Inhibition of IL1β by Canakinumab May Be Effective against Diverse Molecular Subtypes of Lung Cancer: An Exploratory Analysis of the CANTOS Trial Free

Inflammation is the body's natural defense mechanism against infection and injury. However, in a chronic illness such as cancer, a prolonged period of inflammation has the unintended and paradoxical effect of promoting tumor growth. Indeed, tumor-promoting inflammation is now recognized as a hallmark of cancer that enables nascent tumors to acquire other characteristics of the disease (1). Multiple inflammatory pathways have been implicated in oncogenesis, including the IL1β and TNFα pathways (2, 3). IL1β is a cytokine activated by the inflammasome, a multiprotein complex that initiates downstream inflammatory pathways in response to endogenous danger signals. Upon stimulation, the inflammasome activates proteolytic processing of precursor proteins, which in turn produces the biological active forms of key proinflammatory cytokines IL1β and IL18.

Extensive preclinical data exist to support the role of IL1β in various stages of cancer development and progression, including tumor initiation, promotion, angiogenesis, and metastasis (2, 4, 5). In a mouse model of lung cancer metastasis, it was demonstrated that tumors genetically programed to express high levels of IL1β developed lung metastasis more rapidly than controls, and treatment with an anti-IL1β antibody inhibited formation of lung metastasis (6). In addition, it is well-established that smoking, a lung cancer risk factor, induces chronic pulmonary inflammation, which is thought to be one of the mechanisms that drives tumor promotion (7). Consistent with this notion, elevated levels of inflammatory markers, such as C-reactive protein (CRP), in cancer-free individuals are associated with an increased lung cancer risk (8, 9). In addition, polymorphisms in inflammation-related genes, including IL1β, have been shown to be associated with increased lung cancer risk in multiple studies (10). Together, these observations suggest an important role for inflammation in the development of lung cancer, and have raised the hypothesis that IL1β inhibition might be an attractive therapeutic target (11).

Canakinumab Anti-inflammatory Thrombosis Outcomes Study (CANTOS) was a randomized trial designed to assess the role of IL1β inhibition in preventing recurrence of cardiovascular events in patients with atherosclerosis (12–14). All the patients enrolled in CANTOS (N = 10,061) had a history of previous myocardial infarction and a baseline CRP level of 2 mg/L or greater; patients with known current cancer or with a history of cancer other than basal-cell skin carcinoma were excluded from the study. Interestingly, although the study was primarily designed as a cardiovascular outcomes trial (12), a prespecified safety analysis of the trial data also revealed that treatment with canakinumab was associated with a dose-dependent reduction in lung cancer incidence (HR = 0.33; 95% CI, 0.18–0.59; P < 0.0001 for canakinumab 300 mg group compared with placebo) as well as lung cancer mortality (HR = 0.23; 95% CI, 0.10–0.54; P = 0.0002 for canakinumab 300 mg group compared with placebo; ref. 13). This observation provided proof-of-principle that IL1β inhibition may potentially benefit patients with lung cancer.

To gain further insights into the role of IL1β in driving protumor inflammation and lung cancer development, we performed a molecular characterization of plasma samples collected from patients diagnosed with lung cancer during the CANTOS trial. To determine whether these patients might have harbored undiagnosed cancer at or close to trial enrollment, we performed circulating tumor DNA (ctDNA) analysis using plasma samples collected from these patients at baseline and at a time point closer to cancer diagnosis. In addition, plasma and serum from the patients with lung cancer in CANTOS was profiled for circulating inflammatory biomarkers to assess whether the baseline and changes in levels of these biomarkers were associated with the incidence of lung cancer.

Among the 10,061 randomly assigned patients in the CANTOS study, 135 lung cancer cases were previously reported by the site investigators (12). In 19 of these individuals, the diagnosis was not definitive and were therefore excluded from further analysis in this study. Of the remaining 116 patients, plasma samples from 71 patients were available with appropriate consent to allow for ctDNA analysis. EDTA/aprotinin plasma samples collected from two different time points were tested, including the baseline time point collected at trial randomization, and the time point closest to individual patient's lung cancer diagnosis. Samples from additional time points between the two described above were also tested in 28 patients. The CANTOS study was conducted in accordance with the ethical principles laid down in the Declaration of Helsinki and the guidelines for Good Clinical Practice. All patients provided written informed consent before screening.

