MEDICALSCIENCES
`
`Circulating microRNAs as stable blood-based
`markers for cancer detection
`
`Patrick S. Mitchell†‡, Rachael K. Parkin†‡, Evan M. Kroh†‡, Brian R. Fritz†§, Stacia K. Wyman†,
`Era L. Pogosova-Agadjanyan¶, Amelia Peterson†, Jennifer Noteboom储, Kathy C. O’Briant††, April Allen††,
`Daniel W. Lin储††‡‡, Nicole Urban††, Charles W. Drescher††, Beatrice S. Knudsen††, Derek L. Stirewalt¶,
`Robert Gentleman††, Robert L. Vessella储‡‡, Peter S. Nelson†¶, Daniel B. Martin†§§, and Muneesh Tewari†¶,¶¶
`
`Divisions of †Human Biology, ¶Clinical Research, and ††Public Health Sciences, Fred Hutchinson Cancer Research Center, Seattle, WA 98109;
`§§Institute for Systems Biology, Seattle, WA 98103; 储Department of Urology, University of Washington, Seattle, WA 98195;
`and ‡‡Department of Veterans Affairs, Puget Sound Health Care System, Seattle, WA 98108
`
`Communicated by Leland H. Hartwell, Fred Hutchinson Cancer Research Center, Seattle, WA, May 12, 2008 (received for review March 18, 2008)
`
`Improved approaches for the detection of common epithelial
`malignancies are urgently needed to reduce the worldwide mor-
`bidity and mortality caused by cancer. MicroRNAs (miRNAs) are
`small (⬇22 nt) regulatory RNAs that are frequently dysregulated in
`cancer and have shown promise as tissue-based markers for cancer
`classification and prognostication. We show here that miRNAs are
`present in human plasma in a remarkably stable form that is
`protected from endogenous RNase activity. miRNAs originating
`from human prostate cancer xenografts enter the circulation, are
`readily measured in plasma, and can robustly distinguish xe-
`nografted mice from controls. This concept extends to cancer in
`humans, where serum levels of miR-141 (a miRNA expressed in
`prostate cancer) can distinguish patients with prostate cancer from
`healthy controls. Our results establish the measurement of tumor-
`derived miRNAs in serum or plasma as an important approach for
`the blood-based detection of human cancer.
`
`biomarker 兩 miR-141 兩 plasma 兩 serum 兩 prostate cancer
`
`The development of minimally invasive tests for the detection and
`
`monitoring of common epithelial malignancies could greatly
`reduce the worldwide health burden of cancer (1). Although
`conventional strategies for blood-based biomarker discovery (e.g.,
`using proteomic technologies) have shown promise, the develop-
`ment of clinically validated cancer detection markers remains an
`unmet challenge for many common human cancers (2). New
`approaches that can complement and improve on current strategies
`for cancer detection are urgently needed.
`MicroRNAs (miRNAs) are small (typically ⬇22 nt in size)
`regulatory RNA molecules that function to modulate the activity of
`specific mRNA targets and play important roles in a wide range of
`physiologic and pathologic processes (3, 4). We hypothesized that
`miRNAs could be an ideal class of blood-based biomarkers for
`cancer detection because: (i) miRNA expression is frequently
`dysregulated in cancer (5, 6), (ii) expression patterns of miRNAs in
`human cancer appear to be tissue-specific (7), and (iii) miRNAs
`have unusually high stability in formalin-fixed tissues (8–10). This
`third point led us to speculate that miRNAs may have exceptional
`stability in plasma and serum as well. We show here that miRNAs
`are in fact present in clinical samples of plasma and serum in a
`remarkably stable form. Furthermore, we establish proof-of-
`principle for blood-based miRNA cancer detection by using both a
`xenograft model system and clinical serum specimens from patients
`with prostate cancer. Our results lay the foundation for the devel-
`opment of miRNAs as a novel class of blood-based cancer biomar-
`kers and raise provocative questions regarding the mechanism of
`stability and potential biological function of circulating miRNAs.
