Clonal dynamics of circulating tumor DNA during immune checkpoint blockade therapy for melanoma

Abstract

Assessment of treatment efficacy of immune checkpoint inhibitors in melanoma patients is difficult as the response to these therapies varies among patients or lesions. The clonal evolution of cancer during immune checkpoint blockade therapy could cause treatment resistance. We investigated the potential of liquid biopsy in monitoring the mutational profiles of metastatic melanoma during immunotherapy. Plasma samples collected from 21 Japanese metastatic melanoma patients before immune checkpoint blockade therapy were subjected to whole‐exome sequencing (WES). Furthermore, 14 Japanese patients with melanoma were enrolled for longitudinal analysis of circulating tumor DNA (ctDNA). Plasma samples were collected prospectively before and during therapy and sequenced. WES of the pretreatment plasma from Japanese melanoma patients showed detectable ctDNA levels with wide ranges of variant allele frequencies within a sample, suggesting clonal and subclonal mutations in ctDNA. In targeted sequencing using longitudinal samples, ctDNA levels correlated with increased tumor size, while ctDNA content immediately decreased after a surge in a patient exhibiting pseudo‐progression, suggesting the potential of ctDNA analysis in discriminating between pseudo‐ and true progression. Mutant ctDNA levels showed different patterns within the clinical course of specific patients, suggesting that these mutations were derived from different tumor clones with distinct therapeutic responses. During further investigation, WES of plasma samples from 1 patient showed marked differences in the mutational profiles of ctDNA, including expansive tumor evolution during an acute exacerbation. Immunotherapy may induce characteristic clonal evolutions of tumors; longitudinal analysis of ctDNA has the potential of determining these tumor evolution patterns and therapeutic responses.

Abbreviations
• ctDNA
• circulating tumor DNA
• LDH
• lactate dehydrogenase
• PD
• progressive disease
• PR
• partial response
• SD
• stable disease
• SNVs
• single nucleotide variations
• VAF
• variant allelic frequency
• WES
• whole-exome sequencing

INTRODUCTION

Immune checkpoint inhibitors, including anti-PD-1 and anti-CTLA-4 antibodies, have shown promise in treating various cancers.^1-8 However, the efficacy and response to immunotherapy vary among tumor sites in patients, with only a subset of patients responding to these therapies, making it difficult to determine the treatment efficacy of immunotherapy compared with conventional chemotherapy. Patients who initially respond to therapy may exhibit cancer progression during the course of treatment. Cancer cell clones evolve dynamically, and changes in their composition cause acquired resistance to molecular targeted therapies.^9,10 Riaz et al¹¹ used biopsy specimens to investigate changes in the mutational profile of melanomas and tumor immune microenvironments during nivolumab therapy. Tumor samples obtained before and after (23-29 d) initiation of therapy were assessed using WES, transcriptomics, and T cell receptor sequencing. Clonal evolution of cancer, including putative selection due to immunotherapy, is observed during treatment. A similar study demonstrated dynamic changes in the mutation-associated neoantigen landscape upon administering immune checkpoint blockade therapies and analyzed paired pretreatment and resistant tumors obtained from non–small-cell lung cancer patients.¹² However, the longitudinal clonal evolution of melanoma cells subjected to immune checkpoint therapy remains unclear. This can be attributed to the difficulty in repetitively sampling tumor tissues and collecting sufficient cancer tissues from different sites in a metastatic patient.

Analysis of circulating tumor DNA (ctDNA), a major target of liquid biopsy, enables researchers to determine mutational changes in cancer cells’ genome without the need for invasive tissue biopsy. ctDNA harbors mutations arising from multiple cancer clones from different sites of the tumor. Changes in ctDNA (mutations in BRAF and NRAS) in patients with melanoma correlate with response to treatment.^13-17 Clonal changes, such as acquired resistance to targeted therapies, have been analyzed using ctDNA analysis.^18,19

Malignant melanoma is extremely rare in Japan, and acral and mucosal subtypes are common. In this study, we performed WES using plasma DNA from 21 Japanese patients with malignant melanoma before being treated with immune checkpoint inhibitors. Moreover, we performed targeted deep sequencing of plasma DNA extracted from another cohort of 14 Japanese patients with melanoma and assessed mutational changes during immune checkpoint blockade therapies.