EDTA plasma samples collected at trial randomization from 20 noncancer CANTOS patients with appropriate consent were also tested for ctDNA. None of these 20 patients were diagnosed with cancer during the course of the study, and their CRP level at randomization was less than 10 mg/L. These patients were propensity matched with the 71 patients with lung cancer included in this study in terms of sex, age, and smoking status.

ctDNA analysis was performed using the Guardant360 73-gene panel NGS assay (Supplementary Table S1). v2.10 of the assay was used in the testing of the 71 patients with lung cancer, and v2.11 was used for the 20 noncancer patients at a later time. Both versions have a 95% limit of detection ≤0.3% variant allele fraction (15). Minor differences between the versions are listed in Supplementary Table S1. The bioinformatics analysis of both the lung cancer and noncancer patients were standardized using the v2.11 bioinformatics pipeline by the vendor. ctDNA positivity was defined using the assay's default threshold.

Serum or plasma concentrations of IL6, IL1 receptor antagonist (IL1RA), fibrinogen, adiponectin, leptin, IL18, TNFα, and plasminogen activator inhibitor-1 (PAI1) were measured by immunoassays in a subset of patients in CANTOS, whereas CRP was measured in all patients, as described previously (13).

Summary statistics were performed to describe the prevalence of ctDNA, distribution of mutations and longitudinal changes at different time points in the CANTOS patient with lung cancer population. Two-sided Fisher exact test with matching confidence intervals was used to calculate the P value of ctDNA prevalence between the lung cancer and matched noncancer patients. For the comparison of allelic frequency (AF) between lung cancer and matched noncancer patients, the maximum AF was used for patients with multiple mutations identified, and zero was used for patients with no detected ctDNA mutation. Maximum AF was used because each subject had different number of mutated genes. Unlike median AF, maximum AF is independent on the number of mutated genes for each subject. Wilcoxon rank sum distribution test was used to calculate the P value of maximum AF distributions between the two patient populations. Box plots were performed to compare the distribution of time to onset based on patients with lung cancer with mutations and without mutations, as well as based on quartiles of baseline CRP and IL6. Wilcoxon analysis was performed to compare the distribution of time to lung cancer occurrence between lung cancer patients with and without ctDNA at baseline.

Samples from 71 of 116 CANTOS patients with lung cancer were available and with appropriate consent for ctDNA analysis. During the CANTOS trial, blood samples were collected for all patients at a series of predefined time points. Of the 71 patients, 67 had plasma samples available from baseline and 71 patients had samples available from the time point closest to their lung cancer diagnosis, resulting in a total of 67 patients who had samples from both time points. Among the 71 patients with a sample collected at the time point closest to lung cancer diagnosis, 61 patients had their samples collected at a median of 221 days before lung cancer diagnosis and 10 patients had their samples collected at a median of 97 days after lung cancer diagnosis. Samples from additional time points for 28 patients were submitted for ctDNA analysis to confirm ctDNA detection in these patients. All 71 CANTOS patients with lung cancer in this analysis were either current or former smokers, with other baseline clinical characteristics similar to the rest of the CANTOS population. At the time of diagnosis, 41 of these 71 patients with lung cancer (58%) had stage III to IV disease, even though they were not clinically diagnosed with cancer before study enrollment (Table 1). This is consistent with previous observations that up to 70% of patients with lung cancer present with locally advanced or metastatic disease at initial diagnosis due to challenges in lung cancer diagnosis (16, 17). The distribution of histological subtypes in these patients was typical of the general lung cancer population (15). In addition, ctDNA analysis was performed using baseline samples from 20 CANTOS patients who were not clinically diagnosed with cancer during the study. These noncancer patients were propensity matched with the 71 patients with lung cancer included in this study in terms of sex, age, and smoking status.

