DNA Methylation-Based Testing in Liquid Biopsies as Detection and Prognostic Biomarkers for the Four Major Cancer Types

Lung, breast, colorectal, and prostate cancers are the most incident worldwide. Optimal population-based cancer screening methods remain an unmet need, since cancer detection at early stages increases the prospects of successful and curative treatment, leading to a lower incidence of recurrences. Moreover, the current parameters for cancer patients’ stratification have been associated with divergent outcomes. Therefore, new biomarkers that could aid in cancer detection and prognosis, preferably detected by minimally invasive methods are of major importance. Aberrant DNA methylation is an early event in cancer development and may be detected in circulating cell-free DNA (ccfDNA), constituting a valuable cancer biomarker. Furthermore, DNA methylation is a stable alteration that can be easily and rapidly quantified by methylation-specific PCR methods. Thus, the main goal of this review is to provide an overview of the most important studies that report methylation biomarkers for the detection and prognosis of the four major cancers after a critical analysis of the available literature. DNA methylation-based biomarkers show promise for cancer detection and management, with some studies describing a “PanCancer” detection approach for the simultaneous detection of several cancer types. Nonetheless, DNA methylation biomarkers still lack large-scale validation, precluding implementation in clinical practice.


Introduction
During the last decades, efforts have scaled up worldwide to develop more effective biomarkers as an approach to reduce cancer mortality [1]. A cancer biomarker can be defined as an objectively measurable biomolecule, such as a protein, metabolite, RNA, DNA, or an epigenetic alteration, found in body fluids or tissues, that indicates the presence of cancer or provides information on cancer's expected future behavior [2,3].
Tissue biopsy sampling has been the gold-standard approach for patients' diagnosis and prognostication. However, several drawbacks have been pointed out over the years to this approach [4,5]. Lung (LC), breast (BrC), colorectal (CRC) and prostate (PCa) cancers are the most incident and among the deadliest worldwide, despite the efforts for early cancer detection and the emergence of new therapies [13]. Given the shortcomings of current screening methods and prognostic biomarkers, the development and implementation of effective biomarkers for cancer detection at curative, early stages, and for better patients' stratification, is crucial. Thus, for the purposes of this review, a detailed and extensive literature review was conducted aimed to explore the state of the art of ccfDNA methylation blood-based biomarkers for cancer screening, diagnosis, prognosis, prediction, and monitoring of the four most incident cancers worldwide. On a Pubmed database search, the key words "Lung Cancer/Breast Cancer/Colorectal Cancer/Prostate Cancer", "DNA methylation", "Diagnosis/Detection/Prognosis" and "Serum/Plasma" were used ( Figure 2). Lung (LC), breast (BrC), colorectal (CRC) and prostate (PCa) cancers are the most incident and among the deadliest worldwide, despite the efforts for early cancer detection and the emergence of new therapies [13]. Given the shortcomings of current screening methods and prognostic biomarkers, the development and implementation of effective biomarkers for cancer detection at curative, early stages, and for better patients' stratification, is crucial. Thus, for the purposes of this review, a detailed and extensive literature review was conducted aimed to explore the state of the art of ccfDNA methylation blood-based biomarkers for cancer screening, diagnosis, prognosis, prediction, and monitoring of the four most incident cancers worldwide. On a Pubmed database search, the key words "Lung Cancer/Breast Cancer/Colorectal Cancer/Prostate Cancer", "DNA methylation", "Diagnosis/Detection/Prognosis" and "Serum/Plasma" were used ( Figure 2).

Identified
Articles excluded based on abstract, lack of information, sample type, study design and study population: •

Circulating Cell-Free DNA Liquid Biopsies
CcfDNA was firstly described in 1948 when extracellular nucleic acids were found in human blood from healthy individuals by Mandel and Métais [14]. Later on, it was found that, in cancer patients, circulating tumor DNA (ctDNA) fragments between 150 and 1000 base pairs could also be detected due to their release into the bloodstream, either by cell death (apoptosis or necrosis) or active secretion by the release of extracellular vesicles, such as exosomes [10,15]. Depending on several factors including tumor burden, metastatic sites or cellular turnover, ctDNA might account for 0.01% to 90% of the total ccfDNA in the blood of cancer patients [15]. Owing to the fact that ctDNA might represent tumor-specific genetic and epigenetic alterations of all tumor's sub clones present, ccfDNA is an ideal candidate for blood-based liquid biopsies by offering the possibility to test for the presence of cancer and the discrimination of lethally aggressive cancer [16].

