New Concepts in Cancer Biomarkers: Circulating miRNAs in Liquid Biopsies

The effective and efficient management of cancer patients relies upon early diagnosis and/or the monitoring of treatment, something that is often difficult to achieve using standard tissue biopsy techniques. Biological fluids such as blood hold great possibilities as a source of non-invasive cancer biomarkers that can act as surrogate markers to biopsy-based sampling. The non-invasive nature of these “liquid biopsies” ultimately means that cancer detection may be earlier and that the ability to monitor disease progression and/or treatment response represents a paradigm shift in the treatment of cancer patients. Below, we review one of the most promising classes of circulating cancer biomarkers: microRNAs (miRNAs). In particular, we will consider their history, the controversy surrounding their origin and biology, and, most importantly, the hurdles that remain to be overcome if they are really to become part of future clinical practice.


Introduction
Cancer represents the leading cause of morbidity and mortality worldwide, with approximately 14 million new cases and 8.2 million cancer related deaths in 2012, and this number is predicted to rise by approximately 70% over the next two decades according to the World Health Organization [1]. The effective and efficient management of cancer patients relies upon both early diagnosis and the frequent monitoring of patient response to treatment.
The current gold standard of cancer diagnosis is the histological examination of tissue, obtained either by radiologically guided biopsy or surgical excision. However, these procedures are invasive, expensive, and not without risk to the patient. They also take time and need to be consistently evaluated by expert pathologists. Therefore, there is a clear clinical need for alternative diagnostic techniques. In particular, the use of biological fluids such as blood as a source of non-invasive biomarkers of cancer has raised a great deal of interest [2]. So-called "liquidbiopsies" hold great clinical promise, as their non-invasive nature allows for rapid, economical, and repeat sampling-features that permit their use in screening programs and for the close monitoring of treatment response and disease progression, allowing for earlier intervention and dynamic treatment management. Furthermore, there is an increasing awareness of the genetic heterogeneity of tumors and a realization that tissue biopsies may miss this diversity. Liquid biopsies in contrast can capture the entire genetic panorama of the tumoral landscape. Consequently, this technology has the potential to radically improve current treatment regimens and therefore the outcome of cancer patients, allowing for a personalized approach to be taken for each patient. Although the majority of liquid biopsy research to date has focused upon the isolation of circulating tumor cells (CTCs), these cells are relatively rare and require sensitive collection and enrichment technology. Increasingly, the focus of liquid biopsy studies is shifting towards circulating (or cell-free) nucleic acids (cfDNA/RNA) as being easier to collect and analyze. There has been a particular interest in circulating cell-free microRNAs (cfmiRNAs), the subject of this review. For a wider overview of circulating nucleic acids as cancer biomarkers-in particular, mRNA and non-coding RNAs other than microRNAs (miRNAs)-we refer the reader to our previous article [2].
The history of circulating (blood) nucleic acids goes back to a finding in 1947 by Mandel and Metais of RNA and DNA in the plasma of healthy and sick individuals [3]. Remarkably, this report predates the realization that DNA was the molecule responsible for inheritance and the discovery of the double helix structure by Watson and Crick. It was not until the 1960s when scientific interest was aroused by the finding of cfDNA in patients with the autoimmune diseases, systemic lupus erythematosus (SLE) [4] and rheumatoid arthritis [5]. However, it was not until 1977 when the potential of cfDNA as cancer biomarkers was postulated-when Leon et al. reported elevated levels of circulating cfDNA in pancreatic cancer patients [6]. After that, in 1994, cancer-specific DNA mutations in NRAS (myelodysplastic syndrome (MDS)) [7] and KRAS (pancreatic cancer) [8] were found in the blood of cancer patients. cfRNA, in contrast to cfDNA, was not identified until 1999, when Lo et al. first identified (viral) cfRNA in the blood of nasopharyngeal carcinoma patients [9]. Several years later, in 2007, our group reported the presence of miRNAs in the blood of lymphoma patients [10]; the following year, it was demonstrated that miRNAs could be useful as non-invasive biomarkers of cancer [11,12]. miRNAs are endogenous, small (18-24 nt), non-coding (nc) RNA molecules that regulate eukaryotic gene expression post-transcriptionally. miRNAs were unknown in science until just over 20 years ago, and, even then, were not formally recognized until 2001 [13]. There are now over 2500 human miRNAs that have been identified [14], and it is believed that nearly two thirds of all human genes are directly targeted by miRNAs [15]. miRNAs have been shown to play key regulatory roles in virtually every aspect of biology [16], including in the pathogenesis of cancer, and are aberrantly expressed in many diseases ( Figure 1). Indeed, there is now compelling evidence that miRNAs regulate all aspects of the so-called "hallmarks of cancer" that enable tumor growth and metastatic dissemination [17,18] (Figure 2).