PATIENTS AND METHODS Ethics statement

The experimental protocols were approved by the institutional review board at the National Cancer Center (approval number 2014-327). Written informed consent was obtained from all patients. The methods were carried out in accordance with the approved guidelines (The Ethical Guidelines for Medical and Health Research Involving Human Subjects and The Ethical Guidelines for Human Genome/Gene Analysis Research).

Patients and samples

The study involved patients with malignant melanoma treated at the National Cancer Center Hospital, Tokyo, Japan. Patients were staged using the classification of the American Joint Committee on Cancer, 8th edition. Plasma samples were collected before and during treatment.

Blood processing and plasma DNA extraction

For retrospective WES analysis of plasma DNA, plasma samples were obtained from the National Cancer Center Biobank with approval from the National Cancer Center Biobank management committee. Peripheral venous blood samples were centrifuged in EDTA-containing tubes at 1600 g for 10 min at 4°C to isolate plasma, which was stored at −80°C until further use. For longitudinal analysis of plasma DNA, blood was centrifuged in EDTA-containing tubes at 1300 g for 10 min at 4°C. Separated plasma was aliquoted and stored at −80°C within 2 h of blood collection.

Before DNA extraction, plasma samples were centrifuged at 16 000 g for 10 min at 4°C to remove cell debris. Circulating DNA was extracted from 2 to 5 mL of plasma using the QIAamp DNA Circulating Nucleic Acid Kit (QIAGEN) in accordance with the manufacturer's instructions, eluted in 60 µL of elution buffer, and stored at 4°C. Eluted plasma DNA was quantified using SYBR Green I real-time polymerase chain reaction (PCR) and human LINE-1 sequences.²⁰ PCR was performed in a 20-µL reaction volume, containing 3 µL extracted plasma DNA, 0.5 µM each of the forward (5′-TCACTCAAAGCCGCTCAACTAC-3′) and reverse primers (5′-TCTGCCTTCATTTCGTTATGTACC-3′), and 1× iTaq SYBR Green Supermix (Bio-Rad). PCR was performed with the following thermal cycling conditions: 2 min at 94°C and 35 cycles of 10 s at 94°C, 15 s at 58°C, and 15 s at 70°C. A standard calibration curve was plotted using 4-fold serial dilutions of human genomic DNA (Promega) up to a 8 ng/reaction. Each sample was assayed in triplicate.

Genomic DNA extraction

Genomic DNA was extracted from tissue samples using the QIAamp DNA Mini Kit (for frozen tissue samples, QIAGEN) or QIAamp DNA FFPE Tissue Kit (for FFPE samples, QIAGEN) in accordance with the manufacturer's instructions. Eluted genomic DNA was processed using the Covaris Ultrasonicator (Covaris) before preparing the sequencing library.

WES

Whole-exome sequencing libraries were prepared from plasma DNA samples and germline DNA samples using the combination of KAPA Hyper Prep Kit (KAPA Biosystems) and SureSelect Target Enrichment System (Agilent Technologies) as described previously.²¹ Briefly, 7.7-100 ng of input DNA were subjected to end repair, A-tailing, and ligated to the Agilent SureSelect adapter. After (solid-phase reversible immobilization) cleanup, the adapter-ligated DNA was amplified using PCR (6-9 cycles). Target capture and library preparation were performed using the Agilent SureSelect Human All Exon Kit V5 in accordance with the manufacturer's instructions. Sequencing was performed using the Illumina HiSeq2500 system (Illumina).