DNA Methylation
Although the study of tumor mutations was the focus of biomarker research for a long time, their wide diversity has been a challenge for the development of effective diagnostic biomarkers since very large proportions of the genome would need to be examined in order to provide adequate sensitivity [17]. Contrarily, epigenetic alterations seem to be more stable and homogenous in cancer, representing a good alternative for biomarker development [18]. The main studied epigenetic mechanisms are DNA methylation, histone post-translational modifications, histone variants and chromatin remodeling complexes ( Figure 3) [19]. Although these mechanisms are crucial for normal cell development and regulation of specific gene expression patterns, epigenetic dysregulation often leads to inappropriate activation or inhibition of several signaling pathways, which may trigger the development of several pathologies, including cancer [20].
DNA methylation, the most widely studied epigenetic modification in humans, was also the first to be identified in cancer [20,21]. This epigenetic mechanism consists of covalent addition of a methyl group, donated by S-adenosylmethionine (SAM), to the 5-position carbon of a cytosine ring to form 5-methylcytosine (5mC) [22,23]. This modification is catalyzed by DNA methyltransferase enzymes (DNMTs), namely, DNMT3a and DNMT3b that catalyze de novo DNA methylation during embryonic development, establishing tissue-specific DNA methylation, and DNMT1 that is often associated with maintenance of methylation patterns during replication [23]. Typically, this process occurs on cytosine residues present at CpG dinucleotides commonly found in large clusters named CpG islands, which are predominantly located at the 5 end of genes, occupying approximately 60% of human gene promoter regions [20,23,24]. Although gene promoter hypermethylation is associated with transcription repression of the nearby gene (Figure 4), depending on DNA methylation genomic location, it can display different functions [22,25]. Epigenetic gene silencing by DNA promoter methylation may occur either directly, by blocking transcription factors to prevent binding to target sites in or near the promoter, or indirectly, through binding of methyl-CpG-binding proteins (MBP), which can recruit other enzymes like DNMTs and histone deacetylases (HDAC), leading to chromatin conformation changes that further repress gene transcription [20,22]. DNA methylation, the most widely studied epigenetic modification in humans, was also the first to be identified in cancer [20,21]. This epigenetic mechanism consists of covalent addition of a methyl group, donated by S-adenosylmethionine (SAM), to the 5-position carbon of a cytosine ring to form 5-methylcytosine (5mC) [22,23]. This modification is catalyzed by DNA methyltransferase enzymes (DNMTs), namely, DNMT3a and DNMT3b that catalyze de novo DNA methylation during embryonic development, establishing tissue-specific DNA methylation, and DNMT1 that is often associated with maintenance of methylation patterns during replication [23]. Typically, this process occurs on cytosine residues present at CpG dinucleotides commonly found in large clusters named CpG islands, which are predominantly located at the 5′ end of genes, occupying approximately 60% of human gene promoter regions [20,23,24]. Although gene promoter hypermethylation is associated with transcription repression of the nearby gene ( Figure 4), depending on DNA methylation genomic location, it can display different functions [22,25]. Epigenetic gene silencing by DNA promoter methylation may occur either directly, by blocking transcription factors to prevent binding to target sites in or near the promoter, or indirectly, through binding of methyl-CpG-binding proteins (MBP), which can recruit other enzymes like DNMTs and histone deacetylases (HDAC), leading to chromatin conformation changes that further repress gene transcription [20,22].    DNA methylation, the most widely studied epigenetic modification in humans, was also the first to be identified in cancer [20,21]. This epigenetic mechanism consists of covalent addition of a methyl group, donated by S-adenosylmethionine (SAM), to the 5-position carbon of a cytosine ring to form 5-methylcytosine (5mC) [22,23]. This modification is catalyzed by DNA methyltransferase enzymes (DNMTs), namely, DNMT3a and DNMT3b that catalyze de novo DNA methylation during embryonic development, establishing tissue-specific DNA methylation, and DNMT1 that is often associated with maintenance of methylation patterns during replication [23]. Typically, this process occurs on cytosine residues present at CpG dinucleotides commonly found in large clusters named CpG islands, which are predominantly located at the 5′ end of genes, occupying approximately 60% of human gene promoter regions [20,23,24]. Although gene promoter hypermethylation is associated with transcription repression of the nearby gene ( Figure 4), depending on DNA methylation genomic location, it can display different functions [22,25]. Epigenetic gene silencing by DNA promoter methylation may occur either directly, by blocking transcription factors to prevent binding to target sites in or near the promoter, or indirectly, through binding of methyl-CpG-binding proteins (MBP), which can recruit other enzymes like DNMTs and histone deacetylases (HDAC), leading to chromatin conformation changes that further repress gene transcription [20,22]. As aforementioned, DNA methylation is crucial for multiple cellular processes, thus it is understandable that its deregulation has been linked to cancer. Indeed, normal and cancer cells  As aforementioned, DNA methylation is crucial for multiple cellular processes, thus it is understandable that its deregulation has been linked to cancer. Indeed, normal and cancer cells display different methylomes. Usually, a global hypomethylation pattern, which contributes to genomic instability and activation of silenced oncogenes, is observed in cancer [23,26]. Alongside, tumor suppressor genes (TGS) frequently undergo inactivation due to focal promoter hypermethylation [23,26]. Currently, the latter process is considered a major contributor to neoplastic transformation [27].
Interestingly, aberrant DNA methylation is thought to occur at very early stages of cancer development and specific genes seem to be methylated at different tumor stages [23]. Moreover, since these alterations can be assessed in several body fluid samples [23], it is widely accepted that DNA methylation-based liquid biopsies are a promising approach, not only for premalignant/early cancer detection, but also for prognostic assessment. Furthermore, since some genes seem to acquire tissue-specific DNA methylation, it may also be possible to discriminate between different cancer types in the context of metastatic tumors [23] or in liquid biopsies.

