Kinetics of plasma cfDNA predicts clinical response in non-small cell lung cancer patients

Tyrosine kinase inhibitors (TKIs), VEGF/VEGF receptor inhibitors (VEGFIs) and immune checkpoint inhibitors (ICIs) have revolutionized the treatment of advanced cancers including non-small-cell lung cancer (NSCLC). This study aims to evaluate the utility of plasma cell-free DNA (cfDNA) as a prognostic biomarker and efficacy predictor of chemotherapy (CT) with or without these precision therapies in NSCLC patients. Peripheral cfDNA levels in 154 NSCLC patients were quantified before and after the first target cycle of chemotherapy. The correlations of cfDNA with tumor burden, clinical characteristics, progression-free survival (PFS)/disease-free survival (DFS), objective response ratio (ORR), and therapy regimens were analyzed respectively. Baseline cfDNA, but not post-chemotherapeutic cfDNA, positively correlates with tumor burden. Notably, cfDNA kinetics (cfDNA Ratio, the ratio of post-chemotherapeutic cfDNA to baseline cfDNA) well distinguished responsive individuals (CR/PR) from the non-responsive (PD/SD). Additionally, cfDNA Ratio was found negatively correlated with PFS in lung adenocarcinoma (LUAD), but not lung squamous-cell carcinoma (LUSC) which may be due to a limited number of LUSC patients in this cohort. LUAD patients with low cfDNA Ratio have prolonged PFS and improved ORR, compared to those with high cfDNA Ratio. When stratified by therapy regimen, the predictive value of cfDNA Ratio is significant in patients with chemotherapy plus VEGFIs, while more patients need be included to validate the value of cfDNA Ratio in other regimens. Thus, the kinetics of plasma cfDNA during chemotherapy may function as a prognostic biomarker and efficacy predictor for NSCLC patients.

, and large-cell carcinoma (LULC, 5-10%) 3 . The 3-year or 5-year overall survival (OS) of early stage (I and II) NSCLC patients undergoing resection has reached to 83% and 76% respectively 4 . Despite multiple treatment options, the 5-year OS of late stage NSCLC remains extremely low 5 , with over 50% die within one year following diagnosis 6 . Unfortunately, over one third of NSCLC cases are diagnosed at late stage (III and IV) 2 . Advanced NSCLC patients are increasingly benefitting from targeted therapies and immunotherapies 7,8 . These therapies seem to produce some synergistic effects when combined with chemotherapy 9 . Thus, there is an urgent need to discover biomarkers that can assist in selecting optimal treatment, predicting response and prognostics to improve the clinical outcome of NSCLC patients. Genotyping tumor tissue with next generation sequencing (NGS) represents an effective way to capture actionable genetic alterations as potential biomarkers in clinical oncology 10 . However, tissue biopsy may be limited due to insufficiency of sampling or inaccessibility for biopsy and only 25-50% of lung cancer patients have sufficient tissues for genotyping 11 . Complicating biopsy availability, the biopsy represents a single snapshot in time and is often a sample from a heterogenous location within the tumor 12 , obtaining repeat specimens for genetic analysis before and after treatment is logistically difficult 13 . Therefore, liquid biopsy or blood sample becomes an alternative source and promising technology for genotyping. Increasingly, concordance has been established between liquid-and tissue-based genomic screenings 14 . Of note, some studies have suggested that liquid biopsy, specifically cell-free DNA (cfDNA), may better capture the heterogeneity of certain cancer features such as acquired resistance [15][16][17] , and could be useful to monitor tumor burden and metastasis 18 .