Targeted sequencing

Plasma DNA samples, genomic DNA obtained from germline DNA samples from peripheral blood leukocytes, and tumor DNA samples were subjected to targeted capture sequencing. Sequence library preparation, target capture, and WES were performed using the KAPA Hyper Prep Kit (Kapa Biosystems) and SureSelect Target Enrichment System (Agilent Technologies). The whole exons of 74 genes (Table S1) were subjected to targeted deep sequencing using a custom panel. Among the 74 genes, 33 were selected based on the above WES data and are concretely mutated genes listed in COSMIC Cancer Gene Census (https://cancer.sanger.ac.uk/census). Two genes were added as they were reported as significantly mutated genes in melanoma in a previous report,²² even though they were not detected by our WES. Other genes were chosen based on their relationships to the response of immune check point inhibitors^23,24 or BRAF inhibitors,²⁵ or based on the clinical actionability of molecular targets²⁶ (Table S1). Sequencing was performed using the Illumina HiSeq2500 system (Illumina).

Detecting somatic mutations in plasma DNA

Paired-end reads were aligned to the human reference genome (GRCh37) using the Burrows-Wheeler Aligner²⁷ for the plasma DNA and matched germline DNA samples. Probable PCR duplications, for which paired-end reads aligned to the same genomic position, were removed, and pileup files were generated using SAMtools²⁸ and an in-house developed program. The following cut-off values were used for base selection to identify the somatic point mutations (single nucleotide variations [SNVs] and short indels): mapping quality score of at least 20 and base quality score of at least 10. For WES, inactivating mutations and hotspot mutations with at least 10 COSMIC hotspot mutations were selected using the following filtering conditions: (a) the numbers of reads supporting a mutation for plasma DNA were at least 4 with a variant allelic frequency of plasma DNA (TVAF) ≥ 0.004; (b) TVAF of matched germline DNA was < 0.03 and 0.01 for TVAF ≥ 0.15 and 0.15 > TVAF ≥ 0.004, respectively. Other somatic mutations were selected using the following filtering conditions: (a) the numbers of reads supporting a mutation for plasma DNA were at least 4 and 7 with a variant allelic frequency of plasma DNA (TVAF) ≥ 0.15 and 0.15 > TVAF ≥ 0.02, respectively; (b) TVAF of matched germline DNA was < 0.03 and 0.01 for TVAF ≥ 0.15 and 0.15 > TVAF ≥ 0.02, respectively. For targeted deep sequencing, the following filtering conditions were used: (a) the numbers of reads supporting a mutation for plasma DNA were at least 8 and 12 for SNVs and indels, respectively; (b) TVAF of matched germline DNA < 0.03. Due to the sequence-specific nature of sequencing errors, the read information from all germline DNA samples was grouped to reliably discriminate between true and false positives. Subsequently, we used the following filters: (a) the ratio of variant allele frequency of grouped germline DNA (NVAF) and plasma DNA (TVAF), NVAF/TVAF < 0.05; (b) all samples with more than 2 somatic mutations (SNV and/or indel) within any 10 bp window and TVAF < 0.2 were discarded; and (c) mutations must be recognized by the forward and reverse reads of > 5% of all supported reads. Additionally, the following filters were applied for WES: (a) NVAF must be less than 0.004 and 0.006 for SNVs and indels, respectively; (b) P-value of Fisher exact test for the numbers of supported and non-supported reads of a mutation in plasma DNA and grouped germline DNA must be < 0.0001. For targeted deep sequencing, mutations with a root mean square mapping quality score < 40 for reads covering the mutation were discarded. For somatic mutations in the plasma DNA, some mutations were also found in paired normal peripheral lymphocyte DNA at low frequencies. Such mutations were considered somatic mutations in normal lymphocytes associated with clonal hematopoiesis and excluded from ctDNA analysis. ctDNA levels, expressed in human genome equivalents per mL of plasma, were calculated by multiplying the cell-free circulating DNA level based on the percentage of mutant allele fraction measured using sequencing.

