Early cancer detection is an attractive and promising application for liquid biopsy that might revolutionize cancer screenings. In this issue of Cancer Discovery, Foda and colleagues expand the potential utility of a machine learning fragmentome-based model, called DELFI, for detecting liver cancer in high-risk patients.

See related article by Foda et al., p. 616 (5).

Early cancer detection is an attractive and promising application for liquid biopsy that might dramatically change cancer screenings through the introduction of minimally invasive tests. To date, cancer screenings cover a limited set of solid tumors with no approved/effective screening method for most of the cancer types with the exception breast, colorectal, lung, cervical, and prostate cancers. Furthermore, cancer screening adherence in some solid tumors is relatively low due to invasive/uncomfortable procedures and/or financial/organizational issues, limiting the widespread use in the target population. In addition, some currently approved cancer screenings, such as low-dose computed tomography (LDCT) for lung cancer, are limited to high-risk populations, limiting the clinical applications to other clinically relevant subgroups of individuals (e.g., never smokers).

Traditionally, the development of highly sensitive and specific liquid biopsy tests for early cancer screenings has been hampered by biological and technical challenges (1). However, the impressive technological improvements made in the last few years led to the development of novel effective strategies for early cancer detection, exploiting the different components of the large liquid biopsy family, especially cell-free DNA (cfDNA). Multiple approaches have been proposed to date for early cancer detection, including tumor-informed (requiring the knowledge in advance of the tumor mutation profile) or tumor-uninformed (based on de novo mutation and other tumor-derived aberration discovery) approaches, as well as multicancer early cancer detection (MCED) tools or single-tumor tests (2). Different omics have been explored so far (genomics, methylomics, fragmentomics, and metabolomics), and cfDNA fragmentomics is one of the most promising liquid biopsy approaches for early cancer detection. Evaluating the fragmentation patterns of cfDNA across the genome, Foda and colleagues showed that patients with cancer have altered fragmentation profiles, as compared with healthy individuals, and developed a fragmentome-based approach called “DNA evaluation of fragments for early interception,” or “DELFI,” to detect a large number of abnormalities in cfDNA by genome-wide analysis of fragmentation patterns (3). The analysis of fragmentation profiles of 236 patients with cancer with seven different solid tumors (breast, colorectal, lung, ovarian, pancreatic, gastric, or bile duct cancers) and 245 healthy individuals revealed sensitivities of detection for DELFI ranging from 57% to more than 99% among the different cancer types at 98% specificity, with an overall area under the curve value of 0.94 (3). Sensitivity of the DELFI approach was particularly promising for lung cancer, and the performance of this model was recently tested using a prospective cohort of the LUCAS study and an independent validation cohort, including both patients with lung cancer and healthy individuals (4). The DELFI score was able to detect lung cancer across all stages and histologic subtypes and seemed not to be affected by noncancer conditions, which have confounded other potential biomarkers for lung cancer detection. Combining genome-wide cfDNA fragmentation features with clinical risk factors (age, smoking history, and presence of chronic obstructive pulmonary disease) and carcinoembryonic antigen levels in a multimodal model (DELFImulti) followed by LDCT resulted in a reported overall 94% sensitivity (stage I: 87%; stage II: 100%; stage III: 97%; and stage IV: 96%) with 80% specificity (4). In this issue of Cancer Discovery, Foda and colleagues (5) further expand the potential utility of DELFI, applying this machine learning model for detecting liver cancer in high-risk patients (Fig. 1).

Figure 1.

Summary of the study by Foda and colleagues, with key findings. HBV, human hepatitis B virus; HCC, hepatocellular carcinoma; HCV, human hepatitis C virus; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic fatty liver; pts, patients. Created with BioRender.com.

Figure 1.

Summary of the study by Foda and colleagues, with key findings. HBV, human hepatitis B virus; HCC, hepatocellular carcinoma; HCV, human hepatitis C virus; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic fatty liver; pts, patients. Created with BioRender.com.

