Dynamic changes in circulating tumor DNA assessed by shallow whole-genome sequencing associate with clinical efficacy of checkpoint inhibitors in NSCLC Caterina Carbonell1,2, Joan Frigola1,2, Nuria Pardo2,3, Ana Callejo2,3, Patricia Iranzo2,3, Augusto Valdivia2,3, Ilaria Priano3, Susana Cedres2,3, Alex Martinez-Marti2,3, Alejandro Navarro2,3, Laura Lenza , Mireia Soleda , Javier Gonzalo-Ruiz2, Ana Vivancos4, Miriam Sanso 1,2 1,2  5, 6 Enric Carcereny , Teresa Mora n , Ramon Amat 6,7 1,2 and Enriqueta Felip 1,2,3 1 Thoracic Cancers Translational Genomics Unit, Vall d’Hebron Institut d’Oncologia (VHIO), Barcelona, Spain 2 Clinical Research Department, Vall d’Hebron Institut d’Oncologia (VHIO), Barcelona, Spain 3 Oncology Department, Vall d’Hebron Barcelona Hospital Campus, Spain 4 Cancer Genomics Laboratory, Vall d’Hebron Institut d’Oncologia (VHIO), Barcelona, Spain 5 Balearic Islands Health Research Institute (IdISBa), Palma de Mallorca, Spain 6 Medical Oncology Department, Catalan Institute of Oncology Badalona, Hospital Universitari Germans Trias i Pujol, Badalona Applied Research Group in Oncology, Institut Germans Trias i Pujol, Barcelona, Spain 7 Department of Medicine, Universitat Auto noma de Barcelona, Spain Keywords Immune checkpoint inhibitors (ICIs) targeting the PD-1/PD-L1 axis are aneuploidy; cfDNA; immune checkpoint the main therapeutic option for patients with advanced non-small cell lung inhibitors; liquid biopsy; NSCLC; shallow cancer (NSCLC) without a druggable oncogenic alteration. Nevertheless, whole-genome sequencing only a portion of patients benefit from this type of treatment. Here, we Correspondence assessed the value of shallow whole-genome sequencing (sWGS) on plasma R. Amat, Vall d’Hebron Institut d’Oncologia samples to monitor ICI benefit. We applied sWGS on cell-free DNA (VHIO), C/Nazaret 115-117, 08035 (cfDNA) extracted from plasma samples of 45 patients with metastatic Barcelona, Spain NSCLC treated with ICIs. Over 150 samples were obtained before ICI E-mail: ramat@vhio.net treatment initiation and at several time points throughout treatment. From E. Felip, Vall d’Hebron Barcelona Hospital sWGS data, we computed the tumor fraction (TFx) and somatic copy Campus, P. Vall d’Hebron 119-129, 08035 number alteration (SCNA) burden and associated them with ICI benefit Barcelona, Spain E-mail: efelip@vhio.net and clinical features. TFx at baseline correlated with metastatic lesions at the bone and the liver, and high TFx (≥ 10%) associated with ICI benefit. Caterina Carbonell, Joan Frigola and Nuria Moreover, its assessment in on-treatment samples was able to better pre- Pardo are joint first authors of this study dict clinical efficacy, regardless of the TFx levels at baseline. Finally, for a Ramon Amat and Enriqueta Felip are co- subset of patients for whom SCNA burden could be computed, increased senior authors burden correlated with diminished benefit following ICI treatment. Thus, our data indicate that the analysis of cfDNA by sWGS enables the moni- (Received 16 December 2022, revised 17 January 2023, accepted 27 February 2023, toring of two potential biomarkers—TFx and SCNA burden—of ICI bene- available online 21 March 2023) fit in a cost-effective manner, facilitating multiple serial-sample analyses. Larger cohorts will be needed to establish its clinical potential. doi:10.1002/1878-0261.13409 Abbreviations ADK, lung adenocarcinoma; aTI, after treatment initiation; cfDNA, cell-free DNA; ctDNA, circulating tumor DNA; DCB, durable clinical benefit; ddPCR, droplet digital PCR; ICIs, immune checkpoint inhibitors; NDB, no durable benefit; NSCLC, non-small cell lung cancer; OS, overall survival; PD-1, programmed cell death 1; PD-L1, programmed death-ligand 1; PFS, progression-free survival; SCC, lung squamous cell carcinoma; SCNAs, somatic copy number alterations; sWGS, shallow whole-genome sequencing; TFx, tumor fraction; TMB, tumor mutational burden. Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 779 Federation of European Biochemical Societies. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. tumor fraction (TFx) and delivers a genome-wide pro- 1. Introduction file of SCNAs [23], thereby enabling estimation of the Immune checkpoint inhibitors (ICIs) targeting the pro- SCNA burden. Thus, to test the value of these two grammed cell death 1 (PD-1)/programmed death- molecular features (TFx and SCNA burden) on ligand 1 (PD-L1) axis have transformed the clinical cfDNA, we assembled a cohort of 45 patients with management of patients with metastatic non-small cell metastatic NSCLC treated with ICIs, extracted cfDNA lung cancer (NSCLC) whose tumors bear no druggable from plasma samples obtained at different time points oncogenic alteration. As a result, many patients are (before and after treatment initiation until disease pro- treated with this type of ICIs, alone, in combination gression) and applied sWGS. We then examined the with chemotherapy or in combination with other association of these molecular features with the patient checkpoint inhibitors [1]. However, a substantial por- clinical profile (i.e., histology, metastatic lesions loca- tion of patients do not benefit from this treatment, tion) and treatment efficacy. hence many studies have attempted to identify molecu- lar and clinicopathologic biomarkers of benefit to 2. Materials and methods these therapies [2–8]. PD-L1 expression assessed by immunohistochemistry is used to select for pembrolizu- 2.1. Patient and non-cancer donor information mab (an anti-PD1 agent) monotherapy in first-line [9,10], although PD-L1 remains a limited biomarker Fifty-six patients with metastatic NSCLC who were predicting ICIs efficacy [4,11]. Tumor mutational bur- going to be treated with ICI-based therapy den (TMB) determined on tissue biopsy is positively (excluding combinations with chemotherapy) at Hospi- associated with ICI benefit [2,12,13] and long-term tal Vall d’Hebron were prospectively enrolled between benefit in NSCLC [3]. Conversely, a high somatic copy January 2017 and June 2019. Clinical information was number alteration (SCNA) burden has been associated retrieved revising electronic clinical information. Clini- with diminished benefit following ICI treatment in this cal data censoring was May 31st, 2022. Patients were disease [14]. Nevertheless, the availability of tumor tis- treated with ICI-based therapy administered as stan- sue