`
`Results
`Identification and Molecular Cloning of Endogenous miRNAs from
`Human Plasma. Prior reports have suggested that RNA from human
`plasma (the noncellular component of blood remaining after
`
`removing cells by centrifugation) is largely of low molecular weight
`(11). We directly confirmed that human plasma contains small
`RNAs in the size range of miRNAs (18–24 nt) by characterizing the
`size of total RNA isolated from plasma by using radioactive
`labeling. PAGE and phosphorimaging of 5⬘ 32P-labeled plasma
`RNA demonstrated RNA species ranging from 10 to 70 nt in size,
`including a discernable species of size ⬇22 nt characteristic of most
`miRNAs [supporting information (SI) Fig. S1]. The detected signal
`was sensitive to RNase treatment but insensitive to DNase I
`treatment, confirming that the signal originated from RNA
`(Fig. S1).
`To directly determine whether miRNAs are present in human
`plasma, we isolated the 18- to 24-nt RNA fraction from a human
`plasma sample from a healthy donor (see SI Text for details on
`blood collection and plasma RNA isolation) and used 5⬘ and 3⬘
`RNA–RNA linker ligations followed by RT-PCR amplification to
`generate a small RNA cDNA library (Fig. 1A). Of the 125 clones
`sequenced from this library, 27 corresponded to spiked-in size
`marker oligos or linker–linker dimers. Ninety-one of the other 98
`sequences (93%) corresponded to known miRNAs, providing
`direct confirmation that mature miRNAs are present in human
`plasma and indicating that the vast majority of 18- to 24-nt plasma
`RNA species cloned by our protocol are indeed miRNAs (Fig. 1A).
`To quantitate specific miRNAs, we used TaqMan quantitative
`RT-PCR (qRT-PCR) assays (12) to measure three miRNAs (miR-
`15b, miR-16, and miR-24) in plasma from three healthy individuals.
`These three miRNAs, chosen to represent moderate- to low-
`abundance plasma miRNAs (based on the sequencing results
`described above), were all readily detected in the plasma of each
`individual at concentrations ranging from 8,910 copies/␮l plasma to
`133,970 copies/␮l plasma, depending on the miRNA examined
`(Fig. 1B).
`
`Stability of Endogenous miRNAs in Human Plasma. We next sought to
`investigate the stability of miRNAs in plasma, given that this is an
`important prerequisite for utility as a biomarker. Incubation of
`plasma at room temperature for up to 24 h (Fig. 2A Upper) or
`subjecting it to up to eight cycles of freeze-thawing (Fig. 2 A Lower)
`had minimal effect on levels of miR-15b, miR-16, ormiR-24 as
`
`Author contributions: D.W.L., C.W.D., D.L.S., R.L.V., P.S.N., D.B.M., and M.T. designed
`research; P.S.M., R.K.P., E.M.K., B.R.F., S.K.W., E.L.P.-A., A.P., J.N., K.C.O., and A.A. per-
`formed research; N.U., B.S.K., D.L.S., and R.L.V. contributed new reagents/analytic tools;
`P.S.M., R.K.P., E.M.K., B.R.F., S.K.W., R.G., and M.T. analyzed data; and P.S.M., R.K.P., E.M.K.,
`B.R.F., and M.T. wrote the paper.
`
`The authors declare no conflict of interest.
`‡P.S.M., R.K.P, and E.M.K contributed equally to this work.
`§Present address: Illumina, Inc., 9885 Towne Centre Drive, San Diego, CA 92121.
`¶¶To whom correspondence should be addressed. E-mail: mtewari@fhcrc.org.
`
`This article contains supporting information online at www.pnas.org/cgi/content/full/
`0804549105/DCSupplemental.
`
`© 2008 by The National Academy of Sciences of the USA
`
`www.pnas.org兾cgi兾doi兾10.1073兾pnas.0804549105
`
`PNAS 兩
`
`July 29, 2008 兩 vol. 105 兩 no. 30 兩 10513–10518
`
`Downloaded by guest on February 15, 2022
`
`00001
`
`EX1056
`
`

`

`Identification of miRNAs in human
`Fig. 1.
