Liquid biopsies for cancer: From bench to clinic

Abstract

Over the past two decades, liquid biopsy has been increasingly used as a supplement, or even, a replacement to the traditional biopsy in clinical oncological practice, due to its noninvasive and early detectable properties. The detections can be based on a variety of features extracted from tumor‑derived entities, such as quantitative alterations, genetic changes, and epigenetic aberrations, and so on. So far, the clinical applications of cancer liquid biopsy mainly aimed at two aspects, prediction (early diagnosis, prognosis and recurrent evaluation, therapeutic response monitoring, etc.) and intervention. In spite of the rapid development and great contributions achieved, cancer liquid biopsy is still a field under investigation and deserves more clinical practice. To better open up future work, here we systematically reviewed and compared the latest progress of the most widely recognized circulating components, including circulating tumor cells, cell‐free circulating DNA, noncoding RNA, and nucleosomes, from their discovery histories to clinical values. According to the features applied, we particularly divided the contents into two parts, beyond epigenetics and epigenetic‐based. The latter was considered as the highlight along with a brief overview of the advances in both experimental and bioinformatic approaches, due to its unique advantages and relatively lack of documentation.

Liquid biopsy has been increasingly used as a supplement to the traditional biopsy in clinical oncological practice. This review systematically summarizes the discovery histories, features, and clinical values of the key circulating components. The advantages and disadvantages of these circulating markers beyond or based on epigenetics is also discussed.

1. INTRODUCTION

Comprising a bewildering assortment of diseases, cancer ranks among the top leading causes of death worldwide, killing nearly 10 million people each year. ¹ , ² Though the past decades have witnessed an explosion in better understanding of mechanisms underlying tumorigenesis and the development of cancer therapy strategies, it remains a daunting challenge for cancer management, particularly when diagnosed at late stages with poor prognosis. ³ Thereby there is an urgent need to find efficient tools for early detection and prognostic prediction and to identify accurate biomarkers that allow monitoring of cancer status for long‐lasting surveillance.

Tissue biopsy, which relies on obtaining resected tumor samples through a core needle or open surgery, has long been considered as the gold‐standard method for tumor diagnosis. ⁴ Despite its significant contribution, shortcomings do exist, including limitations in early detection and monitoring the ever‐changing complete cancer progression as well as tumor heterogeneity throughout the entire patient journey due to difficulties in repeated specimen acquisition. ⁴ , ⁵ These snags, have given rise to a new approach: liquid biopsy. By definition, it depends on isolation and analysis of the tumor‑derived entities like circulating tumor cells (CTCs), protein, DNA, and RNA that are presented in bodily fluids such as blood or urine. ⁶ Since the complexity and heterogeneity of tumors results from the accumulation of disorders beyond and based on epigenetics, analysis of liquid biopsies has taken into account both these two parts. Specifically, the former usually includes quantitative alterations, somatic mutations, single‐nucleotide variants, and somatic copy‐number aberrations, while epigenetic‐based liquid biopsy includes characterization of DNA methylation, histone posttranslational modification (PTMs), noncoding RNA (ncRNA), and so on.

The feasibility to detect and characterize tumors in such a noninvasive and repeatable way has certainly received tremendous attention, and indeed, liquid biopsy has been demonstrated with considerable clinical implications in tumor diagnosis, prognosis, therapeutic efficacy evaluation, and so on. ⁷ , ⁸ , ⁹ As of April 2023, there are more than 9899 publications listed under the key phrase “cancer liquid biopsies” in PubMed targeting at almost all types of cancers. Meanwhile, more than 3000 clinical trials have been registered at the National Cancer Institute website (http://clinicaltrials.gov). However, the diverse circulating entities, different detection technologies and analysis methods, and complex biological characterizations (e.g., genetic and epigenetic disorders), making it confusing to the cancer research community and clinician. Clear summarizations are always needed to reorganize the massive information for better understanding and utility.

To this end, here, we systemically reviewed the studies and clinical applications of the key components of cancer liquid biopsies (CTCs, circulating tumor DNA [ctDNA], ncRNA, and nucleosomes), along with pointing out the pitfalls and future directions. In particularly, we divided the contents into two parts, beyond epigenetics and epigenetic based. The latter was considered as the highlight due to its immensity clinical potential and relatively lack of documentation. More specifically, we also outlined the recent advances in methodologies for better utility of epigenetic information in liquid biopsy.

2. CANCER LIQUID BIOPSY BEYOND EPIGENETICS

Benefiting from the relatively mature techniques, cancer liquid biopsy beyond epigenetics was relatively better documented and clinical applied up so far. In this section, we summarized the biological features of the most representative entities in body fluids, CTCs and ctDNA, and discussed how their disorders beyond epigenetics were used in cancer biopsy.

2.1. CTCs biology and clinical values

2.1.1. CTCs biology

CTCs were first discovered in 1869 by Ashworth in the blood of a metastatic cancer patient with appearance similar to those in the primary tumors. ¹⁰ Along with the deepened study, they have been observed in different cancer types such as breast, prostate, lung, and so on, and were found to be presented in various biological fluids in addition to blood, including urine, pleural effusion, ascites, and cerebrospinal fluid. ⁴ With that, CTCs were defined nowadays as tumor cells that have been sloughed from either primary or metastatic tumors during an intermediate stage of metastasis and shed into the circulatory systems. ¹¹ , ¹² However, so far, the morphologies of CTCs could not be well‐defined since they may vary between cancer types and developmental stages. ⁴ , ¹³ Not surprisingly, they also varied in genome, transcriptome, proteome, and metabolome. ¹³ , ¹⁴ Besides, CTCs can circulate as singles cells or congregate with parental tumor cells as well as leukocytes, platelets, or endothelial cells, leading to the formation of aggregates with higher tendency of distant metastases. ¹⁵ , ¹⁶ Indeed, it was reported that CTCs clusters appeared to have 23−50 fold enhanced metastatic capacity, along with a shorter half‐life (25–30 min for single cells vs. 6–10 min for clusters) in the circulation. ¹⁷

2.1.2. CTCs clinical values

Since CTCs may be derived from primary tumors and responsible for the formation of new distant metastases, targeting CTCs at different steps of the metastatic cascade could theoretically break off the progression of metastasis. Yet it has taken more than one century to recognize these critical roles of CTCs in cancer metastasis, largely due to difficulties in isolating the very rare CTCs from the massive pool of circulation of patients. ¹⁸ But the good news is, over the past two decades, the rapid advances of technologies such as high‐throughput sequencing, CRISPR/Cas9 editing tools and single‐cell sequencing allow for relatively more efficient isolation of CTCs and facilitate its clinical applications. ¹⁹ , ²⁰ , ²¹ Indeed, as of April 23, 2023, 1147 results were yielded based on the search term "circulating tumor cells" at ClinicalTrials.gov, further reflecting the great interest in clinic of CTCs. Here, we briefly summarized the most prominent CTCs clinical studies and applications, including both cancer prevention and monitoring.

CTCs‐targeting strategies for eliminating cancer metastasis

Intravasation of CTCs into the circulation and extravasation at distant sites are the basic steps of the metastatic cascade. ²² In a preclinical model of breast cancer (BC), Donato et al. ²³ showed that intratumor hypoxia upregulated cell–cell junction, facilitating the formation of CTC clusters with high ability of metastasis. Accordingly, a proangiogenic therapy via EphrinB2 treatment, which typically targeted at the vascular endothelial growth factor pathway, promoted vascularization and tumor growth rate, meanwhile inhibited intratumor hypoxia and intravasation of clustered CTCs, weakening the metastasis formation. ²³ In the follow‐up study, the authors suggested that some of the already available inhibitors such as PLK1 inhibitors could also prevent CTCs intravasation. ²⁴ Besides, it was reported pharmacological or genetic obstruction of invadopodia formation (e.g., cortactin or N‐WSAP), maturation (e.g., Tks5), or function (e.g., Tks4) abrogated the extravasation of cancer cell and the subsequent formation of metastatic colony in lung and breast metastasis mouse models. ²⁵ , ²⁶

Since accumulating evidence has shown that CTC clusters exhibit more robust metastatic properties than single CTCs, dissociation of CTC clusters therefore offers enormous clinical potential in limiting tumor metastasis. For example, inhibition of heparinase either by genetic knockdown or treatment with its inhibitors such as JG6 was sufficiently enough to destroy the cell clusters formation and repress BC metastasis. ²⁷ It was also found that injection of BC metastasis mouse model with a clinical thrombolytic agent, the urokinase‐type plasminogen activator, could effectively prevent CTC cluster assembly and prolong the overall host survival rate by up to 20% relative to the control animals. ²⁸ More recently, in combination of BC patients and mouse models, Gkountela et al. ²⁹ demonstrated that CTC clustering conferred stem‐like features and could be targeted with United States Food and Drug Administration (US FDA)‐approved cardiac glycosides drugs (Na⁺/K⁺ ATPase inhibitors ouabain and digitoxin), which enabled the dissociation of CTC clusters into single cells, leading to metastasis suppression. Based on these results, a proof‐of‐concept clinical trial has been set up to evaluate the effects of cardiac glycosides on disruption of CTC clusters in nine patients with advanced or metastatic BC (NCT03928210). In addition, dissociation of platelet–cancer cell interactions via antiplatelet agents was proved to be another therapeutic option to prevent tumor metastasis. ¹⁵ , ³⁰ Among them, aspirin and several COX‐1 inhibitors have been used routinely in the clinic for decades to centuries. ³¹ , ³² And several adenosine 5′‐diphosphate receptor (P2Y₁₂) antagonists and a thrombin receptor antagonist have also been approved by US FDA. ³¹ , ³³ Similarly, targeting the neutrophils–CTCs association, which contributes to cell cycle progression and metastatic potential of CTCs, also provides opportunity for cancer therapeutic intervention. ³⁴ , ³⁵