RESULTS

Immune checkpoint blockade is usually applicable for patients with advanced and metastatic melanoma. As ctDNA contains tumor DNA from different lesions in a patient, sequence analysis of plasma DNA has the potential to detect mutations in a heterogeneous population of tumor cells. We conducted a retrospective WES analysis of plasma DNA using a cohort of 21 patients with metastatic melanoma treated with nivolumab to determine their ctDNA mutational profiles (Table S2).

The median unique sequence coverage was 439× (range 349-583×). We detected somatic mutations in the ctDNA from all samples (4-163 per sample; Figure 1A and Table S3). Cancer-related genes, such as BRAF (n = 4), GNAS (n = 3), and GRIN2A, NF1, PTEN, and TP53 (n = 2 each), were frequently mutated (Figure 1A), indicating that ctDNA is detectable in patients before treatment.

FIGURE 1. Whole-exome sequencing (WES) of plasma DNA from Japanese patients with metastatic melanoma. A, The top bar plot shows the number of somatic mutations identified in each sample. The recurrently mutated genes listed in the COSMIC Cancer Gene Census are shown in the left-most column. B, Variant allele frequencies of somatic mutations detected in plasma DNA. Box plots indicating the variant allele frequencies of mutations detected in each sample. Each box shows the median (central line), inter-quartile range (box), and ±1.5  ×  inter-quartile range (whiskers)

We then evaluated the VAF of somatic mutations detected in plasma DNA. VAF for each sample was 0.026-0.371 (median 0.039; Figure 1B). Therefore, plasma DNA sequencing analysis identified clonal and subclonal mutations or simultaneous amplification of mutant genes.

To investigate longitudinal changes in ctDNA from patients subjected to immune checkpoint blockade therapy, we prospectively collected plasma samples from melanoma patients to be administered nivolumab or ipilimumab. Among the 19 melanoma patients who were excluded from WES and were prospectively enrolled, plasma samples were available pre- and post-initiation of therapy for 14 cases, including cutaneous (n = 4), mucosal (n = 4), acral (n = 3), and uveal (n = 3) melanomas (Table 1). We collected 2-13 plasma samples/patient (median 7) before and during treatment (range 61-342 d). In cases 4, 17, and 29, immunotherapy was initiated after surgery. Radiotherapy was performed in case 8 before anti-PD-1 therapy. In case 9, the BRAF/MEK inhibitor, vemurafenib, was administered before the immune checkpoint blockade therapy. Quantitative PCR showed time-dependent dynamic changes in DNA content in the plasma for cases 5, 9, 17, 23, and 38 (Figure S1 and Table S4).

TABLE 1 Characteristics of patients subjected to targeted sequencing analysis

Case#	Sex	Age	Number of cfDNA samples	Sampling periods (d)	Type	Location of primary tumor	Stage	Treatment	Overall response	Tumor sample
4	F	42	10	342	Cutaneous	Back	IV	Surgery, anti-PD-1	SD	Not available
5	M	74	9	334	Uveal	Choroid	IV	Anti-PD-1	SD	Available
8	M	73	13	301	Uveal	Choroid	IV	Radiation, anti-PD-1	SD	Not available
9	M	56	4	110	Acral	First toe	IV	Anti-CTLA-4, anti-PD-1	SD	Available
11	M	39	2	78	Cutaneous	Knee	IV	Anti-CTLA-4	SD	Not available
12	M	68	7	111	Mucosal	Nasal cavity	IV	Anti-CTLA-4, anti-PD-1	PD	Not available
16	F	69	8	236	Mucosal	Urethral	IV	Anti-CTLA-4, anti-PD-1	PR	Not available
17	M	85	11	253	Acral	Sole	IV	Surgery, anti-PD-1	SD	Available
18	M	66	10	253	Uveal	Choroid	IV	Anti-PD-1	PR	Not available
23	F	77	6	191	Cutaneous	Femur	III	Anti-CTLA-4	PR	Available
25	M	71	2	61	Mucosal	Esophagus	IV	Anti-PD-1	PD	Not available
29	F	64	7	212	Cutaneous	Head	IV	Surgery, anti-PD-1	PR	Not available
38	M	44	4	133	Acral	Sole	IV	Anti-CTLA-4	SD	Available
39	M	62	3	113	Mucosal	Nasal cavity	IV	Anti-PD-1	PR	Not available

Abbreviations: cfDNA, cell-free DNA; PD, progressive disease; PR, partial response; SD, stable disease.

