Multilayer microfluidic array for highly efficient sample loading and digital melt analysis of DNA methylation

Christine M. O'Keefe a, Daniel Giammanco b, Sixuan Li c, Thomas R. Pisanic II d and Tza-Huei Jeff Wang *acd
aDepartment of Biomedical Engineering, Johns Hopkins School of Medicine, Baltimore, MD 21218, USA
bDepartment of Chemical and Biomolecular Engineering, Johns Hopkins University, Baltimore, MD 21218, USA
cDepartment of Mechanical Engineering, Johns Hopkins University, Baltimore, MD 21218, USA. E-mail: thwang@jhu.edu
dJohns Hopkins Institute for NanoBioTechnology, Baltimore, MD 21218, USA

Received 31st October 2018 , Accepted 28th December 2018

First published on 9th January 2019


Liquid biopsies contain a treasure of genetic and epigenetic biomarkers that contain information for the detection and monitoring of human disease. DNA methylation is an epigenetic modification that is critical to determining cellular phenotype and often becomes altered in many disease states. In cancer, aberrant DNA methylation contributes to carcinogenesis and can profoundly affect tumor evolution, metastatic potential, and resistance to therapeutic intervention. However, current technologies are not well-suited for quantitative assessment of DNA methylation heterogeneity, especially in challenging samples such as liquid biopsies with low DNA input and high background. We present a multilayer microfluidic device for quantitative analysis of DNA methylation by digital PCR and high resolution melt (HRM). The multilayer design facilitates high-density array digitization aimed at maximizing sample loading efficiency. The platform achieves highly parallelized digital PCR-HRM-based discrimination of rare heterogeneous DNA methylation as low as 0.0001% methylated/unmethylated molecules of a classic tumor suppressor gene, CDKN2A (p14ARF).


Introduction

The discovery that disease-associated cell-free DNA (cfDNA) can be readily found in patient plasma has opened new avenues for molecular diagnostics and prognostics in many diseases.1–4 Liquid biopsies provide several advantages over traditional techniques by offering sampling that is noninvasive, heterogeneous, and unbiased, prompting increasingly more research into its clinical utility as a diagnostic and theranostic platform.5–9 Perhaps the most prominent applications of liquid biopsies are aimed at early detection and therapeutic monitoring of cancer, whereby liquid biopsies can provide a simple, noninvasive means of sampling DNA derived from all throughout the body.10–12 In fact, several studies have shown that genetic and epigenetic molecular aberrations, such as mutations, copy number variations and DNA methylation correlate with tumor evolution and can likewise be found in cfDNA in the plasma.13–16

The most well-studied epigenetic modification is DNA methylation, which is often aberrant in cancer and can result in dysregulation of gene expression.17–19 Many tumors exhibit hypermethylation-induced silencing of tumor suppressor genes (TSGs) early in tumorigenesis, prompting the investigation of the use of DNA hypermethylation as a biomarker for early detection of many types of cancer.13,20,21

DNA methylation occurs as a stochastic process, and can thus vary on a patient-by-patient and even cell-by-cell basis.22–24 The precise effects of this variability and heterogeneity are only beginning to be elucidated. Nonetheless, very recent studies have observed that intermediate DNA methylation heterogeneity is predictive of metastatic versus localized clones in Ewing sarcoma,25 and that recurrent methylation reprogramming of numerous CpG sites across the epigenome in acute myeloid leukemia patients occurs throughout the progression of disease.26 However, tumor-derived molecules are rare and exist amongst a high background of healthy cfDNA, often occurring in frequencies as low as 0.01%.10 Consequently, detection of these ultra-rare biomarkers requires extremely sensitive and specific analysis techniques.

While there exist numerous techniques for assessing methylation, most are ill-suited for the assessment of ultra-rare epiallelic fractions. For example, the most commonly employed methods for analyzing locus-specific methylation are based on methylation specific PCR (MSP).27 Although these MSP-based assays can potentially detect methylated fractions as low as 0.01%,28–30 they remain only semi-quantitative and are restricted to a single methylation pattern, making them unsuitable for quantification of DNA methylation heterogeneity.

Digital technologies, such as digital PCR (dPCR) and droplet digital PCR (ddPCR), are capable of quantifying template molecules at very high precision and sensitivity and thus are well-suited for rare molecule detection.31,32 Nonetheless, droplet-based technologies are typically limited to an end-point readout of specific target sequences and consequently they are not amenable to real-time measurements or unknown targets.33 On the other hand, sequencing can provide exact sequence information across the genome, however, its limited sensitivity and high cost render this method impractical to detect rare methylated molecules at fractions below 0.1%.34,35 Therefore, array-based digital technologies are better suited for real-time, multidimensional analysis.

