Circulating TP53 mutations are associated with early tumor progression and poor survival in pancreatic cancer patients treated with FOLFIRINOX

Background: Biomarkers predicting treatment response may be used to stratify pancreatic ductal adenocarcinoma (PDAC) patients for therapy. The aim of this study was to identify circulating tumor DNA (ctDNA) mutations that associate with tumor progression during FOLFIRINOX chemotherapy, and overall survival (OS). Methods: Circulating cell-free DNA was analyzed with a 57 gene next-generation sequencing panel using plasma samples of 48 PDAC patients of all disease stages. Patients received FOLFIRINOX as initial treatment. Chemotherapy response was determined on CT scans as disease control (n = 30) or progressive disease (n = 18) within eight cycles of FOLFIRINOX, based on RECIST 1.1 criteria. Results: Detection of a TP53 ctDNA mutation before start of FOLFIRINOX [odds ratio (OR) 10.51, 95% confidence interval (CI) 1.40–79.14] and the presence of a homozygous TP53 Pro72Arg germline variant (OR 6.98, 95% CI 1.31–37.30) were predictors of early tumor progression during FOLFIRINOX in multivariable analysis. Five patients presented with the combination of a TP53 ctDNA mutation before start of FOLFIRINOX and the homozygous Pro72Arg variant. All five patients showed progression during FOLFIRINOX. The combination of the TP53 mutation and TP53 germline variant was associated with shorter survival (median OS 4.4 months, 95% CI 2.6–6.2 months) compared with patients without any TP53 alterations (median OS 13.0 months, 95% CI 8.6–17.4 months). Conclusion: The combination of a TP53 ctDNA mutation before start of FOLFIRINOX and a homozygous TP53 Pro72Arg variant is a promising biomarker, associated with early tumor progression during FOLFIRINOX and poor OS. The results of this exploratory study need to be validated in an independent cohort.


Introduction
Although some advances have been made in the treatment of pancreatic ductal adenocarcinoma (PDAC), the prognosis of patients remains very poor. 1,2 The standard first-line treatment for locally advanced pancreatic cancer (LAPC) and metastatic PDAC is FOLFIRINOX chemotherapy, a combination of fluorouracil, leucovorin, Circulating TP53 mutations are associated with early tumor progression and poor survival in pancreatic cancer patients treated with FOLFIRINOX irinotecan, and oxaliplatin. With this treatment regimen, improved overall survival (OS) was observed in both LAPC (24.2 months versus 6-13 months) 3 and metastatic PDAC (11.1 months versus 6.8 months) 4 compared with gemcitabine chemotherapy. FOLFIRINOX is also effective in PDAC patients with stage I-II resectable or borderline resectable disease in the adjuvant setting, 5 and several clinical trials are investigating the benefit of neoadjuvant FOLFIRINOX followed by surgical resection. 6 Despite increased survival in patient groups treated with FOLFIRINOX, only a minority of patients will show complete or partial response of the tumor, 7,8 while approximately 20-30% already develop progressive disease during FOLFIRINOX. 4,9 Unfortunately, 60-70% of patients experiences severe, grade ⩾4 toxicity from FOLFIRINOX. 3,4,8 Biomarkers could stratify patients for available therapies. Especially biomarkers that can be measured easily in the circulation, as opposed to tumor tissue, would be ideal. Such a predictive biomarker could prevent non-responding patients from FOLFIRINOXinduced toxicity and select these patients for other treatments.
Circulating cell-free DNA (ccfDNA), including circulating tumor DNA (ctDNA), are short fragments of DNA released into the bloodstream after apoptosis and necrosis of (tumor) cells. CtDNA can be detected in blood serum or plasma, and the presence of tumor mutations in ctDNA is a poor prognostic factor in PDAC patients. [10][11][12] Moreover, increasing ctDNA levels over time and the detection of new mutations during chemotherapy are associated with progression of disease. [13][14][15] However, most studies focus on KRAS mutations only, 13,15 while several other known cancerassociated gene mutations may indicate PDAC progression and treatment response as well.
The aim of this pilot study was to investigate the value of ctDNA mutations in PDAC patients, detected before the start of treatment or after only one cycle of chemotherapy, to predict early tumor progression during FOLFIRINOX and their association with OS.