`plasma. (A) Cloning and sequencing of miRNAs
`from human plasma. The schematic diagram
`depicts the preparation of a small RNA library
`from human plasma. Briefly, the 18- to 24-nt
`fraction from ⬇250 ng of plasma total RNA from
`a single donor (individual 006; described in Ta-
`ble S6) was isolated by PAGE. Purified miRNAs
`were then 3⬘ and 5⬘ligated to single-stranded
`oligonucleotides that contained universal
`primer sequences for reverse transcription and
`PCR. Reverse transcription and PCR generated a
`library of small RNA cDNA molecules that were
`ligated into a plasmid vector (pCR4-TOPO) and
`transformed into Escherichia coli. Inserts from a
`total of 125 individual colonies yielded high-
`quality sequence. Sequences were compared to
`a reference database of known miRNA se-
`quences (miRBase Release v.10.1) (19) and to
`GenBank. Seventy-three percent of sequences
`corresponded to known miRNAs as shown. The
`next most abundant species were matches to
`the sequence of synthetic RNAs spiked in as
`radiolabeled 18- and 24-nt molecular size mark-
`ers during gel isolation steps. When only endog-
`enously derived RNA sequences are considered,
`miRNAs represent 93% (91 of 98) of the recov-
`ered sequences. The miRNA read designated as
`‘‘let-7f G15A’’ denotes a sequence matching the
`known let-7f miRNA except for a G-to-A substi-
`tution at nucleotide position 15. (B) Quantifica-
`tion of representative miRNAs in normal human
`plasma by TaqMan qRT-PCR. The graph indi-
`cates the number of copies of each of three
`representative miRNAs measured in plasma ob-
`tained from three healthy individuals. In each
`case, values represent the average of two repli-
`cate reverse transcription reactions followed by
`real-time PCR. For each miRNA assay, a dilution
`series of chemically synthesized miRNA was used
`to generate a standard curve that permitted
`absolute quantification of molecules of
`miRNA/␮l plasma as shown here (see Fig. S2 for
`standard curve plots). Values were median-
`normalized by using measurements of synthetic
`normalization controls spiked in immediately after addition of denaturing solution during RNA isolation (see SI Text for full details). The absence of amplification in
`reverse transcriptase-negative controls indicated that amplification was originating from an RNA template (real-time PCR plots corresponding to plasma RNA samples
`and negative controls are provided in Fig. S3).
`
`measured by TaqMan qRT-PCR. Given that plasma has been
`reported to contain high levels of RNase activity (13), we sought to
`determine whether the stability of miRNAs is intrinsic to their small
`size or chemical structure, or whether it is caused by additional
`extrinsic factors. We introduced synthetic miRNAs corresponding
`to three known Caenorhabditis elegans miRNAs (cel-miR-39, cel-
`miR-54, and cel-miR-238), chosen because of the absence of ho-
`mologous sequences in humans, into human plasma either before
`or after the addition of a denaturing solution that inhibits RNase
`activity. RNA extraction followed by measurement of synthetic
`miRNAs showed that the synthetic miRNAs rapidly degraded when
`added directly to plasma (the time between addition of synthetic
`miRNAs and subsequent addition of denaturing solution was ⬍2
`min), as compared with their addition after adding denaturing
`solution to plasma (Fig. 2B). These results confirm the presence of
`RNase activity in plasma and the sensitivity of naked miRNAs to
`degradation. The levels of endogenous miRNAs (i.e., miR-15b,
`miR-16, and miR-24) were not significantly altered in any of the
`experimental samples (Fig. 2B), indicating that endogenous plasma
`miRNAs exist in a form that is resistant to plasma RNase activity.
`
`Comparison of miRNA Levels Between Plasma and Serum. Given that
`clinical specimens of serum (the supernatant remaining after whole
`
`blood is permitted to clot) are more plentiful than plasma samples
`in many retrospective clinical sample repositories, we sought to
`determine whether miRNA measurements are substantially differ-
`ent in serum compared with plasma by measuring miR-15b, miR-16,
`miR-19b, and miR-24 in matched samples of serum or plasma
`collected from a given individual at the same blood draw (Fig. 2C).
