Mutations in circulating cell‐free tumour DNA: Predictors of survival in hepatocellular carcinoma

Hepatocellular carcinoma (HCC) incidence is increasing worldwide and prognostic biomarkers are urgently needed to guide treatment and reduce mortality. Circulating cell‐free DNA of tumour origin (ctDNA) is a novel, minimally invasive means of determining genetic alterations in cancer. We determined the utility of ctDNA as a prognostic biomarker of survival in HCC.


| INTRODUC TI ON
Hepatocellular carcinoma (HCC) is the seventh most common cancer and fourth most common cause of cancer-related death worldwide 1 and incidence continues to increase in parallel with the global epidemics of viral hepatitis 2 and non-alcoholic fatty liver disease. 3 Despite important advances in HCC management over the last decade, 4 survival remains poor, contrasting sharply with improvements in mortality for most other cancer types. 1 Prognostic biomarkers are urgently needed to optimise treatment allocation to improve survival, particularly in the era of the rapidly expanding armoury of systemic therapies being approved for advanced stage HCC. 4,5 The genetic pathways of HCC carcinogenesis have been well characterised [6][7][8] and it is now appreciated that there is considerable genetic tumour heterogeneity, even within the same tumour. This genetic heterogeneity, coupled with low rates of tumour biopsy because of radiological criteria for HCC diagnosis, means identification of prognostic mutations has been a major challenge to the field. 4 Circulating cell-free DNA (cfDNA) is an exciting potential alternative to tissue biopsy that may overcome the need for tumour biopsy.
cfDNA is released through apoptosis and necrosis from both healthy and malignant cells into the blood stream. 9,10 Tumour-derived cfDNA (ctDNA) can be distinguished from wild-type cfDNA by identification of somatic alterations present in the tumour of interest, but not in matched somatic tissue. 9 ctDNA quantification and the presence of genetic and epigenetic variants within ctDNA are therefore attractive biomarkers in cancer, 9,11 facilitating dynamic assessment of genetic mutations before, during and after treatment without the need for repeated tumour biopsies. 10,12,13 Several studies have demonstrated the clinical utility of tumourderived plasma ctDNA for guiding prognosis and treatment decisions in various cancer types, including breast, ovarian and lung cancer. 9,14 Preliminary studies have also demonstrated the potential utility of cfDNA levels and specific genetic mutations in ctDNA as clinical biomarkers in HCC. [11][12][13][14][15][16][17][18] For example, Kirk and colleagues 9,10 demonstrated that the Ser249 TP53 mutation could be detected in ctDNA from West African patients with advanced-stage HCC and this mutation had good discriminative accuracy for distinguishing HBV-related HCC from chronic hepatitis B infection and HBVrelated cirrhosis. We have previously shown that mutations in classic carcinogenic pathways described in HCC can be detected in ctDNA in patients at all BCLC stages of the disease and that these mutations can be found in matched HCC tumour tissue. 15 However, to date, there have been few studies investigating the detection of mutations in genes with potential prognostic and therapeutic utility in ctDNA, particularly across all BCLC stages of HCC.
The aim of this study was to determine whether the detection of mutations in well-established carcinogenic driver genes in ctDNA are associated with overall survival in HCC patients.

| Study design and participants
Methods for this study have been described elsewhere. 15 In brief, 51 adult patients (over 18 years of age) with confirmed radiological diagnosis of HCC 1,19 were consecutively recruited from two specialist liver cancer clinics (Imperial College NHS Trust, n = 19 (37.3%), UK; and University of Piemonte Orientale, Italy, n = 32, 62.8%) and 10 adult patients with cirrhosis but no HCC were recruited as controls from a specialist liver clinic (Imperial College NHS Trust, UK).
Patients were recruited prior to commencing treatment for HCC; prior treatment was not a preclusion. Controls were screened for HCC with liver ultrasound within six months of recruitment to the study. The study was approved by the local Institutional ethics committees and was conducted in accordance with the 1975 Declaration of Helsinki.

| Sample collection
A 20 mL blood sample was collected from all subjects on the day of recruitment in an EDTA tube and 1 mL aliquots of plasma and buffy coat were stored in a minus 80°C freezer until further use bosis and metastases) were also collected on the same day as the blood sample. Survival outcomes were prospectively collected after recruitment to the study and recorded prior to analysis of cfDNA samples.

