analysis environment [60]. The specific R packages used were: ggplot2 [61], survival [62], cowplot [63] and survminer [64]. Hypothesis testing included in this paper used non-parametric two-sided Wilcoxon rank-sum tests with continuity correction or chi-squared tests. Survival analyses used log-rank tests to test the null hypothesis that there is no difference in patient outcome between the two groups being compared, or Cox proportional-hazards regression models to test the null hypothesis that there is no dependency of survival time on continuous predictor variables. All survival analyses used ‘overall survival’ (i.e. the length of time from the analysis period start point during which patients were still alive). All Kaplan–Meier curves, log-rank tests and Cox proportional-hazards regression models employed landmark analysis (i.e. the analysis period start point was the time of the last blood sample which contributed the ctDNA information used in the particular analysis) to minimise ‘guarantee time bias’ effects.

Given the small study size, traditional methods such as receiver operating characteristic curve(s) (ROC) analysis were not sufficient when determining the threshold at which to report a ctDNA mutation. Therefore, for each ddPCR assay, the false positive rate (FPR) was determined separately by setting up three independent ddPCR reactions using only wild-type (WT) gBlock constructs as template. For each ddPCR assay, a high concentration of WT DNA template was used to generate between 2000 and 10,000 WT positive droplets per ddPCR reaction. For five ddPCR assays, no false-positive droplets were identified. For the remaining four ddPCR assays, between two and six false-positive droplets were detected. The FPR rate for each ddPCR assay was calculated by dividing the number of false-positive droplets by the total number of WT copies detected. The calculated FPR rate for all melanoma ddPCR assays was relatively low (< 0.03%). However, two FP droplets were generated in a single ddPCR reaction for both BRAF V600E and NRAS Q61K. If two FP droplets were generated in a patient cfDNA sample analysed by ddPCR assay, this would equate to detecting a mutation at approximately 2 copies/mL plasma. On the basis of these considerations and the results of analysis of duplicate isolates of the same samples and freeze–thaw sample pairs (see Sect. 3.7.1 below), a threshold of > 2.5 copies/mL plasma was chosen, below which the result would not be considered positive for the particular mutation. This is the same threshold used by Lee et al., who used ddPCR to quantitate ctDNA as a surrogate marker of response in patients with melanoma undergoing immunotherapy treatment [65].

Another consideration is the management of multiple ctDNA mutations at a single timepoint in a single patient when performing survival analysis. In this study, the ctDNA mutation with the highest concentration in plasma was used since: (i) for many of these patients multiple metastatic tumours may contribute to the measured ctDNA in plasma. As others have shown [66], and as we have recently found in a patient with over 30 metastatic tumours [67], ctDNA mutations with the highest concentrations in plasma commonly represent truncal mutations found in multiple metastases, which may potentially provide a more holistic survey of tumour burden than ctDNAs with lower plasma concentration (which may potentially represent new ‘private’ mutations evolving in single metastatic lesions), and (ii) conversely, when a single clinically important metastatic lesion is rapidly progressing, discordant to other more stable lesions, the measured ctDNA mutation with the highest concentration may be more representative of this rapidly growing lesion, since plasma ctDNA levels have been shown to correlate with metabolically active tumour burden measured by FDG positron emission tomography (PET)/CT [68], and (iii) ctDNA mutations with the highest concentrations in plasma in general are further above the noise threshold of the assay, and may therefore provide more reliable data.

3 Results

3.1 Longitudinal Plasma ctDNA Analysis

Peripheral venous blood samples were collected from 29 patients with stage IV melanoma immediately before each cycle of immune checkpoint inhibitor therapy. Patients received between 4 and 34 cycles of pembrolizumab/nivolumab while participating in this study and each donated between 4 and 12 blood samples. For each blood sample, cfDNA extracted from plasma was analysed using ddPCR alongside at least one other independent genomic technology. From these analyses ctDNA mutations were identified in 28/29 (97%) of the patients (Supplementary Table 6). When NGS was used to analyse participants’ cfDNA, the average raw sequencing depth of amplicons before deduplication by UMT was 13,015. De-duplication of UMT identified a mean UMT family size (the number of repeated sequencing reads of individual DNA molecules) of 16 with a range of 5.1–31.4. Once sequencing data had been de-duplicated by UMTs, the mean sequencing depth of the targeted regions in this panel was 1720 (range 209–3030, Supplementary Table 3).