samples for comprehensive molecular testing in dard therapy or as part of a clinical trial. Responses advanced NSCLC often can be limiting, particularly to to ICIs treatment were assessed using the Response assess disease evolution. To overcome this limitation, Evaluation Criteria in Solid Tumors version 1.1 guide- molecular profiling of analytes using bodily fluids—so- lines [24]. Metastatic lesions were recorded prior to called liquid biopsy—is emerging as a powerful tool, ICIs treatment initiation. owing to its minimal invasiveness and serial testing Twelve donors without cancer gave blood for this capacity. Utilizing this technology to detect circulating study. Four were males and eight were females. Age was tumor DNA (ctDNA) on cell-free DNA (cfDNA) available for eight donors; median age was 54 years. extracted from plasma has generated promising results. This study involves human participants and was Few studies have found a positive association between approved by the ‘Comite de etica de investigacion con baseline TMB, determined on cfDNA, and ICI benefit medicamentos del Hospital Universitario Vall in advanced NSCLC [15–17]. d’Hebron’ (PR(AG)308/2016). Participants gave writ- Furthermore, several studies have indicated that an ten informed consent to participate in the study. The early on-treatment reduction in the levels of the muta- study methodologies conformed to the standards set tions detected on cfDNA is positively associated with by the Declaration of Helsinki. ICI benefit [17–21]. Additionally, undetectable ctDNA at surveillance time points has been useful to identify 2.2. Blood sample collection and plasma and better stratify long-term responders [22]. Neverthe- processing less, these studies are based on relatively expensive gene-panel sequencing or require previous knowledge Peripheral blood from patients and healthy donors was of the genomic landscape of the tumor to follow one collected in BD Vacutainer EDTA Tubes (10 mL). For or several mutations [droplet digital PCR (ddPCR) or lung cancer patients, blood was collected at baseline, and amplicon-sequencing]. approximately every cycle of treatment (2 or 3 weeks, However, shallow whole-genome sequencing (sWGS) depending on the protocol) until progressive disease. on cfDNA is an inexpensive technique, does not Plasma was extracted within 2 h of blood collection require previous knowledge of specific alterations employing two centrifugation steps of 10 min at 200 g (unlike ddPCR), allows investigators to estimate the and stored at 80 °C. 780 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit maxFracCNASubclone: 0.9, maxFracGenomeSubclone: 2.3. Extraction times selection 0.6, chrs: 1 : 22, chrTrain: 1 : 22. A custom panel of nor- Samples were considered baseline (B) when drawn mals built from 12 blood samples served as a reference. before ICI treatment initiation (up to 10 days prior to For each sample, the GC-Map correction median treatment), T1 when collected between 10 and 25 days absolute deviation (MAD) value was retrieved from after treatment initiation (aTI), T2 between days 30 the ‘.params.txt’ file generated by ICHORCNA. Those and 50 aTI and T3 samples between days 60 and 100 whose segmentation profiles had a MAD above 0.1 aTI. Samples categorized as progression were obtained were discarded. ICHORCNA solutions were manually between 15 days before and 50 days after the date of inspected to identify cases where a suboptimal solution radiological progression. Samples at T1, T2, and T3 had been chosen by the algorithm. When this hap- were only considered if they preceded the date of pened, the apparently optimal solution was forced, as progression. recommended in the software documentation. An amplitude filter of 0.05 was applied to the can- didate SCNAs identified. Thus, we considered to be 2.4. cfDNA extraction gained those regions identified as gained or amplified Plasma samples were thawed in a 37 °C water bath and by ICHORCNA and whose amplitude was greater than centrifuged at 4 °C for 10 min at 16 000 g. cfDNA was 0.05, while we considered to be deleted those regions isolated from 1 to 2 mL of plasma using the QIAampÒ called as deleted by ICHORCNA and whose amplitude Circulating Nucleic Acid Kit (QIAGEN Strasse, Hilden, was lower than 0.05. Germany) and quantified with a Qubit Fluorometer SCNA burden was computed as the sum of the (ThermoFisher Scientific, Eugene, OE, USA). number of base-pairs affected by copy number alter- ation events encompassing whole arms and chromo- somes as in Frigola et al. [14]. 2.5. sWGS sequencing Between 4 and 20 ng of cfDNA was used for the prepa- 2.7. Progression-free survival and overall survival ration of barcoded libraries using the NEBNextÒ computing UltraTM DNA Library Prep Kit (New England Biolabs, Ipswich, MA, USA). For patients with less than 4 ng of Progression-free survival (PFS) was computed as the cfDNA available, all cfDNA extracted from 1 to 2 mL number of days between the ICI treatment initiation of plasma was used for library preparation. Samples and the date of progression in those patients who pre- were subjected to low-coverage whole-genome sequenc- sented a radiological progression. In those who did ing on a HiSeq2500 platform to generate single-end not, PFS was computed as the number of days reads (50 bp) at a target mean coverage of 0.29. between the ICI treatment initiation and the date of last follow-up, where patients’ PFS was censored. Durable clinical benefit (DCB) was defined as 2.6. Shallow whole-genome processing remaining without disease progression for 6 months. 2.6.1. Sequence alignment Overall survival (OS) was computed as the number of days between the ICI treatment initiation and the Sequencing reads were aligned to the GRCh38 refer- exitus date, when the event was considered to have ence genome using the mem algorithm of the BWA occurred. For the remainder of the patients, the OS V.0.7.17 software. Duplicates were marked using the was computed as the number of days from the start of MARKDUPLICATES tool from PICARD V.2.21.2. ICI treatment to the date of the patient’s last follow- up. The OS of these patients was censored at this date. 2.6.2. SCNA calling and processing BAM files were transformed into WIG using HMMCOPY’s 2.8. Survival models [25] readCounter with the window parameter set to Multivariate Cox proportional-hazards models were 1 000 000 and the quality set to 20. Next, ICHORCNA [23] built using the lifelines PYTHON library (10.5281/ was used to call SCNAs, with the following parameters: zenodo.4579431) with the step-size parameter set to 0.5. ploidy: 2, 3, normal: 0.5, 0.6, 0.7, 0.8, 0.9, maxCN: 5, Categorical variables were transformed into dummy includeHOMD: False, chrTrain: 1 : 22, estimateNormal: variables, and numerical variables were standardized. True, estimatePloidy: True, estimateScPrevalence: True, Kaplan–Meier curves were produced using the life- scStates: 1, 3, txnE: 0.9999, txnStrength: 10 000, lines PYTHON library, and log-rank tests to compare Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 781 Federation of European Biochemical Societies. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. curves were performed using the lifelines.statistics log- them exhibited statistical significance (Fig. S1A). How- rank_test function. ever, we found an association between certain meta- static lesions and TFx. Specifically, TFx was higher in patients presenting bone or hepatic metastatic lesions 2.9. Statistical analysis (Mann–Whitney Wilcoxon, P = 0.0062 and P = Mann–Whitney Wilcoxon (MWW), Chi-squared, Wil- 0.0011, respectively, Fig. 2A). coxon signed-rank and Kruskal–Wallis tests were per- Next, we examined the correlation between TFx at formed using the SCIPY PYTHON library [26]. baseline and ICI efficacy. When considering two groups of patients, those with detectable ctDNA and those with undetectable ctDNA, there was a nonsignif- 3. Results icant trend toward inferior PFS and overall survival (OS) in those patients with the former (Log rank test, 3.1. Patient cohort characteristics P = 0.18, Fig. 2B; Log rank test, P = 0.22, Fig. S1B). Of the 56 patients prospectively enrolled, 45 patients When we compared patients with durable clinical ben- had confirmed metastatic NSCLC, were treated with efit (DCB) to those with no durable benefit (NDB), ICIs, had follow-up and at least a baseline plasma TFx was higher in the latter group, although it did not sample, thus were included in the analysis. Thirty-two reach statistical significance (Mann–Whitney Wil- patients (71.1%) were male, 43 (95.6%) were current coxon, P = 0.0938; Fig. 2C). To further study the or former smokers, and 32 (71.1%) had adenocarci- value of TFx at baseline, we used a cut-off of 10% as noma histology. Twenty-one patients (46.7%) were suggested by others [27,28], which can be applied treated in first-line therapy and the rest in second. across different cohorts in contraposition to median or Patients were treated with different PD-(L)1 agents tertiles. Thus, we established the following three cate- alone or in combination with other checkpoint inhibi- gories: undetectable ctDNA, detectable TFx < 10% tors. Median progression-free survival (PFS) and over- and TFx ≥ 10%. When considering PFS as outcome, all survival (OS) were 4.6 (139 days) and 12.8 months patients with a TFx ≥ 10% had shorter PFS than (383 days), respectively. Detailed clinicopathologic those patients with undetectable ctDNA or TFx < characteristics can be found in Tables S1 and S2. 10% (Log rank test, P = 0.0021 and P = 0.00089, We retrospectively performed sWGS (average mean respectively), whereas no differences were observed coverage: 0.30, SD: 0.15) on 173 samples, correspond- between the two latter groups (Log rank test, ing to 45 patients and 12 noncancer donors. Eight P = 0.81, Fig. 2D). Although it did not reach statisti- samples were discarded; two due to insufficient cfDNA cal significance, a similar trend was observed when quality and six because of elevated MAD (see using OS as the clinical endpoint (Fig. S1C, Table S3). methods). Thus, 12 noncancer and 153 tumor high- Only one patient, who had a TFx ≥ 10% at baseline quality samples were used in subsequent analysis. achieved DCB, whereas the rest of the patients with a Among the tumor samples, 40 were taken prior to ICI TFx ≥ 10% (n = 6) did not achieve DCB (Fig. S1D). treatment (referred to as baseline (B)) and 113 after Importantly, after adjusting for several clinical param- treatment initiation (aTI). Specifically, 31 samples were eters, TFx ≥ 10% retained a significant association taken at T1, corresponding to between 2 and 3 weeks with diminished PFS in a multivariate Cox regression aTI (median of 14 days aTI), 26 at T2 (median of model (P = 0.003, Table 1), also when incorporating 41 days aTI), 29 at T3 (median of 83 days aTI) and 27 the presence of metastatic lesions on different tissues near radiological progression (Fig. 1). (P = 0.0037, Table S4). A similar trend was observed when using OS as the clinical endpoint (P = 0.066, Table S5). For 30 patients, PD-L1 status as a binary 3.2. Clinical correlates of baseline cell-free DNA variable (positive vs. negative) was available, thus we tumor fraction adjusted for it in the multivariate analysis. In this We detected ctDNA in 19 patients (47.5%) from model, TFx ≥ 10% remained significant (P = 0.04, whom a baseline (B) sample was available and passed Table S6). the different quality controls (Fig. 1). Their median Altogether, our data indicate that, at baseline, high TFx was 9.2% and the maximum was 52.9%. Seven TFx associates with diminished ICI benefit when using patients had a TFx ≥ 10%. a previously established threshold of 10%. Moreover, First, we studied the relationship between TFx at TFx associates with the presence of certain metastatic baseline and different clinical features, such as histol- lesions, suggesting that it might provide useful prog- ogy, sex, smoking status, line of therapy, and none of nostic information. 