Genome‐wide and panel‐based cell‐free DNA characterization of patients with resectable esophageal adenocarcinoma

Circulating tumor DNA (ctDNA) holds promise in resectable esophageal adenocarcinoma (EAC) to predict patient outcome but is not yet sensitive enough to be clinically applicable. Our aim was to combine ctDNA mutation data with shallow whole‐genome sequencing (sWGS)‐derived copy number tumor fraction estimates (ichorCNA) to improve pathological response and survival prediction in EAC. In total, 111 stage II/III EAC patients with baseline (n = 111), post‐neoadjuvant chemoradiotherapy (nCRT) (n = 68), and pre‐surgery (n = 92) plasma samples were used for ctDNA characterization. sWGS (<5× coverage) was performed on all time‐point samples, and copy number aberrations were estimated using ichorCNA. Baseline and pre‐surgery samples were sequenced using a custom amplicon panel for mutation detection. Detection of baseline ctDNA was successful in 44.3% of patients by amplicon sequencing and 10.5% by ichorCNA. Combining both, ctDNA could be detected in 50.5% of patients. Baseline ctDNA positivity was related to higher T stage (cT3, 4) (p = 0.017). There was no relationship between pathological response and baseline ctDNA positivity. However, baseline ctDNA metrics (variant allele frequency > 1% or ichorCNA > 3%) were associated with a high risk of disease progression [HR = 2.23 (95% CI 1.22–4.07), p = 0.007]. The non‐clearance of a baseline variant or ichorCNA > 3% in pre‐surgery samples was related to early progression [HR = 4.58 (95% CI 2.22–9.46), p < 0.001]. Multi‐signal analysis improves detection of ctDNA and can be used for prognostication of resectable EAC patients. Future studies should explore the potential of multi‐modality sequencing for risk stratification and treatment adaptation based on ctDNA results. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.


Introduction
Locally advanced esophageal adenocarcinoma (EAC) can be treated with perioperative chemotherapy or neoadjuvant chemoradiotherapy (nCRT) [1,2].The standard-of-care nCRT regimen used in several countries is based on carboplatin, paclitaxel, and 23 fractions of radiotherapy [2,3].This regimen can induce a complete response (CR) in up to 23% of patients [2].However, despite the induction of CR in a subset of patients, the risk of relapse is 49% within 5 years across the whole group [2].Recently, the CheckMate 577 trial showed disease-free survival benefit of nivolumab postsurgery in patients with residual pathological disease after nCRT [4].Prediction of response and survival outcomes may aid in personalizing treatment decisions such as proceeding to surgery after nCRT and/or tailoring systemic therapy such as adjuvant nivolumab.
The detection of circulating tumor DNA (ctDNA) can be leveraged for clinical purposes in gastroesophageal cancer, as previously observed in stage IV patients [5].In locally advanced EAC, several studies were able to predict recurrence both before and after surgery through next-generation sequencing (NGS) of plasma samples [6][7][8][9].In a longitudinal sampling study based on a pan-cancer panel (77 genes), nine out of ten patients who were ctDNA-positive after surgery experienced disease relapse [6].However, ctDNA positivity after accounting for clonal hematopoiesis with indeterminate potential (CHIP) was only 16% (10/63) after surgery, while more than half of all patients experienced relapse.A recent study reported on a tumor-informed sequencing method where ctDNA was detected in 56% (44/78) of baseline samples, but again a low detection rate of 10% (8/78) was observed in post-treatment samples [8].The detection of ctDNA post-treatment also seems to be specific for incomplete response to nCRT but suffers from poor sensitivity (21%) [10,11].These results suggest that ctDNA can be used to identify risk of recurrence but it is not yet sensitive enough to identify every patient with residual disease.
To improve ctDNA monitoring in EAC, it is important to understand the genomic background of these tumors.Recently, a whole-genome sequencing (WGS) study of 551 EAC samples has unraveled a heterogeneous landscape of driver genes and copy number alterations (CNAs) [12].In total, 77 driver genes were identified with several not reported earlier in EAC.Using the most frequently mutated driver genes in an EAC-specific sequencing panel could potentially improve the sensitivity of ctDNA detection without having to rely on sequencing of tumor cells in parallel.Additionally, combining panel-based sequencing with shallow wholegenome sequencing (sWGS) may aid in the detection of CNAs present in cell-free DNA [13][14][15].Combining an EAC-specific sequencing panel with sWGS may provide synergistic opportunities for clinical use in EAC.
In this study, we investigated the clinical value of ctDNA detection by a custom amplicon panel and sWGS.First, we assessed if serial ctDNA monitoring in EAC patients treated with neoadjuvant therapy can help to identify CR before surgery.Second, we investigated whether ctDNA detection could be used for clinical use cases such as early relapse detection by combining amplicon sequencing and sWGS.