3.2 Comparison of ctDNA Mutations with Mutations Previously Detected in Tumour Tissue

Clinical genomic analysis had previously identified BRAF mutations in the tumour tissue of 11 study patients (Supplementary Table 6). Analysis of ctDNA identified the same BRAF mutation in the blood of all 11 patients (100% concurrence). BRAF ctDNA mutations were also identified in the blood of an additional seven patients for whom clinical genomic analysis of tumour tissue had been conducted, but had not identified mutations in the BRAF gene (Supplementary Table 7), and in an additional four patients who had no previous testing of tumour tissue. All four patients with an NRAS mutation and one patient with a KIT mutation identified in previous clinical genomic analysis of tumour tissue had the corresponding mutations identified in their plasma.

3.3 Longitudinal Monitoring of ctDNA During Immunotherapy

A range of patterns of longitudinal ctDNA concentration over the course of immunotherapy were observed (Fig. 1). These included monotonic ctDNA increases or decreases as have been described previously [14], as well as more complex patterns of change, as have been previously identified [69,70,71]. A full description of the mutations identified, and the quantitation of these over time, is shown in Supplementary Table 8. For ten patients, we observed a clear pattern of decreasing or increasing ctDNA concentration correlated with response to immunotherapy or with progressive disease (examples shown in Fig. 1a and b, respectively). For 11 patients, fluctuating ctDNA concentrations were identified, trending downwards or upwards, consistent with treatment response (examples shown in Fig. 1c and d, respectively). However, two of the patients who demonstrated a downward trend of ctDNA that correlated with initial radiological assessment went on to develop intracranial metastases (MEL0022 and MEL0031). For seven patients who responded to immunotherapy or had stable disease, fluctuating low levels of ctDNA (ranging between undetectable to < 20 copies/mL plasma) were observed (example shown in Fig. 1e). Figure 1f illustrates a longitudinal pattern of BRAF V600E ctDNA detected by ddPCR, which is consistent with an initial response to immunotherapy followed by progression on therapy. In this case, the rising ctDNA profile appears to contradict a CT scan undertaken early in the period of rising ctDNA, but was concordant with a later CT scan undertaken after clinical progression.

Fig. 1
figure 1

Longitudinal analysis of ctDNA levels in six patients with melanoma with different responses to immunotherapy. a Patient MEL0033 demonstrated a decrease in ctDNA concentration was concordant with interval response to therapy confirmed radiologically, b patient MEL0037 demonstrated an increase in ctDNA concentration correlated with progression on therapy confirmed radiologically, c patient MEL0027 demonstrated fluctuating downward ctDNA concentrations that correlated with treatment response, d patient MEL0018 demonstrated an initial decrease in the detectable ctDNA for BRAF V600E and was accompanied by an increasing concentration of ctDNA for NRAS Q61K, e patient MEL0060 demonstrated continuous detection of low concentrations of ctDNA in a patient with clinically stable disease and f patient MEL0017 commenced immunotherapy treatment 8 days after the first collected blood sample. A decrease in ctDNA concentration was concordant with radiological imaging. An increasing level of ctDNA indicated treatment failure, which was confirmed by radiological imaging following patient health complications. Lactose dehydrogenase (LDH) was measured for samples P2–P6 and is plotted as a yellow line. All measurements were within the normal expected range shown with a grey line (<250 U/L). Note the different y-axis scales in the six panels of this figure

Full size image

3.4 Detection of Complex Tumour Heterogeneity Over Time

Concurrent mutations in both BRAF and NRAS genes were detected in ctDNA from 7 of 29 patients (24%). These were identified by ddPCR of ctDNA from at least two plasma samples per patient collected at different timepoints. Since this high prevalence of concurrent BRAF and NRAS mutations was unexpected (see Discussion), all ctDNA mutations detected by ddPCR were subsequently re-assessed, and their presence was confirmed in the same blood plasma sample using at least one other orthogonal technology (either NGS or UltraSEEK MassARRAY). Although six of these seven patients previously had genomic tumour testing performed, mutations had only been detected in the tumours of two of these patients. Additionally, for two of these patients, co-occurring BRAF and NRAS mutations were detected in the baseline plasma sample, but were not detected in the clinical analysis of the tumour tissue conducted two months prior to starting immunotherapy. The detection of concurrent ctDNA mutations is consistent with the presence of multiple genomic tumour cell clones within a patient, with examples shown in Supplementary Fig. 6.