`Measurements obtained from plasma or serum were strongly
`correlated, indicating that both serum and plasma samples will be
`suitable for investigations of miRNAs as blood-based biomarkers.
`
`Tumor-Derived miRNAs Are Present in Plasma. Having demonstrated
`that circulating miRNAs are detectable and stable in blood col-
`lected from healthy individuals, we next sought to determine
`whether tumor-derived miRNAs enter the circulation at levels
`sufficient to be measurable as biomarkers for cancer detection. We
`chose to study a mouse prostate cancer xenograft model system that
`involves growth of the 22Rv1 human prostate cancer cell line in
`NOD/SCID immunocompromised mice (14–17). We established a
`cohort of 12 mice xenografted with 22Rv1 cells injected with
`Matrigel and 12 control mice inoculated with Matrigel alone (Fig.
`3A). Plasma was collected 28 days later (once tumors were well
`established), and RNA was isolated for miRNA quantitation.
`Because of the lack of an established endogenous miRNA control
`
`10514 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0804549105
`
`Mitchell et al.
`
`Downloaded by guest on February 15, 2022
`
`00002
`
`

`

`Characterization of miRNA stability in human plasma. (A) miRNA levels remain stable when plasma is subjected to prolonged room temperature
`Fig. 2.
`incubation or freeze-thawed multiple times. (Upper) The graphs show normalized Ct values for the indicated miRNAs measured in parallel aliquots of human
`plasma samples incubated at room temperature for the indicated times. The experiment was carried out by using plasma from the two different individuals
`noted. Normalization of raw Ct values across samples is based on the measurement of three nonhuman synthetic miRNAs spiked into each sample at known molar
`amounts after initial plasma denaturation for RNA isolation (described in detail in SI Text). (Lower) The graphs show normalized Ct values for the indicated
`miRNAs measured in parallel aliquots of human plasma samples subjected to the indicated number of cycles of freeze-thawing. Raw Ct values were normalized
`across samples by using the same approach as described above. (B) Exogenously added miRNAs are rapidly degraded in plasma, whereas endogenous miRNAs
`are stable. Three C. elegans miRNAs (chosen for the absence of sequence similarity to human miRNAs) were chemically synthesized and added either directly
`to human plasma (from individual 003; described in Table S6) or added after the addition of denaturing solution (containing RNase inhibitors) to the plasma
`(referred to as ‘‘denatured plasma’’). RNA was isolated from both plasma samples, and the abundance of each of the three C. elegans miRNAs was measured
`by TaqMan qRT-PCR (Left), as was that of three endogenous plasma miRNAs (Right). Asterisks indicate that the abundance ratios of cel-miR-39, cel-miR-54, and
`cel-miR-238 added to human plasma directly, relative to addition to denatured plasma, were 1.7 ⫻ 10⫺5, 9.1 ⫻ 10⫺6, and 1.1 ⫻ 10⫺5, respectively and therefore
`too low to accurately display on the plot. (C) Abundance of miRNAs in serum and plasma collected from the same individual is highly correlated. Each plot depicts
`the average Ct values (average of two technical replicates) of the indicated miRNAs measured in serum and plasma samples collected from a given individual
`at the same blood draw. Results from three different individuals are shown. miRNA measurements were highly correlated in both sample types. Results shown
`for synthetic C. elegans miRNAs spiked into each plasma or serum sample (after addition of denaturing solution) demonstrate that experimental recovery of
`miRNAs and robustness of subsequent qRT-PCR is not affected by whether it is plasma or serum that is collected.
`
`MEDICALSCIENCES
`
`for plasma or serum, we introduced three synthetic C. elegans
`miRNAs (described earlier) after the addition of denaturing solu-
`tion to the plasma samples and used these as normalization controls
`to correct for technical variations in RNA recovery (detailed in SI
`Text).
`We first sought to establish that endogenous (murine) miRNAs
`exist in mouse plasma and determine whether the presence of
`cancer may lead to a general increase in plasma miRNAs, whether
`they be tumor- or host-derived. miR-15b, miR-16, and miR-24
`(which are perfectly conserved in mature sequence between human
`and mouse) were all readily detectable in healthy control mice (Fig.