Study design
In total, 271 plasma samples from 111 EAC patients were collected from two clinical cohorts.The first cohort consisted of 71 patients treated with nCRT according to the CROSS regimen from a prospectively collected esophageal and gastric cancer biobank (Bi-OES).The second cohort included 40 patients treated with nCRT combined with atezolizumab (PD-L1 antibody; F. Hoffmann-La Roche, Basel, Switzerland) enrolled in a phase II clinical trial, PERFECT (NCT03087864).Samples were collected at baseline (n = 111), post-nCRT in week 5 (n = 68), and pre-surgery (n = 92).Additionally, baseline white blood cell (WBC) samples from the buffy coat were available.Figure 1 shows the workflow of the study and supplementary material, Figure S1 shows the REMARK flowchart of the included samples.
All patients provided written informed consent to participate.The Bi-OES biobank (METC 2013_241) and the PERFECT trial (METC 2016_325) were both approved by the medical ethics review committee of the Amsterdam UMC, location Meibergdreef.This study was conducted in accordance with the Declaration of Helsinki and the international standards of good clinical practice.

Blood handling and cell-free DNA extraction
Blood samples were collected in 10-ml EDTA tubes to be processed in a two-step centrifugation protocol.Plasma and buffy coat were separated (10 min at 1,300 Â g) and after an ultra-centrifugation step of plasma (10 min at 20,000 Â g), stored in a À80 C freezer [5,16].The extraction of cell-free DNA (cfDNA) was performed by using the QIAamp Circulating Nucleic Acid Kit (Qiagen, Valencia, CA, USA) on the QIAvac Vacuum System.For each sample, between 4 and 8 ml of plasma was used for cfDNA extraction.Quantification of cfDNA was done by the high sensitivity 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA, USA) and Qubit dsDNA High Sensitivity Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA).Genomic DNA was extracted from WBC samples using the QIAamp DNA Mini and Blood kit (Qiagen).

Copy number-derived tumor fractions
The sequencing data were used in downstream analyses to identify CNAs through the ichorCNA software package to generate ctDNA tumor fractions as previously described and validated in tumor tissue [13,20].Exceptions to the software's default settings were the following: (1) From sWGS, an in-house panel of normals was created; (2) non-tumor fraction parameter restart values were set to c(0.95,0.99,0.995,0.999); (3) the ichorCNA ploidy parameter restart value was set to 2; (4) no states were used for subclonal copy number; and (5) the maximum copy number to use was lowered to 3. The tumor fraction with the highest log likelihood was retrieved and reported.Samples were classified as ctDNA-positive based on sWGS data if the ichorCNA threshold was above 3% (as previously reported) [13].