We recently developed a microfluidic digital detection technique called HYPER-melt (high-density profiling and enumeration by melt) for high-throughput assessment of molecular heterogeneity in challenging samples such as liquid biopsies.36 In our previous study, we employed the HYPER-melt platform with our previous assay principle, termed discrimination of rare epialleles by melt (DREAMing),37 to interrogate the methylation status of partially and fully-methylated epialleles on a molecule-by-molecule basis. The HYPER-melt platform enabled absolute quantification and methylation assessment of heterogeneously-methylated epialleles at fractions as low as 1 in 2 million.36 While promising, this method, like most array-based techniques, requires considerably more sample volume than is ultimately analyzed, thereby limiting the number of analyses that can be performed on each sample.38–40

In this study, we present a multilayer microfluidic device for efficient trapping and parallelized DNA methylation analysis of single molecules in picoliter-sized chambers. The multilayer design facilitates efficient digitization of DNA molecules into 13[thin space (1/6-em)]000 wells. We demonstrate the utility of this digital PCR-high resolution melt (HRM) platform through discrimination of partially and fully methylated epialleles of a tumor suppressor gene, CDKN2A (p14ARF) amongst a high background of unmethylated DNA. The ultrahigh sensitivity of this platform provides a means for quantitative assessment of DNA methylation heterogeneity of rare molecules such as those found in liquid biopsies.

Experimental

Device fabrication

Molds were fabricated with SU-8 photoresist (MicroChem) via standard photolithography. Silicon wafers were dehydrated for at least 4 hours at 200 °C, then oxygen-plasma treated at 80 W for 1 minute (Technics PE-IIA). For the well layer, SU-8 3050 was spin-coated at 1800 rpm for 1 min, soft-baked at 95 °C for 27 min, and exposed at 175 J cm−2. For the channel layer, SU-8 3025 was spin-coated at 1800 rpm for 1 min, soft-baked at 95 °C for 14 min, and exposed at 150 J cm−2. Both molds were then baked at 95 °C for 5 min, developed with SU-8 developer (MicroChem), and baked at 200 °C for 1 h.

PDMS devices were fabricated with an adapted ultra-thin soft lithography technique (Fig. S1). PDMS (Ellsworth) at a ratio of 15[thin space (1/6-em)]:[thin space (1/6-em)]1 elastomer base to curing agent was spin-coated on the well and channel pattern layers at 700 and 900 rpm respectively. A sacrificial layer of 6[thin space (1/6-em)]:[thin space (1/6-em)]1 PDMS was spin-coated on a blank wafer at 100 rpm. All were baked for 6 min at 80 °C. The sacrificial layer was then removed, placed on the pattern layer, and temporarily bonded by baking at 80 °C for 6 min. Both joint layers were then removed from the mold. The well layer was bonded to the glass slide by oxygen-plasma bonding at 80 W for 1 min. Next the wells and channel layers were oxygen-plasma treated, aligned, and bonded. Finally, the sacrificial layer was removed, and cover glass and adapters were oxygen-plasma bonded to the top surface.

Modeling and simulation

Two-dimensional multiphase fluid simulations of the sample loading into a single side chamber were conducted with a CFD package (COMSOL Multiphysics® v. 5.2. COMSOL AB, Stockholm, Sweden). The model comprised of a laminar two-phase flow level set and two-phase Darcy's law models. Additional parameters and a detailed model description are given in the ESI (Table S1). The geometric parameters, chamber inlet angle and channel curvature, were varied to assess loading speed of a single well and fluid velocity along the channel (Fig. S2).

Digital PCR and melt

The reaction mix was prepared off-chip to yield final working concentrations of 1.66 mM (NH4)2SO4, 6.7 mM Tris pH 8.8, 2.7 mM MgCl2, 1 mM β-mercaptoethanol, 300 nM primers (IDT), 200 μM dNTPs (ThermoFisher Scientific), 1× ROX (ThermoFisher Scientific), 1.2 U μL−1 Platinum Taq DNA polymerase (ThermoFisher Scientific), 1 mg mL−1 BSA (NEB), 0.01% Tween 20 (Sigma Aldrich), and 1× EvaGreen (Biotium). The device was sealed with adhesive and desiccated for a minimum of 2 hours to produce an internal vacuum. Upon puncturing the inlet, the 8 μL of reaction mix was loaded into the device. Next, a partitioning oil, comprised of 5 g of 100 cst silicon oil and 1 g of 10[thin space (1/6-em)]:[thin space (1/6-em)]1 PDMS was pressurized through the channels to digitize the chambers. The device was placed on a flatbed heater (ThermoFisher Scientific) to undergo dPCR (5 min at 95 °C, followed by 60 cycles of 30 s at 95 °C, 30 s at 61 °C, and 30 s at 72 °C). Digital melt curves were acquired on our previously described digital melt platform,36 in which fluorescent images were captured by a Sony ILCE MILC at 1 s intervals during a thermal ramping at 0.1 K s−1.