CTCs are characterized by rewiring of metabolism and homeostasis to support their growth and metastatic functions. ³⁶ Targeting these vulnerabilities might also be particularly interesting for cancer prevention. For instance, antioxidant defense was reported to be essential for CTCs to evade cell death, and it has indeed been proved that oxidative stress could inhibit the distant metastasis of melanoma. ³⁷ , ³⁸ Meanwhile, constriction of pyruvate metabolism was shown to be sufficient to impair collagen hydroxylation and ultimately the progression of lung metastases derived from BC in mouse models. ³⁹ Likewise, Elia et al. ⁴⁰ discovered that proline catabolism via proline dehydrogenase was required for metastasis formation of BC cells, and hampering Prodh was adequate to block formation of lung metastases in mouse models without adverse effects on healthy tissue and organ function.

More recently, accumulating studies have pointed out that immune checkpoint regulators such as PD‐L1 and CD47 were frequently expressed on metastatic cells, suggesting a new way to reduce the recurrence and metastasis of malignant tumors by blocking the immune checkpoints on CTCs. ⁴¹ , ⁴² , ⁴³ Dual blockage of two checkpoint inhibitors (e.g., PD‐L1 and CD47) or one immune checkpoint inhibitor in combination with EpCAM or HER2 could further enhance immunotherapy against CTCs compared with single‐agent therapy. ⁴⁴ , ⁴⁵ , ⁴⁶

Clinical predictive value of CTCs

Compared with CTCs for disease prevention, usages of CTCs as biomarkers for cancer early diagnosis, prognosis evaluation, and monitoring of therapeutic response were better documented and have been extensively reviewed.

Actually, whether CTC works in cancer early diagnosis is still controversial to date. It is likely that CTCs are very early events in various cancer types, including lung, breast, pancreatic cancer, and so on. ⁴⁷ , ⁴⁸ , ⁴⁹ , ⁵⁰ , ⁵¹ One of the most exciting findings was that, all the five patients with positive CTC count among the 168 patients with chronic obstructive pulmonary disease, developed lung cancer (LC) during follow‐up. Notably, CTCs in this study were detected 1∼4 years earlier than radiological signs of computer tomography‐scan screening. ⁴⁸ In liver cancer, Kalinich et al. ⁵² suggested that in high‐risk populations, the combination of CTC‐scoring assays with the standard serum biomarker α‐fetoprotein (AFP) may offer higher specificity and sensitivity required for hepatocellular carcinoma (HCC) screening, which AFP alone is unable to provide. These proof‐of‐concept results demonstrated the potential of CTCs detection as an early indicator of patients in high risks.

For patients who have been diagnosed with cancers, CTCs can represent as an independent prognostic factor, which offer valuable prognostic and therapeutic response information to help decision‐making during treatment. Most of the studies were based on CTCs count or molecular characteristics, which were performed by the CellSearch system, the only US FDA‐approved technique for CTC detection used clinically. ⁵³ A great deal of clinical trials is currently ongoing or completed with either positive or negative feedback across numerous cancers. For example, the PRODIGE 17 trial (NCT01443065) reported that change in CTC counts between baseline and day 28 could help to tailor treatment to each individual patient with advanced gastric and esophageal cancer, since it was significantly associated with overall survival and progression‐free survival. ⁵⁴ In prostate cancer (PC), given that androgen receptor splice variant 7 (AR‐V7) functions as a very promising biomarker and patients with CTCs AR‐V7(+) metastatic castration‐resistant prostate cancer (CRPC) had a very poor outcome, Armstrong et al. ⁵⁵ conducted a multicenter, prospective‐blinded trial PROPHECY (NCT02269982). AR‐V7 detection in CTCs via two blood‐based assays was shown to be independently linked to worse progression‐free and overall survival with enzalutamide or abiraterone; therefore, alternative treatments should be offered to such patients. ⁵⁵ On this basis, a phase II study that was designed for the determination of the response of patients with metastatic CRPC and AR‐V7(+) CTCs to cabazitaxel (NCT03050866) is ongoing. ⁵⁶ In BC, while the interventional SWOG‐S0500 trial (NCT00382018) failed to show a benefit of CTCs count guided intervention versus physician's choice at disease progression, ⁵⁷ the METABREAST STIC CTC randomized clinical trial (NCT01710605) carried out more recently demonstrated that in hormone receptor‐positive, ERBB2‐negative metastatic BC, CTC count could be used as a reliable marker to guide the decision making for the choice between endocrine therapy and chemotherapy as the first‐line treatment. ⁵⁸ Nevertheless, since STIC CTC was performed before the results of the first‐line phase III trials on CDK4/6 inhibitors, the conclusions could not be extrapolated to CDK4/6 combination therapy. ⁵⁹

In summary, over the last two decades, the biological characteristics of CTCs and clinical value has been widely investigated in most common cancer types, in both localized and metastatic settings. Even though the benefits of CTCs on cancer is not very clear so far according to results of the extensive clinical trials, but just as said by Aceto and coworkers, ⁶⁰ the time is coming for translation of this work into clinical practice thanks to the rapid advances in technology.

2.2. ctDNA biology and clinical values beyond epigenetics

2.2.1. cfDNA and ctDNA biology

The term cell‐free DNA (cfDNA) refers to a mixture of nucleic acids that are released outside of cells that can be detected within bodily fluids, which was originally identified in 1948 by Mandel et al. ⁶¹ in blood samples collected from healthy individuals. However, it was not until 1977 that Leon and colleagues discovered increased levels of cfDNA in patients with pancreas cancer, leading to the hypothesis that some of them were tumor derived. Those were later termed as ctDNA. ⁶² , ⁶³ Further studies showed that ctDNA could be detected in almost 100% of the bladder, colorectal, and ovarian cancer (OC), over 50% of cases with most cancer types, and as little as 10% of glioma cases. ⁶⁴ , ⁶⁵ , ⁶⁶

The concentration of cfDNA in healthy adults is generally quite low, often less than 10 ng/mL of plasma. In cancer patients, it can be anywhere 50 times higher but the proportion of ctDNA in the background of overall cfDNA is highly variable, ranging from <0.05 to 90%. ⁶⁷ Typically, cfDNA is found as double‐stranded fragments of approximately 150−200 base pairs in length, corresponding to the unit size of nucleosome, while ctDNA comprises strands of <145 bp in length. Besides, the short half‐life of ctDNA in the circulation, which was reported to be from 16 min to 2.5 h, making it a very important tool in monitoring the real‐time situation of tumor dynamics. ⁶⁷ , ⁶⁸

2.2.2. ctDNA clinical values beyond epigenetics

In recent years, the clinical application of ctDNA has been well established in routine practice and gained enormous attention from both academic groups and commercial vendors (Table 1). The current assays are based on either genetic (mutations, copy number, fragmentations) or epigenetic aberrations. In this part, we will focus on the former, while a relatively detailed discussion about promise of epigenetics in liquid biopsies is given in the next section.

TABLE 1.

Examples of cancer liquid biopsy products.

Year of approval	Company	Test	Cancer type	Function	Basis	Status
2016	Roche Diagnostics	Cobas EGFR mutation Test v2	NSCLC	Companion diagnostic for targeting treatment	Mutations	US FDA approved
2019	Qiagen Manchester	therascreen PIK3CA RGQ PCR Kit	Breast cancer	Companion diagnostic for targeting treatment	Mutations	US FDA approved
2020	Guardant Health	Guardant 360 CDx	Pan‐cancer	Companion diagnostic for targeting treatment	Mutations	US FDA approved
2020	Foundation Medicine	FoundationOne Liquid CDx	Pan‐cancer	Companion diagnostic for targeting treatment	Mutations	US FDA approved
2019	Invitae Archer DX	Stratafide	Pan‐cancer	Companion diagnostic for targeting treatment	Mutations	FBDD
2017	Natera	Signatera	Pan‐cancer	Molecular residual disease assay	Mutations	FBDD
2020	Thrive Earlier Detection	CancerSEEK	Pan‐cancer	Early detection	Mutations	FBDD
2012	Gen‐Probe	PROGENSA	Prostate cancer	Early detection	lncRNA	US FDA approved
2016	Epigenomics	Epi proColon	CRC	Early detection	DNA methylation	US FDA approved
2020	Bluestar Genomics	Avantect Pancreatic Cancer Test	Pancreatic cancer	Early detection	5hmC	FBDD
2021	AnchorDX	UriFind	Bladder cancer	Early detection	5mC	FBDD

Abbreviations: CRC, colorectal cancer. Data were derived from https://www.hhs.gov/; FBDD, US FDA Breakthrough Devices Designation; NSCLC, non‐small cell lung cancer.