The field of circulating miRNAs has generated a great deal of interest and has been growing at an exponential rate with more than 2000 publications now published on the subject (source: PubMed; Figure 1), and many conferences and commercial entities are involved in this area. Below, we discuss some of the controversies behind the origin of these molecules and their possible functions. We also review some of the major evidence to suggest their potential as cancer biomarkers, but, most importantly, we discuss some the barriers that are still to be overcome if these molecules are to become a part of routine clinical practice.

Origin of Extracellular miRNAs
There are several different hypotheses that have been proposed to explain the presence of circulating miRNAs in biological fluids [19][20][21]. These include the passive release of miRNAs from broken cells after tissue injury, cell apoptosis or necrosis, chronic inflammation, and from cells with a short half-life such as platelets [22][23][24]. For example, specific miRNAs are elevated in blood after myocardial infarction [24][25][26] or hepatobiliar injury [27]. An alternative hypothesis, though not mutually exclusive, is that miRNAs are actively secreted from cells either shuttled via microvesicles such as exosomes or shedding vesicles [12,[28][29][30], or directly in complex with RNA-binding proteins or lipoproteins such as nucleophosmin (NPM1) [31], high-density lipoprotein (HDL) [32], or Argonaute proteins [22,33].
There is some controversy as to which of these represent the true origin of cfmiRNA, or at least the relative proportion of the different routes; until fairly recently, it was believed that most circulating miRNAs were derived from cell-derived vesicles [34].This has been contested by at least two independent reports that suggest that more than 90% of the miRNAs in blood are membrane-free and associated with Ago proteins [22,33]. Irrespective of their origin, the composition of cfmiRNAs appears to differ greatly from their respective donor cells [35]. In fact, some secreted miRNAs are notpresent at all in the parental cells [30].

Cell-Cell Communication (Hormone-Like Molecules)
Aside from their (passive) role as biomarkers, there has been a great deal of interest in the function of cfmiRNAs and in particular their ability to act as signaling molecules that potentially allow tumor cells to modify the bodies response to its own advantage. The first evidence that extracellular miRNAs could act as signaling molecules was discovered in plants in 1996 [36]. There is now emerging evidence of human miRNAs acting in a similar fashion either as paracrine signalers or even as systemic communicators between cells in an endocrine manner (in a hormone-like way) [20,37]. A number of facts support this possibility: miRNAs appear to be selectively packaged and secreted [31,38]; extracellular miRNAs are protected from RNases either by lipoprotein or protein carriers or by microvesicle membranes [33]; and circulating miRNAs are able to alter gene expression in recipient cells and mediate functional changes in them [30,35,39,40]. The first indication that miRNAs could shuttle between cells via exosomes was demonstrated in mast cells [29]. Later, the transfer of miRNAS between different cell types (embryonic stem cells and fibroblasts) was demonstrated [41]. More recently, exosomal miRNA has been shown to be able to modulate inmunological response through modification of the gene expression of antigen presenting cells (APC) by T-cells, B-cells, and dendritic cell-derived miRNAs [35].
Multiple studies suggest that cfmiRNAs could play a role in cancer biology through tumor-derived exosomal miRNA modulating non-tumor cells to the ultimate benefit of the tumor. For example, exosomal-cfmiRNAs have been demonstrated to modulate chemosensitivity [42], angiogenesis, and cell invasiveness [43][44][45][46]. While this is potentially a fascinating phenomena, this is still a contentious issue, and it is worth remembering that the few studies carried out to date have been almost exclusively in vitro. Finally, although Ago2-boundmiRNAs appear to form the majority of cfmiRNA, there is no evidence (or known mechanism) for the active release of vesicle-free AGO2-miRNA complexes in mammals, nor any indication of Ago-2 surface receptors for the uptake by recipient cells. Therefore, the physiological relevance of cfmiRNA as an intercellular signaling mechanism remains to be determined.