Emerging data have demonstrated that liquid biopsy-based biomarkers may serve as indirect indicators for NSCLC diagnosis and treatment monitoring, including circulating tumor cells (CTCs) 19,20 , circulating free tumor DNA (ctDNA) 21,22 , exosomes 23 and tumor-educated platelets (TEP) 24 . However, none of these platforms are perfect. All the above methods still have issues, making none fully satisfactory. For instance, limited CTCs detection efficiency is low, with only 32% of NSCLC patients having ≥ 2 CTCs using CellSearch (the only approved methodology by the U.S. Food and Drug Administration) 25 . Low quantities of ctDNA in blood and sequencing artifacts may debilitate the confidence of NGS applications in detecting the actionable mutations 26 . Both CT and ctDNA are relatively time consuming and not cost-friendly for daily clinical practice. cfDNA, on the other hand, is relatively abundant and easier to quantify in circulating blood. Though the majority of cfDNA is often not of cancerous origin, preliminary studies suggest that cfDNA level and kinetics may still be used to assist in cancer diagnosis, treatment response or prognostic prediction [27][28][29][30][31][32] .
However, the clinical value of cfDNA application in NSCLC has not been well-established due to inconsistent reports [33][34][35][36] . Our recent study confirmed that plasma cfDNA concentration was significantly increased in patients with advanced gastric cancer and can serve as a potential biomarker for chemotherapy monitoring 37 . Here we sought to investigate the predictive value of cfDNA in efficacy of treatment and prognosis for NSCLC patients with chemotherapy, targeted therapy, immunotherapy or combined treatment.

Results
Pathological and demographic characteristics. The pathological and demographic characteristics of the 154 patients were summarized (Table 1). The median age was 62 years (34-79); 107 (69%) of participants were male and 47 (31%) were female. Since our aim is evaluating cfDNA clinical utilization, the most typical Peripheral cfDNA baseline correlates with tumor burden. To assess the relationship between cfDNA and TB, we defined TB_baseline as the pre-treatment TB, and only selected those whose interval between cfDNA test and TB evaluation was within 7 days (N = 80). We defined TB_post-chemotherapy as the post-chemotherapeutic TB and restricted the interval between cfDNA test and TB evaluation to no more than 7 days (N = 47). Overall, a weakly positive correlation between TB and cfDNA was observed at baseline (N = 80, Pearson's coefficient = 0.24; 95% confidence interval (CI) 0.017-0.433; P = 0.03, Fig. 1A), while no significant correlation was found for post-chemotherapy (N = 47, Pearson's coefficient = 0.124; 95% CI − 0.169 to 0.397; P = 0.4, Fig. 1B).
In addition, we also assessed other clinical factors which may be correlated with cfDNA. No significant correlations were found between age and cfDNA either at baseline (P = 0.1) or post-chemotherapy (P = 0.4), stage at baseline (P = 0.9) or post-chemotherapy (P = 0.4), ECOG score at baseline (P = 0.8) or post-chemotherapy (P = 0.8), gender at baseline (Wilcoxon rank sum test, P = 0.5) or post-chemotherapy (Wilcoxon rank sum test, P = 0.4). We also found no significant difference of cfDNA between LUAD and LUSC at baseline (Wilcoxon rank sum test, P = 0.16) or post-chemotherapy (Wilcoxon rank sum test, P = 0.4), or among different therapy regimens (Supplementary Table S1).
Plasma cfDNA relates to objective response rate (ORR) and progression-free survival (PFS)/ disease-free survival (DFS). Since tumor burden usually correlates with clinical outcomes, we then investigated the relationship between clinical outcomes and peripheral cfDNA, we monitored peripheral cfDNA of all available patients (N = 154) at baseline (79% of which were tested before chemotherapy by 0-7 days), postchemotherapy (89% of which were tested after chemotherapy by 20-30 days) and derived cfDNA Ratio (the ratio of post-chemotherapeutic cfDNA to baseline cfDNA) for each patient.