Screening and Diagnosis
Despite advances in new treatment options over the years, the high mortality rate observed in LC patients is mainly related to the fact that more than 75% of LC patients are diagnosed with advanced stage disease [28]. Hence, effective screening options aiming to shift LC diagnosis from advanced to curative early stages are crucial to change the fate of this disease [29]. Currently, low-dose computed tomography (LD-CT) is considered the best LC screening method available [30]. Nonetheless, despite The National Lung Screening Trial has shown a 20% decrease in LC-related mortality rate among high-risk smokers with LD-CT screening comparing to chest X-ray [corroborated by the largest European trial (Dutch-Belgian Lung Cancer Screening Trial)] from the 24% positive test results in this trial, 96.4% were deemed as false-positive [31,32]. Hence, due to the risks related to LD-CT screening, namely, overdiagnosis, radiation exposure, and false positive results leading to unnecessary anxiety and costs [29], the development of specific and accurate screening tools is urgently needed to improve LC survival.
In 2002, Usadel et al. and Bearzatto et al. reported, for the first time, APC me and p16 INK4a me , respectively, in ccfDNA as putative minimally-invasive biomarkers for LC detection [33,34]. Thenceforth, numerous other methylated gene promoters detected in ccfDNA have been proposed for LC detection either individually or in panel (Table 1). RASSF1A me and p16 INK4a me represent the two most frequently reported genes in blood-based liquid biopsies displaying 22-66% sensitivity and 57-100% specificity for LC detection, individually [34][35][36].
After the commercialization in Europe of a test based on SHOX2 me assessment in bronchial aspirates, the methylation of this gene was also evaluated as an LC detection biomarker in plasma. Indeed, plasma SHOX2 me discriminated LC from control samples with 60% sensitivity and 90% specificity, although higher sensitivity was found in stages II (72%), III (55%) and IV (83%) compared with stage I (27%) patients [37]. In line with these results, another study reported that SHOX2 me discriminated LC in subjects undergoing bronchoscopy with 81% sensitivity and 79% specificity [38]. Later on, Weiss et al. reported that SHOX2 me and PTGER4 me panel distinguished LC patients with 67% sensitivity for a fixed specificity of 90%, and with 73% specificity for a fixed sensitivity of 90% [39]. Remarkably, in the end of 2017, the "Epi proLung ® " assay, developed by Epigenomics AG, based on these two genes received the Conformité Européenne (CE) mark for In Vitro Diagnostic (IVD) test. According to their validation study comprising 360 patients from the US and Europe, of which 152 were diagnosed with LC, depending on the Epi proLung ® test score threshold chosen, 85% sensitivity was achieved for 50% specificity, whereas sensitivity decreased to 59% if 95% specificity was considered [40,41].

Prognosis, Prediction, and Monitoring
Besides LC histological subtype, TNM prognostic stage groups remain the most important prognostic feature to predict recurrence and survival, followed by tumor histological grade, gender and age [63,64]. Recently, molecular subtypes have also emerged to provide more personalized genetic information, which may improve prognostic estimates and therapy response prediction [63].
Thus far, only a few small-scaled studies reported the prognostic, predictive and monitoring potential of ccfDNA methylation for LC, most of them performed in patients with advanced disease. DCLK1 me and SOX17 me levels were associated with reduced overall survival (OS) in advanced LC and NSCLC, respectively [42,59]. Likewise, higher SHP1P2 me levels in advanced NSCLC associated with reduced progression-free survival (PFS) and OS [65], whereas BRMS1 me associated with both reduced disease-free survival (DFS) and OS in operable NSCLC, and reduced PFS and OS in advanced NSCLC [66].
Interestingly, after neoadjuvant chemotherapy and surgery with intraoperative radiation therapy, NSCLC patients showed decreased RASSF1A me and RARβ2 me levels, similar to levels in healthy subjects. Moreover, methylation levels increase in at least one of these genes, up to the levels detected before treatment, was observed in all five patients, which disclosed evidence of disease progression [35]. Increased APC me and/or RASSF1A me levels within 24h after cisplatin-based chemotherapy also associated with increased OS [67]. Detectable circulating levels of APC me and RASSF1A me panel at diagnosis was also found to be an independent predictor of increased disease-specific mortality in LC patients, displaying a 3.9-fold risk of dying from LC comparing to those lacking methylation [46].
Remarkably, advanced LC patients that clinically responded to chemo/radiotherapy demonstrated a decrease in SHOX2 me plasma levels, observed at 7-10 days after therapy initiation [68]. Additionally, higher SHOX2 me levels, both before and 7-10 days after therapy beginning were indicative of shorter OS [68]. Similar results were obtained after two cycles of chemotherapy or TKI-based targeted therapy, being SHOX2 me levels before therapy again predictive of OS [69]. Contrarily, 14-3-3σ me levels in stage IV NSCLC patients before treatment with cisplatin-gemcitabine associated with longer survival [70]. Stage IV NSCLC patients with unmethylated CHFR depicted longer OS when treated with EGFR-TKI compared to those treated with chemotherapy, as second-line therapy [71].