miRNAs as Cancer Biomarkers
According to the National Cancer Institute, a biomarker is defined as "a biological molecule found in blood, other body fluids, or tissues that are a sign of a normal or abnormal process, or of a condition or disease." In cancer, they can be divided into three general categories: (1) diagnostic biomarkers, which are used for a differential diagnosis; (2) prognostic biomarkers, which can distinguish tumors with a good outcome from those with a bad outcome; and (3) predictive biomarkers, which are for assessing whether a treatment is likely to be effective for a particular patient or not. An ideal biomarker should have a high specificity, sensitivity, and predictive power. miRNAs have a number of intrinsic characteristics that make them attractive as biomarkers. Firstly, they are highly specific, and it has been shown that miRNA expression profiles differ between cancer types according to diagnosis and the developmental stage of the tumor, with a greater resolution than traditional gene expression analysis [47]. Secondly, unlike other RNA classes, miRNAs are remarkably stable and therefore can be robustly measured not only in biological fluids but also from routinely prepared formalin-fixed paraffin-embedded (FFPE) material [48]. Indeed, unlike other RNA species, miRNAs appear resistant to boiling, pH changes, repeated freeze-thawing cycles, and fragmentation by chemical or enzymes [12,20,49]. It should be noted, however, that cfmiRNAs are not themselves intrinsically resilient to RNase or any other treatment; rather, they are protected by their lipidic or protein-based carrier [12,50,51]. As a result of these characteristics, the use of cfmiRNAs as biomarkers-and in particular as cancer biomarkers-has generated a plethora of publications over the last few years. Due to the limitations of space, we will not attempt to review all of these but instead discuss the more robust studies that identify common cfmiRNA biomarkers in multiple studies. More often than not, these biomarker miRNAs are themselves intimately involved in cancer pathology, as shown in Table 1, which includes their respective experimentally validated targets. While it may be tempting to speculate that these miRNAs may have the same effect while in circulation as intracellularly, there is no evidence that this is indeed the case.   2.1. let-7 Family (let-7a, -7b, -7c, -7e, -7f, -7i) There are 13 different let-7 family members in humans: let-7a-1, 7a-2, 7a-3, 7b, 7c, 7d, 7e, 7f-1, 7f-2, 7g, 7i, miR-98, and miR-202 [244]. Differential expression of let-7 family members has been described to be downregulated in a wide variety of cancers such as melanoma, pancreatic cancer, prostate cancer, and sarcoma, although some, including lymphoma, mesothelioma, and breast cancer, have been shown to be upregulated; thus, the let-7 family is generally regarded as a tumor suppressor [245]. let-7 has been shown to be a direct regulator of some important oncogenes, such as the three RAS genes [246,247], HMGA2 [248,249], STAT3 [250], UHRF2 [251], and MYC [252][253][254]; additionally, let-7 family targets cell cycle and cell proliferation genes [255][256][257]; finally, apoptosis is also shown to be regulated by let-7 family, through CASP3 targeting [258]. RNase III nuclease, known to process pre-miRs, was also confirmed as a direct target of the let-7 family, so they might regulate their own processing [259,260]. let-7 family members have been identified differentially expressed and therefore have been proposed as diagnostic tools in serum/plasma of many cancer types including lung cancer (let-7a,  [263]. In general, low let-7 levels are associated with poor prognosis including overall survival, early recurrence, and tumor size.
Circulating miR-10b levels have been described as being upregulated in patients with ovarian cancer [30], lung cancer [29], oesophageal [75], and glioblastoma [73] compared to healthy controls. Consistent with this, miR-10b is increased in plasma from metastatic breast cancer patients [27,28] and in the cerebrospinal fluid of patients with brain metastasis of both breast and lung cancer [73].