As the model tumor linking genomic disorders to cancer therapy, ctDNA analysis of non‐small cell lung cancer (NSCLC) have paved the road for the Cobas EGFR mutation Test v2 (Roche Diagnostics), the first ctDNA‐based companion diagnostic test to be approved by US FDA and the European Medicines Agency. ⁶⁹ , ⁷⁰ It was based on the results of a multicenter, open‐label, randomized, Phase III study (ENSURE), which aimed to compare the efficacy and safety of erlotinib to gemcitabine plus cisplatin as first‐line treatment for stage IIIB/IV EGFR‐mutation positive NSCLC patients. ⁷¹ Currently, the test contributes to the detection of patients with advanced NSCLC whose tumor carries substitutions (G719X) in exon 18, deletions in exon 19, insertions and substitutions (T790M, S768I) in exon 20, and substitutions (L858R, L861Q) in exon 21. ⁷² In 2020, another companion diagnostic test, FoundationOne Liquid CDx (FoundationOne), which was also based on the detection of EGFR mutations in plasma ctDNA of NSCLC patients, was approved, but using next‐generation sequencing instead of real‐time PCR. It is worth noting that this test also contains ALK rearrangement and METex14‐based detection of NSCLC, PIK3CA mutations‐based detection of BC, BRCA1, BRCA2, and ATM mutations‐based detection of CRPC, and BRCA1 and BRCA2‐based detection of OC. ⁷³ Meanwhile, a series other assays used for adjuvant therapy were sprang up, such as Guardant 360 CDx assay (Guardant Health) based on simultaneous mutation detection of 55 tumor genes for pan‐cancer, and therascreen PIK3CA RGQ PCR Kit (QIAGEN Manchester) based on PIK3CA mutations for BC.

Another example should be mentioned is the CancerSEEK platform, the first screening methodology for cancer early detection through assessment of the levels of circulating proteins and mutations in cfDNA. ⁷ , ⁷⁴ It was reported by Cohen et al. ⁷ that the CancerSEEK tests were positive in around 70% of the eight cancer types (breast, lung, ovary, stomach, liver, pancreas, colorectum, esophagus). The sensitivities ranged from 69 to 98% for the detection of the five cancer types (esophagus, stomach, ovary, liver, and pancreas) without screening test available for average‐risk individuals. ⁷ As a follow‐up study, an exploratory prospective, interventional study, called DETECT‐A was performed to evaluate the feasibility and safety of CancerSEEK coupled with PET‐CT imaging of 10,006 women not previously known to have cancer. The specificity of CancerSEEK alone was estimated to be 98.9%, and increased to 99.6% when combined with PET‐CT. ⁷⁵ CancerSEEK is now commercially available since the corresponding clinical trial (NCT04213326) has already completed, but its approvement by US FDA is still awaited.

For the surveillance of cancer recurrence after surgical resection, the first ctDNA‐based marketed assay is Signatera (Natera), a highly personalized and sensitive molecular residual disease assay. Specifically, it is custom designed for each patient to help them to detect relapse earlier than the standard care methods such as carcinoembryonic antigen (CEA) level measurement and CT imaging, and with more than 98% overall positive predictive value of relapse across multiple solid tumors (colorectal, breast lung, bladder cancer, etc.). ⁷⁶ , ⁷⁷ , ⁷⁸

Despite the contributions achieved, negative comments on these assays do exist. For instance, Fakih et al. ⁷⁹ recruited 48 patients with resected CRC to compare the specificity and sensitivity of ctDNA surveillance (Signatera) with measurement of CEA levels, imaging, and combination of imaging with CEA levels as suggested by the National Comprehensive Cancer Network guidelines. Results indicated that ctDNA did not achieve better sensitivity than imaging in detecting recurrence (53.3 vs. 60%). And combination of imaging with CEA measurement even showed greater superiority than ctDNA, with sensitivity of 73.3%. ⁷⁹ The findings brought into questions the reliability and safety of ctDNA assay regarding the risk of disease recurrence in CRC and the authors therefore called for additional investigations before the adoption of ctDNA solely in clinical practice. Besides, Sullivan et al. ⁸⁰ advised that ctDNA is unreliable to detect somatic gene alterations in gastrointestinal peritoneal carcinomatosis with a very low concordance rate of 18%, according to a retrospective analysis of available panel‐based ctDNA results (Guardant 360 CDx) of patients with gastrointestinal cancers treated between 2015 and 2020.

Overall, in addition to the promising potential, caution is warranted in utilizing ctDNA as a therapeutic decision‐making tool in patients, as the detection of ctDNA can be influenced by multiple factors such as disease location, burden, treatment, and tumor vascularization. More investigations are still needed.

3. EPIGENETIC‐BASED CANCER LIQUID BIOPSY

As discussed above, genetic disorders have made great contributions in the field of cancer liquid biopsy. However, it is noteworthy that entities within the circulation of one body share the same genetic information, thus limiting the ability for the detection of tissue‐of‐origin to pinpoint specific cancer type. ¹ , ⁸¹ Meanwhile, genetic information usually varies from case to case, challenging the identification of sensitive and generalizable biomarkers. ⁸² Epigenetic profiles could cover perfectly these shortages, as it differs between cells from different tissue types or those with different pathologies but is generalizable across individuals. Even more, the frequency, reversibility, and accessibility of epigenetic alterations in body fluids, endow them the potential to become excellent candidates as clinically valuable cancer biomarkers (Figure 1). ⁸³ , ⁸⁴ In the following sections, we focused particularly on the most widely recognized circulating epigenetic biomarkers, including DNA methylation/hydroxymethylation, histone PTMs, and ncRNAs.

FIGURE 1.

Examples of key circulating components beyond or based on epigenetics. Their advantages and disadvantages are indicated, respectively. The elements were downloaded from https://www.vecteezy.com/ and the figure was created with Adobe Illustrator 2020.

3.1. ctDNA methylation and hydroxymethylation

3.1.1. ctDNA 5‐methylcytosine and clinical values

5‐Methylcytosine (5mC) is one of the most deeply studied epigenetic modifications that happens by the addition of a methyl group to the C‐5 position of DNA cytosine ring by DNA methyltransferases, occurring predominantly in CpG dinucleotides. ⁸⁵ Cumulative evidence has demonstrated that abnormal DNA methylation plays a critical role in cancer development, mainly through focal hypermethylation at multiple genomic regions (mostly CpG sites), global hypomethylation, or mutagenesis at methylated cytosines. ⁶⁶ , ⁸⁵ , ⁸⁶ Compared with other markers, DNA methylation imbalance typically precedes tumor formation in a high percentage of cancer cases, facilitating early detection. ⁸⁷ , ⁸⁸ Besides, they are chemically stable and the methylation patterns are often tissue specific, which are helpful to unmask the original primary tumor site of cancer of unknown primary. ⁸⁹ These superiorities make cancer‐associated DNA methylation alteration quite attractive for the development of cancer biomarker tests with substantive clinical utility.

The earliest proof that cancer‐related aberrant DNA methylation can be measured in the serum of patients came in 1999, when Wong et al. and Esteller et al. showed promoter hypermethylation of tumor suppressor genes such as p16 in the circulation of HCC and NSCLC patients, respectively (Figure 2). ⁹⁰ , ⁹¹ Notwithstanding the limited sample size and number of targeted genes, these two studies that were published back‐to‐back opened the window into taking advantage of DNA methylation profiles in noninvasive detection. ⁹⁰ , ⁹¹ Since then, a considerable number of studies have been carried out to characterize changes in methylation of ctDNA, both at the global or gene‐/locus‐ specific level (Table 2). Common hallmarks of various cancers, including global hypomethylation, hypomethylation of DNA repeat elements, oncogene hypomethylation and hypermethylation of CpG islands in the gene promoter, were all demonstrated to be detectable in ctDNA. ⁸⁴ , ⁹² The identified biomarkers have been proven to provide considerable aid across the whole patient journey from early diagnosis, to treatment and recurrence monitoring (Table 2). A very representative example of ctDNA methylation‐based biomarker is SEPT9 in colorectal cancer (CRC). Occurring in over 90% of CRC tissues, DNA hypermethylation at the promoter region of SEPT9 is recognized as a hotspot and is considered to be a specific biomarker of CRC early stages. ⁹³ In 2008, Sledziewski and coworkers ⁹⁴ measured levels of DNA methylation in three marker genes (SEPT9, NGFR, TMEFF2) in human plasma samples from CRC patients or healthy controls, of which SEPT9 was proved to be the most specific one. After validation by a series of follow‐up case–control studies, the plasma‐based SEPT9 gene methylation assay, developed as the “Epi proColon” test, was ultimately approved by the US FDA in 2016 as the only blood‐based CRC screening test (Table 1). ⁶⁶ , ⁸⁴ , ⁹⁵ This is an important milestone of bringing circulating methylation biomarkers from the laboratory to the clinical practice. More recently, SEPT9 was shown to be also effective for CRC postsurgical assessment and prognosis prediction. ⁹⁶ , ⁹⁷ Soon afterwards, another two commercially available tests for LC detection in plasma‐cfDNA levels namely “Epi proLung” and “lung EpiCheck” were used in clinical practice in 2017 and 2021, respectively. Epi proLung is based on aberrant DNA methylation of stature homeobox 2 (SHOX2) and prostaglandin E receptor 4 (PTGER4) genes with a fixed specificity of 90% and sensitivity of 67% for LC, ⁹⁸ while lung EpiCheck is based on the measurement of methylation level of six genes, enabling detection of up to 85% of early‐stage LC with a specificity of 64% in the high‐risk population. ⁹⁹

FIGURE 2.