Screening and Diagnosis
BrC is estimated to be the most incident and deadly cancer in females, worldwide [13]. BrC incidence increased after the implementation of mammography-based screening [72]. Screening mammography endures a sensitivity around 82% and 91% specificity [73]; however, sensitivity decreases with high breast density [74]. Furthermore, although BrC screening features several benefits, it presents some important disadvantages, including overdiagnosis and false-positive results. Indeed, about 11% of cases detected in a population invited to screening would probably not be clinically relevant in the woman's lifetime, still are treated [75]. Additionally, false-positive results lead to extra imaging exams and eventually biopsy procedures, which can cause discomfort and anxiety to the subjects [75,76]. The diagnosis of BrC comprises clinical examination and biopsy procedures either by image-guided core needle biopsy or fine-needle aspiration, to confirm the diagnosis by histopathological analysis [72]. Since these are invasive procedures, advances in BrC detection are needed, specifically in pre-screening methods, which might select patients to invasive/costly screening tests, avoiding overdiagnosis and unnecessary exams.
Interestingly, Shan et al. reported a six-gene panel that detects BrC with a tumor size ≤ 1cm with higher sensitivity (85.82%) than mammography (81.82%) [87]. Even though numerous studies report the feasibility of DNA methylation to detect BrC, its validation and transfer to the clinical setting is still overdue. Nonetheless, DNA methylation shows promise to complement BrC standard screening and diagnosis methods.

Prognosis, Prediction, and Monitoring
In addition to staging, several biomarkers are used to predict BrC patients' response to therapy and prognosis, including histological grade-a high histological grade is associated with poorly differentiated tumors and worse prognosis [96]. Furthermore, 75% of BrC cases are estrogen receptor (ER)-positive, being less aggressive, and associated with a better prognosis [97]. Accordingly, progesterone receptor (PR)-positive tumors are frequently ER-positive. On the contrary, ER-negative and PR-negative tumors are frequently grade 3, often recur, and do not respond to hormonotherapy [96]. Erb-B2 receptor tyrosine kinase 2 (ERBB2) [or human epidermal growth factor receptor 2 (HER2)] is a growth factor overexpressed in 15% of all BrC cases [97]. Hence, therapies targeting ERBB2 such as trastuzumab are currently available for BrC patients, improving patient outcome [97]. Moreover, gene expression profiles have emerged to assist in adjuvant treatment decision [64] Oncotype DX ® , MammaPrint ® and PAM50 (Prosigna) are examples of gene expression profiles that allow for BrC classification and prognostic stratification [64]. Nonetheless, their usefulness in clinical practice is limited and therefore their implementation remains restricted [98]. Thus, new and reliable biomarkers to aid in defining a BrC patient prognosis is of major importance.
Chimonidou et al. analyzed CST6 me in ccfDNA from plasma samples and, during follow-up time, CST6 was methylated in 13 of the 25 BrC patients that relapsed and in 3 of the 9 patients which died, however not reaching statistical significance [99]. Furthermore, RASSF1A me and PITX2 me were found to be independent biomarkers of poor OS and RASSF1A me together with lymph node and ER status indicated poor distant DFS in BrC patients' ccfDNA [100]. Recently, a study analyzing methylation patterns of several genes in ccfDNA showed that BrC patients with positive SOX17 me and WNT5A me displayed shorter OS, whereas KLK10 me associated with higher number of relapses and shorter disease-free interval [101]. Apart from being a detection biomarker, SOX17 me associated with lymph node metastasis, poor DFS and OS in plasma samples from BrC's patients [102]. Visvanathan et al. described a cumulative methylation index (CMI) comprising the methylation of 6 genes (AKR1B1 me , HOXB4 me , RASGRF2 me , RASSF1 me , HIT1H3C me , and TM6SF1 me ) that associates with PFS and OS, i.e., PFS and OS were significantly shorter in metastatic BrC patients with high CMI [103]. Additionally, 14-3-3-σ me methylation was significantly associated with the response to metastatic BrC treatment [104]. APC me and RASSF1A me were also associated with poor outcome, with a relative risk of death of 5.7 [105].