Materials and methods
This article was written according to the reporting recommendations for tumor marker prognostic studies (REMARK) guidelines. 16 Patient selection All patients were selected from two multicenter, prospective trials in the Netherlands. Patients with resectable or borderline resectable PDAC participated in the randomized clinical trial PREOPANC-2 (Dutch trial register NL7094) comparing neoadjuvant FOLFIRINOX with neoadjuvant gemcitabine-based chemoradiotherapy, followed by surgical resection of the primary tumor if applicable. Patients with locally advanced or metastatic PDAC were selected from the prospective cohort study iKnowIT (Dutch trial register NL7522) focusing on the predictive value of circulating biomarkers. Both trials were approved by the ethics committees of all participating hospitals with patients included in this article: (homozygous) were considered germline mutations, not ctDNA mutations. All base changes and accompanying amino acid changes were annotated on the forward strand.

Statistical analyses
Continuous data with a non-normal distribution were compared with either Mann-Whitney U tests, or with Wilcoxon Signed Rank tests for paired data.
Categorical data, such as detection rates of mutations, were compared using Fisher's exact or Pearson's Chi-squared tests where appropriate.
Univariable and multivariable binary logistic regression was performed to analyze the predictive value of ccfDNA concentrations and ctDNA mutations for tumor progression during FOLFIRINOX chemotherapy, adjusted for known predictive patient characteristics: stage of disease, and CA19-9 levels. Multicollinearity between variables was checked using the variance inflation factor (VIF). Factors with p < 0.10 and VIF < 3 were selected for multivariable analysis.
OS was calculated as the time between the start of FOLFIRINOX and death. All patients included in this cohort died of cancer progression. Differences in median OS were derived from Kaplan-Meier curves whereby groups were compared using log-rank tests. The prognostic value ctDNA mutations was also tested with univariable and multivariable Cox regression analyses, including known prognostic factors: age, stage of disease, and CA19-9 levels. Multicollinearity between variables was checked using the variance inflation factor (VIF). Factors with p < 0.10 and VIF < 3 were selected for multivariable analysis.
Only two-sided tests were used and p-values < 0.05 were considered statistically significant. Data were analyzed using SPSS Statistics for Windows (version 25.0; IBM, Armonk, NY, USA).

Patient characteristics
In total, ccfDNA isolated from plasma from 48 patients was sequenced both before and after the first cycle of FOLFIRINOX, resulting in 96 samples. A Figure S1. There was no statistically significant difference in ccfDNA concentration before chemotherapy between disease control and progressive disease patients (p = 0.074). After one cycle of chemotherapy, the median ccfDNA concentration was significantly lower in patients with progressive disease (p = 0.018) (Supplemental Figure S1).
In 27 out of 96 (28.1%) of the sequenced plasma samples, at least one ctDNA mutation was detected, corresponding to samples from 21 out of 48 patients. Supplemental Figure S2 shows an overview of each ctDNA mutation detected per plasma sample in 21 out of 48 patients, including their variant allele frequency (VAF).
ctDNA mutation detection rates did not differ between samples collected before start of FOLFIRINOX (35.4%) and samples collected after one cycle of FOLFIRINOX (20.8%, p = 0.112). There were no differences in ctDNA mutation detection rates before start of FOLFIRINOX between patients with disease control (30.0%) and patients with progressive disease (44.4%, p = 0.361), or after one cycle of chemotherapy between patients with disease control (16.7%) and patients with progressive disease (27.8%, p = 0.468).
The most frequently detected ctDNA mutations were KRAS (17/96 samples in total, 12/48 before chemotherapy, 5/48 after 1 cycle of chemotherapy), TP53 (12/96 samples in total, 8/48 before chemotherapy, 4/48 after 1 cycle of chemotherapy), and PIK3CA (4/96 samples in total, 2/48 before chemotherapy, 2/48 after 1 cycle of chemotherapy) mutations. Differences in detection rates between patients with disease control and patients with progressive disease are presented in Table 2. Before start of FOLFIRINOX TP53 ctDNA mutations were significantly more often detected in patients with progressive disease (33.3%) compared with patients with disease control (6.7%, p = 0.040). After chemotherapy no statistically significant differences between disease control and progressive patients were found in detection rates of any of the major tumor mutations. For this reason, only results retrieved from samples collected before FOLFIRINOX will be further discussed.
There were no differences in detection rates of ctDNA mutations between the different stages of disease (Supplemental Table S2).