To accurately detect somatic mutations in low abundance ctDNA in plasma using deep sequencing, we designed a gene panel, including mutated genes from our WES data, previously identified mutated genes in melanoma, genes related to response to immune check point inhibitors^23,24 or BRAF inhibitors,²⁵ and clinical actionable driver genes associated with potential targeted therapies.²⁶ Table S1 lists the exons of 74 genes targeted.

Targeted deep sequencing of the extracted plasma DNA had a median unique sequence coverage of 2021× (range 1139-3005×, Table S4). We detected somatic mutations in ctDNA with a threshold of VAF > 0.006. Subsequently, we monitored the changes in the amounts of mutant DNA during treatment (Table S5). We considered identical mutations detected at different time points in the same patient to be true mutations, although with low VAF, and included mutations with 0.003 < VAF < 0.006.

We identified changes in the repeatedly detected, identical mutant ctDNA in cases 4, 8, 9, 12, 16, 17, 18, and 39 (Figure 2). As the total amount of plasma DNA changed over time, we determined the amounts of mutant DNA by multiplying the level of plasma DNA with the percentage of mutant alleles measured using sequencing.

FIGURE 2. Longitudinal changes in mutant circulating tumor DNA (ctDNA) in immunotherapy-treated melanoma patients. A, The amounts of mutant DNA and lactate dehydrogenase (LDH) in plasma are shown on the left, and magnetic resonance images are shown on the right for case 4. The red arrowheads and circles indicate the tumor location. In case 4, brain metastasis was observed 3 mo after surgery, and nivolumab therapy was initiated following radiation therapy. Evaluation at week 11 showed radiographic response accompanying undetectable ctDNA. Mutant ctDNA levels increased during disease progression with new brain metastasis. B, Case 8 was treated with nivolumab following radiotherapy for bone metastases, and the disease was stable at weeks 9, 22, and 33. Two mutations were highly elevated during radiotherapy and 17 wk after the initiation of nivolumab. C, Case 9 underwent 4 cycles of ipilimumab. Radiographic progression with the spread of subcutaneous tumor was documented at week 13. Therapy was discontinued due to immune-related adverse effects. Increased levels of ctDNA were observed at week 14. D, Case 12 had subcutaneous, hepatic portal lymph node, adrenal gland (red arrowhead), and peritoneal metastases at the beginning of immunotherapy. After treatment with ipilimumab for 2 cycles, we initiated nivolumab therapy. Although the radiographic response was noted at week 9 as the adrenal gland metastasis decreased, peritoneal metastases progressed at week 14. While ctDNA decreased to undetectable levels at week 9, multiple mutant ctDNAs were detected at weeks 14 and 16. E, Case 16 was treated with nivolumab after discontinuation of ipilimumab. Mutant ctDNA levels were elevated at the beginning of nivolumab treatment, with increases in lung metastasis. ctDNA levels were increased after a temporary decrease in week 13. Imaging showed a new brain metastasis at week 24. F, In case 17, after surgery for inguinal lymph node metastasis, the residual tumor was treated with nivolumab for 4 cycles, and therapy was discontinued due to immune-related adverse effects. However, computed tomography showed an SD. ctDNA and inguinal lymph node metastasis were consistently observed during follow-up. G, In case 18, pseudo-progression was noted as liver metastasis slightly increased 9 wk after initiation of nivolumab therapy. Thereafter, the lesion gradually decreased and indicated a PR at week 63. While transient elevation of ctDNA was observed at week 15, it rapidly decreased at week 18. H, Case 39 was suspected of having new metastasis in the sternal lymph node 9 wk after initiation of nivolumab therapy. This patient underwent radiotherapy for sternal lymph node in addition to nivolumab, which resulted in PR at week 19. hGE, human genome equivalent; Ipi, ipilimumab; Nivo, nivolumab; PD, progressive disease; PR, partial response; RT, radiation therapy; SD, stable disease