Timeline of main discoveries about epigenetic‐based cancer liquid biopsy with a focus on clinical applications. (A) cfDNA and ctDNA methylation. (B) Circulation nucleosomes and histone PTMs. (C) Circulating miRNA. The elements were downloaded from https://www.vecteezy.com/ and the figure was created with Adobe Illustrator 2020.

TABLE 2.

Summary of recent representative ctDNA methylation‐based liquid biopsy studies in breast, lung, colorectal, ovarian cancer, and multiple cancer types.

Year	Cancer type	Clinical sample size	Type of measured DNA methylation	Markers identified	Function	References
2022	BC	123 patients, 40 controls.	Global	15 markers	Diagnosis and subtyping	100
2022	BC	59 patients, 20 controls.	Gene specific	ESR1	Therapeutic efficacy evaluation	101
2021	BC	204 patients, 129 controls	Gene specific	26 markers	Diagnosis	102
2021	BC	101 patients, 102 controls	Global	10 hypomethylation markers	Diagnosis	103
2019	BC	200 patients, 35 controls.	Gene specific	KLK10, SOX17, WNT5A, MSH2, GATA3	Prognosis	104
2018	BC	230 patients, 60 controls	Gene specific	ESR1	Therapeutic efficacy evaluation	105
2017	BC	31 tissues and 1869 serum samples	Global	Six markers, of which EFC#93, was validated for clinical use.	Diagnosis	106
2016	BC	56 micro‐dissected breast tissue specimens, 29 blood samples from healthy volunteers	Global & gene specific	12 novel epigenetic markers (JAK3, RASGRF1, CPXM1, SHF, DNM3, CAV2, HOXA10, B3GNT5, ST3GAL6, DACH1, P2RX3, and chr8:23572595) for detecting BC and four internal control markers (CREM, GLYATL3, ELMOD3, and KLF9).	Diagnosis	107
2013	BC	319 patients, 237 controls	Gene specific	ITIH5, DKK3, and RASSF1A	Diagnosis	108
2013	BC	193 patients, 60 controls	Gene specific	SOX7	Prognosis	109
2022	LC	44 patients, 39 controls	Gene specific	BCAT1	Diagnosis	110
2022	LC	64 patients, 12 controls	Gene specific	APC, RASSFIA1, FOXA1, SLFN11, SHOX2	Prognosis	111
2021	LC	91 patients, 60 controls	Gene specific	KMT2C	Prognosis	112
2021	LC	311 patients, 417 controls	Gene specific	6 hypermethylated markers	Diagnosis	99
2019	LC	132 patients, 118 controls	Global	9 hypermethylated markers	Diagnosis	113
2017	LC	83 patients, 40 controls	Gene specific	CDO1, HOXA9, AJAP1, PTGDR, UNCX, and MARCH11	Diagnosis and prognosis	114
2016	LC	118 patients, 212 controls	Gene specific	SHOX2, PTGER4	Diagnosis	98
2014	LC	122 patients, 24 controls	Gene specific	BRMS1	Prognosis	115
2022	CRC	15 patients, 5 controls	Global	CIMP subtypes	Diagnosis and prognosis	116
2022	CRC	55 patients	Gene specific	SFRP2 and SDC2	Therapeutic efficacy evaluation	117
2021	CRC	272 patients, 402 controls	Gene specific	MYO1‐G	Diagnosis and therapeutic efficacy evaluation	118
2021	CRC	96 patients, 78 controls	Gene specific	10 markers	Therapeutic efficacy evaluation	119
2020	CRC	801 patients, 1021 controls	Global & gene specific	cg10673833	Diagnosis and prognosis	120
2019	CRC	147 patients, 136 controls	Global	Dozens of DNA hypermethylation markers including known (e.g., SEPT9 and IKZF1) and novel (e.g., EMBP1, KCNQ5, CHST11, APBB1IP, and TJP2) genes	Diagnosis	121
2018	CRC	184 patients, 224 controls	Global	SEPT9 and SHOX2	Diagnosis and therapeutic efficacy evaluation	122
2018	CRC	182 patients, 50 controls	Global & gene specific	EYA4, GRIA4, ITGA4, MAP3K14‐AS1, MSC	Therapeutic efficacy evaluation	123
2014	CRC	3078 patients, 3796 controls	Gene specific	SEPT9	Diagnosis	124
2022	OC	249 patients, 288 controls	Global	15 markers	Diagnosis and prognosis	125
2020	OC	85 patients, 35 controls	Gene specific	HOXA9, HIC1	Diagnosis	126
2015	OC	114 patients, 80 controls	Gene specific	RUNX3, TFPI2, OPCML	Diagnosis	127
2021	Multiple	8542 patients, 6712 controls	Global	>100,000 informative methylation regions	Diagnosis	128
2020	Multiple	414 patients, 614 controls	Global	10,613 CpG sites across 477 genomic regions	Diagnosis	129
2020	Multiple	2482 patients, 4207 controls	Global	>100,000 informative methylation regions	Diagnosis	130

Abbreviations: BC, breast cancer; LC, lung cancer; OC, ovarian cancer.

Certainly, studies of ctDNA methylation to evaluate tumor occurrence, progression, and recurrence have spread to other types of cancers, such as OC, BC, PC, ¹⁰³ , ¹²⁵ , ¹³¹ , ¹³² , ¹³³ and so on. Specifically, higher frequencies of methylated RASSF1A and BRCA1 in blood‐derived DNA from patients with BC than from healthy females were reported repeatedly. Other tumor suppressor genes, such as DAPK, GSTP1, RARB, APC, SFN, and ESR1 were also found to be more frequently methylated in BC cases than in controls. ¹³⁴ , ¹³⁵ , ¹³⁶ In OC, well‐described aberrant methylation‐based markers include BRCA1, RASSF1A, OPCML, HOXA9, RUNX3, TFPI2, and so on. A case in point is HOXA9, a homeobox gene normally expressed during differentiation of Mullerian ducts into the female reproductive tract, whose promoter hypermethylation has been suggested as a biomarker for noninvasive diagnosis of OC as well as an indicator of disease grade, since it can be detected in most high‐grade serous OC tissue and plasma samples. ⁸⁸ Yet it gave a relative unimpressive result when works alone. For optimization, Singh et al. ¹²⁶ developed a novel marker panel with combined analysis of HOXA9 and HIC1 hypermethylation, which achieved enhanced sensitivity (88.9%) in distinguishing individuals with OC from controls. Besides, HOXA9 showed potential in informing clinical decision‐making, as detection of HOXA9 meth‐ctDNA, at baseline and during treatment with the PARP inhibitor veliparib, was strongly associated with patient outcomes in BRCA1/2‐mutated, platinum‐resistant, and intermediate‐resistant OC. ¹³³

In fact, in addition to hypermethylation at the promoter of specific tumor suppressor genes, the most dominant epigenetic alteration should be global hypomethylation, particularly in the repetitive DNA sequences, which make about half of the human genome. ¹³⁷ , ¹³⁸ One of the most well studied examples is the long interspersed nuclear element‐1 (LINE‐1) sequences. Based on analysis of LINE‐1 methylation status in the cfDNA from 21 patients with esophageal adenocarcinoma (n = 19) or Barrett's esophagus (n = 2), Saggioro and coworkers ¹³⁷ showed that hypomethylated LINE‐1 sequences were present in EADC cfDNA and correlated with tumor progression. In CRC, the cfDNA LINE‐1 hypomethylation index (LHI) in patients with large tumors (≥6.0 cm), advanced N stage (≥2), and distant metastasis (M1) was statistically significantly higher than other CRC patients. Additionally, either CRC patients at early stage I/II (n = 57) or advanced stage III/IV (n = 57) had significantly higher LHI than healthy controls (n = 53), suggesting the potential of plasma cfDNA LINE‐1 hypomethylation as a novel biomarker for CRC, particularly for early stage detection. ¹³⁷ In NSCLC, Azhikina and coworkers ¹³⁹ observed a significant difference of LINE‐1 promoter methylation index between cancer patients (n = 56) and healthy controls (n = 44) (ROC‐curve analysis: n = 100, AUC = 0.69, p = 0.0012). Besides, in another study, they indicated a tight association between this serum marker and the pathological process of LC, since changes in the LINE‐1 methylation index of cell surface‐bound csb‐criDNA depended on the patient response to chemotherapy. ¹⁴⁰

There were also a series of arrays targeted at multicancers based on either hyper‐ or hypomethylation regions, or in combination of both (Table 2). For example, it was reported that the DNA methylation‐based blood test, PanSeer assay, was able to detect five common cancer types including stomach, esophageal, colorectal, lung, or liver cancer up to four years prior to conventional diagnosis regardless of tissue‐of‐origin in a robust manner. ¹²⁹ A larger case–control study, with 6689 participants [2482 cancer (>50 cancer types), 4207 noncancer] and targeting a panel of >100,000 informative methylation regions from plasma cfDNA was performed by the healthcare company, GRAIL. They successfully detected a wide range of cancer types at both nonmetastatic or metastatic stages, with sensitivity and specificity performance closing to the goal of population‐level screening. Two corresponding clinical trials in intended‐use populations are ongoing (NCT02889978, NCT03085888) ¹³⁰ and one of the products Galleri has already been commercially used (Table 1).