Targeted sequencing and data analysis
A custom amplicon sequencing panel was developed using publicly available whole exome and WGS data from EAC cases to assess the baseline, pre-surgery, and WBC samples similar to our previously validated approach in metastatic gastroesophageal cancer patients [5].The Catalogue Of Somatic Mutations in Cancer (COSMIC), cBioPortal, International Cancer Genome Consortium (ICGC), and The Cancer Genome Atlas (TCGA) databases were screened for EAC datasets.An additional dataset was identified which included n = 551 EAC cases with WGS data from Frankell et al [12].In this dataset, several previously unknown driver mutations in EAC were described.The design of our custom panel was based on the prevalence of cases with a mutation in an EAC driver gene.If a driver gene was mutated in more than 5% of EAC cases in the Frankell cohort, we included it in our panel.In total, 23 genes were eligible: ABCB1, APC, ARID1A, ARID2, CDKN2A, DNAH7, EPHA3, KCNQ3, KRAS, LRRK2, MUC6, NAV3, NIPBL, PIK3CA, RNF43, SCN3A, SLIT2, SMAD4, SMARCA4, TRPA1, TSHZ3, ZFHX3, and TP53.All missense, nonsense, insertion, and deletion mutations from the aforementioned genes present in the publicly available EAC sequencing datasets predicted as (likely) benign by the online database ClinVar were excluded [21].Custom panel design was done in the Ion Ampliseq Designer (Thermo Fisher Scientific) environment for the Ampliseq HD NGS platform.The panel was 4.02 kbp in size, consisting of two pools with in total 698 amplicons.This EAC amplicon panel was used both for the cfDNA and for the WBC samples.Library preparation was performed according to the standard operating procedure of the Ion AmpliSeq HD Library Kit and the Ion AmpliSeq HD Dual Barcode kit with 5 ng of input per pool.Sequencing of libraries was done on the Ion S5 NGS system with nine samples submitted per 540 Chip.The mean base coverage depth for cfDNA samples was 10,524Â and for WBC samples 9,647Â.Workflow of cfDNA sequencing from EAC patients derived from three time-points.Baseline and pre-surgery samples were sequenced with both a genome-wide approach and an amplicon panel.Post-nCRT samples were only sequenced using the genome-wide approach.

288
T van den Ende, Y van der Pol et al Data analysis was performed in the Ion Reporter online software environment.A baseline variant was called if it was present in at least three functional families, with three reads in reverse and forward primers per family.Filtering out of non-ctDNA variants was done through the Oncomine Extended (5.18v2) filter chain after a manual check.This included removing known germline variants present in dbSNP, 5000 Exomes, ExAC, and UCSC common single-nucleotide polymorphisms.Variants of interest should be either missense, nonsense, insertions, or deletions in exonic or splice site regions.Thereafter, called variants were screened for CHIP by first comparing them with the WBC sequencing results of the same patient.If the variant was not present in the WBC sample, the level of pathogenicity was assessed with the following online databases: Ensembl, ClinVar, OncoKB, and the National Cancer Institute TP53 Database.Likely pathogenic, pathogenic variants, and nonsense mutations with a predicted nonsense-mediated decay (NMD) were classified as ctDNA-derived.Insertion and deletions were only regarded as ctDNA-derived if they were previously reported and predicted to be (likely) pathogenic.In the pre-surgery samples, we tracked the baseline mutations classified as derived from ctDNA.A baseline mutation was called in the pre-surgery sample if it was present in both the forward and the reverse primers and in two functional families (at least three reads per family).

Statistical analysis
The χ 2 test was used to test proportions between two groups of categorical variables.Continuous metrics between two groups were tested using the Mann-Whitney U-test.Correlations were assessed through the Pearson correlation coefficient.The Cox proportional-hazards model and the log-rank test were used to compare cancer-specific death (CSD) and time to progression (TTP).Multivariable Cox-regression analyses were conducted with a dichotomous ctDNA variable, age (continuous), sex (male versus female), cT stage (cT1-2 versus cT3, 4), cN stage (cN0 versus cN1-3), tumor location (esophagus versus junction), and differentiation grade (well/moderate versus poor versus unknown).CSD was defined as the days elapsed between start of treatment and cancer-related death or end of follow-up.TTP was defined as the days elapsed from start of treatment until disease progression, recurrence, or end of follow-up.Data cut-off for the survival analyses was 14 January 2022.GraphPad Prism version 9.1 (GraphPad Software Inc., San Diego, CA, USA) and SPSS version 26.0 (IBM Corp., Armonk, NY, USA) were used for statistical analyses.An α below 0.05 was regarded as statistically significant.All statistical tests were conducted two-sided.