782 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit No ctDNA ctDNA < 10% ctDNA ≥ 10% Ongoing Treatment unrelated exitus Radiological progression Individual patient 1084 1160 1162 1203 1225 1342 1810 1810 0 10 40 80 100 200 300 400 500 600 Days since ICIs start Fig. 1. Swimmer plot showing the progression-free survival time, vital status, time points assessed by sWGS and the corresponding value of Tumor Fraction. ctDNA, circulating tumor DNA; ICIs, immune check point inhibitors. undetectable TFx at T1 had substantially longer PFS 3.3. Value of on-treatment determinations than the other two groups (Log rank test, P = 0.000017 Several studies, employing gene-panel sequencing, have and P = 0.12, undetectable vs. TFx ≥ 10% and established the importance of ctDNA kinetics to pre- TFx < 10%, respectively, Fig. 3B), whereas shorter PFS dict therapeutic efficacy [17–20]. We, therefore, applied was observed in the TFx ≥ 10% compared to sWGS on plasma samples drawn at different on- TFx < 10% (Log rank test, P = 0.004, Fig. 3B). Similar treatment time points (Fig. 1). As specified above, we data were obtained when considering OS (Fig. S2B, analyzed samples taken 2–3 weeks aTI (T1), around Table S3). Interestingly, none of the patients with objec- 40 days aTI (T2), and around 80 days aTI (T3). tive response (partial response) had detectable TFx at To address the value of TFx at T1, as we had done T1. When comparing patients who achieved DCB to for baseline, we initially considered two groups of those who did not, TFx was significantly higher in the patients according to whether we detected ctDNA or latter group (Mann–Whitney Wilcoxon, P = 0.0047, not. Contrary to what we had observed at baseline, Fig. 3C), and indeed none of the patients with a patients positive for TFx had shorter PFS and OS TFx ≥ 10% at T1 achieved DCB (Fig. 3D). (Log rank test, P = 0.0023 and P = 0.00028, respec- We then addressed the value of TFx dynamics tively; Fig. 3A, Fig. S2A, Table S3). Next, we exam- between B and T1, therefore, we established three ined ICI benefit across the three groups of patients groups: patients with undetectable TFx at both time using TFx = 10% as the cut-off. Patients with points, patients who had TFx at baseline but cleared Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 783 Federation of European Biochemical Societies. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. (A) P=0.1473 P=0.0062 P=0.0011 P=0.4131 P=0.1663 P=0.3471 P=0.1101 0.5 0.4 Tumor Fraction 0.3 0.2 0.1 0.0 Y N Y N Y N Y N Y N Y N Y N Brain Bone Liver Lung Extrathoracic Pleura Adrenal lymph nodes gland (B) (C) (D) P=0.0938 1.0 1.0 Progression-free probability Progression-free probability TFx(B) Dectectable 0.5 TFx(B) ≥ 10% TFx(B) Undetectable TFx(B) < 10% 0.8 0.4 0.8 Tumor Fraction TFx(B) Undetectable 0.6 0.3 0.6 0.4 0.2 0.4 0.2 0.1 0.2 0.0 0.0 0.0 500 1000 1500 0 NDB DCB 500 1000 1500 0 Time (days) Time (days) No. at risk (censored) No. at risk (censored) 19 3 2 2 7 0 0 0 21 6(1) 6(1) 0(7) 12 3 2 2 21 6(1) 6(1) 0(7) P=0.18 P=0.00089 P=0.81 P=0.0021 Fig. 2. Clinical correlates of Baseline Tumor Fraction (A) Tumor fraction levels according to the presence (Y) or absence (N) of the indicated metastatic lesion. Mann–Whitney Wilcoxon, P-values are shown. The boxes show the quartiles of the dataset, whereas the whiskers extend to show the rest of the distribution, except for points that are determined to be ‘outliers’ by the seaborn library. (B) Kaplan–Meier survival curves of progression-free survival (PFS) in patients with undetectable ctDNA versus detectable ctDNA at baseline. Log rank test, P-value is shown. (C) Tumor fraction according to no durable benefit (NDB) and durable clinical benefit (DCB). Mann–Whitney Wilcoxon, P-value is shown. The boxes show the quartiles of the dataset, whereas the whiskers extend to show the rest of the distribution, except for points that are determined to be ‘outliers’ by the seaborn library. (D) Kaplan–Meier survival curves of progression-free survival (PFS) in patients with undetectable ctDNA, patients whose Tumor Fraction (TFx) was < 10% and patients whose TFx was ≥ 10%. Log rank test, P-values are shown. at T1 and patients with detectable TFx at both time test, P = 0.92 and P = 0.89, PFS and OS, respectively, points. Of note, none of the patients who had undetect- Fig. 3E,F). These data indicate that TFx at T1 may able TFx at B exhibited detectable TFx at T1. Patients reflect ICI benefit more accurately than baseline with detectable TFx at T1 had diminished PFS (and OS) levels. Similar results were obtained when the groups compared to the two groups of patients with TFx unde- were established based on the 10% cut-off (Fig. S2C, tectable at T1, (B detectable and T1 detectable vs. B Table S3). detectable and T1 undetectable, and B detectable and T1 It is worth highlighting that at T1, there was no detectable vs. B undetectable and T1 undetectable, Log association between TFx and the metastasis site, as rank test, P = 0.032 and P = 0.0083, respectively, it was observed at baseline (Fig. 2A, Fig. S2D). In Fig. 3E,F). Interestingly, PFS was similar for those fact, for few patients with liver metastasis (all had patients with undetectable TFx at T1 regardless of their detectable TFx at B) we did not detect ctDNA at TFx at baseline; OS followed a similar trend (Log rank T1. 784 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit Table 1. Multivariate Cox proportional-hazards survival model of have lower TFx (B)), were only included in the progression-free survival at baseline. unpaired analysis. Variable HR HR lower 95% HR upper 95% P-value Histology 3.4. High SCNA burden correlates with Other diminished ICIs benefit ADK 0.71 0.19 2.64 0.61 SCC 1.24 0.26 6.01 0.76 We had previously shown that high SCNA burden (of Smoking those events encompassing arms and chromosomes, No that is aneuploidies), determined by sWGS on tissue Yes 0.79 0.18 3.54 0.76 biopsy, is associated with diminished PFS in patients Sex with NSCLC treated with ICIs [14]. We, therefore, Male assessed its value when determined on baseline plasma Female 2.42 0.95 6.18 0.065 samples. As we had detected reliable ctDNA in only ICIs treatment line First 19 patients, we assessed its value as a continuous vari- Second 4.08 1.52 10.93 0.0052 able, using a univariate cox model. Our data indicate Tumor Fraction (TFx) that SCNA burden of those events encompassing arms TFx < 10% and chromosomes negatively associated with ICI bene- TFX ≥ 10% 4.44 1.66 11.88 0.0030 fit is significant for OS (P = 0.03) and close to signifi- cant for PFS (P = 0.07). When considering two other time points, T2 and T3, 4. Discussion we obtained similar results to those obtained for T1 (Fig. S2E,F, Table S3), although the number of In this study, we performed sWGS on plasma samples patients with plasma samples was lower, since some to evaluate the value of two distinct molecular param- patients had already progressed. Considering these eters—TFx and SCNA burden—as biomarkers of ICI data and because patients may display different benefit in metastatic NSCLC. Owing to the minimally dynamics of response to ICIs, we addressed the value invasive nature