Firstly, we compared the baseline cfDNA and post-chemotherapeutic cfDNA between responsive group (PR/ CR, N = 56) and non-responsive group (SD/PD, N = 80). Overall, the responsive group trended toward higher baseline cfDNA (median 17.68 ng/mL) than the non-responsive (median 13.70 ng/mL) (P = 0.058, Wilcoxon rank-sum test, Fig. 2A). However, we found no significant difference in post-chemotherapeutic cfDNA between the two (P = 0.6, Wilcoxon rank-sum test, Fig. 2B), although the median post-chemotherapeutic cfDNA in the responsive (17.18 ng/mL) was modestly lower than that of the non-responsive (19.15 ng/mL). Notably we found www.nature.com/scientificreports/ a significantly lower ratio in the responsive group (median 0.87) than that of the non-responsive (median 1.21) (P = 0.012, Wilcoxon rank-sum test, Fig. 2C). These data suggested that cfDNA can be used to discriminate responsive patients from the non-responsive well, especially with cfDNA ratio which reflected the dynamic change of plasma cfDNA.
To better evaluate the utility of cfDNA as a predictive tool, we divided this cohort into Ratio_low and Ratio_ high group by the median of cfDNA Ratio (1.0271). Comparative analysis was then carried out between these two groups. Similar procedure was also performed between Baseline_low and Baseline_high group (cut-value: the median of cfDNA baseline, 15.43 ng/mL) and between Post-chemotherapy_low and Post-chemotherapy_high group (cut-value: the median of post-chemotherapeutic cfDNA, 18.42 ng/mL), respectively.
Stratification analysis by subtype. Since LUAD and LUSC are two main pathologic subtypes of NSCLC with different clinical managements and prognostics, we further analyzed the prognostic significance of the cfDNA baseline, cfDNA post-therapy, and cfDNA ratio in these two subgroups, respectively.
The LUSC group, as a minority of our cohort (N = 26), no significantly improved PFS/DFS benefit was found for the Ratio_low group (HR: 1.12 (95% CI 0.27-4.91); P = 0.85, Fig. 4A) compared with Ratio_high group. With only 2 patients followed past 300 days, both in the Ratio low group, the median PFS/DFS was 4.9 months in the Ratio low compared to 6.8 months in the Ratio_high group. However, ORR of Ratio_low group (41.1%) was trended higher than that of the Ratio_high group (22.2%).
In addition, no significant difference of PFS/DFS was found between Baseline_low and Baseline_high group (Fig. 5B) or Post-chemotherapy_low and Post-chemotherapy_high group (Fig. 5C) when stratified by these therapy regimens, respectively.

Discussion
Over the past two decades, important advancements have been achieved in the treatment of advanced NSCLC with our increasing understanding of the disease biology, tumorigenesis, early detection and multimodal care 2 . Notably, the utilization of targeted therapy and immunotherapy has brought about remarkable survival benefits in selected patients 7,8 . However, there is a lack of universal and reliable biomarkers to predict or evaluate the treatment response and prognosis of different managements. In this study, we confirmed that the kinetics of plasma cfDNA (Ratio, post-/pre-) is well correlated with clinical response (ORR) and progression free survival (PFS)/disease free survival (DFS) at least in chemotherapy with VEGF inhibitor targeted therapy.