Results
In total, 111 patients treated with either neoadjuvant chemoradiotherapy (nCRT) according to the CROSS regimen (n = 71) or nCRT with anti-PD-L1 (n = 40) were included.Baseline cfDNA samples were available from 111 patients, post-nCRT from patients and presurgery from 92 patients (total n = 271 samples).Baseline characteristics were as expected from a resectable EAC cohort with a median age of 64 years, 88% male, 75% cT3 disease, and 70% cN+ (Table 1).
From sWGS, ichorCNA tumor fractions per baseline sample were calculated (median 1.09%, range 0.33-20.08%;Figure 2C).In total, 10.5% (11/105) of samples were ctDNA-positive based on the ichorCNA threshold of 3% (Figure 2C).In supplementary material, Table S3, associations between ichorCNA, clinical variables, and mutation status are given.Supplementary material, Table S4 gives an overview of CNAs for ichorCNA-positive patients.A correlation was observed between the maximum mutation VAF per ctDNA positive sample and the matched ichorCNA values (r = 0.66, p < 0.0001, Figure 2D).Thereafter, we assessed ctDNA positivity in baseline samples with both amplicon sequencing and ichorCNA data available (n = 103).In total, 50.5% (52/103) were ctDNA-positive based on these criteria.Based on the combined classification, a significantly higher proportion of cT3, 4 patients was observed in the ctDNA + group compared with the ctDNA À group ( p = 0.017, Figure 2E) but not for N stage (Figure 2F).In conclusion, combining both metrics is thus complementary for the detection of ctDNA in resectable EAC and positivity is related to T stage.

ctDNA classification not sufficient to predict pathological response
We next assessed if the ctDNA detection status (ctDNApositive versus ctDNA-negative) can be used in a clinically relevant scenario for the detection of CR before surgery using longitudinal sampling.Two groups were compared: patients with a CR (TRG1) and patients with residual cancer (TRG2-5 or pre-operative progression).At baseline, samples were used with both mutation, ichorCNA data and a tumor response status (n = 101).Two samples were excluded from the initial n = 103 as no response status was available (refused surgery).Based on the combined classification, the proportion of CR patients was comparable between the ctDNAnegative and ctDNA-positive groups at baseline (χ 2 p = 0.36, Figure 3A).
Longitudinal sampling based on ctDNA detection post-nCRT (n = 37 nCRT + anti-PD-L1, n = 27 nCRT) and pre-surgery (n = 30 nCRT + anti-PD-L1, n = 54 nCRT) by sWGS was not able to predict pathological response.ctDNA positivity based on ichorCNA was seen both in CR and in non-CR samples at the three sampling time-points (Figure 3B).Moreover, no subgroup analysis was conducted for treatment regimen due to the low number of ichorCNA-positive patients in each cohort (see supplementary material, Table S5).
Furthermore, we assessed the value of mutation tracking which identified four ctDNA + patients in the pre-surgery sample, of whom n = 3 had a non-CR and cfDNA sequencing from rEAC 291 n = 1 refused surgery.Next, we assessed if mutation tracking combined with ichorCNA detection presurgery in matched samples could predict pathological response (n = 75; one patient excluded from n = 76 due to surgery refusal).There was no statistically significant relationship between ctDNA status pre-surgery and response status (χ 2 p = 0.75, Figure 3C).There was also no predictive signal when comparing total and subtotal responders (TRG 1, 2) with poor responders (TRG 3-5 or pre-operative progression).
Based on these results, it was not possible to accurately characterize pathological response using ctDNA mutation or ichorCNA analysis.