Differential expression of miR-29 family members in plasma/serum has been observed in several cancer types. The expression levels of all the miR-29 family members were upregulated in sera of patients with osteosarcoma. In particular, miR-29a and miR-29b were associated with poor prognosis [162]. miR-29a has been shown to be upregulated in colorectal cancer and therefore has been proposed as a potential non-invasive biomarker for early detection of colorectal cancer [101,103,164,231], also involving liver metastasis [163]. It has also been found to be upregulated in breast cancer [165,166] and downregulated in oral and ovarian cancer, compared with healthy controls [76,102]. Similarly, serum levels of miR-29b have been proposed as potential biomarkers for diagnosis and prognosis of colorectal cancer [167], whereas miR-29c could be useful as a predictor of postoperative early relapse [168]. However, it was found to be downregulated in serum of nasopharyngeal carcinoma patients, compared with controls [170].
Members of the miR-30 family have been identified differentially regulated in body fluids, but their potential as biomarkers has mostly been reported in combinations with other miRNAs. For example, a blood test based upon a combination of the levels of five miRNAs including miR-30c has been described to effectively differentiate prostate cancer patients from benign prostatic hyperplasia (BPH) patients and healthy controls [173]. In addition, the combination of four plasma circulating miRNAs, including miR-30c and serum PSA, has a greater potential to be used as a noninvasive diagnostic biomarker for prostate cancer screening than PSA testing alone [174]. Similar studies have been reported for miR-30c and miR-30a-3p in lung adenocarcinoma [171,175], miR-30a in esophageal adenocarcinoma [172], and miR-30d in lung cancer [176]. In hepatocellular carcinoma, miR-30e has been recently found to be downregulated in serum when compared with healthy controls [178], and, in lung cancer, high levels of miR-30d in serum have been associated with a shorter overall survival [176,177].
Elevated levels of miR-34a in serum can discriminate between breast cancer patients and healthy controls, and are also associated with the presence of overt metastasis [72,237]. High levels of circulating miR-34a have also been observed in ovarian and lung cancer [72,74,179], and miR-34b has been found to be upregulated in serum from prostate cancer patients [77]. In osteosarcoma patients, miR-34b levels were found to be downregulated when compared with controls, and these expression levels were significantly decreased in the metastatic patients [180]. Similarly, downregulation of circulating miR-34c in serum of NSCLC patients and miR-34b/c in serum of breast cancer patients has been reported and might have potential as biomarkers for the diagnosis of these pathologies [181,182].
Deregulated levels of miR-125a were present in the saliva of oral squamous cell carcinoma (OSCC) patients and in serum of NSCLC patients compared with healthy controls [183,184]. In a similar way, miR-125b levels were significantly lower in glioma patients and in serum-derived exosomes of melanoma patients [192,193]. In addition, low circulating levels of miR-125a have been associated with poor prognosis in both breast cancer and hepatocellular carcinoma [185,186]. In contrast, miR-125b was found to be upregulated in the plasma and serum of metastatic prostate cancer patients [356], breast cancer [125], OSCC [190], colorectal cancer [191], and NSCLC [188], in comparison with healthy controls, and to be associated with poor prognostic outcome and chemotherapeutic resistance in this cancer [187][188][189].
In many studies, differentially expressed levels of circulating miR-155 have been identified, including breast cancer [90,[194][195][196], colorectal cancer [198], lung cancer [57,199], AML [201], diffuse large B-cell lymphoma (DLBCL) [11,202], and esophageal cancer [200], making it a potential non-invasive diagnostic biomarker for early detections in these pathologies. In fact, a biosensor for miR-155 detection in plasma has recently been developed for the diagnosis of breast cancer [368]. Elevated levels of miR-155 are also related to overt metastasis in breast cancer [72,197], and these high levels have also been identified not only in blood but also in the urine of breast cancer patients [126]. In addition, miR-155 has been also suggested as a prognostic biomarker in chronic lymphocytic leukemia (CLL) and adult T-cell leukemia (ATL) [203,204], and as a predictive biomarker to response to therapy in CLL [203]. The miR-200 family is believed to play crucial roles in both cancer initiation and metastasis-in particular, in epithelial-mesenchymal transition (EMT)-primarily through the targeting of ZEB1 and ZEB2 transcription factors [369,370]. It has also been associated with angiogenesis by the targeting of VEGFA and VEGF receptors [371,372] and pro-angiogenic ligands such as CXCL8 and CXCL1 [373].