In one patient (MEL0035), a BRAF V600E mutation was detectable in all eight plasma samples (P1–P8), with additional BRAF V600K and NRAS Q61R ctDNA mutations becoming detectable by ddPCR after the fourth immunotherapy cycle (analysis of their fifth plasma sample P5 is shown in Fig. 2). NGS quantified the variant allele frequency (VAF) of the BRAF V600E, BRAF V600K and NRAS Q61R ctDNA mutations P5 as 4.6%, 6.5% and 0.2%, respectively. Given the unexpected presence of an additional BRAF V600K mutation, which could have potentially evolved from a BRAF V600E-mutant tumour or reflect the selection of minor pre-existing sub-clones, ctDNA from the fifth (P5) plasma sample was reanalysed using a third orthogonal technology, UltraSEEK MassARRAY, which also confirmed the presence of all three mutations. These observations are consistent with the BRAF V600K and NRAS Q61R mutations emerging in sub-clones from the original BRAF V600E-mutant tumour.

Fig. 2
figure 2

Detection of co-occurring BRAF and NRAS gene mutations in the ctDNA from a single patient. Analysis of the fifth plasma sample from patient MEL0035 is shown. a An Integrated Genome Browser (IGV) image of NGS sequence reads (grey bars) shows three concurrent mutations in this sample: NRAS Q61R (c.182 T > C, 25,050 depth, 0.2% VAF), BRAF V600K (c.1798 AC > TT, 47,133 depth, 6.5% VAF) and BRAF V600E (c.1799A > T, 47,089 depth, 4.6% VAF). The sequencing results are supported by concordant ddPCR analysis (b) and UltraSEEK MassArray results (c)

Full size image

In three patients, no mutations were detected in either tumour tissue or ctDNA isolated from plasma, taken at the time of the first cycle of immunotherapy. For one patient (MEL0022), a BRAF V600E ctDNA mutation only became detectable after the fourth immunotherapy cycle (P5 plasma sample), even though NRAS Q61R was detected from the first pre-treatment sample onwards (Supplementary Fig. 6).

3.5 Tumour Heterogeneity in Patients with Stage III Melanoma

To assess whether this mutational complexity was restricted to immunotherapy-treated patients, two patients with stage III melanoma not receiving immunotherapy were also analysed. In these patients, similar degrees of ctDNA mutational heterogeneity was observed. In the first patient (MEL0026), ddPCR analysis of both a lymph node metastasis and plasma cfDNA sample identified concurrent TERT C228T, BRAF V600E and BRAF V600K mutations, which were confirmed by NGS analysis. The relative VAF of these three mutations were consistent with the presence of a BRAF V600E-mutant sub-clone arising within a BRAF V600K-mutant tumour (Supplementary Fig. 7). In the second patient (MEL0011), BRAF V600E, V600K and K601E mutations were identified in ctDNA with the BRAF V600K and BRAF K601E becoming more prevalent 18 months after surgery. ddPCR and NGS analysis of cfDNA from this patient were concordant, with the NGS analysis revealing the consistent presence of variants for two independent tumour sub-clones (Supplementary Fig. 8).

3.6 Association Between ctDNA and Clinical Outcome

While identifying associations between ctDNA level and immunotherapy response in patients with melanoma was not the core purpose of this study, some interesting associations were observed. We found that those patients who were categorised as having no durable clinical benefit (defined here as partial or stable response lasting > 6 months, following the classification of Rizvi et al. 2015 [56]) from their immunotherapy, or patients with progressive disease as their best RECIST response, had no significant differences in concentrations of ctDNA in their first (baseline) longitudinal blood sample compared with other patients (Supplementary Fig. 9a,b). However, when this analysis was repeated using the final (last) longitudinal blood sample from each patient, patients categorised as having no durable clinical benefit from immunotherapy or patients with progressive disease as their best RECIST response had significantly higher concentrations of ctDNA than other patients (P = 0.00003 and P = 0.0003, respectively, Supplementary Fig. 9c-d). Receiver-Operating Characteristic curves (ROCs) are shown for the baseline and last cfDNA samples collected from each patient in Supplementary Fig. 9e.