`3B) and were not expressed at substantially different levels in
`xenograft-bearing mice, indicating that the presence of tumor does
`not lead to a generalized increase in plasma miRNAs.
`We next sought to identify candidate tumor-derived miRNAs for
`examination in plasma by first profiling the expression of 365 known
`miRNAs in 22Rv1 cells by using a microfluidic TaqMan low-density
`miRNA qRT-PCR array (Applied Biosystems). We identified
`two miRNAs, miR-629* and miR-660, that (i) were expressed in
`these cells as indicated by low cycle threshold (Ct) values and (ii) did
`not have known mouse homologs and therefore would be expected
`to be tumor-specific markers in this setting (Table S1). We next
`analyzed plasma samples from control and xenograft mice for the
`levels of miR-629* and miR-660 by TaqMan qRT-PCR. Levels of
`miR-629* and miR-660 were generally undetectable in the control
`mice, whereas they were readily detected (ranging from 10 to 1,780
`copies/␮l plasma for miR-629* and 5,189–90,783 copies/␮l for
`miR-660) in all of the xenografted mice (Fig. 3C). Levels of both
`
`miR-629* and miR-660 were able to independently differentiate
`xenografted mice from controls with 100% sensitivity and 100%
`specificity. These data establish proof of the principle that tumor-
`derived miRNAs reach the circulation where their measurement in
`plasma can serve as a means for cancer detection.
`To understand the basis for the wide variation in miRNA
`abundance observed among the different xenografted mice, we
`compared plasma levels of miR-629* and miR-660 with tumor mass
`in each mouse. Levels of these miRNAs were moderately correlated
`with tumor mass (Fig. S4 and see Table S7), indicating that variation
`in miRNA abundance across animals reflects, at least in part, the
`differences in tumor burden.
`
`Tumor-Derived miRNAs in Plasma Are Not Cell-Associated. We sought
`to further explore the mechanism of protection of tumor-derived
`miRNAs from plasma RNase activity by testing the hypothesis that
`they are present inside circulating tumor cells that might have
`escaped pelleting during primary centrifugation for plasma isola-
`tion. We took two approaches to address this hypothesis. In the first
`experiment, we filtered pooled plasma generated from the xeno-
`graft or control groups through a 0.22-␮m filter, followed by RNA
`extraction from the filtrate and the material retained on the filter
`(referred to as the retentate). Measurement of miR-629* and
`miR-660 by qRT-PCR in each of the samples demonstrated that
`virtually all of the tumor-derived miRNAs passed through the
`0.22-␮m filter (Fig. S5A). As expected, tumor-derived miRNAs
`were essentially undetectable in all of the samples from the control
`group. In a second independent experiment, we subjected plasma
`
`Mitchell et al.
`
`PNAS 兩
`
`July 29, 2008 兩 vol. 105 兩 no. 30 兩 10515
`
`Downloaded by guest on February 15, 2022
`
`00003
`
`

`

`Tumor-derived miRNAs are detectable in plasma. (A) Schema for 22Rv1 human prostate cancer xenograft experiment. (B) MiRNAs are present in plasma
`Fig. 3.
`of healthy control mice and their levels are not nonspecifically altered in cancer-bearing mice. Plasma levels of miR-15b, miR-16, and miR-24 were measured in
`12 healthy control mice and 12 xenograft-bearing mice. The mature sequence of these miRNAs is perfectly conserved between mice and humans. Ct values were
`converted to absolute number of copies/␮l plasma by using a dilution series of known input quantities of synthetic target miRNA run on the same plate as the
`experimental samples (dilution curves are provided in Fig. S2). Values shown have been normalized by using measurements of C. elegans synthetic miRNA controls
`spiked into plasma after denaturation for RNA isolation (details of the normalization method are provided in SI Text). (C) Tumor-derived miRNAs are detected
`in plasma of xenograft-bearing mice and can distinguish cancer-bearing mice from controls. Plasma levels of miR-629* and miR-660 (two human miRNAs that
`are expressed in 22Rv1 cells and do not have known murine homologs) were measured in all control and xenografted mice. Ct values were converted to absolute
`number of copies/␮l plasma and normalized as described for B (see Table 5) threshold. Given that homologous miRNAs are not believed to exist in mice, the low
`level of signal detected for a few mice in the control group, particularly for the miR-660 assay, is likely to represent nonspecific background amplification. As
`expected, in the control (nontumor-bearing) mice group, qRT-PCR for miR-629* ormiR-660 in plasma from most animals could not detect any appreciable signal.