The changes in ctDNA levels were correlated with the patients' clinical response and serum concentrations of lactate dehydrogenase (LDH) as a clinical biomarker for melanoma. The amounts of ctDNA decreased during the PR in cases 12, 18, and 39. Mutant ctDNA was undetectable during PR in cases 12 and 18 (Figure 2D,G). Conversely, the amounts of ctDNA increased during PD in cases 4, 9, 12, and 16 (Figure 2A,C-E).

Transient increases in ctDNA after the initiation of treatment were observed in cases 8 and 18. In case 18, ctDNA levels were transiently elevated at week 15 and then maintained at a very low or undetectable level during follow-up. At 9 wk after the initiation of nivolumab, computed tomography showed pseudo-progression based on the increase in a target lesion size. Subsequently, liver metastasis gradually slowed and resulted in PR at week 63 (Figure 2G).

Some cases showed different patterns of changes in mutant ctDNA levels, even in the same patient. In case 4, a mutation in MLL3 (the only mutation detected before surgery for cutaneous melanoma on the back) disappeared after resection (Figure 2A). Although mutant MLL3 was undetectable at the time of appearance of brain metastasis (11 wk), the amount of ctDNA with mutant MLH1 increased until the initiation of nivolumab treatment. Subsequently, the levels of both mutations elevated during PD, suggesting different mutational profiles in the different lesions, such as primary and metastatic tumors (Figure 2A). Case 8 was administered nivolumab after radiotherapy for bone metastases, and SD was monitored at weeks 9, 22, and 33. Two MSH3 mutations were detected in the plasma throughout the observation period. However, mutations in ARID1B transiently increased during and 3 mo after irradiation, indicating the differential nature of acquiring MSH3 and ARID1B mutations (Figure 2B). Therefore, ctDNA analysis has the potential to detect mutations derived from different tumor clones with altered reactivity to therapy.

We sequenced tumor DNA from archival tumor tissues from cases 5, 9, 17, 23, and 38 using the same gene panel and sequencing platform for ctDNA analysis (Table S5). None of the somatic mutations identified in case 5 was common between the tumor tissues and plasma DNA samples. Considering that VAFs of mutations detected in the tumor tissue were low (0.006-0.025), they were presumably subclonal mutations and could not be detected in the plasma. Tumor tissue and plasma DNA from case 9 showed identical mutations in BRAF, suggesting mutant BRAF to be the driver of melanoma progression that was derived from melanoma. Therefore, the BRAF/MEK inhibitor, vemurafenib, was administered before the immune checkpoint blockade therapy in case 9. After 3 mo with vemurafenib, progression of the disease was observed using computed tomography. Moreover, BRAF mutations were detected in the tumor tissue from case 17. However, this mutation was not detected in the plasma DNA. Instead, we identified a mutant MET in the plasma before and after surgery. As mutant MET was not detected in the tumor samples used for sequencing, ctDNA derived from different subclones harboring MET amplification might be dominant in the plasma (Figure 2F). Targeted sequencing analysis did not detect somatic mutations in the plasma DNA from cases 23 and 38.

The data from longitudinal targeted sequencing of ctDNA showed the importance of liquid biopsy in determining clonal changes in patients with melanoma undergoing therapy. Several mutations that were undetectable before treatment were identified in plasma DNA collected at the time points corresponding to acute exacerbation with massive ascites in case 12 (Figure 2D). This indicates the expansive evolution of tumor clones during immunotherapy.