Elevated serum levels of miR-200a, miR-200b, miR-200c, and miR-141 have been suggested as good biomarkers for diagnosis and prognosis in ovarian cancer [205,206,208], and serum levels of miR-429 were associated with poor overall survival in NSCLC [169]. In addition, elevated levels of circulating miR-141 have been identified to show diagnostic potential in patients with upper urinary tract urothelial cancer [216], lung cancer [74], prostate cancer [12,[217][218][219][220], breast cancer [207], and bladder cancer [374]. In breast and bladder cancer, this upregulation is also associated with prognosis. Furthermore, miR-200c was found significantly elevated in the plasma of patients with colorectal cancer [210], gastric cancer [212,213], and breast cancer [207], and this upregulation was associated with poor prognosis. In colorectal cancer, it has also been identified as a metastasis predictive biomarker [211] as well as miR-141 [221]. Similarly, miR-200c can be useful to predict prognosis in NSCLC [214] and in esophageal cancer. In the latest research, the serum level of miR-200c, as well as miR-200b, can be useful for predicting response to chemotherapy [215], and also has prognostic value in prostate cancer and predictive value in docetaxel chemotherapy outcomes [209].

Extracellular miRNAs in Other Biological Fluids
In addition to blood, other biological fluids such as urine, saliva, cerebrospinal fluid, vitreous humor of the eye, breast milk, seminal fluid, and tears have been studied as potential sources of miRNA biomarkers [422,423] (Figure 3). The majority of these studies concern tumor types associated with the source of the biological fluids. For example, saliva has been studied in head and

Extracellular miRNAs in Other Biological Fluids
In addition to blood, other biological fluids such as urine, saliva, cerebrospinal fluid, vitreous humor of the eye, breast milk, seminal fluid, and tears have been studied as potential sources of miRNA biomarkers [422,423] (Figure 3). The majority of these studies concern tumor types associated with the source of the biological fluids. For example, saliva has been studied in head and neck squamous cell carcinoma [183,[424][425][426][427], tumors of the parotid gland [183,424,428], esophageal cancer [154], and pancreatic cancer [142,429]. Urine is another well studied source of cfmiRNAs associated with cancer in particular urological cancers including prostate and bladder cancer (reviewed in [430]). In addition, several studies have looked at the potential of urine for miRNA biomarkers in ovarian, breast, and liver cancer [126,431,432]. miRNAs in cerebrospinal fluid have been described as potential biomarkers for the diagnosis and monitoring of disease in brain tumors such as glioblastoma but also in CNS lymphomas and in brain metastases of non-neuronal origin [73,[433][434][435][436]. In a similar vein, miRNAs, in the vitreous humor of the eye, have been identified in ocular cancers including vitreoretinal lymphoma or uveal melanoma [437,438]. It also has been suggested that the miRNA profile of breast milk could be a more sensitive biomarker for breast cancer than blood-associated miRNAs [439] and that seminal fluid-associated miRNAs can serve as biomarkers of prostate cancer [440].

Challenges in Studying cfmiRNA
A major obstacle to the translation of cfmiRNAs from laboratory studies into the clinic is the lack of consistent and robust results with many apparently contradictory reports in the literature. A likely reason for this lack of reproducibility is that there are very few multi-center studies, and cohorts are often insufficiently powered. Another confounding factor is the fact that there is a high degree of inter-individual variability in the levels of cfmiRNAs, even when focusing only on healthy populations [441]. Moreover, there is a technical source of variation between studies, such as the starting material used for the experiments (e.g., the purification of cells, the cell types, the control populations used, the RNA extraction method, etc.), the technological platforms (e.g., microarray, qRT-PCR vs. next generation sequencing (NGS) etc.), and the differing statistical methodologies used.
The blood collection and processing represent critical points of variability in cfmiRNA studies. In the first instance, miRNA contamination can occur at the venopuncture site itself [442]. After extraction, the elapsed time between blood collection and processing should be minimized to prevent lysis and cellular contamination, which can be a major source of variability between samples [443][444][445][446]. In addition, the choice of anti-coagulant used in plasma collection can influence downstream detection technologies, such as qRT-PCR and heparin-coated tubes, should be avoided [447]. Another major source of difference in cfmiRNA profiles comes from the choice of whether to use serum or plasma, and whether to purify exosomes or use whole serum/plasma [448][449][450][451].