3.1.2. ctDNA 5hmC and clinical values

With the mediation of ten‐eleven translocation (TET) enzymes, 5mC can be oxidized to 5‐hydroxymethylcytosine (5hmC), and further to 5‐carboxylcytosine and 5‐formylcytosine. ¹⁴¹ , ¹⁴² 5hmC is the most abundant and stably oxidized product among them, ¹⁴³ but when compared with the 8% methylated cytosine in the human genome, number of hydroxymethylated cytosines is really tiny with only 0.5 to 1%. ¹⁴⁴ Meanwhile, it was reported that in contrast to 5mC, 5hmC‐based profiles are relatively more stable and robust, providing better specificities. ¹⁴⁵ Since 5hmC are found to be enriched in enhancers, promoters, and other regulatory elements, it is understandable that changes in 5hmC are associated with changes in gene expression. ¹⁴⁶ , ¹⁴⁷ , ¹⁴⁸ More recently, aberrations of 5hmC have been linked to various cancers, such as malignant melanoma, BC, bladder cancer, and NSCLC, and have been considered as potential cancer biomarkers and therapeutic targets. ¹⁴⁹

The first proof‐of principle global analysis of hydroxymethylome in cfDNA was carried out in 2017 by Quake et al., who developed a low‐input whole‐genome 5hmC sequencing method based on chemical labelling and applied it on cfDNA from 49 patients of seven different cancer types (LC, hepatocellular carcinoma, pancreatic cancer, glioblastoma, gastric cancer, colorectal cancer, BC). Results showed that cell‐free 5hmC display distinct features that could be used to predict cancer types and stages with high accuracy, proposing a new dimension of information to liquid biopsy‐based diagnosis and prognosis. ¹⁵⁰ Subsequently in 2020, they presented a noninvasive early detection study of pancreatic ductal adenocarcinoma (PDAC) based on 5hmC alterations in cfDNA with AUC of 0.92 in discovery dataset (n = 79) and 0.92–0.94 in two independent test sets (n = 228). ¹⁵¹ The corresponding product belongs to Bluestar Genomics, a company dedicated to cancer early detection. Besides, just a few days ago, Levy and coworkers ¹⁵² published their latest research about 5hmC‐based early detection of pancreatic cancer using cfDNA. In combination of 5hmC sequencing with machine learning algorithm, they developed a precise, efficient, and scalable assay, which needs only 10 ng of cfDNA, enabling the effective measurement of cancer presence in individuals at high risk for PaC. The results were based on the case–control clinical trial carried out from June 2018 to May 2022 (NCT03869814), in which cancer and noncancer subjects were recruited from 146 sites across the United States. To provide further supporting evidence on clinical performance of the assay, the authors have launched another interventional study by enrolling up to 6500 subjects from multisite and across multiyear (NCT05188586).

Another team must be introduced is Zhang and his colleagues, ¹⁵³ who have also been devoted to the field of 5hmC‐related liquid biopsy for many years. In fact, just one month after Quake et al.’s publication in 2017, Zhang et al. ¹⁵³ reported their findings that cfDNA biomarkers based on 5hmC were highly predictive for gastric and CRCs and were superior to other conventional cfDNA biomarkers and meanwhile comparable to 5hmC biomarkers from tissue biopsies. Later, they established a 5hmC‐based diagnostic model, which could accurately discriminate early HCC (stage 0/a) from non‐HCC (validation set: area under curve (AUC) = 88.4%; (95% CI 85.8–91.1%)). Meanwhile, the model showed high capacity for distinguishing early HCC from high‐risk subjects, which were with liver cirrhosis or chronic hepatitis B virus infection (validation set: AUC = 84.6%; (95% CI 80.6–88.7%)). Even more surprisingly, this approach showed superior performance over AFP, the most common serum test applied for HCC screening and diagnosis. ¹⁴⁸

3.2. Circulating nucleosome and their PTMs

3.2.1. Nucleosome and histone PTMs

As the basic unit of chromatin in eukaryotes, each nucleosome contains an octamer with two copies of each of the core histones H2A, H2B, H3, and H4, joined together by a linker histone H1 with approximately 146 base pairs of DNA wrapped in 1.67 left‐handed superhelical turns around the octamer. ⁸⁷ Histone tails are decorated by a plethora of covalent PTMs on distinct sites and even to several degrees (mono‐, di‐, tri‐) at a single residue, including methylation (me), acetylation (ac), ubiquitylation (ub), and so on. Histone PTMs have been extensively proven to play fundamental roles in the regulation of the underlying genes, but their functions are remarkably specific due to the highly diverse types, in which each mark all has its own message to convey and therefore was suggested to serve as an epigenetic code. ¹⁵⁴ For example, acetylation is generally associated with “relaxed” chromatin and activation of gene transcription, while according to the degree of methylation and the residue methylated, methylation can have various functional consequences. For example, H3K27me3 and H3K9me3 are related to gene silencing but H3K4me2 and H3K4me3 are linked to gene activation. ¹⁵⁵

Aberrations of histone PTMs patterns, which are usually a consequence of the deregulation of enzymes responsible for the deposition (known as writers, e.g., histone acetyltransferase, methyltransferase) and removal (known as erasers, e.g., deacetylases, demethylases) of modifications, were identified as general hallmarks of cancer. ¹⁵⁶ The typical histone PTMs as cancer biomarkers are illustrated here in Figure 3. More importantly, PTM players did have shown their excellent capacity as targets of cancer therapy, with several drugs targeting at normalization of H3K27 methylation or histone lysines acetylation having been successfully approved by US FDA. ¹⁵⁷

FIGURE 3.

Typical histone posttranslational modifications implicated in cancer. Only canonical histones H2A, H2B, H3, and H4 are involved here. The figure was created with Microsoft Office PowerPoint.

3.2.2. Circulating histone PTMs and clinical values

Similar to cfDNA, nucleosomes can also be released into biological fluid following cell death or may as well as active mechanisms, which are referred to as circulating nucleosomes. ¹⁵⁸ Actually, it is said some proportion of cfDNA may circulate as nucleosomes or chromosomes, rather than as free DNA. ¹⁵⁹ Since many studies have to be performed retrospectively on serum and plasma samples collected earlier, Sozzi et al. ¹⁶⁰ and Holdenrieder et al. ¹⁶¹ investigated the loss of DNA and nucleosomes during long‐term storage, respectively. A dramatic decrease of DNA concentration of about 30% per year was reported, while the annual loss of circulating nucleosome is only about 7%, reflecting its relatively more stable structure. ¹⁶⁰ , ¹⁶¹ Though many studies have demonstrated a generally increased level of circulating nucleosomes in cancers including breast, prostate, colorectal, and LC compared with healthy controls, especially in cases with advanced tumors. However, similar phenomenon can also be observed in many benign diseases, thus reducing their clinical utility for cancer monitoring to a certain extent. ¹⁵⁵

In 2008, Deligezer et al. ¹⁶² found that H3K9me1 was detectable in the promoter region of CDKN2A in a third of the plasma samples from 21 multiple myeloma patients. It was the first study to show evidence of the presence of PTMs in circulating nucleosomes, nearly ten years after the discovery of cfDNA methylation (Figure 2). ¹⁶² Subsequently, a growing number of investigations pointed to the fact that the characterization of PTMs on circulating nucleosomes is an exciting new area that could provide valuable sources for cancer early detection and treatment response monitoring. For instance, CA 19‐9 is regarded as the gold standard marker for PC detection with an average clinical sensitivity of 79 at 82% specificity, while when it was combined with a panel of four epigenetic circulating nucleosome biomarkers (5mC, H2A.Z, H2A1.1, H3K4me2), the sensitivity increased to 92 at 90% specificity. ¹⁶³ In CRC, Rahier et al. ¹⁶⁴ suggested an age‐adjusted panel based on three cf‐PTMs (H2AK119Ub, H3K9Ac, H3K27Ac) and the global level of cf‐nucleosomes, which reached an AUC of 0.97 for the distinguish between patients and healthy controls with sensitivity of 75 and 86 at 90% specificity for stages I and II, respectively. More recently, in response to the problem that the concentration of circulating nucleosomes is extremely low but the concentration of native antibodies is high in plasma, Sadeh et al. ¹⁶⁵ devised a protocol for chromatin immunoprecipitation (ChIP)‐seq of cf‐nucleosome from less than 2 mL of plasma and applied it to around 250 samples from more than one hundred subjects, including 61 self‐declared healthy donors, 29 patients suffering from autoimmune, metabolic, or viral liver diseases, 56 patients with metastatic CRC, and four patients with acute myocardial infarction. Levels of four histone PTMs associated with active transcription including H3K4me1/2/3 and H3K36me3 were measured. In addition to the determination of cells of origin, the cfChIP‐seq showed a high sensitivity for detection and identification of subgroups of CRC patients with distinct cancer‐related transcriptional programs. ¹⁶⁵ Later that same year, another study established EPINUC as a novel liquid biopsy approach, allowing analysis of multiple histone PTMs and DNA modifications at single‐molecule precision, which could distinguish CRC patients from healthy controls at high sensitivity and specificity, even at early stages. ¹⁶⁶