Previous studies have suggested that ctDNA becoming undetectable in at least one treatment timepoint is significantly associated with patient survival [72]. Results from our cohort of immunotherapy-treated patients with stage IV melanoma concorded with this observation; undetectable ctDNA in one or more blood samples taken at the start of the first five immunotherapy cycles had a significant association with overall survival (Cox proportional-hazards model P = 0.0004, Fig. 3). In line with this observation, undetectable ctDNA in one or more blood samples taken at the start of the first five immunotherapy cycles also had a significant association with durable clinical benefit from immunotherapy (chi-squared test P = 0.0009).

Fig. 3
figure 3

Association between undetectable ctDNA and patient outcome. Log-rank tests identified a significant positive association (Cox proportional-hazards model P = 0.0004) between overall survival and ctDNA being undetectable (< 2.5 copies/mL plasma) in one or more blood samples taken at the start of the first five immunotherapy cycles

Full size image

3.7 Assessment of Technical Parameters of ctDNA Analysis Relevant to Clinical Implementation

3.7.1 Consistency

Several pre-analytical and analytical variables can affect ctDNA analysis [56,57,58]. These include time to plasma processing, cfDNA extraction protocol and platform selected for ctDNA analysis. To assess some of these factors, first, cfDNA was extracted on two different days from 38 paired aliquots of plasma, each pair of aliquots originating from the same blood sample. Although there was an overall correlation between the concentration of cfDNA isolated in the paired preparations (Spearman’s rank ρ = 0.89, Supplementary Fig. 10), some variability in cfDNA concentrations was observed (mean coefficient of variation 13.4%, range 0.3–41.3%), which is concordant with previous reports [73].

Next, we assessed how freeze–thawing of cfDNA stocks might impact the final ctDNA concentration. For 65 cfDNA samples, the ctDNA concentration was determined after one and two freeze–thaw cycles, using custom ddPCR assays. The same ctDNA variant was detected in the replicate ddPCRs for 62/65 (95%) of samples, and the ctDNA concentrations following one and two freeze–thaw cycles showed very good concordance (Spearman’s rank ρ  = 0.96, Supplementary Fig. 11). For these 62 samples, the concentration of ctDNA was > 4 copies/mL plasma. For the remaining 3/65 cfDNA samples, the ctDNA concentration was < 4 copies/mL plasma, so the variability observed may be attributable to other factors, including random sampling. On the basis of this analysis, we set the threshold for reporting the presence of ctDNA mutations in blood plasma as > 2.5 copies/mL.

3.7.2 Assessment of the Concordance Between Genomic Platforms for ctDNA

As shown above, in general, the ctDNA mutations detected by ddPCR assays and NGS using the AmpliSeq melanoma panel were concordant (Supplementary Table 8). Therefore, we next assessed the relative sensitivity of these two technologies in more depth for a set of mutations commonly found in melanoma. NGS analysis was performed on 42 cfDNA samples collected from 18 patients with melanoma, and 62 mutations were identified. For these mutations, the cfDNA sample was reanalysed using one of seven different mutation-specific ddPCR assays matching the variants detected by NGS. All 62 ctDNA mutations were identified by both technologies. The ctDNA levels quantitated by ddPCR (copies/mL plasma) and NGS (VMT, the number of individual DNA molecules identified that carry each ctDNA mutation) correlated moderately well (Spearman’s rank ρ = 0.928; Supplementary Fig. 12). The UltraSEEK mass spectrometry method was used to assess a subset of samples (Supplementary Table 9) and this method in general provided concordant results with ddPCR (e.g. Fig 2 and Supplementary Table 6). However, the version of UltraSEEK used in the current study to detect BRAF V600E ctDNA mutations may be less sensitive than ddPCR, since it failed to detect a subset of BRAF V600E mutations that were detected by ddPCR. This lack of detection using UltraSEEK tended to occur in plasma samples when ddPCR indicated low ctDNA concentration (Supplementary Fig. 13).

