1. Introduction
Although malignant melanoma accounts for ~4% of skin cancer cases, it accounts for ~75% of all associated mortalities. In the US alone, it is estimated there were 87,000 new cases and nearly 10,000 deaths due to melanoma in 2017 [
1]. Furthermore, the incidence of melanoma has been increasing at faster rate than any other cancer type, having doubled since 1973 [
2]. The clinical case for early diagnosis of melanoma is compelling, as if detected early enough, (stage I) 5 year survival is >95%, whereas in advanced melanoma (stage IV) survival is just 10–20% [
3].
Non-invasive diagnostics, or liquid biopsies, represent a major advance towards earlier diagnosis and disease monitoring of cancer patients, including those with melanoma. As a consequence, there has been a great deal of interest in recent years in the potential of circulating nucleic acids, and in particular circulating microRNAs (miRNAs) [
4,
5]. In melanoma, several studies have implemented circulating miRNA in metastasis and risk of recurrence [
6,
7]. However, outside of miRNAs there has been little research on other cell-free (cf) RNA species in the circulating transcriptome. Part of the reason for this paucity of knowledge is the presence of high levels of RNase activity in blood, which typically results in fragmentation of longer RNA species such as mRNA [
8,
9]. This makes detection of these molecules particularly challenging. The advent of next-generation sequencing (NGS) technology, however, presents new opportunities for the field, as sequences can be elucidated in a ‘bottom-up’ manner without the need of a priori probe sequence knowledge. We used next generation sequencing (NGS) to characterize and compare the circulating transcriptomes of plasma from melanoma patients with different stage disease, along with sex-/age-matched healthy individuals, in order to identify novel biomarkers for this cancer.
We identified not only circulating miRNAs with biomarker potential, but also mRNA gene fragments and non-coding YRNAs. YRNAs (Ro-associated Y) are poorly characterized, non-coding RNAs, which were initially identified in the blood of rheumatic autoimmune disease patients [
10]. They are a family based around four highly conserved sequences (
RNY1,
RNY3,
RNY3,
RNY5) involved in Ro60 inhibition, DNA replication, and quality control of non-coding RNAs [
11,
12,
13]. Our possible biomarkers were validated in an independent cohort of 327 plasma samples from melanoma patients, collected retrospectively and prospectively. This study provides evidence that the largely unexplored circulating transcriptome could provide a valuable source of liquid biopsy biomarkers for melanoma in particular, and cancer in general.
3. Discussion
The presence and relative stability of cfRNA in biological fluids has led to a great deal of interest in their use as ‘liquid biopsies’ for disease, in particular for cancer. However, with the exception of miRNAs, the circulating transcriptome remains largely unexplored. While NGS offers researchers the ability to elucidate the circulating transcriptome in its entirety, and therefore to identify novel biomarkers of disease, the application of RNAseq to biofluids such as plasma poses many challenges, not least the low quantity and quality of RNA present in these samples. As a consequence, studies to date have focused on the technical optimization of these techniques [
19,
20,
21]. However, very few studies to date have sought to assess the potential usefulness of their findings through validation in independent cohorts.
In order to fully explore the complexity and biomarker potential of the melanoma circulating transcriptome, we pooled samples to maximize the starting quantity of cfRNA. As a result, we were able to obtain 40–50 million reads per pooled sample, an order of magnitude higher than comparable studies [
20,
22]. In contrast to exosomal cfRNA [
22], we found that miRNAs only represented a minor component (<3%) of the whole plasma circulating transcriptome, levels similar to other plasma NGS studies [
19,
20,
21]. We identified 442 different miRNAs in our samples, somewhat higher than that reported in comparable studies [
23,
24], probably as a result of the pooled design and the higher quantity of RNA that we used. Consistent with other studies, we found that
let-7b,
miR-423, and
miR-320a-3p were the most highly expressed miRNAs in our plasma samples [
20]. We identified eleven miRNAs that were differentially expressed between healthy controls and the different clinical stages of melanoma (
Table 2). This included
miR-21, which has previously been shown to be upregulated in melanoma plasma samples [
25], and
miR-92b and
miR-628, both of which are more highly expressed in plasma from monosomy 3 uveal melanoma patients [
26].