Screening and Diagnosis
Due to its slow progression time and the opportunity to easily remove precancerous and early stage cancerous lesions, if caught on time, CRC entail minimal risk to the patient with about 90% long-term survival [106,107]. Therefore, CRC screening programs are of great interest. Currently, analysis of trace blood in stool by fecal occult blood test (FOBT)/fecal immunochemical test (FIT), and internal imaging of the colon by colonoscopy are the main screening options available for CRC early detection [106,108]. Additionally, biopsy samples during colonoscopy are mandatory for histological diagnosis [109]. Nevertheless, fecal screening tests have limited sensitivity to detect precancerous lesions, whereas colonoscopy is very precise and can be used to remove the lesion during the examination, although it is a costly and invasive procedure with low patient compliance [110]. Hence, despite being recognized that screening reduces CRC incidence and mortality, the availability and compliance to the current screening tests remain suboptimal [111].
The increasing knowledge of the influence of epigenetic alterations in malignant transformation in the gut gave rise to an opportunity for development of sensitive and specific minimally invasive epigenetic-based biomarkers for CRC. Hence, plentiful studies have investigated the detection value of these biomarkers (Table 3). SEPT9 me is the mostly reported methylated gene in blood from CRC patients. Remarkably, this marker was the first blood-based IVD assay for detection of occult cancer based on an epigenetic alteration, approved by the US Food and Drug Administration (FDA) in 2016, under the designation "Epi ProColon ® 2.0" (Epigenomics AG) [40]. Additionally, this CE-IVD marked test is also commercially available in Europe and China [112]. A meta-analysis published in 2017 reported that SEPT9 me sensitivity for CRC detection varies between 73-78% depending on the algorithm used to consider a positive result, while specificity varies between 84-96% [113]. Nevertheless, when the biomarker performance of this gene's methylation was assessed in a multicenter screening setting (PRESEPT clinical trial) with asymptomatic individuals older than 50 years old, the results from 53 CRC cases and 1457 subjects without CRC yielded 48% sensitivity and 92% specificity [114].
Given the importance of detecting pre-malignant conditions, several studies have not only studied the CRC detection performance, but also the ability to detect adenomas. Disappointingly, SEPT9 me performance to detect advanced adenomas ranged between 8-31% [115][116][117], being reported to be 11% in the previously mentioned screening setting study [114]. Thus, the usefulness of this gene for population-based screening is questionable.
As expected, gene panels have improved the performance to detect both adenomas and CRC. Remarkably, APC me , MGMT me , RASSF2A me and WIF1 me panel discriminated adenomas and early stage CRC with 75% and 87% sensitivity, respectively, and 92% specificity [118], whereas SFRP1 me , SFRP2 me , SDC2 me , and PRIMA1 me panel detected adenomas with 89% sensitivity and 87% specificity, and CRC with 92% sensitivity and 97% specificity [119]. Nevertheless, the performance of these panels in a large screening setting remains to be elucidated. Interestingly, BCAT1 me and IKZF1 me panel performance has been evaluated in large multicenter studies, displaying 62-66% sensitivity and 92-95% specificity for CRC detection, although with a limited 6-9% sensitivity for adenoma detection [120,121].

Prognosis, Prediction, and Monitoring
TMN prognostic stage groups, in addition to cancer location, patient's age and comorbidities are the basis for CRC prognostication and patient management [64,109]. Regarding ccfDNA methylation-based biomarkers, higher RUNX3 me levels associated with lymphatic invasion, advanced pathological stage and tumor recurrence [151]. TFPI2 me and SFRP2 me levels associated with poorly differentiated carcinoma, deep invasion, lymph node and distant metastasis [131,152], the former also associating with tumor size [131], and the latter with shorter OS [152]. SST me was independently associated with higher recurrence risk and shorter DSS in patients who underwent curative surgical resection [153].
A small-scale study suggested that p16 INK4a me could reflect the recurrence status during follow-up after surgery [154]. Interestingly, in a prospective cohort study including 150 stage I-III CRC patients from whom serum samples were obtained 1 week before, as well as 6 months and 1 year after surgery, high levels of TAC1 me after 6 months and SEPT9 me after 1 year were independent predictors of tumor recurrence and shorter DSS [155]. Additionally, the increment of TAC1 me and SEPT9 me levels independently predicted disease recurrence, whereas NELL1 me at both 6 months and 1 year associated with disease-specific survival (DSS) [155]. In the same line, SEPT9 me was suggested as follow-up marker for recurrence and metastasis detection, also associating with tumor size, histological grade and histological type [117]. GATA5 me , SFRP2 me , ITGA4 me , SHOX2 me and SEPT9 me levels were also associated with tumors' histological grade, TNM stage and lymph node metastasis [141,156], whereas GATA5 me also associated with large tumor size [141]. Furthermore, APC me, SEPT9 me , SHOX2 me and SOX17 me levels were also shown to be significantly higher in female patients with metastatic disease [36], whereas higher RARB2 me , SEPT9 me and SOX17 me levels were disclosed in metastatic CRC male patients [46].
Interestingly, various studies have demonstrated the prognostic value of HLTF me and HPP1 me levels in CRC patients. Indeed, serum HLTF me , HPP1 me and hMLH1 me were significantly correlated with tumor size, and the two former genes (HLTF me and HPP1 me ) further associated with metastatic disease and tumor stage, as well as with worst outcome [125]. Later, Philipp et al., in a study involving 311 serum samples of CRC patients, reported that HLTF me and HPP1 me associated with tumor size, stage, grade and metastatic disease, and HPP1 me also associated with nodal status. Moreover, in stage IV patients, high levels of both these genes associated with reduced OS [157]. In patients curatively resected for CRC, pre-therapeutic serum HLTF me levels were associated with increased relative risk of disease recurrence [158]. In a clinical trial including 467 metastatic CRC patients treated with a combination therapy containing fluoropyrimidine, oxaliplatin and bevacizumab, patients with detectable plasmatic HPP1 me before the start of treatment showed a significantly poorer OS [159]. Moreover, patients in which it reduced to undetectable levels 2-3 weeks after treatment, showed a better OS compared to patients that maintained detectable plasma HPP1 me levels [159], suggesting the usefulness of HPP1 me as a prognostic and early response biomarker. Recently, Barault et al. also suggested that plasmatic methylation changes of EYA4 me , GRIA4 me , ITGA4 me , MAP3K14-AS1 me and MSC me panel over time correlate with tumor response in metastatic CRC patients treated with chemoor targeted therapy [160]. Interestingly, using the same panel, Amatu et al. reported that circulating methylated DNA normalized to the total amount of circulating DNA dynamics can reflect clinical response to treatment with regorafenib [161].
Since 2017, COLVERA™, a Laboratory Developed Test (LDT), based on a two-gene panel (IKZF1 me and BCAT1 me ) is commercialized in the USA, for post-surgery residual disease detection and CRC patient surveillance [162]. The detection of these two genes in blood associated with stage [120] and showed a rapid reversion after surgical resection (35 of 47 positive patients at diagnosis, became negative after surgery) [163]. Moreover, in patients undergoing surveillance after primary CRC treatment, this panel was positive in 68% plasma samples of the 28 patients with clinically detectable recurrent CRC, whereas CEA was positive in only 32%, although specificity was similar in both tests (87% and 94%, respectively) [164]. Hence, this panel doubled the sensitivity of the current gold-standard marker for CRC monitoring. Recently, positivity of this panel after surgery also independently associated with increased risk of recurrence [165].