We hypothesized that ctDNA would be more readily detectable in the samples collected from CANTOS patients who developed lung cancer during the study, compared with those who were not clinically diagnosed with cancer within the study follow-up period. Interestingly, when including all of the mutations reported, 40 of the 67 (60%) patients with lung cancer samples from the baseline time point were found to be ctDNA positive with a median AF of 0.29%, whereas 14 of 20 (70%) of noncancer patients samples from baseline were also ctDNA positive with a median AF of 0.25% (P = 0.44 for the percentage of ctDNA positive patients; P = 0.45 for AF). Because the Guardant360 ctDNA assay does not differentiate between tumor mutations and clonal hematopoiesis of indeterminate potential (CHIP) mutations, it is possible that some of the mutations found were CHIP mutations. To prioritize somatic mutations detected in each set of samples that are more likely to have a strong clinical significance, we reanalyzed the ctDNA results including only mutations commonly associated with cancer from the Catalogue of Somatic Mutations in Cancer (COSMIC) database. Using this approach, 29 of the 67 (43%) patient with lung cancer samples were ctDNA positive at baseline with a median AF of 0.44%, and only 4 of the 20 (20%) of the noncancer patients samples were ctDNA positive with a median AF of 0.18% (P = 0.07 for the percentage of ctDNA positive patients; P = 0.03 for AF; Table 2).

We used the same approach of including only COSMIC mutations in the analysis of ctDNA prevalence in samples collected at other time points from the CANTOS patients with lung cancer. We found that of the 71 patient samples from the time point closest to lung cancer diagnosis, 36 (51%) were ctDNA positive with a median AF of 0.41% (Table 2). Overall, of the 71 patients assayed, 46 (65%) had detectable ctDNA in at least one of the time point samples, with a median AF of 0.46%. When all patients, time points, and mutation types were considered, the number of ctDNA mutations detected per patient ranged from 1 to 8, with an average of 1.7 mutations detected per patient.

Further, we compared the mutation profiles of the 58 CANTOS patients with lung cancer who had detectable COSMIC mutations in ctDNA to The Cancer Genome Atlas (TCGA) database, according to the histology of their disease (18, 19). Given the small sample size, it was not feasible to perform meaningful statistical comparison of the prevalence of each individual mutation between the CANTOS and TCGA patients with lung cancer. Nonetheless, we noted that the COSMIC mutations identified in the CANTOS patients with lung cancer were largely consistent with mutations commonly found in lung cancer as previously reported by TCGA (Table 3), with the most notable exception of mutations in the TP53, EGFR, HNF1A, and KRAS genes. For TP53, 84% of patients with squamous cell carcinoma in TCGA carried TP53 mutations, whereas only 57% and 64% of the CANTOS patients tested positive for TP53 mutations at baseline and time closest to cancer diagnosis, respectively. For EGFR, 14% of patients with adenocarcinoma in TCGA carried EGFR mutations, whereas none of the CANTOS patients with adenocarcinoma had EGFR mutations at baseline. In the patients with squamous cell lung cancer, although only 4% in TCGA carried EGFR mutations, 14% in the CANTOS patients were positive for EGFR mutations both at baseline and at the time closest to cancer diagnosis. In addition, 22% of patients with adenocarcinoma had HNF1A mutations at baseline, but none was reported in TCGA. On the other hand, 33% of patients with adenocarcinoma in TCGA had KRAS mutations, but none was found at baseline and only 5% at time closest to cancer diagnosis in the CANTOS patients. Overall, no difference was observed in the mutation distribution between the baseline and the time closest to cancer diagnosis time points (Supplementary Table S2), or between different treatment arms (Supplementary Table S3).

To assess the clinical relevance of ctDNA, we compared time to onset of lung cancer diagnosis in patients with and without detectable COSMIC ctDNA mutations at baseline (Fig. 1). Patients who did not have a detectable COSMIC ctDNA mutation had a longer time to lung cancer diagnosis compared with patients with ctDNA mutation (median time to lung cancer diagnosis 837 days vs. 407 days; Wilcoxon P value = 0.011).