It is worth mentioning that the prominent advantage of circulating nucleosomes as biomarkers is the diversiform variety of potential epigenetic features available, which enables fine‐tuning of sensitivity and specificity. Indeed, when assessed individually, the ability of these epigenetic markers in discriminating cancer patients from controls seemed always unremarkable. However, the performance was substantially improved upon combination of multimarkers. Even though neither of these reported panels has been put into clinical practice up to now due to the relatively later starting, a breakthrough is clearly foreseeable in the near future.

3.3. Circulating ncRNA

Asides from DNA and proteins, another encouraging source of epigenetic‐based liquid biopsy biomarkers is RNA. In human, approximately 80% of transcripts are noncoding (ncRNA), but functional. ¹⁶⁷ It has been extensively reported that ncRNAs are frequently overexpressed in cancers and suitable for the monitoring of cancer progression as well as recurrence. ¹⁶⁸ , ¹⁶⁹ Accordingly, a series of investigations have suggested the utility of circulating ncRNAs for cancer detection, especially microRNAs (miRNAs) and long ncRNAs (lncRNAs). ¹⁷⁰ , ¹⁷¹

3.3.1. Circulating miRNA and clinical values

miRNAs are a class of endogenous, single‐stranded, small ncRNAs with an average length of 22 nucleotides. Ever since their discovery in 1993, miRNA was closely watched due to its pivotal role in the gene regulatory network at the posttranscription level. ¹⁷² , ¹⁷³ Mechanically, miRNAs function via base pairing to complementary sequences within the target mRNA molecules, resulting in gene silencing by mRNA cleavage or translational repression depending on the degree of complementarity. ¹⁷⁴ As a single miRNA could target up to 400 different mRNAs, it is predicted that more than 30% of the human genes could be directly regulated by miRNAs, involving in almost all fundamental cell processes. Not surprisingly, dysregulations of miRNAs are recognized as features of many pathological processes including cancer. ¹⁷⁵ In general, those significantly overexpressed miRNAs in cancer patients, which accelerate tumor occurrence, development, or metastasis, were considered as oncogenes, and those decreased were regarded as tumor suppressors. ¹⁷⁶ Furthermore, the expression of miRNAs is usually tissue and disease specific, making them potential promising biomarker candidates.

While a majority of miRNAs remain intracellular, about 10% of the known human miRNAs can be secreted into extracellular space and transported to the circulating body fluid such as peripheral blood, which are termed as circulating miRNAs or cell‐free miRNAs. One mind‐blowing point is that unlike cfDNA or circulating nucleosomes, these miRNAs can be easily detected in plasma or serum in a fairly stable form, thanks to the protection from the extracellular environment by binding to special lipid proteins and encapsulation by extracellular vesicles, therefore being resistant to RNase digestions. ¹⁷⁶ , ¹⁷⁷ The half‐life of circulating miRNA is reported to be varied widely between miRNAs, with an average of up to 12 h. ¹⁷⁸

The significance of circulating miRNAs in cancer was first highlighted in 2008 by Lawrie et al., ¹⁷⁹ who performed a retrospective study to compare levels of tumor‐associated miRNAs (miR‐155, miR‐210, and miR‐21) from the serum of healthy controls and patients with diffuse large B‐cell lymphoma (DLBCL). Both of them were found to be elevated in DLBCL, among which miR‐21 high expression was associated with relapse‐free survival. ¹⁷⁹ Soon afterwards, Mitchell et al. ¹⁸⁰ indicated that serum levels of miR‐141 can robustly discriminate patients with PC from the healthy controls. In the next few years, there was a burst of studies, correlating circulating miRNAs to multiple cancers including breast, bladder, pancreatic, lung, colorectal, and gastric cancers.

Overexpression of miR‐17, miR‐21, miR‐25, miR‐93, miR‐103, miR‐106b, miR‐151, and miR‐181a in serum specimens was shown to be superior to current clinical serum markers in distinguishing early stage esophageal squamous cell carcinoma patients from healthy controls. ¹⁸¹ Significantly increased signatures of miR‐21, miR‐200a, miR‐200b, miR‐200c, miR‐141, miR‐203, miR‐205, miR‐214, miR‐221, and miR‐622 were observed in the serum of OC with respect to those of the control, as well as decreased miR‐132, miR‐26a, let‐7b, miR‐145, miR‐143, and so on. ¹⁸² , ¹⁸³ , ¹⁸⁴ In NSCLC, serum miR‐126, miR‐182, miR183, miR‐210, miR‐19, miR‐20, miR‐21, miR‐125, miR‐145, and miR‐205 were found to possess early detective value, exhibiting similar or even better sensitivity and specificity with traditional tumor marker CEA. ¹⁸⁵ , ¹⁸⁶

Despite its worth in cancer early diagnosis, the utility of circulating miRNA as a tool for segregation of specific cancer subtypes was also well appreciated, which contributes to the determination of tumor mechanisms and selection for more optimized therapeutic approach. Taking the well‐known heterogeneous disease, BC as an example, it can be subclassified according to the presence of estrogen receptor, HER2/neu receptor, and progesterone receptor. Tumors expressing none of these three receptors are named triple‐negative breast cancer (TNBC), which tends to be more advanced and aggressive than other subtypes. ¹⁸⁷ Shin et al. showed that compared with non‐TNBC, the expression of plasma miR‐16, miR‐21, and miR‐199a‐5p was lower in TNBC, among which miR‐199a‐5p expression exhibited the highest value to distinguish TNBC from non‐TNBC and healthy controls. ¹⁸⁸ In contrast, circulating miR‐373 was significantly increased in TNBC than in receptor‐positive tumors. ¹⁸⁹ Another study reported that the serum levels of miR‐16, miR‐155, miR‐21, and miR‐195 not only can discriminate patients with early‐stage BC (stage I or II) from healthy individuals, but also can make an distinguish between those with different molecular subtypes of BC from healthy controls. ¹⁹⁰

Levels of circulating miRNAs also showed a tight association with cancer prognosis and chemotherapy or radiotherapy outcomes. For example, in metastatic CRC, grouping the patients treated with first‐line bevacizumab + chemotherapy based on the changes of circulating miR‐126 disclosed a borderline significant separation of the groups in the progression‐free survival analysis favoring patients with decreasing miR‐126 levels, suggesting that it is worthy of consideration as a possible biomarker for the resistance to anti‐angiogenic containing treatments. ¹⁹¹ In radiochemotherapy‐treated neck and head squamous cell carcinoma patients, miR‐93‐5p and miR‐425‐5p were identified as the top candidates of commonly deregulated therapy‐responsive circulating miRNAs, but the authors declared that additional independent patient cohort that includes clinical follow‐up data is still required for the further confirmation of the prognostic value of the exciting observation above. ¹⁹²

The list of circulating miRNA benefits cancer monitoring can go on, we could not cover it all here. Instead, a figure is presented to summarize the frequently altered circulating miRNAs, which have been proven to with diagnostic, prognostic, or predictive relevance in several common human cancers (Figure 4 and Table 3). Of note, the specificity depends on the types of miRNAs to a large extent. For instance, as one of the most abundant and conserved miRNAs, miR‐21 is significantly highly expressed across most cancer types. Interestingly, statistics by Jenike et al. ¹⁹³ claimed that miR‐21 has been recognized as a specific predictive or prognostic biomarker by at least 29 diseases; in other words, it has no specificity to any one disease. This surely raises the question that whether it could be considered a viable candidate to be a biomarker for specific cancer, despite its continued evaluation as such. ¹⁸⁶ , ¹⁹³

FIGURE 4.

Multicancer‐shared circulating miRNA biomarkers. The list of significantly altered circulating miRNAs in nine common cancers was made to our best knowledge. Error may exist due to incomplete collection. The figure was created with R (version 4.1.0).

TABLE 3.

Potential cancer type‐specific circulating miRNA biomarkers in nine common cancers.