Circulating cfDNA is derived from a combination of apoptosis, necrosis and active secretion from both cancer cells and normal cells which are subjected to harsh stimuli such as chemotherapy 38 or driven by inflammatory process 39 . It was found at higher levels in patients with advanced cancer than in early stage disease or healthy individuals 40,41 . In the present study, we found a positive correlation between tumor burden and cfDNA baseline in NSCLC (Fig. 1). Although the tumor-derived fraction of these total cfDNA (ctDNA) has been widely investigated as a prognostic biomarker in various cancer types including breast, colon and lung cancer 18,[42][43][44] , the main challenges are low amount of ctDNA, detection cost and reproducibility limitations. For example, some typical difficulties of NGS application in this scenario include inadequate analytical sensitivity and specificity, such as detection limit of low allelic frequencies, and sequencing false positive 41,45 . The total cfDNA with its higher feasibility has become an attractive alternative biomarker [15][16][17] . Our study utilized proven fluorescent probes and quick turnaround time of the SuperbDNA technology to measure plasma total cfDNA 37,46 . This technology enables the cfDNA in patients' plasma to be detected directly without any isolation procedures, which avoids any cfDNA loss during conventional isolation process and make the assay more accurate. Indeed, the Ratio (post-chemotherapy/pre-chemotherapy) of cfDNA shows a correlation with clinical response. The responsive patients obviously have a much lower Ratio than those with no response (Fig. 2). In addition, the Ratio, but not baseline or post-chemotherapy level of cfDNA, has a reversed correlation with PFS/DFS evaluated by RECIST1.1 (p = 0.05), combining all cases regardless of therapy regimens they received (Fig. 3). With Ratio cutoff-value set at the median (1.03), Ratio_low group has a significantly improved PFS with 2 months longer than that of Ratio_high group (4.1 months) (Fig. 3) 33,36,47 , our data revealed that the response of cfDNA is an effective treatment efficacy indicator. To avoid potential effects of predefined cutoff-value, we also performed the correlations of PFS/DFS with each individual cfDNA baseline, post-chemotherapy and Ratio. Similarly, cfDNA Ratio, but not baseline or post-chemotherapy, was significantly negatively related with PFS/DFS (P = 0.01) in univariate analysis. Interestingly, when stratified by pathohistology, the predicted value of the cfDNA Ratio was only significant in the LUAD group. While the LUSC group was much smaller with only 2 patients exhibiting tumor progression, larger studies are needed to determine the utility of the cfDNA Ratio in LUSC patients. Among different therapy regimens, the strong negative correlation between PFS/DFS and Ratio was reproduced in patients with chemotherapy plus VEGFIs (P = 0.02), but not chemotherapy only (P = 0.74), chemotherapy plus TKIs (P = 0.68) or ICIs (P = 0.59). It may be attributed to a relative short term of follow-up, insufficient case number or non-molecular preselection, since reports have shown that TKIs mostly benefit NSCLC patients with driver (such as EGFR) mutations 48,49 and ICIs usually take a longer time to be clinically effective 50 . Further study targeting molecularly selected patients with a larger scale and longer follow-up is needed for validations. In terms of clinical treatment response, objective response rate (ORR) showed a similar pattern as PFS/DFS. Only cfDNA Ratio, not baseline or post-chemotherapy, distinguished the subgroups who had a better clinical response and beneficial outcomes. Our results are consistent with a previous report that monitoring plasma DNA during chemotherapy can identify patients who are likely to exhibit a therapeutic response 51 . Other studies, however, suggested that cfDNA concentration is not reliable enough to predict treatment response in NSCLC when treatment is chemotherapy 36,52 . One possible explanation is that sensitivity to chemotherapy and cfDNA levels during treatment may vary among individuals, or depend on timing of the sample acquisition 34 . We selected evaluating cfDNA level after one cycle of chemotherapy (post-treatment 20-30 days) based on the consideration that cfDNA would remain relatively stable during cycles and early evaluation could allow for therapeutic adjustment if needed. PFS/DFS can be measured but the results are too late to allow for therapeutic modification. ORR is a quicker index but still requires an imaging cycle and detailed image evaluation. The cfDNA Ratio is measured after the first cycle and is immediately interpretable, allowing for real time treatment adjustments. The Ratio-low group enjoyed an ORR more than 1.5 times higher than that of Ratio-high group (42.8% vs 28.5%) regardless of treatment regimen. The effect was most pronounced in the chemotherapy plus VEGFIs group, only 8.6% of patients in Ratio-low group had disease progressed (PD), while in Ratio-high group the proportion increased to 23.3%. These data support the predictive role of cfDNA Ratio in efficacy of chemotherapy.
To our knowledge, this is the first study to utilize the concept of cfDNA Ratio to better aid personalized medicine management. The concentration of cfDNA varies among individuals based on personalized nuances of the physiology and tumor characteristics 53 . Using a cfDNA ratio, captured in an appropriate time interval, normalizes the physiological effects leading to an estimate of tumor response. Indeed, from our data, the snapshot of baseline cfDNA did correlate with some clinical parameters like tumor burden. Yet for clinical response (ORR) and prognostic prediction, the cfDNA ratio seems to provide a better measure of tumor response.