Of the 67 patients who had samples for ctDNA analysis from multiple time points, 18 had the same COSMIC mutations detected across at least two time points, thus enabling the assessment of change in AF over time in these individuals. If a patient harbored multiple mutations, the mutation with the largest change in AF was used in the analysis. Also, if a patient had ctDNA results from multiple time points, the two time points that presented the biggest difference in AF were included. Of the 18 patients, 6 had a change in AF of greater than ±1%, with 4 patients showing an increase in AF over time and two patients showing a decrease (Table 4). In addition, 13 of the 67 patients did not have detectable COSMIC mutation at baseline but became positive at a later time point, suggesting disease progression.

To gain further insights on the role of inflammatory cytokines in lung cancer, we performed additional analysis on cytokine data collected during the CANTOS trial, focusing on patients who were diagnosed with lung cancer during the study (13). We first examined the baseline levels of five inflammatory serum biomarkers (CRP, IL6, IL1RA, IL18, and TNFα), comparing results from the patient with lung cancer subset with the overall population enrolled in CANTOS. As previously reported, the baseline CRP level of the patient with lung cancer subset was elevated to a median of 6.0 mg/L (n = 116), compared with the already elevated median baseline CRP level of 4.2 mg/L in the rest of the CANTOS population (n = 9,945; ref. 13). The baseline IL6 level also was higher in the patients with lung cancer, with a median of 3.1 ng/L in patients with lung cancer (n = 79) compared with 2.6 ng/L in the rest of the CANTOS population in whom it was measured (n = 4,979; ref. 13). In contrast, the baseline levels of the other inflammatory biomarkers IL1RA, IL18, and TNFα were not significantly different between the patients with lung cancer and the rest of the CANTOS population.

As a next step, we sought to understand the relationship between baseline levels of inflammatory biomarkers (CRP, IL6, IL1RA, IL18, leptin, TNFα, adiponectin, fibrinogen, and PAI1) and time to onset of lung cancer. For each biomarker, we separated the patients with lung cancer into quartiles of baseline levels, looking for potential associations with time to lung cancer diagnosis. Among the nine inflammatory biomarkers, only the baseline levels of CRP and IL6 trended with the time to lung cancer diagnosis (Fig. 2A and B). For baseline CRP, patients in the highest quartile (CRP > 12 mg/L) had a median time to lung cancer diagnosis of 14.9 months compared with 20.7 months in patients in the lowest quartile (CRP < 3.4 mg/L). For baseline IL6, patients in the highest quartile (IL6 > 6.0 ng/L) had a median time to lung cancer diagnosis of 14.9 months compared with 27.0 months in patients in the lowest quartile (IL6 < 2.3 ng/L). None of the other protein biomarkers examined showed strong trends with time to lung cancer diagnosis, including the inflammatory cytokines IL18 and TNFα.

As reported previously, in the CANTOS population, IL6 and CRP were downregulated by canakinumab in a dose-dependent manner consistent with canakinumab's role in targeting the IL1β inflammatory pathway specifically (12, 13, 20). To examine how these cytokines along with IL18 and TNFα responded to canakinumab in patients destined to be diagnosed with lung cancer, the longitudinal changes in the levels of CRP, IL6, IL18, and TNFα were analyzed in these patients. It is important to note that, per protocol, canakinumab treatment was stopped as soon as the patient was diagnosed with lung cancer. Similar to the whole CANTOS population, CRP and IL6 levels decreased upon canakinumab treatment in those patients destined to be diagnosed with lung cancer. For CRP, in the combined canakinumab treatment group, the median % change from baseline at 3 and 12 months was −43.4% and −51.5% respectively, compared with −11.8% and −17.2% in the placebo group. Similarly, for IL6, in the combined canakinumab treatment group, the median % change from baseline at 3 and 12 months was −22.5% and −17.4% respectively, compared with +13.0% and +24.5% in the placebo group. No change in IL18 and TNFα levels over this time period was observed (Fig. 3A–D).