Close modal

Early diagnosis of hepatocellular carcinoma (HCC) is an urgent unmet medical need and represents a global challenge. International guidelines recommend HCC surveillance for high-risk patients with abdominal ultrasound with or without serum alpha-fetoprotein (AFP) dosage (6). However, overall adherence to HCC surveillance is suboptimal, with only ∼50% of the high-risk patients receiving appropriate screening (7) with relatively low specificity and sensitivity. For these reasons, multiple liquid biopsy approaches, mostly cfDNA-based, have been proposed for early detection of HCC in order to increase the performance of currently available surveillance procedures, using not only plasma, but also other bodily fluids, such as urine. Recently, a multiomics approach with a simultaneous detection of genetic and epigenetic alterations and genome-wide discovery of methylation markers, called Mutation Capsule Plus, has been reported. After de novo screening of methylation markers on cfDNA samples of patients with HCC and mutational profiles on HCC and non-HCC cases, an HCC multiomics detection model was elaborated and validated both retrospectively and prospectively with sensitivities ranging from 80% to 90% at a 94% sensitivity (8). In addition to single-tumor tests, other studies have evaluated in multiple solid tumors, including HCC, the potential of cfDNA analyses using MCED tests. One of the MCED assays, Galleri, has recently reached the market after promising results in a large, prospective, case-controlled observational study, called Circulating Cell-free Genome Atlas study (CCGA; NCT02889978). In the first large substudy, whole-genome bisulfite sequencing (WGBS; detecting genome-wide DNA methylation status) was demonstrated to be superior to other methods for MCED; in the second substudy, the WGBS assay was refined into a targeted methylation assay, and machine learning classifiers for cancer detection and cancer signal origin prediction were developed; and, finally, in the third substudy, clinical validation of the MCED test was performed (9). The specificity for cancer signal detection was 99.5% with an overall sensitivity for cancer signal detection of 51.5%. Sensitivity increased with stage and was 67.6% in 12 prespecified stage I to III cancers that account for ∼2/3 of annual U.S. cancer deaths, including liver and bile duct cancers (93.5% sensitivity with cancer signal detection in 43/46 liver/bile duct cancer). Albeit limited by small numbers, sensitivity in liver/bile ducts cancers was high across the different stages (stage I: 100%; stage II: 70%; stage III: 100%; stage IV: 100%; ref. 9). This MCED assay is being actively studied as a screening tool in several ongoing studies, which will define the clinical utility in the current cancer screening scenario.

Others have investigated the potential role of urine cfDNA analyses for HCC detection. The performance of a urine panel including mutated TP53, methylated RASSF1a, and GSTP1 was recently reported. Albeit this panel was inferior to serum AFP in detecting HCC cases, the combined use of urine cfDNA and serum AFP was associated with 79.6% sensitivity at 90% specificity and seemed particularly useful for low AFP cases (10).

In this issue of Cancer Discovery, Foda and colleagues report on the development of a genome-wide fragmentome approach for early detection of liver cancer (5). In this study, cfDNA samples of 47 patients with HCC and 344 noncancer patients, including 52 individuals at high-risk for HCC, from the United States/European Union underwent whole-genome sequencing to identify genomic and chromatin features associated with fragmentation changes. Using a machine learning approach, called DELFI, to determine if changes in cfDNA fragmentomes could distinguish patients with HCC from those without cancer, they reported an overall sensitivity for detecting cancer, including early-stage disease, of 94% among all individuals and 79% among high-risk individuals at 90% specificity. The DELFI model was externally validated in an East Asian population with and without HCC with similar results, suggesting that this approach could be generalizable across different high-risk populations worldwide. As expected, serum AFP was associated with a 47% sensitivity for HCC detection, but the genome-wide cfDNA fragmentation had improved performance. Interestingly, the combination of DELFI and AFP might improve HCC detection over the DELFI approach alone, with a combined sensitivity of 85% at a combined specificity of 90% (5). Despite their promise, these results should be evaluated with caution due to the small sample size. Furthermore, the incremental value of the DELFI approach as compared with abdominal ultrasound (with or without serum AFP) for HCC surveillance in high-risk individuals should be further evaluated, as should the cost-effectiveness of this approach. As the fragmentation profile is related to overall tumor burden, as confirmed in the present study, the performance of this approach in resectable HCC should be further explored. Therefore, before clinical implementation and widespread use of the DELFI approach, large prospective studies are needed to confirm the promising results of this study and further validate genome-wide cfDNA fragmentation for early HCC detection.