`These points are therefore not shown on the graph, even though plasma samples from the entire group of 12 mice in the control group were studied.
`
`pools from the xenograft and control groups to a series of two
`centrifugations (one at 2,000 ⫻ g, which should pellet any intact cells
`remaining after the initial centrifugation used to collect plasma,
`followed by another at 12,000 ⫻ g, which should pellet any large cell
`fragments). We assayed for miRNA expression in the starting
`material, any pelleted material obtained from each centrifugation,
`and the supernatant remaining after the 12,000 ⫻ g centrifugation.
`As shown in Fig. S5B, virtually all of the tumor-derived miRNA was
`present in the supernatant of the 12,000 ⫻ g spin. Taken together,
`the data indicate that tumor-derived miRNAs are not associated
`with intact cells or large cell fragments. These results do not exclude
`the possibility that circulating tumor cells or fragments of the same
`may have been lysed during the process of blood collection or
`plasma processing. Even if that is the case, however, our results
`show that miRNAs that may have been released are ultimately
`present in a stable, protected form of size much smaller than that
`of a typical epithelial cell.
`
`Detection of Human Prostate Cancer Based on Measurement of a
`Prostate Cancer-Expressed miRNA in Serum. We next sought to
`extend this approach to cancer detection in humans. We reasoned
`that an ideal marker would be (i) expressed by the cancer cells at
`moderate or high levels and (ii) present at very low or undetectable
`levels in plasma from healthy individuals. We established a list of
`likely blood-based miRNA biomarker candidates for prostate can-
`
`cer by (i) compiling a list of miRNAs expressed in human prostate
`cancer specimens based on published miRNA expression profiling
`data (7, 18) and (ii) filtering out miRNAs detected in healthy
`donor-derived plasma in our miRNA cloning experiment (Fig. 1A)
`or detected on a microfluidic TaqMan qRT-PCR array analysis of
`plasma from a normal healthy individual (details provided in SI
`Text and see Table S8). This process generated a list of six leading
`candidates (miR-100, miR-125b, miR-141, miR-143, miR-205, and
`miR-296) for further investigation.
`We chose to analyze these candidates in a case-control cohort of
`serum samples collected from 25 individuals with metastatic pros-
`tate cancer and 25 healthy age-matched male control individuals.
`To efficiently screen multiple miRNA biomarker candidates, we
`first generated two pools of serum aliquots derived from the
`individuals in the case and control groups, respectively. We isolated
`RNA from both pools and screened them for differential expression
`of the six candidate biomarker miRNAs by TaqMan qRT-PCR
`assays. Results of this screen indicated that five of six of these
`candidate miRNA biomarkers showed increased expression, al-
`though to varying degrees, in the prostate cancer serum pool
`compared with the healthy control group serum pool (Table S2).
`For one of the candidates (miR-205), no conclusion could be
`reached because miRNA levels in both pools were lower than the
`limit of detection of the assay (as determined by a standard curve
`using a dilution series of a synthetic miR-205 RNA oligonucleotide).
`
`10516 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0804549105
`
`Mitchell et al.