This pioneer study has several limitations. (1) Neither ORR nor PFS/DFS are fully predictive of overall survival (OS). However, both ORR and PFS/DFS are used clinically to alter therapy, and an even earlier measure of tumor response would be advantageous. (2) Basal release and accumulation of cfDNA in the plasma, as mentioned above, is not an identical for every tumor or every patient. We have provided data supporting its imperfect potential for measuring basal TB in gastric cancer 37 and now in NSCLC. As other common tumor markers (ex. CEA, PSA), more tumor subtypes should be screened for further validations. (3) cfDNA quantification is quick, accurate, and inexpensive, but it is not specific for cancer. Other pathological conditions such as inflammation and tissue necrosis can also affect cfDNA level. Logically ctDNA or other tests, if they can be made quantitative and reliable would be useful as adjuncts to calibrate the cfDNA test. (4) Small sample size in some groups (e.g. females and LUSC) could be a potential limitation for not observing significant differences. More female and/or LUSC patients are needed to validate the significance of cfDNA kinetics in different clinical settings. (5) Genetic variability could limit the such cfDNA based measurements. Nevertheless, in clinical settings, the genetic variability measurement currently relies on either NGS panel or specific mutation quantitation (qPCR), which requires prior knowledge of the disease or/and individual status. The cfDNA measurement in current study aims to overcome these inconvenience and provide an alternative way to predict clinical response and prognostics.
As with most initial discoveries, this is a single institution study that promises to advance a simple test that can provide an early indicator of NSCLC response to a number of different systemic therapies. We believe it should advance to a larger, multi-center trial.  Assessment of peripheral cfDNA. All patients were subjected to peripheral blood samples collection before (baseline) and after (post-therapy) the first target cycle of chemotherapy. The cfDNA concentration was determined by QuantiDNA Direct cfDNA Test Kit (Diacarta. Inc., CA, USA) according to the manual and our previous publication 37 . The method is based on a patent technology with convenience and cost-effective. In brief, 2-3 ml peripheral blood was drawn and subjected to 10 min centrifugation in 1900×g for plasma isolation. The plasma sample were centrifuged 10 min at 13,000×g in 4C. Plasma samples were first diluted at tenfold by adding 10 µL of plasma into 90 µL of 1 × PBS (pH7.4). Diluted plasma samples were heated at 95 °C for 5 min for DNA denaturation and then immediately chilled on ice. Next, 20 µL of prepared plasma samples were loaded to a 96-well microplate (Greiner Bio-One, USA) together with 80 µL of Working Probe Solution containing Lysis buffer, DNA probe set, Blocking reagent, and Proteinase K. The microplate was incubated at 55 °C overnight (15-18 h) with shaking at 600 rpm followed by sequential hybridization with Pre-amplifier probe ( 55 was the primary outcome that was defined as the days from the date of initial chemotherapy until the date of progressive disease, recurrence, death, or the last follow up if progression or death had not occurred. www.nature.com/scientificreports/ Statistical analysis. We stratified the treatment evaluation by dug combination regiment which consisted of four groups: (1) chemotherapy only; (2) chemotherapy plus VEGF/VEGF receptor inhibitors (VEGFIs); (3) chemotherapy plus tyrosine kinase inhibitors (TKIs); and (4) chemotherapy plus immune checkpoint inhibitors (ICIs). The primary outcome was (1) progression-free survival (PFS)/disease-free survival (DFS); and secondary outcomes was (2) objective response ratio (ORR), defined as the proportion of CR and PR in all subjects. An initial model without interactions was used to identify the prognostic impact of baseline cfDNA, post-therapy cfDNA, and the cfDNA ratio respectively. Other demographic or clinical factors which may be associated with PFS/DFS were also evaluated via univariate Cox model separately and multivariate Cox model together. Survival curves were plotted by the Kaplan-Meier method with R package 'survival' and 'survminer' .