The choice of RNA purification procedure can also critically affect the results of cfmiRNA studies and should be considered carefully in terms of experimental design. For example, small RNA molecules with low GC content are known to be selectively lost during Trizol-based extraction (the most popular method) when present in low concentrations, such as in biological fluids, and thus should be avoided if possible; specific commercially available kits should be used instead [452]. Many researchers use non-human miRNAs (e.g., C. elegans sequences) as spike-in controls to control for variability between the miRNA extraction efficiency between samples [12]. Another important issue is that it is almost impossible to accurately quantify RNA in samples from biological fluids due to the low quantities of RNA present and the high levels of contaminating salts and protein that can interfere with spectrophotometric measurement. Therefore, studies often use fixed volumes of plasma to standardize, even if it is evident that they may contain different amounts of RNA [453].
There are many different methods available to measure cfmiRNAs, including qRT-PCR (LNA-based, Taqman or other proprietary technologies), digital PCR (dPCR), microarrays, and next generation sequencing (NGS) techniques. The choice of platform depends largely on the experimental design required (Figure 4). Importantly, it should be borne in mind that the choice of technique can massively influence measurements; indeed, several studies show a lack of concordance between platforms when using the same sample source [454,455]. Another challenging issue in cfmiRNA studies is the lack of consensus about a suitable endogenous reference to use in biological fluids, as the small nucleolar RNAs (snoRNAs) generally used as reference genes in miRNA cell-based studies is not present in biological fluids due to degradation [456,457]. As an alternative, individual miRNAs themselves are frequently used [11,12,458]. However, it has been shown that the expression levels of the most commonly used housekeeping miRNAs in cfmiRNA studies vary significantly between samples depending upon the pathology that is being studied [77,459]. Therefore, miRNAs to be used as reference genes have to be chosen with care, determining empirically for each experiment which miRNAs are more stable (using geNorm and/or NormFinder algorithms), an approach taken by some studies [458,459] but not always possible when sample volumes are limited. Alternatively, a more economical option is to include at least two (preferably three) miRNAs as reference controls for cfmiRNAs studies.

Conclusions
It is clear that there is a great deal of interest in liquid biopsies, and in cfmiRNA in particular, as a viable alternative to tissue-based sampling in the clinic. Such an approach would bring a fundamental change to cancer patient management by allowing repeated sampling for treatment response monitoring, an assessment of tumor heterogeneity, and even cancer screening programs. cfmiRNAs are particularly attractive candidates for non-invasive cancer biomarkers due to their surprising degree of stability in biological fluids; as we have outlined above, there is now a wealth of literature to suggest that this class of molecules holds great clinical promise. The caveat is that, as a very recently discovered field, there appears to be little agreement between seemingly identical studies, presumably due to many different factors outlined above between studies. In others words, Another challenging issue in cfmiRNA studies is the lack of consensus about a suitable endogenous reference to use in biological fluids, as the small nucleolar RNAs (snoRNAs) generally used as reference genes in miRNA cell-based studies is not present in biological fluids due to degradation [456,457]. As an alternative, individual miRNAs themselves are frequently used [11,12,458]. However, it has been shown that the expression levels of the most commonly used housekeeping miRNAs in cfmiRNA studies vary significantly between samples depending upon the pathology that is being studied [77,459]. Therefore, miRNAs to be used as reference genes have to be chosen with care, determining empirically for each experiment which miRNAs are more stable (using geNorm and/or NormFinder algorithms), an approach taken by some studies [458,459] but not always possible when sample volumes are limited. Alternatively, a more economical option is to include at least two (preferably three) miRNAs as reference controls for cfmiRNAs studies.

Conclusions
It is clear that there is a great deal of interest in liquid biopsies, and in cfmiRNA in particular, as a viable alternative to tissue-based sampling in the clinic. Such an approach would bring a fundamental change to cancer patient management by allowing repeated sampling for treatment response monitoring, an assessment of tumor heterogeneity, and even cancer screening programs. cfmiRNAs are particularly attractive candidates for non-invasive cancer biomarkers due to their surprising degree of stability in biological fluids; as we have outlined above, there is now a wealth of literature to suggest that this class of molecules holds great clinical promise. The caveat is that, as a very recently discovered field, there appears to be little agreement between seemingly identical studies, presumably due to many different factors outlined above between studies. In others words, there is a clear need of setting standardized approaches to be put into practicein future cfmiRNA biomarker studies if these molecules are to ever make their way into routine clinical practice.