Germline variant detection
Five germline variants were found in multiple patients: TP53 p.Pro72Arg, KDR p.Gln472His, KIT p.Met541Leu, ERBB2 p.Ile625Val, and  There were no differences in the distribution of germline variant genotypes between the different stages of disease (Supplemental Table S2).   (Figure 1d). Patients with any ctDNA mutation before the start of FOLFIRINOX,   The presence of any detectable ctDNA mutations before start of FOLFIRINOX remained a significant prognostic factor for OS after adjusting for age, stage of disease, and baseline CA 19-9 level with HR 4.29 (95% CI 1.40-13.12, p = 0.011). For the presence of the TP53 Pro72Arg germline variant, prognostic value could not be demonstrated.

Discussion
This multicenter pilot study describes the predictive and prognostic value of ctDNA mutations in PDAC patients, detected with next generation sequencing (NGS) before and after one cycle of FOLFIRINOX. We found that circulating TP53 mutations detected before the start of FOLFIRINOX predict tumor progression during FOLFIRINOX. These mutations include both TP53 ctDNA mutations and a homozygous TP53 Pro72Arg germline variant. Furthermore, circulating TP53 mutations were found to be a poor prognostic factor for OS in PDAC patients treated with FOLFIRINOX. The results of our study suggest that PDAC patients could be spared from ineffective FOLFIRINOX and its side effects by a simple blood draw before the start of treatment. To our knowledge, TP53 ctDNA mutations and the common TP53 Pro72Arg variant have not been described previously for their predictive value for FOLFIRINOX response in PDAC. However, both types of mutations have been described to play a role in cancer development and progression, including in PDAC. [19][20][21] It is important to distinguish the prognostic and predictive value of ctDNA mutations from tumor tissue-specific mutations. KRAS mutations are present in almost all PDAC tumors, and ~40% of PDAC tumors have TP53 mutations. 22 Only a limited number of the mutant PDAC patients, however, have detectable ctDNA mutations: in our cohort, 44%. It is known that PDAC patients with mutant KRAS and mutant TP53 tumors have a worse prognosis compared with patients with wild-type KRAS and TP53 tumors. 23,24 The prognosis is even worse for patients with detectable ctDNA mutations. The underlying hypothesis for this negative prognostic effect of mutant ctDNA is that aggressive tumors with extensive metastatic properties grow faster, have a high cell death rate, infiltrate (large) blood vessels, and lead to a higher tumor load, thus shed more ctDNA into the circulation. 25 The prognostic value of detectable ctDNA mutations in patients with PDAC has been described in several studies. The presence of any and/or specific ctDNA mutations, such as KRAS and TP53, are associated with poor OS and progression-free survival. [26][27][28] Our results support this hypothesis: the detection of any ctDNA mutations before the start of FOLFIRINOX was a negative prognostic factor for OS (HR 4.29, 95% CI 1.40-13.12, p = 0.011). Moreover, all patients with a detectable TP53 ctDNA mutation before the start of FOLFIRINOX died from PDAC progression within 10 months. In our cohort, KRAS ctDNA mutations were associated with OS in univariable analyses, but not statistically significant in multivariable analyses.
The TP53 gene is an important tumor suppressor gene. Wild-type TP53 regulates the cell cycle, initiates apoptosis and senescence, and activates DNA repair in situations of DNA damage and cellular stress, thus inhibiting tumorigenesis. 29 TP53 is the second most frequently mutated gene in PDAC and is likely responsible for the susceptibility to cancer development. 22 In human cancers, missense mutations in TP53 are the most common type, often leading to gain-of-function and promotion of tumorigenesis. 29,30 Most somatic TP53 mutations are located in codons 175, 245, 248, 249, 273, and 282. 30,31 Others have shown that restoration of wild-type TP53 in PDAC cell lines with gain-of-function TP53 mutations enhances the sensitivity to 5-fluorouracil (5-FU), irinotecan, cisplatin, and gemcitabine. 32 These data support the results of our study: patients without detectable TP53 ctDNA mutations showed a better response to FOLFIRINOX.
The germline TP53 Pro72Arg SNP is a wellknown variant in the human population. With the replacement of a guanine base by a cytosine base, the accompanying amino acid changes from a proline (Pro) into an arginine (Arg). This amino acid change affects the structure of the protein and might thereby influences its function. 33 The TP53 Pro72Arg variant shows varying allele-frequencies in different populations, according to the 1000 Genomes Project. 34 Because of its high frequency in humans, the SNP TP53 Pro72Arg has been studied for its association with cancer risk and cancer development in a multitude of studies. 19 Allele frequencies of this SNP are known to be different in European compared with Asian populations and the combination with ethnicity-specific genetic makeup could lead to different phenotypes. In our Dutch Caucasian PDAC cohort, the Arg/Arg variant is most prevalent: 62.5% of patients show this homozygous variant. Since no healthy controls were included in this experiment, we were not able to assess whether this frequency is different from the healthy Dutch population, and whether the Arg variant is associated with an increased risk of PDAC. For the interpretation of clinical trials, it is very important to keep in mind that genotypes might influence the response to treatment. It cannot be assumed that results are directly applicable to patients from other ethnicities or elsewhere in the world, since allele frequencies of SNPs may differ considerably.