3.7.3 Impact of Clonal Haematopoiesis on ctDNA Analysis

Where clinical genomic analysis of tumours had been performed and identical mutation(s) identified in tumour samples and in blood plasma, the tumours are clearly the most likely source of the mutation(s) detected in the plasma. However, when mutations were detected in the plasma of patients for whom there had been no genomic analysis of tumour tissue, or no mutations were found in analysed tumour tissue, it is possible that the mutations observed in blood plasma were derived from clonal haematopoiesis of indeterminate potential (CHIP) occurring in bone marrow, rather than ctDNA from the tumour. CHIP is the age-related clonal expansion of haematopoietic stem cells carrying mutations, which can be perpetuated by chemotherapy and can generate circulating DNA variants that mimic low-frequency ctDNA mutations [74,75,76]. CHIP can occasionally involve mutations in cancer driver genes such as BRAF [77]. Therefore, for all patients where ctDNA was detected without any evidence of mutations in tumour tissue, ddPCR analysis was performed on 100 ng peripheral blood leucocyte (PBL) gDNA isolated from the same blood collection(s) as that in which ctDNA mutations were detected (Supplementary Table 10). For four of the seven patients, no variant ddPCR droplets were detected in the PBL genomic DNA template. While in the other three patients, corresponding variants were identified in PBL genomic DNA, and their concentration was well below the cut-off set in this study for reporting a ctDNA mutation of > 2.5 copies/mL plasma (range 0.0754–0.179 copies/μL), suggesting that these may be false positive droplets generated during ddPCR, or due to a minimal amount of ctDNA co-purified with the gDNA (Supplementary Table 10).

3.7.4 Modification of Plasma Collection Processes for Patients in Geographically Remote Locations

Finally, we investigated the use of cell-stabilising blood collection tubes, which are reported to allow up to 14 days delay in laboratory processing for plasma, for ctDNA analysis. This is important to assess in a NZ context, to implement ctDNA monitoring of patients located in geographically remote regions. Thirty-three blood samples from 18 donors were collected at each timepoint in both EDTA and cell-stabilising tubes. For each patient, the EDTA tube was processed to extract plasma within 4 h or after 7 days for the cell-stabilising tubes. cfDNA was isolated from frozen aliquots, and ddPCR assays were performed to measure the concentration of ctDNA in the samples. This identified the same ctDNA mutations in 30/33 (91%) of the patient samples collected using the cell stabiliser tubes, as were detected using the standard EDTA protocol, and the concentrations showed high concordance (Spearman’s rank ρ 0.81, Supplementary Fig. 14). For three samples, ctDNA was only detected in either the EDTA tube (n = 2) or the cell-stabilising tube (n = 1). Since the measured ctDNA levels in these samples were low (< 13 copies/mL), factors other than the collection protocol may have contributed to this variability.

4 Discussion

Plasma cfDNA analysis in patients with advanced-stage melanoma detected complex mutational patterns over the course of immunotherapy treatment. These are summarised schematically in Supplementary Fig. 15. Some of these mutations appeared to be clinically actionable. ctDNA analysis identified 100% of the BRAF mutations previously found by clinical genomic testing of tumour tissue, as well as BRAF mutations in an additional seven patients (24%), where prior clinical testing had not detected these mutations in the tumour. Potentially, the BRAF ctDNA mutations detected in plasma, but not in tumour tissue, may have evolved since the tumour tissue was tested or may have been present in low cellular fractions or unsampled regions of heterogeneous tumours/metastases.

Previous studies have suggested that plasma ctDNA levels reflect therapeutic responses in real time [6, 12, 14,15,16,17]. However, our study was not designed to quantify this type of real-time temporal relationship between ctDNA concentration and treatment response, since clinical practice at the time of the study did not allow clinical imaging and venesection for ctDNA analysis to be precisely coordinated. Nevertheless, trends in ctDNA concentration for a subset of patients concorded with clear response or progression assessed according to RECIST criteria (e.g. Fig 1).

In some patients, we observed longitudinal variability in ctDNA levels. Although this could have been due to technical artefacts [78] we showed that in general our ctDNA analysis methods were technically reproducible (Supplementary Figs. 10–12). In particular, our results suggest that the quantitation of ctDNA mutations is relatively robust to variability in cfDNA preparation, but that the number of freeze–thaw cycles may affect the detection of very low concentration ctDNAs. On the basis of this analysis, for the assays used here, we would re-test any result of < 2.5 ctDNA copies/mL plasma. CHIP could also conceivably affect measured ctDNA concentrations [74,75,76]; however, we were unable to find clear evidence that CHIP contributed significantly to the longitudinal variability observed here (Supplementary Table 10). Therefore, as previously suggested [79, 80], we speculate that the dynamic mutational complexity seen in blood samples from some patients may, in addition to technical factors, reflect authentic biological changes in multiple evolving metastatic lesions with differing functional states and variable responses to treatment. Levels of detectable ctDNA are also reportedly affected by variable first pass metabolism of cfDNA [81], by progressive tumour desmoplasia blocking the export of ctDNA [81, 82], and by anatomic compartmentalisation such as the blood–brain barrier [83]. In our study, patients whose lack of response to treatment was not accompanied by a clear increasing trend in ctDNA levels had intracranial metastases (for example, patient MEL0022, Supplementary Fig. 15).