Based on our sequencing results, we measured the expression of
miR-320a-3p and
miR-134-5p, the two miRNAs most differentially expressed between samples, in a validation cohort of 96 melanoma patients and 28 controls. Both miRNAs were significantly down-regulated (
p < 0.0001) in plasma from all stages of melanoma patients when compared to samples from healthy controls.
MiR-320a has also been found to be down-regulated in melanoma tumor cells when compared to heathy skin samples [
27]. Furthermore, this miRNA was shown to function as an inhibitor of cell proliferation. The down-regulation of
miR-320a has been observed in the blood of several cancers including colorectal cancer [
28], gastric cancer [
29], and retinoblastoma [
30]. Moreover,
miR-320a is up-regulated in melanoma cells after treatment with bevacizumab or rapamycin + bevacizumab [
31]. Furthermore,
miR-134 has been characterized as a tumor suppressor, acting to regulate proliferation, apoptosis, and invasion and migration in a wide range of cancer types, including melanoma [
32,
33]. ROC analysis of miRNA expression gave AUC values of 0.798 and 0.788 respectively. The
miR-320a miRNA had a sensitivity of 90%, whereas
miR-134 had a specificity of 96%, suggesting these two miRNAs in combination could be useful biomarkers for melanoma.
Even though circulating extracellular mRNA was first detected in 1999 (in melanoma) [
34], as the vast majority of circulating mRNA is degraded by blood RNase activity [
35], this potential source of biomarkers has largely been overlooked, even though mRNA fragments can represent up to 75% of total cfRNA [
19]. In our study, just over 5% of mapped reads corresponded to protein-encoding mRNA fragments. We detected 3672 probes (<500 bases in length) that had at least 25 mapped reads in our samples. Nearly 40% of the probes mRNA fragments mapped to the 5´-terminus (i.e., first 10%) of their respective gene transcripts, probably reflecting the 3´to 5´ cleavage activity of RNase A, the major RNase species in blood [
26]. We did not notice a corresponding shift in the length profile between healthy and melanoma patient samples [
36].
Pathway analysis of the genes corresponding to differentially-expressed mRNA fragments showed significant enrichment in the angiopoietin, p21-activated kinase (PAK) and Eukaryotic Initiation Factor 2 (EIF2) pathways. It has been previously reported that circulating level of Angiopoietin-2 (Ang-2) protein in melanoma patient sera closely correlates with disease progression [
37]. Similarly, amplification of the PAK (p21-activated kinase) pathway is characteristic of BRAF-wild type melanoma [
38], while in BRAF-mutant melanoma it is responsible for resistance to MAPK-inhibitor treatment [
39]. Interestingly, both
SOS1 and
ATM1, which were the third and fourth most differentially expressed probe sets in our analysis, form part of the angiopoietin, PAK, and EIF2 pathways. We measured levels of
SOS1,
ATM1,
CD109, and
AMFR mRNA fragments in plasma samples from 173 melanoma patients and 47 healthy controls. With the exception of
CD109, these mRNA fragments mapped to regions corresponding to the 5′-terminus of the reference gene transcript and included the initiation codon. Consistent with the NGS data (
Table A1), levels of all these mRNA fragments were up-regulated in melanoma patient samples compared to samples from healthy controls. Particularly intriguing was the up-regulation of
CD109 and
AMFR in stage 0 samples compared to samples from more advanced stage melanoma, suggesting that these mRNA fragments could be used for early diagnosis of melanoma, although we do not have data on how many of these patients went on to develop advanced disease. (
Figure 3b–e).