of liquid biopsy, biomarkers that can of TFx determination for any of the surveillance time be assessed with this approach have a great potential, points. None of the patients, who had a TFx ≥ 10% since they can be easily determined over the course of at any time point, experienced DCB. In fact, this the treatment. Thus, we applied sWGS on samples patient group had substantially shorter PFS and OS obtained prior to ICI initiation and at several time compared to those patients who did not present a points during treatment until disease progression. TFx ≥ 10% at any of the time points assessed (Log Several studies have explored different analytes rank test, P = 0.000012 and P = 0.000017, respectively, employing liquid biopsy as a tool to evaluate treat- Fig. S3A,B, Table S3). ment efficacy. Specific to NSCLC and ICI benefit, Collectively, these data indicate that the study of most studies have utilized gene-panel sequencing on ctDNA, using sWGS at different on-treatment time cfDNA from plasma. For example, baseline blood points (even at 2–3 weeks aTI), provides valuable TMB (bTMB) has been shown to associate with ICI information with regard to the patient’s clinical benefit benefit [15–17], although its determination requires rel- following ICI treatment, regardless of the patient’s atively large gene-panels and may require correction values at B. for clonal hematopoietic mutations [29,30]. Others Finally, as we had samples drawn at baseline and at used gene-panels to monitor ctDNA kinetics to assess progression, we examined possible changes in the levels ICI benefit in NSCLC [17–20] and found that an of TFx between both time points as the result of the early-on reduction of ctDNA levels is associated with treatment. We observed that samples at progression more favorable clinical outcome. In fact, what has (n = 27) had a trend toward higher TFx than those at emerged from these and other studies is the concept of baseline (n = 40), although the increase did not reach molecular response, as a decrease in the detection of statistical significance (Mann–Whitney U test, P = 0.2, ctDNA (using different thresholds). Fig. S3C). In fact, this trend was not observed when However, fewer studies have employed sWGS on restricting the analysis to the 24 patients with paired cfDNA—particularly in the context of NSCLC or samples from both timepoints (Wilcoxon Signed- ICIs. This approach was used to study PI3K inhibitors Ranked test, P = 0.73, Fig. S3D); this is likely because and chemotherapy efficacy in squamous NSCLC [27] patients, who had not progressed (and who tend to and also in a pan-cancer cohort of patients to study Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 785 Federation of European Biochemical Societies. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. (A) (B) 1.0 1.0 Progressoin-free probability TFx(T1) ≥ 10% Progresion-free probability TFx(T1) Detectable TFx(T1) <10% 0.8 TFx(T1) Undetectable 0.8 TFx(T1) Undetectable 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 500 1000 1500 0 500 1000 1500 0 Time (days) Time (days) No. at risk (censored) No. at risk (censored) 10 1 0 0 5 0 0 0 21 7(1) 7(1) 2(6) 5 1 0 0 P=0.0023 21 7(1) 7(1) 2(6) P=0.040 P=0.12 P=0.00017 (C) (D) P=0.0047 1.0 TFx(T1) ≥ 10% Fraction of patients (T1) 0.3 TFx(T1) < 10% Tumor Fraction (T1) 0.8 7 TFx(T1) Undetectable 0.2 0.6 14 0.4 5 0.1 0.2 3 2 0.0 0.0 NDB DCB NDB DCB (E) (F) TFx(B) Det - TFx(T1) Det TFx(B) Det - TFx(T1) Det 1.0 1.0 Progression-free probability TFx(B) Det - TFx(T1) Und TFx(B) Det - TFx(T1) Und TFx(B) Und - TFx(T1) Und TFx(B) Und - TFx(T1) Und Survival probability 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 500 1000 1500 0 500 1000 1500 0 Time (days) Time (days) No. at risk (censored) No. at risk (censored) 9 1 0 0 9 1 0 0 7 2 2 2 7 5 4 3 12 5 5 0(5) 12 10 5 0(5) P=0.033 P=0.0070 P=0.92 P=0.89 P=0.0083 P=0.00069 786 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit Fig. 3. On-treatment value of Tumor Fraction determination. (A) Kaplan–Meier survival curves of progression-free survival (PFS) in patients with undetectable ctDNA versus detectable ctDNA at T1. Log rank test, P-value is shown. (B) Kaplan–Meier survival curves of progression- free survival (PFS) in patients with undetectable ctDNA, patients whose Tumor Fraction (TFx) was < 10% and patients whose TFx was ≥ 10% at T1. Log Rank test, P-values are shown. (C) Tumor fraction according to no durable benefit (NDB) and durable clinical benefit (DCB) at T1. Mann–Whitney Wilcoxon, P-value is shown. The boxes show the quartiles of the dataset, whereas the whiskers extend to show the rest of the distribution, except for points that are determined to be ‘outliers’ by the seaborn library. (D) Fraction of patients with Undetect- able ctDNA, TFx < 10% and TFx ≥ 10% according to NDB and DCB. (E, F) Progression-free survival (PFS) and overall survival (OS) in patients with undetectable ctDNA at baseline and undetectable ctDNA at T1 (TFx(B) Und - TFx(T1) Und), detectable ctDNA at baseline and undetectable at T1 (TFx(B) Det - TFx(T1) Und), or detectable ctDNA at both (TFx(B) Det - TFx(T1) Det). Log rank test, P-values are shown. ICI benefit [31]. In fact, to the best of our knowledge, biopsy, and ICI benefit. As sWGS on cfDNA enables our study is the first ever to assess the value of apply- us to estimate the SCNA burden, we have assessed its ing sWGS on cfDNA in a cohort of patients with met- value as a biomarker of ICI benefit, using plasma sam- astatic NSCLC treated with ICIs. ples. Although we could only determine the SCNA Our results suggest that the value of TFx, deter- burden in half of our cohort, our data indicate that mined by sWGS, at baseline and in on-treatment sam- the SCNA burden is negatively associated with ICI ples may provide different information. While at benefit, even when adjusting for different features in a baseline, TFx might have a prognostic role, its value multivariate model. Interestingly, a recent study in in the case of on-treatment samples might be predic- urothelial cancer, applying an alternative technique to tive of treatment efficacy, a concept already suggested identify chromosomal aberrations on plasma samples, by Zhang et al. [32] in a pan-cancer cohort of patients also found a negative association between aneuploidy with advanced disease who were treated with check- burden (alterations in the copy number of chromo- point