Identifying clinical associations between ctDNA level and immunotherapy response in patients was not the core purpose of this study. Nevertheless, we observed that ctDNA concentration in each patient’s final blood sample of the study was significantly associated with patient outcome and durable clinical benefit, which is not surprising given that for many patients the study endpoint coincided with patients leaving the study due to either clear treatment response or progression on therapy. A small number of previous trials [5, 72] have found that baseline ctDNA levels are associated with patient outcome; however, we did not see this association in our cohort. We did observe that the frequency of low or undetectable ctDNA in each patient’s set of longitudinal blood samples was associated with both overall survival and durable clinical benefit of immunotherapy, which concords with previous studies [72]. Potentially, ctDNA ‘clearance’ from blood indirectly indicates the ability of an immune response to be mounted following checkpoint inhibitor treatment. We also observed that undetectable ctDNA in any blood sample taken during the first five therapy cycles had a significant association with patient outcome. Potentially, this information could be used in the future to guide escalation of infusion therapy (for example, selection of combination therapy over a single agent) and/or guide escalation in imaging to monitor treatment effect (which may, for example, trigger switching to a salvage management option). The general clinical utility of ctDNA will be informed by current international clinical trials, and the utility of ctDNA in the specific NZ clinical context described here now needs to be assessed in local prospective studies.

Our NGS analysis identified substantial genomic heterogeneity in samples from individual study patients, in line with the complex mutational landscape of advanced metastatic melanoma identified in previous ctDNA studies [1, 13]. Interestingly, longitudinal ctDNA analysis in our study revealed the apparent evolution of this genomic heterogeneity over the course of treatment, which may indicate sub-clonal mutations progressively superseding one another through the selective pressure of multiple cycles of immunotherapy. The detection in one patient each of multiple concurrent BRAF ctDNA mutations and multiple concurrent NRAS ctDNA mutations, as well as concurrent BRAF and NRAS mutations in 24% of patients, is interesting. Despite the high mutational burden of melanomas from sun-exposed tissues, concurrent BRAF and NRAS hotspot mutations appeared relatively uncommon and appeared to be anti-correlated in the initial TCGA analysis of melanoma [1], as well as in more recent analyses of primary melanoma [84]. However, tumours carrying both BRAF and NRAS hotspot mutations are found [1, 85, 86]. NRAS mutations may develop as resistance mechanisms in patients undergoing BRAF inhibitor therapy [87, 88]; they may co-occur in the same cell [88] and indeed some NRAS mutations may enhance BRAF binding affinity and oncogenic MAPK /ERK signalling [89]. For some patients in this study, we used multiple orthogonal ctDNA analysis methods, which concorded in their identification of multiple driver mutations in the same plasma sample (e.g. Fig. 2). However, previous studies have suggested that multiple driver mutations observed in ctDNA may be an artefact of CHIP, based on detection of the observed cfDNA mutation in PBL but not in tumour tissue [76]. In the current study, we were unable to confirm the presence of CHIP above a reasonable detection threshold in any of the patients studied here. Nevertheless, the high frequency of multiple driver mutations identified here requires further study before firm conclusions can be drawn. Another important finding of this study relates to the selection of ctDNA mutation(s) to assay, when the intention of ctDNA analysis is to follow treatment response. In some studies, a single oncogenic driver mutation present at an early timepoint is selected for tracking treatment response [17, 23]. However, our results suggest that, for a proportion of patients with advanced melanoma, the monitoring of a single target ctDNA with ddPCR may not fully represent the tumour dynamics occurring in the patient, and the use of NGS intermittently could allow the detection of any novel mutations arising.