CD109 has been identified as an important regulator of the Epithelial–mesenchymal transition (EMT pathway), and has also been found to be down-regulated in more advanced stage hepatocellular carcinoma [
40]. The product of the
AMFR gene, gp78, also regulates EMT, and increasing levels of
AMFR are associated with metastatic melanoma [
41,
42]. Intriguingly,
CD109 is a predicted target of
miR-134 by the Targetscan algorithm; we are currently carrying out experiments to confirm this. ATM1 is a serine/threonine kinase induced by DNA damage and associated with risk in many cancer types [
43]. SOS1 is a guanine nucleotide exchange factor for RAS proteins frequently mutated in melanoma [
44]. Interestingly, all four of these gene fragments were more highly expressed not only in advanced stage disease, but also stage 0 disease; indeed, levels of
CD109 and
AMFR were higher in plasma from stage 0 disease than more advanced stage disease, suggesting that these biomarkers maybe non-tumoral in origin. Consistent with this hypothesis, the release of
CD109 by bone marrow mesenchymal stem cells has recently been shown to attenuate EMT in skin squamous cell carcinoma [
45], and
AMFR plays an important role in regulation of the anti-cancer immune Stimulator of interferon genes (STING) pathway [
46].
To test the potential diagnostic ability of these biomarkers, we carried out ROC analysis, however the results from individual mRNA fragments were poor (AUC range 0.722 (
SOS1) to 0.767 (
ATM1). In contrast, a combination of
ATM1,
SOS1, and
AMFR resulted in an AUC value of 0.825 with a sensitivity of 75% and specificity of 92%. Although these findings need to be confirmed independently, this combination compares very favorably with existing sera markers such as LDH and S100B with reported sensitivities/specificities of 41.6/84.2% and 36.3/96.5%, respectively [
47].
By far the largest class of circulating cfRNA that we identified in the samples corresponded to YRNA sequences, accounting for close to 50% of sRNA mapped reads. Remarkably, despite the prevalence of cfYRNAs in the blood, there is very little known about this class of ncRNA. YRNAs are short 80–110 nt ncRNAs, first identified in the early 1980s as an RNA component of the soluble Ro60 ribonucleoprotein particle found in the blood of patients with autoimmune diseases [
48]. The function of YRNAs is still poorly understood; they appear to be essential for DNA replication [
49] and are up-regulated in cancers [
50], presumably as a result of their association with apoptosis [
51]. The first description of circulating cfYRNAs came in 2013 from Dhahbi et al., who observed that 33% of mapped reads from sera of healthy individuals were YRNA sequences [
52]. The same group later reported that YRNA accounts for 38% of cfRNA in sera from breast cancer patients [
53], and subsequently, in the sera of head and neck squamous cell carcinoma patients [
54]. More recently, a study of 183 plasma samples from healthy individuals found that YRNAs accounted for 63% of cfRNA [
55]. As far as we are aware, apart from a recent study that measured YRNAs in the sera of 30 renal carcinoma patients [
56], this the first study to look at the biomarker potential of YRNAs in cancer patients. Interestingly, we found that levels of YRNAs were significantly higher in samples from patients with stage 0 disease, maybe pointing to increased levels of tumor-associated apoptosis [
51] even despite the small tumor sizes compared to more advanced disease stages.
In summary, we have elucidated the circulating transcriptome of plasma samples from melanoma patients and found a number of novel RNA biomarker species that we validated independently using qRT-PCR and ddRT-PCR in combined retrospective and prospective cohorts, detected from only 1mL of serum. These findings have potential clinical utility as new tools for early detection of melanoma, particularly as our results suggest that these biomarkers can detect disease much earlier than current diagnostic techniques. Furthermore, as blood-based biomarkers, there is potential for screening of non-symptomatic individuals. Whilst it is clear that much further validation is required, this study provides strong evidence that the circulating transcriptome holds much promise as a source of liquid biopsies for melanoma that surely merits further exploration.