In the CANTOS trial, canakinumab showed a dose-dependent reduction in risk of lung cancer of up to 67% (13). In this analysis, we further show that ctDNA bearing COSMIC mutations was present at entry in close to half of the patients destined to be diagnosed with lung cancer in CANTOS study and that the effects were most related to changes in CRP and IL6 levels. We found that ctDNA containing COSMIC mutations was present in 43% of patients (29 of 67) at baseline, before any of the patients were clinically diagnosed with lung cancer. Further, those with detectable COSMIC ctDNA at baseline had a shorter time to lung cancer diagnosis than those patients without detectable COSMIC ctDNA, suggesting that the mutations detected in the patients with lung cancer were likely of tumor origin. The COSMIC mutations identified in CANTOS patients with lung cancer were consistent with mutations commonly found in lung cancer. The distribution of mutations was largely similar to previously reported by TCGA (18, 19), with some noticeable differences in EGFR, TP53, HNF1A, and KRAS. These differences may in part be due to the small sample size in this study, as well as concordance between ctDNA and tumor tissue mutation detection, the latter of which was used in the TCGA studies. Taken together, these data suggest that many patients might have harbored undiagnosed tumors at enrollment in the CANTOS trial, even though we could not confirm the ctDNA-detected mutations in tumor tissues, as no biopsy was performed in the CANTOS study.

Growing evidence supports the utility of ctDNA analysis as a “liquid biopsy” technology to diagnose and monitor cancer disease status (21–23). With the development of more sensitive ctDNA assay technologies such as Cancer Personalized Profiling by Deep Sequencing (CAPP-seq; ref. 24), mutations of low AF (e.g., <0.5%) can now be reliably detected (25, 26). In fact, using CAPP-seq technology, Chaudhuri and colleagues were able to detect ctDNA in pretreatment samples from stage I to III patients with lung cancer with a MAF of 0.62%, which was nearly 10-fold lower than what they previously observed in metastatic lung adenocarcinoma (24, 27). This is comparable with the MAF of 0.44% that we detected in the 29 CANTOS patients with lung cancer at baseline, up to 4 years prior the diagnosis of lung cancer in these patients. Our report of ctDNA detection in CANTOS patients with lung cancer before cancer diagnosis suggest that perhaps the current ctDNA technology has reached the sensitivity required for early cancer detection. Nonetheless, when we analyzed the ctDNA data across multiple time points, we noted that only 18 of 67 patients had the same COSMIC mutations detected across at least two time points. This is not surprising as most of the mutations were detected at AF close to the detection limit of the assay, suggesting that reliability of ctDNA detection at low AF still needs to be improved.

Advancements in sequencing technology also lead to new discoveries in CHIP, the age-dependent clonal expansion of hematopoietic cells harboring acquired somatic mutations. Using deep sequencing technology, recent studies found that CHIP mutations might be present at very low allele frequencies in blood cells of as high as 95% of individuals ages 50 to 70 years (28, 29). Circulating cell-free DNA contains DNA fragments shed from all nucleated cells in blood. Typically, these cells contain predominantly normal white blood cells, but may also include small populations of blood cells harboring CHIP mutations, as well as dislodged tumor cells in the case of patients with cancer. Thus, cell-free DNA analysis in patients with cancer may detect both ctDNA mutations from the tumor and CHIP mutations from the hematopoietic lineage, especially in an older patient population (30). Sequencing a matched normal sample will enable differentiation between ctDNA and CHIP mutations. Nonetheless, this approach is costly, and matched normal samples are not always available. Instead, new methods are being developed to help identify CHIP mutations using bioinformatics. This is especially important for assays designed to detect ctDNA in precancer or patients with early-stage cancer, as the amount of ctDNA in circulation is extremely low. In fact, the new generation of ctDNA assays tend to prioritize on different aspects of analytical assay performance based on the intended application. For example, a ctDNA assay designed to detect oncogenic driver mutations in patients with late-stage cancer may prioritize sensitivity over specificity due to the relative abundance of ctDNA compared with CHIP mutations. In these patients, the detection of CHIP mutations is unlikely to interfere with the identification of clinically actionable mutations to guide the patient's cancer treatment. On the other hand, a ctDNA assay designed for early-stage cancer detection must prioritize specificity over sensitivity. In these patients, detection of CHIP mutations may easily be mistaken for ctDNA mutations, which may potentially lead to a false-positive cancer diagnosis.