Cancer type	Potential cancer type‐specific circulating miRNA biomarkers
Breast cancer	miR‐100‐5p, miR‐106‐3p, miR‐133a, miR‐140‐5p, miR‐148‐3p, miR‐190, miR‐197‐3p, miR‐206, miR‐20b‐5p, miR‐24‐3p, miR‐296, miR‐3160‐5p, miR‐320c, miR‐328‐3p, miR‐331, miR‐34‐5p, miR‐362‐3p, miR‐373, miR‐374a‐5p, miR‐376c‐3p, miR‐4483, miR‐4710, miR‐4755‐3p, miR‐5698, miR‐575, miR‐652‐3p, miR‐8089, miR‐874‐3p, miR‐92a‐3p, miR‐93‐5p, miR‐98‐5p
Colorectal cancer	let‐7b‐3p, miR‐1260b, miR‐1290, miR‐130a, miR‐135, miR‐135b, miR‐139‐3p, miR‐144‐5p, miR‐145‐3p, miR‐150‐3p, miR‐17‐3p, miR‐17‐5p, miR‐18a‐5p, miR‐18b‐5p, miR‐19a, miR‐200b‐3p, miR‐21‐3p, miR‐320d, miR‐338‐5p, miR‐378, miR‐409‐3p, miR‐425‐5p, miR‐4772‐3p, miR‐497, miR‐548c‐5p, miR‐6803‐5p, miR‐7, miR‐720, miR‐92‐3
Gastric cancer	miR‐10b‐5p, miR‐1225‐5p, miR‐143‐3p, miR‐20a‐3p, miR‐217, miR‐296‐5p, miR‐34, miR‐376c, miR‐423‐5p, miR‐501‐3p, miR‐588
Hepatocellular cancer	miR‐122, miR‐122‐5p, miR‐122a, miR‐1247‐3p, miR‐125b, miR‐130b, miR‐142‐3p, miR‐199a, miR‐224, miR‐26a, miR‐34b, miR‐34c, miR‐483, miR‐500, miR‐519d, miR‐548‐2‐3p, miR‐595, miR‐665, miR‐718, miR‐939
Esophageal cancer	miR‐148a‐3p, miR‐151, miR‐20b, miR‐30a‐5p, miR‐339‐5p, miR‐34a‐5p, miR‐584
Pancreatic cancer	miR‐100, miR‐182, miR‐196, miR‐200, miR‐221‐3p, miR‐3679‐5p, miR‐3976, miR‐423‐3p, miR‐4306, miR‐4644, miR‐940
Lung cancer	miR‐125, miR‐1254, miR‐126‐3p, miR‐134, miR‐16‐5p, miR‐181‐5p, miR‐182‐5p, miR‐183‐5p, miR‐185, miR‐19, miR‐198, miR‐20, miR‐22, miR‐23b‐3p, miR‐30a‐3p, miR‐30e‐3p, miR‐320b, miR‐361‐5p, miR‐4257‐3p, miR‐574‐5p, miR‐9‐5p
Prostate cancer	miR‐107, miR‐1845, miR‐26b, miR‐484, miR‐874
Ovarian cancer	miR‐214, miR‐30‐5p

The list of significantly altered circulating miRNAs in nine common cancers was made to our best knowledge. Error may exist due to incomplete collection.

3.3.2. Circulating lncRNA and clinical values

Among the 60,000 ncRNAs in the human genome, >70% are lncRNAs, which comprise a heterogeneous group of transcripts with more than 200 nucleotides in length. ¹⁶⁷ They are located either within intergenic, intronic, antisense stretches or even overlap with protein coding genes. ¹⁶⁷ , ¹⁹⁴ lncRNA has been recognized to be involved in multiple essential biological processes, such as chromatin remodeling, transcriptional and posttranscriptional modification control, and so on. ¹⁷¹ , ¹⁹⁵ , ¹⁹⁶ In particular, some lncRNAs, such as PVT1, MALAT1, and HOTAIR, were shown to be consistently aberrant in multiple cancers and appear to play key roles in player in carcinogenesis as well as metastasis. ¹⁷¹

During the past two decades, emerging evidence has indicated that lncRNAs are detectable in various human biofluids and resistant to degradation by ribonucleases, since they are encapsulated inside extracellular vesicles or in association with proteins. ¹⁶⁸ Therefore, similar to miRNA, the consideration of lncRNA as one of the promising sources for noninvasive cancer biomarkers is not surprising. A very successful example is PROGENSA®, the only US FDA‐approved lncRNA used in the diagnostic process for PC based on urine determination of PCA3 (Table 1). ¹⁹⁷ It has to be mentioned that the test took about 14 years from the 1999 basic‐research publication to the 2012 US FDA approval and related‐research continues. For example, Schalken et al. ¹⁹⁷ and Lin et al. ¹⁹⁸ (NCT00756665) indicated that even though PCA3 was shown to be associated with reclassification at sBx1 in a multivariable model, while PCA3, or PCA3 together with TMPRSS2‐ERG, indicated minimal improvement to the clinical utilization of a multivariable model. Therefore, implementing PCA3 and TMPRSS2‐ERG into clinical practice would considerably reduce the number of prostate biopsies. ¹⁹⁸ , ¹⁹⁹ , ²⁰⁰

Besides, the implication of circulating lncRNA has been extensively studied in many other cancer types. For more information, please refer to the recent reviews that summarized findings regarding lncRNA‐based liquid biopsy, either for several types of cancer ¹⁷¹ , ¹⁹⁵ , ²⁰¹ or focused only on one type of cancer such as NSCLC, ¹⁹⁶ prostate, ²⁰² gastric, ¹⁶⁹ breast, ¹⁷⁰ DLBCL, ¹⁹⁴ and oral squamous cell carcinoma. ²⁰³

3.4. Technical advances for decoding circulating epigenetic information

Despite the growing bodies of work and resources dedicated to this topic, inadequacies still hamper the identification of more reliable promising circulating epigenetic markers to be used as tools to guide cancer prevention and therapy, including: (1) extremely low concordance between different studies; (2) tiny amount in the bloodstream and lack of tools to discriminate those derived from nontumors; (3) a deeper knowledge of the core mechanisms and dynamics of circulating epigenetic markers is needed; (4) conventional analytical and computational strategies are time and resource consuming. Enlarging the cohort size, a more precise patient inclusion criteria, selecting better internal controls, and making discrimination between serum and plasma could be several relatively easier choices to alleviate these issues, nevertheless, to fundamentally figure out relies on the optimization of experimental and analytical methodologies, which we are going to discuss in more detail below.

3.4.1. Experimental techniques

Current methods for the detection of cfDNA methylation can be divided into two categories: or bisulfite conversion‐free or bisulfite conversion‐based methods. Whole‐genome bisulfite sequencing is the most informative and widely‐used cfDNA‐applicable bisulfite‐based one, as it can be performed with 125 pg–30 ng of DNA, but with high prices. Even though methylated CpG tandems amplification and sequencing is more cost friendly and is able to detect hypermethylated CpG sites in ctDNA samples as little as 7.5 pg, its preference for CGGCGG‐rich sites may miss other potential biomarkers. ¹³⁰ , ²⁰⁴ Most importantly, bisulfite treatment is DNA damaging, especially for the tiny amounts of cfDNA, making the detection more challenging. Thus, bisulfite conversion‐free methods based on antibody or magnetic beads enrichment and restriction enzyme were suggested to protect DNA from degradation and adapt to the low cfDNA input. One representative example is TET‐assisted pyridine borane sequencing (TAPS), a newly developed approach based on mild chemistry. Compared with bisulfite sequencing, it improves sequence quality, mapping rate, and coverage while reducing sequencing cost by half. In addition to these advantages in DNA methylation analysis, TAPS is also ideal for simultaneous genetic analysis due to the nondestructive nature. ²⁰⁵ Indeed, application on HCC and PDAC cfDNA demonstrated that TAPS has the benefit of providing a wealth of information including differential methylation, fragmentation profiles, and tissue of origin, enabling integrated multimodal epigenetic and genetic analysis to accurately discriminate samples from patients with HCC or PDAC from controls. Though the current TAPS protocol has been optimized to work with 10 ng cfDNA purified from 1 to 3 mL of plasma, it remains urgent to lower more the input. ²⁰⁶

The classical ELISA assay and ChIP followed by quantitative or whole genome sequencing enable gene‐specific or genome‐wide mapping of histone modifications, providing the first glimpse into the rich histone PTMs information present in plasma. ¹⁶² , ¹⁶³ However, a large amount of input materials is required while only a single layer of information can be measured, that is, not a cost‐effective way. Sadeh et al. ¹⁶⁵ devised a ChIP‐seq protocol for analysis of cf‐nucleosome from less than 2 mL of plasma to overcome the extremely low concentration of circulating histones as well as the high concentration of native antibodies in plasma, achieving highly specific results that are comparable to the reference ChIP‐seq in tissues. Briefly, ChIP antibodies were covalently immobilized to paramagnetic beads, therefore can be incubated directly in plasma to avoid competition with native antibodies. Meanwhile, prior to the isolation of DNA, the barcoded sequencing‐DNA adaptors are designed to directly ligate to chromatin fragments, forming the so‐called on‐bead adaptor ligation. ¹⁶⁵ Another recently developed system, EPINUC, allows high‐resolution detection and integration of multiple parameters including six active and repressive histone PTMs, their ratios, and combinatorial patterns, from less than 1 mL of plasma sample by single‐molecule imaging, which further expands the burgeoning field of liquid biopsies. Nevertheless, the authors indicated that more studies are still needed to test if this system can make a distinction between different cancer types based solely on the epigenetic profiles. ¹⁶⁶

Circulating ncRNAs are generally considered rather stable, and the related approaches are more sensitive and easier to handle; thus, in my point of view, their extraction and detection are relatively simpler. The most‐used methods are RT‐qPCR and RNA‐Seq. RT‐qPCR is regarded as the gold standard for ncRNA analysis, which offers high sensitivity and high specificity with a large dynamic range, particularly for low RNA input samples, but is only limited to the analysis of known ncRNAs. Instead, RNA‐seq allows for the detection of a larger panel RNA with high sensitivity, including novel ncRNAs. However, relatively time‐consuming protocols, high amount of starting material, and high costs are required. ¹⁷⁵ , ²⁰⁷

In summary, there is no perfect technology when considering all the tested parameters. Utilization of complementary approaches would make the identification of circulating epigenetic markers more robust.