The combination of the TP53 Pro72Arg variant with a somatic TP53 mutation in tumor tissues has been described for its poor prognostic value in other cancer types 35 ; however not yet in PDAC. It would be relevant to further analyze whether in this population more poor prognostic features could be found with the use of large public databases.
The variety in ctDNA mutation detection methods between different study results makes it difficult to draw general conclusions. For example, Droplet Digital PCR has a higher sensitivity for ctDNA mutation detection than NGS, but can 12 journals.sagepub.com/home/tam only be used to search for pre-determined specific mutations, such as in KRAS codons 12 and 13. 36 We used a broad 57 gene cancer panel, including more amplicons than only the major PDAC hotspots and covering the entire TP53 coding region.
For this exploratory study, we included patients with PDAC from all disease stages. We specifically chose this study design for a couple of reasons. First, the distinction between disease stages is often difficult and we believe that the disease stage determined with radiography is a less important factor for the choice of treatment and prognostic outcomes than the molecular biology of PDAC. Second, with the increasing number of clinical trials investigating neoadjuvant (FOLFIRINOX) chemotherapy, including resectable PDAC patients in biomarker studies is required for future personalized treatment. However, we do acknowledge that including patients from different disease stages is also a limitation of the study. Treatment schedules differ between stages of disease. Resectable patients might undergo surgical resection of the tumor, and LAPC patients sometimes receive additional stereotactic body radiation therapy. Although our data does not show differences in detection rates of ctDNA mutations or germline variants between the different disease stages, treatment schedules might have impact on survival differences between or even within disease stages.
The low amount of ccfDNA/ctDNA in PDAC patients compared with patients with other solid cancers, such as lung cancer, 37 is another important limitation when using broader sequencing techniques. Furthermore, the sensitivity to detect ctDNA mutations using NGS in this study can be improved. It is important, however, to emphasize the stringent method we used to measure ccfDNA concentrations. Instead of the more commonly used, less accurate methods, such as NanoDrop (spectrophotometric) or Qubit (fluorometric), we have used RT-qPCR with Alu115 primers to determine the true ccfDNA concentration, preventing overestimation of ccfDNA quantity. 38 Moreover, due to the study design of this pilot study, including a relative small number of patients, we did not consecutively select patients for NGS.
This study was designed as a broad, exploratory pilot study, since there is limited literature on possible predictive biomarkers for FOLFIRINOX response in patients with PDAC. 39 The next step would be to conduct a validation study including a larger patient cohort, focused on TP53 mutations alone, comparing different treatment regimens to FOLFIRINOX and implementing a more sensitive NGS protocol. By increasing the amount of plasma for DNA isolation, the ccfDNA yield will increase. A higher DNA input amount for sequencing library preparation would increase the probability to detect rare tumor mutations. Last, a molecular barcoding technique with unique molecular identifiers (UMIs) could be used to be able to detect ctDNA mutations at lower allele frequencies without the risk of false positive mutation calling, because errors introduced during library preparation, target enrichment, or sequencing can be filtered out easily. 40 When including more patients, subgroup analyses on the different tumor stages can be performed in order to check if the predictive value is the same in all PDAC patients receiving FOLFIRINOX. It would be important to see whether treatment response can also be predicted with circulating TP53 mutations for other types of chemotherapy in PDAC, and if patients that are not responding to FOLFIRINOX could benefit from, for example, gemcitabine-based chemotherapy.
In summary, the combination of a TP53 ctDNA mutation with a homozygous TP53 Pro72Arg germline variant is a marker for early tumor progression during FOLFIRINOX and is associated with poor OS. Before translating these results to clinical practice and adjusting treatment decisions, additional cohort studies will be necessary to validate our findings.