In this study, when including all reported variants, mutations were found in 60% (40 of 67) of CANTOS patients with lung cancer with a median AF of 0.29% and 70% (14 of 20) of matched noncancer patients with a median AF of 0.25% at the time of study randomization (P = 0.44 for percentage of ctDNA positive patients; P = 0.45 for AF). However, when the analysis was limited to known cancer-associated mutations from the COSMIC database, mutations were reduced to 43% (29 of 67) of patients with lung cancer with a median AF of 0.44% and only 20% (4 of 20) of noncancer patients with a median AF of 0.18% (P = 0.07 for percentage of ctDNA positive patients; P = 0.03 for AF). There are several possible explanations for the mutations detected in CANTOS patients who were not diagnosed of cancer during the study. First, the fact that most of the mutations identified were not COSMIC mutations suggests that these were likely CHIP mutations at low AF, given the CANTOS population was older with an average age of 61. Because there are known overlap between COSMIC and CHIP mutations, such as TP53 (30), including only COSMIC mutations in our analysis will not completely eliminate all confounding CHIP mutations. Second, the CANTOS patient population is in general at a higher risk for developing cancer due to their advanced age, smoking history, and elevated CRP level. Even though the noncancer patients were not clinically diagnosed with cancer during the course of the study, it was still possible they had harbored undiagnosed tumors, especially with a median follow-up duration of only 3.7 years. Thus, some of the COSMIC mutations detected in the 20% of noncancer CANTOS patients could possibly be of actual tumor origin. Third, most of the mutations detected in noncancer patients had an AF of <0.5%, near the assay's detection limit. In conjunction with the small sample size of the non-cancer patients in this study, it is possible that some of the mutations detected were false positives. These possibilities highlight the complexity of ctDNA analysis and data interpretation. Further advancements in technology are necessary to improve the reliability of ctDNA analysis, especially for early cancer detection.

For the CANTOS patients who may already have had undiagnosed cancer at baseline, it is possible that IL1β inhibition by canakinumab reduced the lung cancer incidence by slowing disease progression. In addition, the distribution of mutations in the CANTOS patients with lung cancer was largely similar to that previously reported, with no obvious differences observed between different time points and treatment arms, perhaps except EGFR and TP53. The small patient number, high percentage of current and former smokers, as well as the enrichment of high-risk patients with elevated CRP may potentially account of the minor differences between the observed mutations in CANTOS patients with lung cancer compared with TCGA results. The lack of enrichment of any specific mutation in the treatment arms at the time closest to diagnosis suggests that canakinumab's effect was not restricted to a specific molecular subtype of lung cancer.

In addition to ctDNA analysis, we also assessed circulating protein biomarkers in the CANTOS patients with lung cancer, including the inflammatory biomarkers IL6, CRP, IL18, and TNFα. Of these biomarkers, IL6 and CRP are directly downstream of the IL1β pathway, IL18 belongs to a different branch of the NLRP3 inflammasome pathway, and TNFα belongs to an entirely different inflammatory pathway. All of these inflammatory cytokines have been implicated in the development and progression of cancer (3, 31–34). Further, previous epidemiologic studies have suggested that patients with elevated CRP are at a higher risk for developing lung cancer (8, 9). Consistent with this notion, patients who developed lung cancer during the CANTOS trial had higher baseline (i.e., prediagnosis) levels of CRP and IL6 when compared with patients who did not develop lung cancer, as previously described (12, 13, 20). Furthermore, among nine different inflammatory protein biomarkers, (i.e., CRP, IL6, IL1RA, IL18, Leptin, TNFα, adiponectin, fibrinogen, and PAI1) only high baseline levels of CRP and IL6 trended toward shorter median time to lung cancer diagnosis. Thus, similar to the presence of ctDNA at baseline, elevated CRP and IL6 levels show an association with quicker progression to lung cancer diagnosis.

