Comparison of a healthy miRNome with melanoma patient miRNomes: are microRNAs suitable serum biomarkers for cancer?

Abstract

MiRNAs are increasingly recognized as biomarkers for the diagnosis of cancers where they are profiled from tumor tissue (intracellular miRNAs) or serum/plasma samples (extracellular miRNAs). To improve detection of reliable biomarkers from blood samples, we first compiled a healthy reference miRNome and established a well-controlled analysis pipeline allowing for standardized quantification of circulating miRNAs. Using whole miRNome and custom qPCR arrays, miRNA expression profiles were analyzed in 126 serum, whole blood and tissue samples of healthy volunteers and melanoma patients and in primary melanocyte and keratinocyte cell lines. We found characteristic signatures with excellent prognostic scores only in late stage but not in early stage melanoma patients. Upon comparison of melanoma tissue miRNomes with matching serum samples, several miRNAs were identified to be exclusively tissue-derived (miR-30b-5p, miR-374a-5p and others) while others had higher expression levels in serum (miR-3201 and miR-122-5p). Here we have compiled a healthy and widely applicable miRNome from serum samples and we provide strong evidence that levels of cell-free miRNAs only change significantly at later stages of melanoma progression, which has serious implications for miRNA biomarker studies in cancer.

Keywords: circulating miRNAs; healthy miRNome; melanoma biomarkers; miRNA qPCR arrays; serum/tissue samples.

Figure 1. Overview of study design

Numbers in brackets indicate number of separate samples included in pools. Individual melanoma patient serum samples analysed on custom qPCR arrays comprise 4 stage 0, 11 stage I, 17 stage II, 11 stage III and 9 stage IV samples.

Figure 2. The secreted healthy miRNome

A) Stability of the healthy miRNome. Mean-standard deviation plots of averaged raw Cq values or normalized expression for 1066 miRNAs (qPCR arrays of 3 serum pools (from 19 individual healthy donors) and circadian serum samples from 4 individuals). 30 miRNAs (list) with a STD above 1 for Cq values (red dots) were determined as unstable. MiRNAs absent in all samples as well as the averaged healthy miRNome are shown in Supplementary Table S2. B) Analysis of circadian miRNomes. Twelve serum samples of 4 volunteers collected at different times of day were compared to the healthy miRNome. Fold changes +/− SD were calculated for miRNAs that had a measurable higher or lower expression at one time point only (midday). C) Pie chart summarising data of the healthy miRNomes. NA: not expressed (Cq values >30 in all samples), MC: bad melt curves in more than 80% of samples, low expression: Cq values >27 in more than 80% of samples, high expression: Cq < 27 in more than 80% of samples, unstable: list of Fig. 2A.

Figure 3. MiRNA profiling of melanoma samples

A) Heatmap of whole miRNome qPCR array data of all 44 samples (see Fig. 1) normalized by global plate mean (miRNAs absent across all samples are not depicted). WB: whole blood, NS: normal skin tissue, MET: metastatic tissue, PM: primary melanoma tissue, st: stage; NHEK: normal human epidermal keratinocyte cell line; NHEM: normal human epidermal melanocyte cell line, HY: healthy, HYcd: healthy circadian, MP: melanoma patient, HP: healthy pool, PP: patient pool. * represent normal skin, arrows metastatic tissue and “P” indicates primary tumour. B) Heatmap and C) PCA and t-SNE plots of 82 individual serum samples (30 healthy volunteers and 52 melanoma patients) analysed by custom qPCR arrays. Data were normalized using the 5 most stable miRNAs. Colours: blue-male; pink-female; green-healthy; red-patient.

Figure 4. Biomarker studies

A) Heatmap and PCA showing the averages of samples belonging to different melanoma stages and healthy serum samples (based on Fig. 3B and 3C). S: stage; hy: healthy. B) ROC curves on pairwise comparisons between indicated sample groups (30 healthy, 4 stage 0, 11 stage I, 17 stage II, 11 stage III, 9 stage IV samples; early: stage 0+I+II; late: stage III+IV). Underlined miRNAs were upregulated while all others were downregulated compared to healthy or early stage samples. MiRNAs in grey shade should not be considered as potential biomarkers because they were highly expressed (Cq < 15) in whole blood samples and might therefore derive from contaminating blood cells (miR-1260a, miR-22-3p, miR-1280, miR-451a and miR-16-5p, also see Supplementary Table S3), or they were part of the unstable miRNAs (miR-432-3p, miR-373-5p; see Fig. 2A). ACC: accuracy value, AUC: area under the curve.

Figure 4. Biomarker studies

A) Heatmap and PCA showing the averages of samples belonging to different melanoma stages and healthy serum samples (based on Fig. 3B and 3C). S: stage; hy: healthy. B) ROC curves on pairwise comparisons between indicated sample groups (30 healthy, 4 stage 0, 11 stage I, 17 stage II, 11 stage III, 9 stage IV samples; early: stage 0+I+II; late: stage III+IV). Underlined miRNAs were upregulated while all others were downregulated compared to healthy or early stage samples. MiRNAs in grey shade should not be considered as potential biomarkers because they were highly expressed (Cq < 15) in whole blood samples and might therefore derive from contaminating blood cells (miR-1260a, miR-22-3p, miR-1280, miR-451a and miR-16-5p, also see Supplementary Table S3), or they were part of the unstable miRNAs (miR-432-3p, miR-373-5p; see Fig. 2A). ACC: accuracy value, AUC: area under the curve.

Figure 5. Comparison of tissue and serum samples

Bar diagram showing averaged relative expression values from available serum (blue), normal skin (red), primary tumour lesion (green) and metastatic lesions (purple) samples. Arrows indicate gradually increasing expression of miRNA with progression of disease, stars highlight miRNAs that were only detected in tissue samples but not in serum, dots mark miRNAs with serum expression only. Expression levels of single samples can be seen in Supplementary Fig. S3.