`
`Downloaded by guest on February 15, 2022
`
`00004
`
`

`

`MEDICALSCIENCES
`
`a Receiver Operating Characteristic plot (Fig. 4B) reflects strong
`separation between the two groups, with an area under the curve
`(AUC) of 0.907. Comparison of miR-141 levels to prostate-specific
`antigen (PSA) values among the prostate cancer patients demon-
`strated Pearson and Spearman (rank) correlation coefficients of
`⫹0.85 and ⫹0.62, indicating that miR-141 and PSA levels are
`moderately correlated (Table S3). Serum levels of nonbiomarker
`candidate miRNAs miR-16, miR-19b, and miR-24 were not signif-
`icantly different between cases and controls, supporting the notion
`that miR-141 is specifically elevated in prostate cancer, as opposed
`to reflecting a nonspecific, generalized increase in serum miRNA
`levels in the setting of cancer (Fig. 4C). Taken together, the results
`extend to human cancer the concept that circulating miRNAs can
`serve as markers for cancer detection.
`
`miR-141 Is an Epithelial-Associated miRNA Expressed by Several
`Common Human Cancers. miR-141 is a member of an evolution-
`arily conserved family of miRNAs that includes, in humans,
`miR-141, miR-200a, miR-200b, miR-200c, and miR-429 (19).
`The expression of zebrafish homologs of this family, when
`studied by in situ hybridization, was found to localize to various
`epithelial tissues (20). To gain more insight into the potential
`biological role of miR-141, we explored the large miRNA
`expression profiling dataset generated by Lu et al. (7), who
`profiled a diverse range of human cancer types. Consistent
`with findings from the zebrafish studies, the expression of
`miR-141 was tightly associated with expression in epithelial
`samples compared with nonepithelial samples (Fig. S6), and
`miR-141 was expressed in a wide range of common epithelial
`cancers including breast, lung, colon, and prostate.
`To determine the relative expression of this miR-141 specifically
`between the epithelial and stromal compartments of prostate tissue,
`both comparatively between the two cell types and relative to all
`other known miRNAs within a cell type, we generated small RNA
`libraries from primary cultures of human prostate epithelial and
`stromal cells and subjected them to massively parallel sequencing
`(detailed in SI Text). We found that miR-141 (and two of its family
`members, miR-200b and miR-200c) was readily detected in the
`prostate epithelial cell dataset but strikingly absent in the prostate
`stromal cells (Table S4). In fact, of all miRNAs in this analysis,
`miR-141 and miR-200b were the two most overexpressed in prostate
`epithelial cells relative to prostate stromal cells (Table S4). Taken
`together, the data are consistent with the notion that miR-141 is an
`epithelial-restricted miRNA that can be detected in the circulation
`as a prostate cancer biomarker.
`
`Discussion
`Our Results Establish That Tumor-Derived miRNAs, Detected in Plasma
`or Serum, Can Serve as Circulating Biomarkers for Detection of a
`Common Human Cancer Type. Although there is a long history of
`investigation of circulating mRNA molecules as potential biomar-
`kers (21), blood-based miRNA studies are in their infancy. Re-
`cently, Chim et al. (22) reported the detection by qRT-PCR of
`miRNAs of presumed placental origin in the plasma of pregnant
`women, and Lawrie et al. (23) reported detecting elevations in
`miRNAs in serum from lymphoma patients. Beyond confirming the
`early reports, our study yielded (i) a more comprehensive view of
`plasma miRNAs by direct cloning and sequencing from a plasma
`small RNA library, (ii) unique results on miRNA stability that
`provide a firm grounding for further investigation of this class of
`molecules as blood-based cancer biomarkers, and (iii) evidence that
`tumor-derived miRNAs can enter the circulation even when orig-
`inating from an epithelial cancer type (as compared with hemato-
`poietic malignancies like lymphoma). Most importantly, our study
`of miR-141 in prostate cancer patients demonstrates that serum
`levels of a tumor-expressed miRNA can distinguish, with significant
`specificity and sensitivity, patients with cancer from healthy
`controls.
`
`Detection of human prostate cancer by serum levels of tumor-
`Fig. 4.