To further understand the changes in the ctDNA mutational profiles of case 12, we performed WES using plasma DNA. Plasma DNA samples obtained before therapy (Pre), 12 wk (12w), and 16 wk (16w) after initiation of therapy were subjected to WES analysis. We obtained sequencing reads for Pre, 12w, and 16w plasma DNA samples, with mean unique sequencing coverages of 286.3×, 446.3×, and 531.3×, respectively. Somatic mutations in the plasma DNA were initially detected with a threshold of VAF > 0.006, and the numbers of reads supporting a mutation for plasma DNA were at least 8. Using these criteria, we identified 226, 391, and 746 somatic mutations (SNVs and short indel) in the Pre, 12w, and 16w plasma DNA samples, respectively (Table S6). We then modified the threshold of the number of mutant reads to normalize the detectability of somatic mutations in plasma DNA by adjusting for sequencing coverage to 8, 12.47, and 14.84 for Pre, 12w, and 16w plasma DNA, respectively. We detected 226, 112, and 220 somatic mutations using the Pre, 12w, and 16w plasma samples, respectively.

We detected 78 identical mutations at 2 or more time points. Figure 3A shows the changes in VAF of the identical mutations. Among these, only 3 mutations were detected at 3 time points. Thirteen mutations detected in the Pre and 12w plasma samples were undetectable in the plasma collected at 16w, suggesting the reduction of subclones harboring these mutations during therapy. VAFs of 59 mutations detected in the plasma before therapy were undetectable at 12w before increasing at 16w. Moreover, 3 mutations that were not detected in the pretreated plasma were found in the 12w and 16w samples, suggesting that those mutations were derived from expanding subclones during cancer progression.

FIGURE 3. Somatic mutations detected in plasma DNA from case 12 using WES. A, Chronological changes in variant allele frequencies of mutations detected in plasma DNA using WES. In total, 78 somatic mutations identified at multiple time points are shown in the plot. B, Venn diagram of somatic mutations detected in plasma DNA using WES. Somatic mutations detected in samples collected before therapy (Pre) are in green, at week 12 (12w) are in blue, and at week 16 (16w) are in yellow. WES, whole-exome sequencing

Figure 3B shows the commonalities of the somatic mutations detected in the plasma. Although 78 mutations were detected across 2 or 3 time points, the vast majority of mutations were detected at a single time point. As much as 162 mutations detected in the 16w plasma DNA could not be detected in the earlier plasma DNA samples. This implies the evolution of multiple subclones after 12w and before acute exacerbation (eg, massive ascites) in this patient. Therefore, the changes in the composition of cancer subclones during therapy might reflect a patient's clinical disease status.

DISCUSSION

In this study, we investigated the mutational profiles of ctDNA and its changes during immune checkpoint blockade therapy in Japanese patients with malignant melanoma. Retrospective WES analysis confirmed the presence of ctDNA with somatic mutations, including subclonal mutations in the pretreatment plasma. BRAF and NRAS are frequently mutated in melanoma; therefore, hotspot mutations in these genes have been used as ctDNA biomarkers in the Western population.^13,15-17 However, Asian populations have lower rates of mutation in BRAF or NRAS than do Caucasian populations.^29-31 The majority of mucosal or uveal melanomas do not have mutant BRAF.^32-35 In our 21 plasma DNA samples subjected to WES, we observed that 33% of the samples had mutant BRAF or NRAS. Therefore, ctDNA analysis of these hotspot mutations might be insufficient for Japanese melanoma cases.

Prospective longitudinal analysis of plasma DNA in immune checkpoint inhibitor-treated melanoma patients showed that the amounts of plasma DNA dynamically changed based on treatment and patient status. Plasma DNA levels are regulated by various physiological conditions, including injury and inflammation. Elevation of plasma DNA levels after surgery was observed in cases 4 and 17. Cases 8, 16, and 18 showed transient increases in the amounts of plasma DNA 6-10 wk after immunotherapy initiation. Immune responses associated with immune checkpoint blockade, such as pseudo-progression or immune-related adverse events, are observed within 12 and 10 wk of therapy initiation, respectively.^36,37 The activation of immune responses by immune checkpoint inhibitors affects the total amounts of plasma DNA.