Screening and Diagnosis
Curable PCa is mostly asymptomatic, and, thus, patients with clinically detected disease are mostly at advanced stage, resulting in worse outcome and limited treatment options [166]. Digital rectal examination (DRE), in combination with serum prostate-specific antigen (PSA) quantification, remains the gold standard PCa screening tools [167], however both methods present drawbacks. A DRE positive result is dependent on clinicians' expertise and the majority of cancers detected by this method are at advanced stage [168], besides compliance is rather low. The widespread adoption of PSA screening since the late 1980s has facilitated the shift to detection of PCa at early stages [168]. Nonetheless, despite being highly sensitive, since benign prostatic hyperplasia (BPH) and other benign conditions also cause PSA elevation, the lack of cancer-specificity of this approach entails a high false-positive rate and overdiagnosis of non-life threatening PCa [167,169]. Indeed, only less than one third of the patients undergoing transrectal ultrasound-guided (TRUS) biopsy (standard diagnostic approach) due to elevated PSA levels or abnormal DRE are diagnosed with cancer [170]. In parallel, a negative result does not completely rule out the existence of cancer, leading to a large number of unnecessary invasive tissue biopsies that might be repeated due to the uncertainty of diagnosis if elevated PSA levels persist [170]. Thus, the introduction of more specific alternatives is urgently sought.
Besides blood-based liquid biopsies, aberrant DNA methylation in urological cancers can also be detected in urine, which is a non-invasive, easily accessible source of exfoliated cells and ccfDNA from diverse sites of the urinary system [171]. Hence, several studies have been carried out using this source [171]. Nonetheless, in PCa, higher sensitivities are achieved after manipulation of the prostate, either by prostate massage or DRE, increasing the invasiveness of this procedure [171]. Therefore, blood-based liquid biopsies might represent the better minimally invasive procedure for PCa detection. A summary of the currently reported ccfDNA methylation PCa detection biomarkers is depicted in Table 4.
Since epigenetic alterations are usually multiple and not necessarily overlapped, multigene panels are pivotal to increase the modest sensitivities of individual genes. Indeed, Ellinger et al. reported that using a gene panel comprising GSTP1 me , PTGS2 me , RPRM me and TIG1 me , PCa diagnostic coverage increased from 42% (GSPT1 me alone) to 47% (panel), maintaining 93% specificity [172]. Furthermore, Sunami et al. reported that GSTP1 me , RASSF1A me and RARβ2 me were hypermethylated in 13%, 24% and 12% of serum samples from PCa patients, respectively, whereas the three gene panel increased the detection rate to 29%, with 100% specificity [173]. Remarkably, the addition of FOXA1 me to the previous panel increased sensitivity to 72% although at the expense of lower specificity (72%) [46]. More recently, other panels without comprising GSTP1 me have also been tested. Indeed, MCAM me , ERα me and ERβ me panel disclosed 75% sensitivity and 70% specificity for early PCa detection [177]. Likewise, ZNF660 me , CCDC181 me , ST6GALNAC3 me and HAPLN3 me in serum displayed 22%, 26%, 31% and 44% sensitivity, respectively, and 100% specificity for PCa. Remarkably, the best gene panel (ST6GALNAC3 me , CCDC181 me and HAPLN3 me ) increased sensitivity to 67%, keeping 100% specificity [178].
Interestingly, given the modest sensitivities obtained with ccfDNA methylation even with gene panels, studies have also attempted to understand whether ccfDNA might have a better performance by complementing it with serum PSA levels. Indeed, in a Mexican cohort with biopsy-confirmed PCa, a panel comprising GSTP1 me and RASSF1A me allowed for cancer detection with 73% positive predictive value (PPV) and 59.6% negative predictive value (NPV), increasing to 81% and 66%, respectively, when serum PSA was also considered [179]. Likewise, serum GADD45a me increased sensitivity from 38% to 94% when PSA and free circulating DNA levels were also considered, although specificity decreased from 98% to 88% [180].