blockade. First, we found that at baseline only some and arms) and pembrolizumab (anti-PD1) benefit high levels of TFx (≥ 10%) correlated (negatively) with [37]. Taken overall, results obtained with tissue sam- ICI clinical efficacy. Conversely, its determination in ples and cfDNA, point to the value of SCNA burden on-treatment samples seems to reflect ICI efficacy (aneuploidy) as a pan-cancer indicator of diminished more closely. In fact, our results indicate that regard- ICI benefit. Nevertheless, its value in other ICI-based less of the levels of TFx at baseline, early-on treatment therapies, such as ICIs plus chemotherapy, remains TFx values were associated with ICI benefit. For unexplored and should be addressed in future studies. instance, none of the patients that presented a As there are no currently reliable biomarkers identified TFx ≥ 10% at any of the surveillance time points that would definitively point to the best therapy for a achieved DCB. Secondly, we found that at baseline, given patient, the value of SCNA burden as a specific there is a positive correlation between TFx and certain biomarker in this respect could, in fact, differ between metastatic lesions (bone and liver), but this association ICI-based therapies. Thus, the search for reliable bio- was lost when treatment was initiated, indicating that markers to support such clinical decisions is essential. TFx on-treatment would reflect treatment efficacy. It is fair to say that our study has several limita- These data align with previous observations using tions. After appropriate quality controls and (blind) gene-panel sequencing, as discussed by Zhang et al. manual curation of the SCNA profiles to maximize [32]. Furthermore, when using sWGS, other investiga- reliability of the data, we detected ctDNA in ~ 50% of tors found a positive association between bone metas- patients in our cohort. This finding aligns with the tasis and TFx in prostate cancer [33] and TFx and concept that sWGS is less sensitive than other tech- liver metastases in breast cancer [28]. niques (gene-panel, amplicon-sequencing or ddPCR). Thus, sWGS may indeed represent a cost-effective Also, it is important to mention, that due to the approach (compared to more expensive gene-panels) to nature of the approach taken (sWGS plus IchorCNA monitor the disease course, allowing multiple serial- [23]), the TFx and the SCNA burden are interdepen- extractions without extensively increasing the financial dent, and thus a combined analysis of both of these costs. Furthermore, it does not require previous parameters was not possible. From a clinical perspec- knowledge of the tumor’s genome, nor is it limited by tive, the patients included in this study received differ- the detection of few mutations such as small panels or ent anti-PD(L)1, and some were treated with ddPCR. combinations of checkpoint inhibitors. Furthermore, Importantly, we in NSCLC [3,14] and others [34–36] owing to the enrollment period of our patients, PD-L1 in other tumor types, have described a negative corre- expression was not available for all the patients and so lation between SCNA burden, determined in tissue we could only evaluate positive versus negative levels. Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 787 Federation of European Biochemical Societies. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. Nevertheless, when adjusting for PD-L1 expression in Conflict of interest a multivariate model, TFx retained its significance. Although it would be beyond the scope of this work, EF reports consulting fees or speaker’s bureau: it would certainly be interesting to establish the value AMGEN, AstraZeneca, Bayer, Bristol-Meyers Squibb, of combinatorial biomarkers and jointly assess TFx Daichi Sankyo, Eli Lilly, F Hoffmann-La Roche, dynamics, SCNA burden, TMB and some immune- Glaxo Smith Kline, Janssen, Medical Trends, Meds- related biomarkers in plasma samples. The basis for cape, the healthcare business of Merck KGaA, Darm- such an assessment would be the suggestion that stadt, Germany, Merck & Co., Kenilworth, NJ, SCNA burden and TMB are likely independent bio- Novartis, Peervoice, Peptomyc, Pfizer, Sanofi, Takeda, markers of response—at least in the case of tumor tis- Touchtime Oncology. Board: Grifols, independent sue samples [3], and that combining some biomarkers member. Research funding: Fundaci on Merck Salud, appeared to effectively predict DCB [17]. Finally, our Madrid, Spain; Grant for Oncology Innovation-the analysis was retrospective and was based on our previ- healthcare business of Merck KGaA, Darmstadt, Ger- ous findings of the value of SCNA burden as a bio- many to Vall d’Hebron Institut of Oncology. CC and marker of ICI benefit, as shown with tissue samples JF: partial support through the Grant for Oncology [14]. Innovation, the healthcare business of Merck KGaA, Darmstadt, Germany. AVi reports consultant or advi- sory role: Guardant Health, Merck, Roche, Bristol- 5. Conclusions Myers Squibb, Incyte, Bayer. Research funding: Bris- Employing sWGS on cfDNA, our analysis suggests tol Myers Squibb, Roche, Incyte. Stock and Other that TFx and SCNA burden are potential biomarkers Ownership Interests Reveal Genomics. EC reports of ICI benefit. TFx determination in on-treatment Consultant or Advisory Role: AstraZeneca, Boehringer samples may be more informative of therapy efficacy, Ingelheim, Bristol-Myers Squibb, Merck & Co., Kenil- whereas its determination at baseline may have more worth, NJ, Novartis, Roche, Takeda. Speaking: Astra- prognostic value (i.e., for certain metastatic lesions). Zeneca, Boehringer Ingelheim, Bristol-Myers Squibb, Considering this and owing to its cost-effectiveness, Merck & Co., Kenilworth, NJ, Novartis, Pfizer, sWGS of plasma samples may very well be a useful Roche, Takeda. Grant support: Merk. Other: Bristol- tool to complement other biomarkers, especially for Myers Squibb, Pfizer, Roche, Takeda. TM declares monitoring the disease course. Moreover, our data consulting/advisory role fees from Roche, Bristol confirm previous observations regarding the negative Myers, Boeringher, AstraZeneca, Lilly, and Research correlation between SCNA burden and ICI benefit. funding from Kyowa Kirin and Janssen, all of them Further studies with larger cohorts and combinatorial unrelated with the present project. SC Consulting or biomarkers will potentially help us to establish