3.4.2. Computational approaches

It should be pointed out that a single epigenetic biomarker in liquid biopsy usually cannot accurately diagnose or prognosticate cancer due to the heterogeneity of phenotypes and the variability of biomarker expression across individuals. ²⁰⁴ Integration of multiple parameters based on various aspects of epigenetics as well as mutation‐based, or protein biomarkers, will help to define more robust signatures for specific cancer states. Therefore, there is an unmet need to search for powerful analytical and computational tools for multiparameter analysis since the conventional methods are time and resource consuming, requiring large cohorts and analytical power. Thankfully, the recently rapidly developed machine learning coupled with big data offers an unprecedented opportunity to automatically discover and classify various cancer‐specific signatures from liquid biopsies. ²⁰⁸

Currently, the traditional machine learning models such as support vector machine (SVM), linear models, and random forest (RF) still account for a major position in early cancer detection due to their training speed and robustness on small dataset. ²⁰⁹ One of the nice examples was carried out by Bahado‐Singh et al. ²¹⁰ on LC. When using plasma CpG biomarkers combined with multiple machine learning algorithms including SVM, RF, generalized linear model, prediction analysis for microarrays, linear discriminant analysis, and deep learning, they achieved highly accurate detection of LC (AUC = 0.90–1.0) with high sensitivity and specificity values. ²¹⁰

Meanwhile, algorithms are under updating with leaps and bounds to fill in the information gaps over current methods and further improve the performance. For example, the most‐used DNA methylation analysis focuses on the methylation rate (β value) of an individual CpG site in a cell population. Yet, such population‐average measures are shown to be not sensitive enough to capture abnormal methylation signals, which affect only a small number of cfDNAs, thus challenging for the extraction of tiny tumor signals in cfDNA data with low tumor fraction and sequencing coverage. To figure out, Li et al. averaged the methylation values of all CpG sites in a given read and denoted as α value. Interestingly, a striking difference (0 and 1) between the normal cfDNAs and abnormally methylated cfDNAs was obtained (α tumor = 0% and α normal = 100%). They termed this novel, read‐based probabilistic approach as "CancerDetector," which can sensitively identify a trace amount of tumor cfDNAs out of all cfDNAs in plasma. ²¹¹ Nevertheless, the prediction results of CancerDetector can be affected by different depths of sequencing data via introducing systematic deviation, which may reduce the accuracy of cancer diagnosis. ²¹² Soon, another approach called DISMIR was proposed, providing ultrasensitive and robust HCC detection, in particularly at low sequencing depths. The outperformance was mainly benefited from the novel design of deep learning model, which combined the DNA sequence together with methylation information for each read, thus grasping sequence motifs related to tumors and extracting the joint patterns of DNA sequence and methylation across different regions from the whole genome to ensure the source prediction of individual reads more accurate. In contrast, the performance of methods based on only methylation information such as RFs, SVM, and FSR was much worse than DISMIR, suggesting the superiority of integrating the information of DNA sequence and methylation. ²¹² In addition, Liu et al. ¹⁰³ suggested that combining the characteristics of plasma epigenetic signatures with the traditional diagnostic imaging by machine learning could improve the current clinical practice of BC early detection, by reducing the false positive rate and avoiding unnecessary harms.

Rapid technical innovation in the field of epigenetics definitely promotes the development of liquid biopsy, but followed by a waste of the huge amount and multifariousness of data produced from various laboratories measured by different experimental and bioinformatic technologies, because of lacking efficient collection, standardized quality control and analysis procedures. Even though several public databases such as cell‐free epigenome atlas have been introduced to allow easy querying, visualization, and comparison of the collected data by users, ²¹³ more informative and systematic tools are still needed for efficient integration and reuse of these data.

4. CONCLUSIONS AND FUTURE DIRECTIONS

Over the past decades, the emerging field of liquid biopsy has provided exciting new avenues for cancer monitoring and intervention. Indeed, as summarized in this review, there is a growing list of valid circulating indicators, which showed promising ability in clinic or even have been practiced such as Cobas EGFR mutation Test v2, CancerSEEK, Galleri, Epi proColon, and so on. Due to the lack of space, we apologize here to those whose excellent work has not been included. However, to achieve more and better clinical applications, there is still a long and winding road to go, as limitations still exist.

First and foremost, it is unlikely that a single circulating alteration can fulfill all requirements to accurately refine clinical decision‐making; therefore, the combination of different types of biomarkers might provide more reliable results. ⁸³ , ²¹⁴ Cancer liquid biopsy was first introduced from CTCs and rapidly extended to ctDNA and other tumor‐derived products such as circulating nucleosomes, ncRNA, and so on. The features used for discrimination of tumor are also diversifying over the years, from quantitative differences, to genetic disorders and now, plus epigenetic abnormalities. As indicated in Figure 1, each approach has its unique advantages and disadvantages. For instance, CTCs are closely related with cancer metastasis, thus targeting CTCs possesses both the capacities of elimination and prediction of cancer metastasis. ⁵⁹ , ⁶⁰ While unlike cfDNA or circulating nucleosomes, circulating ncRNAs are usually presented in a fairly stable form, due to the protection by extracellular vesicles. ¹⁷⁶ , ¹⁷⁷ From another dimension of view, though genetic disorders extracted from tumor‑derived entities such as somatic mutations have been relatively well applied in clinic, limitations do exist. For example, entities within the circulation of one body share the same genetic information, but meanwhile varies from individual to individual, challenging the detection of tissue‐of‐origin and identification of generalizable biomarkers. ¹ , ⁸¹ Epigenetic alterations that are reversible, tissue specific, and with higher frequencies thus could nicely balance these shortages. ⁸³ , ⁸⁴ Therefore, since most of the current applications are based on one of the dimensions, it is strongly recommended to make them complement to each other to achieve better performance.

Second, there is a lack of standardized methodology of sampling, isolation, detection and analysis that is sensitive and specific enough to detect sparse circulating biomarkers against the complex substrate of clinical samples, which leads to divergencies between studies and also waste of information source. ²⁰⁴ At the very beginning, the field focused mainly on development of minimally isolation approaches, which have the capacity to enrich circulating tumor‑derived entities, as they are extremely small in quantity and highly dynamic. As the performance of isolation has improved, it has been driven to the establishment of computational tools for the maximize extraction of useful information. Technological innovation is absolutely the major contributors for promoting clinical practice and should never be prevented, but meanwhile, a motley variety of data was generated from various approaches, which has low mutual compatibilities. To improve data utilization and enhance reliability, it is of great importance to standardize and better organize the existing data, and also establish normative methods for the future work. Besides, as suggested by Keller and Pantel et al., ¹⁸ development of appropriate reference materials will also contribute to more standardized quality controls, quantification, and reporting among laboratories.

Besides, many studies in the field of cancer liquid biopsy have so far been developed for research or investigational purposes only, since they are lack of standard design of interventional studies and continuous monitoring to demonstrate the clinical utility. Another more important link for the implementation of liquid biopsy from bench to clinical practice is the establishment of evaluation system.

Regarding the assessment of solid tumors in response to traditional cancer therapy, the Response Evaluation Criteria In Solid Tumors (RECIST) provides radiological evaluation as the gold standards. ²¹⁵ However, no such guidelines exist to date for the evaluation of liquid biopsy, which is an issue requiring urgent solution. Last, current efforts and clinical practice on cancer liquid biopsy are mainly focused on blood, other bodily fluids such as pleural effusions, urine, and cerebrospinal fluid deserve more attention.

AUTHOR CONTRIBUTION

J. S., Z. C., and M. W. conceived and designed the study. Z. C., C. L., Y. Z., Y. Y., J. L., and J. S. contributed to the management of data collection and analysis. Z. C., M. W., and J. S. wrote the manuscripts. All authors reviewed and approved the final manuscript.

CONFLICT OF INTEREST STATEMENT

The authors declare no competing interests.

ETHICS STATEMENT

Not applicable.

Chen Z, Li C, Zhou Y, et al. Liquid biopsies for cancer: From bench to clinic. MedComm. 2023;4:e329. 10.1002/mco2.329

DATA AVAILABILITY STATEMENT

Not applicable.