C. Rolfo reports personal fees from Roche, MSD, Inivata, Archer, Boston Pharmaceuticals, MD Serono, and Novartis, grants from Pfizer and MD Serono, and nonfinancial support from GuardantHealth outside the submitted work. A. Russo reports personal fees (advisory board honoraria) from AstraZeneca, MSD, Novartis, and Pfizer, and personal fees (writing engagement honoraria) from AstraZeneca, MSD, and Novartis outside the submitted work.

1.
Rolfo
C
,
Cardona
AF
,
Cristofanilli
M
,
Paz-Ares
L
,
Diaz Mochon
JJ
,
Duran
I
, et al
.
Challenges and opportunities of cfDNA analysis implementation in clinical practice: Perspective of the International Society of Liquid Biopsy (ISLB)
.
Crit Rev Oncol Hematol
2020
;
151
:
102978
.
2.
Rolfo
C
,
Russo
A
.
Liquid biopsy for early stage lung cancer moves ever closer
.
Nat Rev Clin Oncol
2020
;
17
:
523
4
.
3.
Cristiano
S
,
Leal
A
,
Phallen
J
,
Fiksel
J
,
Adleff
V
,
Bruhm
DC
, et al
.
Genome-wide cell-free DNA fragmentation in patients with cancer
.
Nature
2019
;
570
:
385
9
.
4.
Mathios
D
,
Johansen
JS
,
Cristiano
S
,
Medina
JE
,
Phallen
J
,
Larsen
KR
, et al
.
Detection and characterization of lung cancer using cell-free DNA fragmentomes
.
Nat Commun
2021
;
12
:
5060
.
5.
Foda
ZH
,
Annapragada
AV
,
Boyapati
K
,
Bruhm
DC
,
Vulpescu
NA
,
Medina
JE
, et al
.
Detecting liver cancer using cell-free DNA fragmentomes
.
Cancer Discov
2023
;
13
:
616
31
.
6.
Heimbach
JK
,
Kulik
LM
,
Finn
RS
,
Sirlin
CB
,
Abecassis
MM
,
Roberts
LR
, et al
.
AASLD guidelines for the treatment of hepatocellular carcinoma
.
Hepatology
2018
;
67
:
358
80
.
7.
Zhao
C
,
Jin
M
,
Le
RH
,
Le
MH
,
Chen
VL
,
Jin
M
, et al
.
Poor adherence to hepatocellular carcinoma surveillance: a systematic review and meta-analysis of a complex issue
.
Liver Int
2018
;
38
:
503
14
.
8.
Wang
P
,
Song
Q
,
Ren
J
,
Zhang
W
,
Wang
Y
,
Zhou
L
, et al
.
Simultaneous analysis of mutations and methylations in circulating cell-free DNA for hepatocellular carcinoma detection
.
Sci Transl Med
2022
;
14
:
eabp8704
.
9.
Klein
EA
,
Richards
D
,
Cohn
A
,
Tummala
M
,
Lapham
R
,
Cosgrove
D
, et al
.
Clinical validation of a targeted methylation-based multi-cancer early detection test using an independent validation set
.
Ann Oncol
;
2021
;
32
:
1167
77
.
10.
Kim
AK
,
Hamilton
JP
,
Lin
SY
,
Chang
TT
,
Hann
HW
,
Hu
CT
, et al
.
Urine DNA biomarkers for hepatocellular carcinoma screening
.
Br J Cancer
2022
;
126
:
1432
8
.