Discussion
In this study, we investigated the clinical value of measuring ctDNA in plasma samples from patients with resectable EAC using a custom-designed amplicon sequencing panel for the characterization of mutations and ichorCNA tumor fraction estimates derived from sWGS.Combining both methods was complementary to improve the detection of ctDNA in baseline samples and ctDNA positivity was related to disease stage.The identification of pathological response based on longitudinal sampling was insufficient and hampered by the very low abundance of ctDNA in non-baseline samples of poor responders.ctDNA metrics were prognostic both at baseline and pre-surgery for disease progression and recurrence.
The detection of ctDNA in resectable EAC is improved by combining panel-based sequencing with CNA-estimated tumor fractions from sWGS.In our panel, TP53 was the most commonly found variant.This is in line with results from tumor sequencing of EAC samples and ctDNA detection [6,12].Interestingly, the genes that we selected based on publicly available tumor sequencing data were able to additionally identify ctDNA-positive patients without having to rely only on TP53.This is especially relevant for the further development of non-tumor informed mutation sequencing panels to include a wide array of EAC-specific genes to detect also tumors without a TP53 mutation.Also, the presence of CNAs as measured by ichorCNA, and which are hard to detect by panel-based sequencing, was able to identify ctDNA-positive patients who were not detected by panel-based sequencing.This could be related to either the absence of mutations in the panel sequenced regions or the fact that the CNAs are the main cancer drivers in these patients.Widespread CNAs also in cancer driver genes are important in EAC tumor biology [12,25].A word of caution should be noted in that the detection of CNAs by sWGS in cfDNA samples could have been affected by the polyploidy frequently observed in EAC, although the effect is probably limited for the highly diluted cfDNA samples included in our study.Furthermore, we cannot exclude that stochastic noise was due to the low tumor fraction and sensitivity of both assays; a patient can be ctDNA-positive by chance in only one assay and not in the other.By combining these two approaches mentioned above, we were able to classify 50.5% of baseline samples as ctDNA-positive.This is comparable to previously published studies with baseline detection rates of between 36% and 56% with panel-based sequencing and 30% with ichorCNA in an advanced-stage cohort [6,8,11,14].For the future development of ctDNA assays, combining multiple metrics may thus improve detection.As previously observed, higher cT stage was related to ctDNA positivity.In an EAC cohort where a tumor-informed assay was used, higher T stage (cT3 versus cT2) was also associated with ctDNA positivity [26].Further improvement in sensitivity and specificity is needed to also detect lowerstage patients.Prediction of CR was not possible by baseline ctDNA profiling due to the fact that responders also shed ctDNA before the start of neoadjuvant treatment.In a tissue informed panel-based EAC sequencing study (n = 78), several complete responders were also

292
T van den Ende, Y van der Pol et al positive at baseline, in line with our findings [8].Several studies have found that ctDNA dynamics of EAC patients predicts pathological response to neoadjuvant therapy [10,11,27].For example, non-clearance of ctDNA pre-surgery is associated with an incomplete response [10,11].Our results also showed that tracking cfDNA sequencing from rEAC 293 mutations is specific for an incomplete response but the number of positive classified patients was limited.The detection of ctDNA after nCRT seems to be hampered by lower shedding and possibly more background biological noise as we observed higher cfDNA concentrations in non-baseline samples.To accurately predict CR pre-surgery, ultra-sensitive techniques may be necessary or other methods of detection including sequencing for tumor methylation signatures [28][29][30][31].Personalized tumor-informed assays could also help to boost sensitivity by providing high reliability and a reduction in the amount of targets that need to be sequenced [26].In bladder cancer, an ultra-sensitive approach (median 105,000Â) based on tumor-informed sequencing of a personalized mutation assay was able to classify 45% of patients with residual disease after neoadjuvant chemotherapy as ctDNA-positive and all CR patients as ctDNA-negative [32].Other promising methods include sequencing for cfDNA methylation signatures which already can be leveraged for the detection of cancer [33][34][35][36].Moving forward, sensitive innovative ctDNA approaches in EAC will most likely also need to be combined with, for example, radiologic imaging and endoscopic evaluation of the tumor site to predict pathological response before surgery [37,38].
In line with previous findings in resectable EAC, the detection of ctDNA has prognostic value [6,8,11,26].In our study, the baseline presence of a ctDNA fraction (VAF > 1% or ichorCNA > 3%) was prognostic for early progression and recurrence.Baseline ctDNA status was also associated with recurrence in two recently published EAC sequencing studies [11,26].In other tumor types, such as colorectal carcinoma and melanoma, baseline measurement of ctDNA and high VAF were prognostic for survival and disease progression [22,39,40].The presence and burden of ctDNA thus