its Advisory Role: Bristol-Myers Squibb Recipient, F. value in NSCLC. Hoffmann La Roche AG, Pfizer, Boehringer Ingel- heim, Merck & Co., Kenilworth, NJ, Amphera. Travel Accommodations Expenses: F. Hoffmann La Roche Acknowledgements AG, Pfizer, Boehringer Ingelheim. AC reports advisory We acknowledge the CELLEX Foundation for provid- role and/or travel compensation: Bristol-Myers Squibb ing research facilities and equipment. We thank our Recipient, F. Hoffmann La Roche AG, Pfizer, Boeh- patients and donors for agreeing to participate in this ringer Ingelheim, Merck & Co., Kenilworth, NJ, study. This work was partially supported by the Kyowa Kirin, Celgene, Leo Pharma, Medscape, Kern healthcare business of Merck KGaA, Darmstadt, Pharma. Astra Zeneca. PI reports advisory role and/or Germany (CrossRef Funder ID: 10.13039/100009945; travel compensation: Bristol-Myers Squibb Recipient, Grant for Oncology Innovation to the Vall d’Hebron F. Hoffmann, La Roche AG, Merck & Co., Kenil- Institute of Oncology (VHIO), Barcelona, Spain); the worth, NJ, Boehringer Ingelheim, Merck & Co., Kenil- Insituto de Salud Carlos III through the projects PI20/ worth, NJ, Rovi, Yowa Kirin, Grunenthal Pharma 00987 to EF and PI22/01585 to RA (Co-funded by S.A., Pfizer, Medscape, Kern Pharma. AN reports European Union). The healthcare business of Merck advisory role, speaker’s bureau or travel compensa- KGaA, Darmstadt, Germany reviewed the manuscript tion: Bristol-Myers Squibb, F. Hoffmann La Roche for medical accuracy only before journal submission. AG, Pfizer, Boehringer Ingelheim, Oryzon Genomics, The authors are fully responsible for the content of Pfizer, AstraZeneca. AVa reports consulting or advi- this manuscript, and the views and opinions described sory role: Bristol-Myers Squibb Recipient, F. Hoff- in the publication reflect solely those of the authors. mann La Roche AG, Pfizer, Boehringer Ingelheim, 788 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit Merck & Co., Kenilworth, NJ, Amphera. Travel, advanced non-small cell lung cancer. Mol Oncol. Accommodations, Expenses: F. Hoffmann La Roche 2021;15(4):887–900. AG, Pfizer, Boehringer Ingelheim. NP reports advisory 4 Lu S, Stein JE, Rimm DL, Wang DW, Bell JM, role and/or travel compensation: Merck & Co., Kenil- Johnson DB, et al. Comparison of biomarker modalities worth, NJ, Bristol-Myers Squibb Recipient, F. Hoff- for predicting response to PD-1/PD-L1 checkpoint mann La Roche AG, Pfizer, Boehringer Ingelheim, blockade: a systematic review and meta-analysis. JAMA Grunenthal Pharma S.A Kern Pharma. A.M.M. pro- Oncol. 2019;5(8):1195–204. vided consultation, attended advisory boards and/or 5 Diem S, Schmid S, Krapf M, Flatz L, Born D, Jochum W, et al. Neutrophil-to-lymphocyte ratio (NLR) and speaker’s bureau for the following organizations: platelet-to-lymphocyte ratio (PLR) as prognostic Bristol-Myers Squibb, Lilly, F. Roche, Merck & Co., markers in patients with non-small cell lung cancer Kenilworth, NJ, Pfizer, Boehringer Ingelheim, Astra- (NSCLC) treated with nivolumab. Lung Cancer. Zeneca. All remaining authors declare no conflict of 2017;111:176–81. interest. 6 Alban TJ, Chan TA. Immunotherapy biomarkers: the long and winding road. Nat Rev Clin Oncol. 2021;18 Author contributions (6):323–4. 7 Callejo A, Frigola J, Iranzo P, Carbonell C, Diaz N, Conception and design of the study: RA, JF, CC, and Marmolejo D, et al. Interrelations between Patients’ EF. Investigation: CC, JF, LL, RA, and EF. Acquisi- Clinicopathological characteristics and their association tion of data: CC, LL, and NP. Resources: NP, AC, with response to immunotherapy in a real-world cohort PI, AVi, MSa, AM-M, AN, SC, TM, EC, MSo, AVa, of NSCLC patients. Cancers (Basel). 2021;13(13):3249. and EF. Reviewed the clinical data: NP, AC, AV, PI, 8 Cristescu R, Mogg R, Ayers M, Albright A, Murphy E, IP, EF, AM-M, AN, SC, TM, and EC. Data curation: Yearley J, et al. Pan-tumor genomic biomarkers for CC, AC, MSa, PI, NP, IP, MSo, and JG-R. Formal PD-1 checkpoint blockade-based immunotherapy. analysis: JF. Project Administration: JG-R, CC, and Science. 2018;362(6411):eaar3593. RA. Interpretation of the data: CC, JF, and RA. Writ- 9 Reck M, Rodriguez-Abreu D, Robinson AG, Hui R, ing—review and editing: All authors. Writing—original Cs€oszi T, F€ul€ op A, et al. Pembrolizumab versus draft: RA. Funding acquisition: EF and RA. chemotherapy for PD-L1-positive non-small-cell lung cancer. N Engl J Med. 2016;375(19):1823–33. 10 Reck M, Remon J, Hellmann MD. First-line Peer review immunotherapy for non–small-cell lung cancer. J Clin The peer review history for this article is available at Oncol. 2022;40:586–97. 11 Patel SP, Kurzrock R. PD-L1 expression as a predictive https://publons.com/publon/10.1002/1878-0261.13409. biomarker in cancer immunotherapy. Mol Cancer Ther. 2015;14(4):847–56. Data accessibility 12 Rizvi H, Sanchez-Vega F, La K, Chatila W, Jonsson P, Halpenny D, et al. Molecular determinants of response The data that support the findings of this study are to anti-programmed cell death (PD)-1 and anti- available on request from the corresponding authors. programmed death-ligand 1 (PD-L1) blockade in The data are not publicly available due to privacy or patients with non-small-cell lung cancer profiled with ethical restrictions. targeted next-generation sequencing. J Clin Oncol. 2018;36(7):633–41. References 13 Hellmann MD, Nathanson T, Rizvi H, Creelan BC, Sanchez-Vega F, Ahuja A, et al. Genomic features of 1 Grant MJ, Herbst RS, Goldberg SB. Selecting the response to combination immunotherapy in patients optimal immunotherapy regimen in driver-negative with advanced non-small-cell lung cancer. Cancer Cell. metastatic NSCLC. Nat Rev Clin Oncol. 2021;18 2018;33(5):843–52.e4. (10):625–44. 14 Frigola J, Carbonell C, Iranzo P, Pardo N, Callejo A, 2 Rizvi NA, Hellmann MD, Snyder A, Kvistborg P, Cedres S, et al. High levels of chromosomal aberrations Makarov V, Havel JJ, et al. Mutational landscape negatively associate with benefit to checkpoint determines sensitivity to PD-1 blockade in non-small inhibition in NSCLC. J Immunother Cancer. 2022;10(4): cell lung cancer. Science. 2015;348(6230):124–8. e004197. 