Prognosis, Predicition, and Monitoring
PCa is a very heterogeneous disease, ranging from small, low grade, clinically indolent tumors to large, lethally aggressive ones [184]. Thus, the main goal after establishing the presence of cancer is to evaluate its extension and aggressiveness through grading and staging, to assess prognosis and plan the treatment strategy. Currently, PCa prognostic stage groups are based on nomograms combining TNM classification, preoperative serum PSA levels, and histological International Society of Urological Pathology (ISUP) Grade Group [185,186], which is based on the evaluation of the two most common differentiation patterns in a tumor [187]. Furthermore, it is estimated that 30-50% patients treated with curative intent may show rising serum PSA levels (biochemical recurrence) within 10 years after treatment, with clinical disease progression occurring in up to 40% of these [188,189]. Thus, considering these uncertainties, there is an urgent need for development and implementation of more accurate PCa biomarkers to assist clinicians and patients in decision-making.
Besides the study of its detection value, the potential of methylation-based biomarkers in ccfDNA to predict disease progression and therapy response have also been tackled. GSTP1 me [181,190], MDR1 me , EDNRB me and RARβ2 me [181] were reported to be more frequent in castration-resistant prostate cancer (CRPC) patients than in early-stage PCa patients. GSTP1 me levels also associated with Gleason score and presence of metastasis [190] as well as with reduced DSS along with APC me [191] in CRPC patients. Furthermore, GSTP1 me and RASSF2A me were more frequently detected in men with non-organ confined compared to organ-confined disease and both associated with increased Gleason score [175]. Interestingly, preoperative serum GSTP1 me was also reported as an independent predictor of biochemical recurrence following radical prostatectomy [192]. Sunami et al. reported that GSTP1 me , RASSF1A me and RARβ2 me associated with Gleason score and serum PSA levels, whereas GSTP1 me and RARβ2 me also associated with advanced stages of disease [173]. Moreover, individually, serum PCDH17 me , PCDC10 me and PCDH8 me were also associated with advanced clinical stage, higher preoperative serum PSA, as well as lymph node metastasis and shorter biochemical recurrence-free survival [193][194][195]. Besides being also associated with Gleason score, advanced tumor stage and high PSA, CDH13 me was further associated with shorter OS [183].
Remarkably, a recent phase III multicenter trial enrolling 600 CRPC patients showed that detectable serum GSTP1 me levels prior and after two cycles of chemotherapy were both independently associated with decreased OS [196]. In the same study, undetectable serum GSTP1 me after two cycles of docetaxel further associated with longer time to PSA progression [196].