Previous reports have demonstrated that canakinumab can significantly reduce serum levels of CRP and IL6 levels in patients with atherosclerosis (12, 20, 35). This is consistent with the known specificity of canakinumab against the IL1β inflammatory pathway. In this study, we assessed the modulation of inflammatory cytokines in the subset of CANTOS patients who developed lung cancer, as these pathways might behave differently in atherosclerosis patients who also developed lung cancer. Nonetheless, we found that similar to the whole CANTOS population, compared with other inflammatory biomarkers such as IL1RA, IL18, and TNFα, only CRP and IL6 were modulated by canakinumab in patients with lung cancer. These suggest that specifically targeting the IL1β inflammatory pathway might be paramount to reducing protumor inflammation in lung cancer. Whether inhibiting multiple inflammatory pathways might be more effective in treating lung cancer remains to be explored.

C.C. Wong reports personal fees from Novartis (employee and stock) outside the submitted work; in addition has a patent for use of IL1β binding antibodies pending. Y.A. Wang reports other from Novartis (employee and stock) outside the submitted work. M.F. Prescott reports personal fees from Novartis Pharmaceuticals (employee and stock) during the conduct of the study and personal fees from Novartis Pharmaceuticals (employee and stock) outside the submitted work. L. Krajkovich reports personal fees from Novartis (employee and stock) outside the submitted work. M. Dugan reports other compensation from Novartis (employee and stock) outside the submitted work; in addition, has a patent for use of IL1 in cancer pending. P.M. Ridker reports grants from Kowa, Novartis, NHLBI, and Amarin (research grants) and personal fees from Novartis, Flame, Agepha, CiviBioPharm, Jansen, AstraZeneca, Inflazome, Corvidia (consultant) during the conduct of the study; in addition, Dr. Ridker has a patent pending. (While Dr. Ridker is listed as a co-inventor on patents held and owned by Novartis that relate to the use of IL1 inhibition in cancer, he has no financial interest in these patents.) A-M. Martin reports employment with and is a recipient of stocks from Novartis. E.C. Svensson reports employment with the Novartis Institutes for Biomedical Research. No potential conflicts of interest were disclosed by the other authors.

C.C. Wong: Conceptualization, data curation, formal analysis, supervision, writing-original draft, project administration, writing-review and editing. J. Baum: Conceptualization, data curation, formal analysis, supervision, methodology, project administration, writing-review and editing. A. Silvestro: Formal analysis, writing-review and editing. M.T. Beste: Conceptualization, data curation, formal analysis, methodology, project administration, writing-review and editing. B. Bharani-Dharan: Conceptualization, formal analysis, methodology, writing-review and editing. S. Xu: Formal analysis, writing-review and editing. Y.A. Wang: Formal analysis, writing-review and editing. X. Wang: Formal analysis, writing-review and editing. M.F. Prescott: Formal analysis, project administration, writing-review and editing. L. Krajkovich: Conceptualization, data curation, formal analysis, project administration, writing-review and editing. M. Dugan: Formal analysis, project administration, writing-review and editing. P.M. Ridker: Formal analysis, supervision, project administration, writing-review and editing. A.-M. Martin: Conceptualization, formal analysis, supervision, methodology. E.C. Svensson: Conceptualization, formal analysis, supervision, writing-review and editing.

This study was sponsored by Novartis Pharmaceutical Corporation. The authors thank Shiva Krishna Rachamadugu of Novartis Healthcare Pvt. Ltd., for providing medical editorial assistance with this manuscript, as well as Guardant Health for helpful discussions on ctDNA analysis.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.