294
T van den Ende, Y van der Pol et al seems to be a marker for hematogenous dissemination.Prognostic utility was also observed based on pre-surgery sampling in our EAC cohort with 14% positivity (nonclearance of baseline mutation or ichorCNA > 3%).Previous tumor-informed and tumor-agnostic studies support the strong relationship between ctDNA status posttreatment (10-16% positive) and disease recurrence [6][7][8]26,27].Our results are thus comparable to the literature, and could indicate that further improvement in post-treatment detection may require, for example, repeated sampling to improve detection rates.In this regard, sWGS offers a relatively inexpensive detection method for CNAs, both at baseline and at later timepoints to identify patients at risk of recurrence.Taken together, these results support the exploration of a standardized and validated ctDNA assay in a clinical trial.This assay could be used to guide treatment decisions comparable to the design of a recently performed randomized study in stage II colorectal cancer [41].Measurement of ctDNA post-surgery in resectable EAC could be used to guide adjuvant nivolumab instead of pathological response status [4].Pathological complete responders at risk of recurrence could, if they are ctDNA-positive, still receive nivolumab.On the other hand, patients who are ctDNA-negative will be spared a year-long intensive treatment regimen.Other applicabilities pre-surgery may include omitting surgery in ctDNA-positive patients and offering them treatment escalation in the form of immuno(chemo)therapy. Based on our findings, ctDNA seems to be a useful marker for identifying a small subgroup of patients at high risk of future disease progression irrespective of pathological response status and is therefore suited for the above-mentioned purposes.At this moment, it is not yet sensitive enough to identify every pathological responder in the locally advanced setting, as recently reported, and therefore can not yet serve as a marker to spare surgery in ctDNA-negative patients posttreatment [10].To adequately evaluate and explore the clinical value of ctDNA in EAC and build upon these results, we need clinical trials to explore standardized assays focused on patients at high risk of progression.

Conclusions
Multi-signal analysis from mutation sequencing and ichorCNA tumor fraction estimates can improve the identification of ctDNA in plasma from resectable EAC patients.Longitudinal sampling was not able to accurately identify complete responders pre-surgery.
Early progression or recurrence in a subset of patients could be identified from combining baseline or presurgery ctDNA metrics.Considering these results together, ctDNA positivity in resectable EAC is strongly related to disease recurrence and could be explored to guide treatment decisions.
FM and HWMvL contributed to data interpretation and scientific discussions regarding the results.TvdE drafted the manuscript.All authors contributed to the manuscript and read and approved the final manuscript.

Figure 1 .
Figure 1.Treatment schedule including blood sampling time-points and study workflow.Treatment regimen of the nCRT cohort and the PERFECT trial.Patients enrolled in PERFECT also received anti-PD-L1 immune checkpoint inhibition (ICI) therapy in weeks 1, 4, 7, 10, and 13.Workflow of cfDNA sequencing from EAC patients derived from three time-points.Baseline and pre-surgery samples were sequenced with both a genome-wide approach and an amplicon panel.Post-nCRT samples were only sequenced using the genome-wide approach.

Figure 2 .
Figure 2. Baseline results from panel sequencing and shallow whole-genome sequencing (sWGS).(A) Mutation oncoplot of baseline mutations detected in plasma from resectable esophageal adenocarcinoma (EAC) patients.(B) The variant allele frequency (VAF) per detected mutation in ctDNA-positive patients in order of frequency mutated.(C) The baseline ichorCNA values of EAC patients derived from sWGS.(D) The maximum VAF of a baseline detected variant in ctDNA-positive patients was correlated to their corresponding baseline ichorCNA value.(E) The distribution of T stage according to baseline ctDNA status.Patients were classified as ctDNA-positive if a mutation was detected or the ichorCNA value was above 3%.(F) The distribution of N stage according to baseline ctDNA status.Patients were classified as ctDNA-positive if a mutation was detected or the ichorCNA value was above 3%.

Figure 4 .
Figure 4. Longitudinal ctDNA tracking in relation to survival and disease progression.The colors of the circles and squares indicate the time-point of ctDNA measurement: red, baseline; purple, post-neoadjuvant chemoradiotherapy; and green, pre-surgery.

Table 1 .
Baseline characteristics and clinical data of the 111 included patients.
*In total, 25 patients had a pathological complete response and one patient a ypT0N+ response.†Two patients refused surgery and one patient died before surgery.GEJ, gastroesophageal junction; nCRT, neoadjuvant chemoradiotherapy; PD-L1, programmed death ligand 1; TRG, tumor regression grade.