Cell-Free DNA Methylation as a Candidate "PanCancer" Screening Biomarker
The screening of these four cancer types faces different challenges. On the one hand, LC and CRC demand an early cancer detection that is still not fully accomplished with the current screening tools available. On the other hand, BrC and PCa diagnosis should not only be focused on early disease detection, but also be restrictive to clinically significant disease detection. Hence, new screening biomarkers have been exhaustively searched (Tables 1-4). One may wonder if a minimally invasive "PanCancer" detection approach based on liquid biopsies with good sensitivity and specificity for simultaneous cancer detection would increase screening effectiveness. In fact, a study performed in the context of CRC screening revealed that only 37% of the 172 subjects were compliant with screening colonoscopy [197]. Interestingly, 97% of the subjects who refused this modality accepted a non-invasive alternative, of which 83% selected SEPT9 me blood-test and only 15% the stool test, being the primary reason for this choice the convenience of the procedure [197]. Regardless of being small scaled, this study clearly demonstrates that screening compliance can dramatically increase if a convenient minimally invasive option, such as a liquid biopsy, is offered. Nevertheless, the establishment of a "PanCancer" panel for a simultaneous detection of several cancer types is a challenge given their heterogeneity.
Regarding ccfDNA methylation-based biomarkers, although only RARβ2 me , and RASSF1A me levels were reported in all four cancer types (LC, BrC, CRC, and PCa), several other genes have already been reported in at least two of the four different cancer types ( Figure 5). Interestingly, a study published in 2007 that included 70 serum samples from metastatic BrC, NSCLC, gastric, pancreatic, colorectal and hepatocellular carcinoma and 10 healthy serum controls demonstrated that a gene panel hypermethylation (RUNX3me, p16me, RASSF1Ame and CDH1me) detected cancer samples with 89% sensitivity and 100% specificity, using MSP [198], suggesting the putative value of using a single panel to detect several malignancies.
Remarkably, genome-wide DNA methylation studies have also been performed aiming to detect the presence of cancer and underlying cancer type [199][200][201]. Indeed, Li et al. and Kang et al. developed "CancerDetector" and "CancerLocator" that may detect cancer using probabilistic approaches based on ccfDNA methylation sequencing [200,201]. Moreover, Moss et al. using Illumina methylation arrays demonstrated that plasma methylation patterns can be used to identify cell typespecific ccfDNA in healthy and pathological conditions, including different types of cancer [202].
Interestingly, in 2018, the Laboratory for Advanced Medicine launched the LTD IvyGene ® Test in the USA for detection of BrC, CRC, liver cancer and LC [203]. This test utilizes a multi-target approach derived from sequencing methods to detect ccfDNA methylation profile in 40mL of whole blood samples [203,204]. Notwithstanding, according to their website, the test detected cancer with 84% sensitivity and 90% specificity, it was only validated in 197 samples obtained from subjects with either no history of cancer or diagnosis of one of the four cancers [203]. Thus, the performance evaluation of IvyGene ® Test in a large screening setting is still warranted.
Although epigenome-wide approaches allow for simultaneous screen of several hundreds of genes and might be advantageous to boost the discovery of new differentially methylated DNA regions, they are not still widely available in clinical laboratories and require high-level bioinformatics expertise, fast data processing and large data storage capabilities [205,206]. Additionally, depending on the number of samples and genes to be analyzed, targeted approaches, such as qMSP or droplet digital MSP (ddMSP) might be more cost-effective [207]. Indeed, two recent studies aimed to develop a "PanCancer" panel to detect the most incident malignancies, evaluated methylation levels of several genes in plasma samples using multiplex qMSP. Remarkably, APCme, FOXA1me and RASSF1Ame panel was able to detect BrC, CRC and LC in female patients with 72% sensitivity and 74% specificity [36], whereas FOXA1me and RARβ2me panel detected LC and PCa in males with 64% sensitivity and 70% sensitivity [46]. Although both studies included a "CancerType" panel to indicate the most likely cancer topography, reaching specificities above 80%, sensitivity to this end was rather limited [36,46]  Interestingly, a study published in 2007 that included 70 serum samples from metastatic BrC, NSCLC, gastric, pancreatic, colorectal and hepatocellular carcinoma and 10 healthy serum controls demonstrated that a gene panel hypermethylation (RUNX3 me , p16 me , RASSF1A me and CDH1 me ) detected cancer samples with 89% sensitivity and 100% specificity, using MSP [198], suggesting the putative value of using a single panel to detect several malignancies.
Remarkably, genome-wide DNA methylation studies have also been performed aiming to detect the presence of cancer and underlying cancer type [199][200][201]. Indeed, Li et al. and Kang et al. developed "CancerDetector" and "CancerLocator" that may detect cancer using probabilistic approaches based on ccfDNA methylation sequencing [200,201]. Moreover, Moss et al. using Illumina methylation arrays demonstrated that plasma methylation patterns can be used to identify cell type-specific ccfDNA in healthy and pathological conditions, including different types of cancer [202].
Interestingly, in 2018, the Laboratory for Advanced Medicine launched the LTD IvyGene ® Test in the USA for detection of BrC, CRC, liver cancer and LC [203]. This test utilizes a multi-target approach derived from sequencing methods to detect ccfDNA methylation profile in 40mL of whole blood samples [203,204]. Notwithstanding, according to their website, the test detected cancer with 84% sensitivity and 90% specificity, it was only validated in 197 samples obtained from subjects with either no history of cancer or diagnosis of one of the four cancers [203]. Thus, the performance evaluation of IvyGene ® Test in a large screening setting is still warranted.
Although epigenome-wide approaches allow for simultaneous screen of several hundreds of genes and might be advantageous to boost the discovery of new differentially methylated DNA regions, they are not still widely available in clinical laboratories and require high-level bioinformatics expertise, fast data processing and large data storage capabilities [205,206]. Additionally, depending on the number of samples and genes to be analyzed, targeted approaches, such as qMSP or droplet digital MSP (ddMSP) might be more cost-effective [207]. Indeed, two recent studies aimed to develop a "PanCancer" panel to detect the most incident malignancies, evaluated methylation levels of several genes in plasma samples using multiplex qMSP. Remarkably, APC me , FOXA1 me and RASSF1A me panel was able to detect BrC, CRC and LC in female patients with 72% sensitivity and 74% specificity [36], whereas FOXA1 me and RARβ2 me panel detected LC and PCa in males with 64% sensitivity and 70% sensitivity [46]. Although both studies included a "CancerType" panel to indicate the most likely cancer topography, reaching specificities above 80%, sensitivity to this end was rather limited [36,46].
Despite the significant advancements, standardization of technical procedures, such as amount of blood collected, type of samples to be used (plasma vs. serum), and DNA extraction and methylation analysis are still necessary to allow reproducibility between different studies and boost the translation of these markers into clinical practice. Nevertheless, although validation in large independent prospective cohorts by clinical trials in asymptomatic individuals to access the real clinical value of a screening method to detect several cancer types is still required, the data suggests that the hypothetical use of a "PanCancer" panel based on ccfDNA methylation is amenable to increase patient compliance to screening programs and decrease patient morbidity and mortality as well as healthcare systems costs.

Conclusions
In conclusion, these data indicate that LC, BrC, CRC, and PCa detection and patients' stratification according to prognosis can be achieved by analyzing gene methylation levels in ccfDNA extracted from plasma or serum, constituting a minimally invasive approach. Methylation levels' analysis in ccfDNA may not only complement current screening methods, but also aid in stratifying cancer patients according to recurrence risk and response to therapy. Nevertheless, these findings still lack validation in larger multicenter studies to enable their implementation in clinical practice.

Conflicts of Interest:
The authors declare no conflict of interest.