`associated miRNA miR-141. (A) Serum levels of miR-141 discriminate patients
`with advanced prostate cancer from healthy controls. Serum levels of the
`prostate cancer-expressed miRNA miR-141 were measured in 25 healthy con-
`trol men and 25 patients with metastatic prostate cancer (clinical data on
`subjects is provided in Table S3). Ct values were converted to absolute number
`of copies/␮l serum by using a dilution series of known input quantities of
`synthetic target miRNA run simultaneously (on the same plate) as the exper-
`imental samples (dilution curves are provided in Fig. S2). Values shown have
`been normalized by using measurements of C. elegans synthetic miRNA
`controls spiked into plasma after denaturation for RNA isolation (details of
`normalization method are provided in SI Text). The dashed line indicates a
`100% specificity threshold. (B) Receiver Operating Characteristic (ROC) plot.
`The data shown in A were used to draw the ROC plot shown. (C) Serum levels
`of nontumor-associated miRNAs are not substantially different between pa-
`tients with prostate cancer and controls. Serum levels of miR-16, miR-24, and
`miR-19b were measured as negative controls as they are not expected to be
`cancer-associated in the serum. Absolute quantification of miRNAs and data
`normalization were carried out as described for A.
`
`Of all of the candidates, miR-141 showed the greatest differential
`expression (46-fold overexpressed) in the prostate cancer pool
`compared with the control pool (Table S2). We therefore focused
`our study on miR-141 by measuring the abundance of this miRNA
`in all of the individual serum samples comprising the case and
`control groups. Consistent with results from the analysis of pooled
`samples, serum levels of miR-141 were, in general, substantially
`higher in cancer cases compared with controls (Fig. 4A). Compar-
`ison of the two groups by a Wilcoxon two-sample test yielded W ⫽
`63 with a P ⫽ 1.47 ⫻ 10⫺7, confirming a significant difference in
`miR-141 levels between the two groups. Furthermore, serum levels
`of miR-141 could detect individuals with cancer with 60% sensi-
`tivity at 100% specificity (Fig. 4A). Representation of the data using
`
`Mitchell et al.
`
`PNAS 兩
`
`July 29, 2008 兩 vol. 105 兩 no. 30 兩 10517
`
`Downloaded by guest on February 15, 2022
`
`00005
`
`

`

`The available data indicate that miR-141 is expressed in an
`epithelial cell type-specific manner in a range of common human
`cancers. Given this, we speculate that it could have value in the
`setting of detecting cancer recurrence for cancer types for which
`clinically validated blood biomarkers are lacking (e.g., lung cancer,
`breast cancer, etc.). We also anticipate that advances in miRNA
`qRT-PCR assay design and assay optimization, and the application
`of alternative miRNA quantitation strategies, will substantially
`improve the approach and will likely be needed to detect cancer at
`lower tumor burdens (i.e., early-stage disease). It is likely that other
`blood-based miRNA markers that are specific for particular cancer
`types will be discovered. The results presented here establish the
`foundation and rationale to motivate future global investigations of
`miRNAs as circulating cancer biomarkers for a variety of common
`human cancers.
`
`and quantitative assay development for validation of biomarker
`candidates (2). In addition, the inherent regulatory function of
`miRNAs makes it likely that many miRNAs expressed in tumor
`tissue influence the biological behavior and clinical phenotype of
`the tumor. As the functional roles of miRNAs in tumor biology are
`unraveled, we envision that blood-based miRNA biomarkers that
`predict clinical behavior and/or therapeutic response will be
`identified.
`
`Materials and Methods
`Clinical Samples. Human plasma and serum samples from healthy donors or
`patients with cancer were obtained with informed consent under institutional
`review board-approved protocols. Samples were derived from the Pacific Ovarian
`Cancer Research Consortium, local healthy donors from the Seattle area, or the
`Department of Urology, University of Washington. Details of sample collection
`and processing and relevant corresponding clinical data are provided in SI Text.
`
`Our Study Raises Intriguing Questions Regarding the Mechanism of
`miRNA Stability in Plasma and Potential Biological Roles of Circulating
`miRNAs. The remarkable stability of miRNAs in clinical plasma
`samples raises important and intriguing questions regarding the
`mechanism by which miRNAs are protected from endogenous
`RNase activity. One tantalizing hypothesis is that they are packaged
`inside exosomes that are secreted from cells. Exosomes are 50- to
`90-nm (24), membrane-bound particles that have been reported to
`be abundant in plasma (25) and that