The use of multiple orthogonal ctDNA analysis methods provides a higher level of confidence for reporting ctDNA variants detected at low levels, as suggested previously [90]. The custom Ampliseq NGS panel used in this study, which was selected due to its high reported sensitivity [91], proved only marginally less sensitive than ddPCR. We found that the results of ctDNA analysis using this NGS panel were both internally reproducible and highly concordant with ddPCR analysis. Our study included analysis using the UltraSEEK mass spectrometry method (Supplementary Table 9), which also provided a high level of concordance with ddPCR and NGS, however, may have lower sensitivity than ddPCR for some mutations (Supplementary Fig. 13). In general, previous studies of advanced cancer, as in our study, report a high frequency of ctDNA detection and high concordance between different analysis platforms [90]. Conversely, some studies of smaller early-stage tumours, which may generate lower concentrations of ctDNA, have identified lower concordance between different ctDNA technologies and between different commercial test vendors [92, 93]. Despite our custom NGS panel’s apparent sensitivity, we suggest that alternative ctDNA analysis methods could further increase sensitivity in the future, such as utilising DNA fragment length differences [94], analysis of ctDNA methylation [95] and reduced amplification of wild-type DNA through blocking oligonucleotides [96]. Further improvements in sensitivity will be especially important when ctDNA analysis is used for relapse surveillance, detection of minimal residual disease or diagnostic screening.

Similar to other studies [97], we found that ctDNA detection in blood samples collected in cell-stabilising tubes was possible without significant loss of test sensitivity. Even though these tubes can allow a delay in processing of up to 14 days, a delay of 7 days to processing was selected and tested in this study as a clinically suitable timeframe for NZ. Based on these results here and by others, there is sufficient evidence that cell-stabilising tubes can be applied to support ctDNA testing for geographically remote patients [40], including some Māori [46, 47] and Pacific People [48] who suffer significant cancer outcome inequities in NZ. In addition, ctDNA analysis undertaken in the intervals between surveillance imaging may provide valuable supplementary information to guide clinical care during health service disruptions, such as those introduced during the coronavirus disease 2019 (COVID-19) pandemic.

Limitations of this study include the relatively small number of patients in the cohort, as well as their recruitment from a single centre. While the clinical care for these patients was not unduly affected by the COVID-19 pandemic, restrictions within clinical services meant that the sample collection timeframe was impacted. While two different clinical analyses were performed for these patients (best RECIST response and durable clinical benefit), the currently available clinical descriptors for immunotherapy response, which are suitable for research such as this, remain somewhat limited. Further work to generate more appropriate clinical descriptors for this type of research is required. The small size of this study limited the value of analyses traditionally used to calibrate the trade-off between sensitivity and specificity for laboratory assays, such as ROC analysis (Supplementary Fig. 9e). Nevertheless, calibration based on real-world data is important for ddPCR in particular, since the frequency of false-positive droplets may be determined by several difficult-to-quantify experimental factors, including non-specific hybridisation of probes, optical errors, high template concentrations and biological sources, such as template contamination. Therefore, in this study, the threshold for reporting a ctDNA mutation was determined largely by a sensitivity/specificity analysis using positive control ‘gBlock’ DNAs. Larger studies will be needed to formally determine the thresholds at which to report ctDNA in these particular assays and to determine which ctDNA (or combination of ctDNAs) to follow longitudinally in any one patient. A further limitation was the potential reduction in sensitivity of ddPCR quantitation for patients who had multiple mutations present within the same region of a gene (as shown in Supplementary Figs. 4–5). The prevalence of multiple concurrent driver mutations (especially in BRAF and NRAS genes) is surprising and as noted above, has not been reported consistently in large melanoma tumour sequencing studies [1, 84]. Whether multiple concurrent sub-clonal driver mutations, including those that appear to emerge over the course of treatment as identified here, are in fact more common than previously recognised, or are an artefact of CHIP that for some reason we could not detect, requires further investigation in larger cohorts.

In summary, we have shown that plasma cfDNA analysis can detect complex tumour mutational patterns and their evolution over time during treatment, a subset of which appear to be clinically informative and potentially actionable. We found that orthogonal ctDNA technologies generate largely concordant results and can allow a higher level of sensitivity by increasing confidence in ctDNA variants detected at low levels. The potential for ctDNA to be used alongside clinical imaging in a stretched health service, and the suitability of delayed processing protocols for geographically remote patients, positions ctDNA testing as a valuable technology in our drive towards cancer care equity.