K E Y P O I N T S
• Detection of tumour mutations in plasma cell-free DNA is feasible.
• A novel mutation in CSMD3 was associated with reduced survival.
• These findings require further validation.

| Data analysis
Analysis of matched plasma ctDNA and HCC DNA sequences was performed using Mutect v.1 (Broad Institute) using matched germline PBMC DNA as a reference, as outlined previously. 15 Genetic variants were identified after filtering with additional quality control steps including removal of all variants with low read coverage and samples with low GC quality (outlined in Supplementary Methods, Figure S1). Non-coding variants were excluded to focus on variants of potential clinical significance. Mutations identified in ctDNA, but not in matched germline DNA were determined. We then applied additional filters to remove variants in order to maximise specificity for clinically significant variants in ctDNA. Non-coding genetic variants, variants that were also identified in cirrhotic controls, and recognised variants listed in human SNP databases (dbSNP, 1000 Genomes) were excluded from analysis, unless these variants were also listed in the COSMIC cancer mutation database (see Supplemental Methods and Figure S1). Finally, we excluded variants identified in the cirrhosis controls.

| Statistical analysis
Clinical variables were described using mean ± standard deviation or median (IQR) as appropriate to distribution. Associations between categorical variables were assessed by chi-square test.
Associations with overall survival were assessed using Cox proportional hazards modelling, including all variables with P < .05 in association with overall survival. We included genetic markers associated with survival on univariable analysis (P value <.05), adjusted for BCLC stage, Child-Pugh class and age (known determinants of survival in HCC). Treatment was not included in the model because (a) numbers for individual treatment modalities were small, (b) treatment modality was not found to be independently associated with survival when adjusted for BCLC stage and (c) sample size precluded more than five parameters in the model. Proportional hazards assumption was tested using log-log plots and Schoenfeld residuals. The significance threshold was a two-sided P-value of <.05. All statistical analyses were performed using STATA version 12.1 (Statacorp).    On univariable analysis, the presence of CSMD3 gene mutations was associated with shorter overall survival (Logrank HR 3.18, 95% CI 1.14-8.86, P = .027; Table 3 and Figure 1). There was also a positive association between the presence of   15 In this study, we additionally report mutations in genes that are less common but described in HCC, including KEAP1, MET, AMP and APC, 20 and in several genes that are described in other malignancies but have not frequently been reported in HCC, including CSMD3, IGF2R, BAZ2B and DSE. Whilst putative mutations in these genes have been described in the COSMIC database and in other studies, their role in HCC is less certain. It has been previously reported that the mutational rate in HCC increases with advancing tumour stage. 21 Thus, the propensity for mutations to accumulate as checkpoint and DNA repair mechanisms are lost may mean non-driver mutations also become more common at mutation 'hotspots' that may still serve as useful biomarkers in HCC.

| Mutations in key drivers of hepatic
In this study, CSMD3 mutations were associated with higher levels of circulating cell-free DNA which is seen in cancers with greater tumour burden and greater inflammatory component. 10 It is important that the biological significance of CSMD3 in  none of which identified CSDM3 as a driver mutation in HCC. 6,7,28,29 This may be because liver tissue sequencing studies have traditionally focussed on HCC tissue specimens, not adjacent liver tissue specimens. of all HCC cases (Table 2), making CSMD3 more likely to be useful clinically given the relatively high frequency of mutation detection in HCC cases. TERT promotor mutations, one of the most common genetic alterations found in HCC, 7 were not assessed in this study and would likely add greater sensitivity to HCC mutation detection in ctDNA in future studies. 21,22 There are several limitations in this pilot study of ctDNA prog-

| CON CLUS ION
Detection of CSMD3 mutations in plasma ctDNA was associated with reduced overall survival in our pilot HCC patient cohort.
These preliminary data highlight the exciting potential of ctDNA as a prognostic biomarker in HCC and further studies evaluating its clinical utility in large prospective cohorts of HCC patients are warranted.

ACK N OWLED G EM ENTS
We

CO N FLI C T O F I NTE R E S T
The Institute for Pathology (University Hospital of Cologne, Germany) is an official testing institution of Qiagen Inc (Hilden, Germany) and Reinhard Büttner is a member of the advisory board of Qiagen Inc Furthermore, Qiagen Inc has supported the study by providing reagents for cfDNA extraction and NGS library construction.

AUTH O R S ' CO NTR I B UTI O N S
JH lead the study, contributed to study design, recruited UK study participants, collected UK clinical samples and data, contributed Rohini Sharma https://orcid.org/0000-0003-2441-549X