Targeted deep sequencing analysis of plasma DNA showed dynamic changes in mutant ctDNA levels during therapy, demonstrating the correlation between ctDNA and response to therapy. However, discrimination between pseudo- and true progression has been challenging in cancer immunotherapy. Case 18 experienced pseudo-progression. Interestingly, ctDNA levels decreased within 12 wk of nivolumab initiation despite a slight increase in tumor size. Although ctDNA levels increased after pseudo-progression (15 wk), mutant ctDNA was undetectable at the next time point, and the target lesion gradually decreased, thereby resulting in PR and suggesting the utility of ctDNA analysis in differentiating between pseudo- and true progression. The transient increase in ctDNA in case 18 could be because of damaged cancer cells by the cytotoxicity of immune cells. Case 8 also showed transient increases in ctDNA after radiotherapy. As the effects of radiation are observed several months after irradiation, the transient elevation of ctDNA could be due to the radiotherapy-induced death of tumor cells. The elevation of ctDNA levels is not always associated with disease progression in patients undergoing immunotherapy.

Another challenge in cancer therapy is characterizing refractory clones to therapy. Liquid biopsy can detect mutations derived from different tumor clones from multiple lesions with different reactivities to therapy. ctDNA analysis could be valuable in characterizing responsive and refractory clones. In case 8, durable SD was documented throughout follow-up. While MSH3 mutations were constantly detected in ctDNA, mutations in ARID1B transiently appeared 13-17 wk after radiotherapy. As no other malignancy was documented in this case, these mutations might be derived from different tumor cell clones. Notably, suppression of ARID1B impairs DNA repair and sensitizes cells to ionizing radiation.³⁸

Interestingly, WES of plasma DNA from case 12 revealed a dramatic change in mutational profiles of ctDNA during immunotherapy. While a variety of somatic mutations in ctDNA increased with progressing disease, WES showed an abrupt increase in somatic mutations exclusively detected in plasma at a single time point. In accordance with the reported changes in mutational profiles in tumor tissues during therapy,^11,12 our results suggest a drastic alteration in the composition of tumor subclones in a patient during immunotherapy.

However, there are some limitations to this study. First, the patient cohort was limited, and a larger cohort should be studied to understand the utility of ctDNA analysis in monitoring cancer during immunotherapy. Second, the 74-gene panel designed for targeted deep sequencing might be insufficient to assess chronological changes in ctDNA in some patients. No identical mutation was detected at different time points in 6 of 14 patients in targeted sequencing of plasma DNA, suggesting the presence of clonal mutations outside the target region of our gene panel. A more comprehensive analysis, such as WES, may help determine the mutational profiles of ctDNA in a wide range of patients undergoing immune checkpoint blockade therapy.

In conclusion, we showed that longitudinal ctDNA analysis could potentially monitor response to therapy, including pseudo-progression, and help understand clonal evolution during therapy in melanoma patients. The timely assessment of ctDNA mutational profiles might help devise treatment strategies.

ACKNOWLEDGMENTS

We thank all the patients and their families who contributed to this study. This work was supported by grants from the National Cancer Center Research and Development Fund (25-A-4 and 28-A-4 to SY and 29-A-6 to TS); Japan Agency for Medical Research and Development (JP20ck0106558 to SY); Joint Research Project of the Institute of Medical Science, University of Tokyo (SY and TS); Integrated Frontier Research for Medical Science Division, Institute for Open and Transdisciplinary Research Initiatives, Osaka University (SY); Yasuda Medical Foundation (SY); Yakult Bio-Science Foundation (SY); Princess Takamatsu Cancer Research Fund (SY); Takeda Science Foundation (SY); and Kobayashi Foundation for Cancer Research (KN). The National Cancer Center Biobank is supported by the National Cancer Center Research and Development Fund, Japan.

DISCLOSURE

The authors declare no conflict of interest.