3 Frigola J, Navarro A, Carbonell C, Callejo A, Iranzo P, 15 Wang Z, Duan J, Cai S, Han M, Dong H, Zhao J, Cedres S, et al. Molecular profiling of long-term et al. Assessment of blood tumor mutational burden as responders to immune checkpoint inhibitors in a potential biomarker for immunotherapy in patients Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 789 Federation of European Biochemical Societies. sWGS on cfDNA to monitor ICI benefit C. Carbonell et al. with non-small cell lung cancer with use of a next- 28 Stover DG, Parsons HA, Ha G, Freeman SS, Barry generation sequencing cancer gene panel. JAMA Oncol. WT, Guo H, et al. Association of cell-free DNA tumor 2019;5(5):696–702. fraction and somatic copy number alterations with 16 Gandara DR, Paul SM, Kowanetz M, Schleifman E, survival in metastatic triple-negative breast cancer. J Zou W, Li Y, et al. Blood-based tumor mutational Clin Oncol. 2018;36(6):543–53. burden as a predictor of clinical benefit in non-small- 29 Weber S, van der Leest P, Donker HC, Schlange T, cell lung cancer patients treated with atezolizumab. Nat Timens W, Tamminga M, et al. Dynamic changes of Med. 2018;24(9):1441–8. circulating tumor DNA predict clinical outcome in 17 Nabet BY, Esfahani MS, Moding EJ, Hamilton EG, patients with advanced non–small-cell lung cancer Chabon JJ, Rizvi H, et al. Noninvasive early treated with immune checkpoint inhibitors. JCO Precis identification of therapeutic benefit from immune Oncol. 2021;5:1540–53. checkpoint inhibition. Cell. 2020;183(2):363–76.e13. 30 Slavin TP, Banks KC, Chudova D, Oxnard GR, 18 Goldberg SB, Narayan A, Kole AJ, Decker RH, Teysir Odegaard JI, Nagy RJ, et al. Identification of J, Carriero NJ, et al. Early assessment of lung cancer incidental germline mutations in patients with immunotherapy response via circulating tumor DNA. advanced solid tumors who underwent cell-free Clin Cancer Res. 2018;24(8):1872–80. circulating tumor DNA sequencing. J Clin Oncol. 19 Ricciuti B, Jones G, Severgnini M, Alessi JV, Recondo G, 2018;36(35):3459–65. Lawrence M, et al. Early plasma circulating tumor DNA 31 Weiss GJ, Beck J, Braun DP, Bornemann-Kolatzki K, (ctDNA) changes predict response to first-line Barilla H, Cubello R, et al. Tumor cell-free DNA copy pembrolizumab-based therapy in non-small cell lung cancer number instability predicts therapeutic response to (NSCLC). J Immunother Cancer. 2021;9(3):e001504. immunotherapy. Clin Cancer Res. 2017;23(17):5074–81. 20 Raja R, Kuziora M, Brohawn PZ, Higgs BW, Gupta 32 Zhang Q, Luo J, Wu S, Si H, Gao C, Xu W, et al. A, Dennis PA, et al. Early reduction in ctDNA predicts Prognostic and predictive impact of circulating tumor survival in patients with lung and bladder cancer dna in patients with advanced cancers treated with treated with durvalumab. Clin Cancer Res. 2018;24 immune checkpoint blockade. Cancer Discov. 2020;10 (24):6212–22. (12):1842–53. 21 Anagnostou V, Forde PM, White JR, Niknafs N, Hruban 33 Choudhury AD, Werner L, Francini E, Wei XX, Ha G, C, Naidoo J, et al. Dynamics of tumor and immune Freeman SS, et al. Tumor fraction in cell-free DNA as responses during immune checkpoint blockade in non– a biomarker in prostate cancer. JCI Insight. 2018;3(21): small cell lung cancer. Cancer Res. 2019;79(6):1214–25. e122109. 22 Hellmann MD, Nabet BY, Rizvi H, Chaudhuri AA, 34 Davoli T, Uno H, Wooten EC, Elledge SJ. Tumor Wells DK, Dunphy MPS, et al. Circulating tumor aneuploidy correlates with markers of immune evasion DNA analysis to assess risk of progression after long- and with reduced response to immunotherapy. Science. term response to PD-(L)1 blockade in NSCLC. Clin 2017;355(6322):eaaf8399. Cancer Res. 2020;26(12):2849–58. 35 Roh W, Chen PL, Reuben A, Spencer CN, Prieto PA, 23 Adalsteinsson VA, Ha G, Freeman SS, Choudhury AD, Miller JP, et al. Integrated molecular analysis of tumor Stover DG, Parsons HA, et al. Scalable whole-exome biopsies on sequential CTLA-4 and PD-1 blockade sequencing of cell-free DNA reveals high concordance reveals markers of response and resistance. Sci Transl with metastatic tumors. Nat Commun. 2017;8(1):1324. Med. 2017;9(379):eaah3560. 24 Eisenhauer EA, Therasse P, Bogaerts J, Schwartz LH, 36 Lu Z, Chen H, Li S, Gong J, Li J, Zou J, et al. Tumor Sargent D, Ford R, et al. New response evaluation copy-number alterations predict response to immune- criteria in solid tumours: revised RECIST guideline checkpoint-blockade in gastrointestinal cancer. J (version 1.1). Eur J Cancer. 2009;45(2):228–47. Immunother Cancer. 2020;8(2):1–11. 25 Lai D, Ha G, Shah S. HMMcopy: copy number 37 Mendelaar PAJ, Robbrecht DGJ, Rijnders M, de Wit prediction with correction for GC and mappability bias R, de Weerd V, Deger T, et al. Genome-wide for HTS data. R package version 1.28.1 2020. aneuploidy detected by mFast-SeqS in circulating cell- 26 Virtanen P, Gommers R, Oliphant TE, Haberland M, free DNA is associated with poor response to Reddy T, Van Der Walt SJ, et al. Computing in pembrolizumab in patients with advanced urothelial python. 2020;17. cancer. Mol Oncol. 2022;16:2086–97. 27 Chen X, Chang CW, Spoerke JM, Yoh KE, Kapoor V, Baudo C, et al. Low-pass whole-genome sequencing of circulating cell-free DNA demonstrates dynamic Supporting information changes in genomic copy number in a squamous lung Additional supporting information may be found cancer clinical cohort. Clin Cancer Res. 2019;25 online in the Supporting Information section at the end (7):2254–63. of the article. 790 Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of Federation of European Biochemical Societies. C. Carbonell et al. sWGS on cfDNA to monitor ICI benefit Fig. S1. Overall survival results based on ctDNA Table S3. Kaplan–Meier pairwise statistics for the dif- detection at baseline. ferent conditions related to supplementary figures. Fig. S2. On-treatment values of Tumor Fraction. Table S4. Multivariate Cox proportional-hazards sur- Fig. S3. On-treatment values of Tumor Fraction and vival model (Progression-free survival at baseline). Tumor Fraction comparison between baseline and pro- Table S5. Multivariate Cox proportional-hazards sur- gression samples. vival model (Progression-free survival at baseline). Table S1. Summary of clinical characteristics of the Table S6. Multivariate Cox proportional-hazards sur- cohort. vival model (Progression-free survival at baseline). Table S2. Detailed clinical characteristics of individual Data S1. Legends. patients. Molecular Oncology 17 (2023) 779–791 Ó 2023 The Authors. Molecular Oncology published by John Wiley & Sons Ltd on behalf of 791 Federation of European Biochemical Societies.