Imaging

End-point fluorescent images were captured by a Typhoon Scanner (ge Amersham). Two channels were detected: excitation and emission of 488 nm and 526 nm for the EvaGreen signal as well as 532 nm and 610 nm for the ROX reference signal. Post-processing of the images was performed in Matlab. RGB fluorescent images from melt acquisition were converted to grayscale. A binary mask image was aligned with the reference image to identify the pixel-space definition of each well following our previously outlined procedure.36 Positive wells were identified by thresholding of the EvaGreen signal of each well within its neighborhood.

Results and discussion

Overview

To achieve high-efficiency and high-density methylation profiling, we developed a multilayer microfluidic chip for digitization and interrogation of individual DNA sequences (Fig. 1A). The assay design follows the principles set by our previously developed technique, DREAMing (discrimination of rare epialleles by melt). Briefly, sequences undergo bisulfite conversion to convert unmethylated cytosine residues to uracil, thereby creating a sequence change that is representative of the template methylation status. Next, methylation-preferred primers amplify all epialleles via PCR. To interrogate the methylation density of the amplicons, the device undergoes high resolution melt (HRM), a technique that observes sequence-specific denaturation via an intercalating dye during temperature ramping (Fig. 1B). The point at which half of the amplicon is denatured is termed its “melt temperature.” The methylation density is then determined from melt temperature of each amplicon, producing a methylation heterogeneity profile of the sample (Fig. 1C).
image file: c8lc01189c-f1.tif
Fig. 1 (A) Heterogeneously methylated epialleles are loaded into the device and digitized. (B) All epiallelic patterns are amplified and the amplicon in each positive well is discriminated by melt temperature. (C) The melt temperature is used to calculate the methylation density of the template. Aggregating melt temperatures from all positive wells provides a molecule-by-molecule methylation profile of the sample.

The device consists of two PDMS pattern layers oriented such that the features are in contact. The lower PDMS layer, containing 13[thin space (1/6-em)]000 750 pL-sized chambers or wells, sits below the channel layer, overlapping at the inlets to each chamber (Fig. 2A). Two glass slides sandwich the device, and PDMS adapters provide the loading interface (Fig. 2B). The reaction mix rapidly enters the desiccated device through the channels and fills the wells (Fig. 2C). Next, a partitioning oil flows through the evacuated channels, thus isolating and digitizing the individual molecules.


image file: c8lc01189c-f2.tif
Fig. 2 (A) The device consists of 2 PDMS layers sandwiched between glass slides. The curved channel sits above the wells. (B) The device is fully loaded with dye from a single inlet and semi-circle bifurcation. (C) Reagents are loaded into a vacuumed device, filling the wells. Oil is pressurized through the channel for partitioning and to eject excess air.

Multilayer device architecture

Most array-based technologies suffer from sample loss in loading or digitization.38–40 Previously, our device design comprised of parallelized straight channels with perpendicular inlet orientation.36 During vacuum-assisted loading, the fluid rapidly filled the channels before saturating the wells, disqualifying approximately 89% of the sample for analysis. To minimize loss due to channel bypassing, we sought to implement a design that retards flow through the channels while maintaining or increasing the fill velocity of each well.

The multilayer design promotes efficient loading through two primary mechanisms. First, the curved channel design directs fluid to flow towards the walls of the channels where the inclined inlets for the chambers are located. The presented curved, single-channel design constantly redirects flow, increasing flow resistance in the direction of the channel. Second, the ratio of the well height (hw) to the channel height (hc) can be tuned to adjust the relative force of the vacuum in the direction of the wells (Fvacuum,w) with respect to the channel (Fvacuum,c).

Modeling and simulation of 2-dimensional device geometries

To model the effects of the channel and inlet device geometries on the flow profile, two-dimensional multiphase fluid simulations of the sample loading into different channel and inlet geometries were conducted with COMSOL Multiphysics. First, we compared the curved channel to a straight channel under conditions simulating vacuum-assisted loading. Initially, the inlet of the main channel was filled with water (aqueous phase) and the rest of the geometry was occupied by low pressure air (gas phase). The modeled curved channel is a 180° semi-circle. Both channels were 10 mm in overall length and 0.1 mm wide.

The simulation results showed an approximately 5-fold retardation in travel through the curved channel with respect to the straight channel over the 10 mm path (Fig. S2A). Furthermore, the curved channel demonstrates a more parabolic velocity profile than the straight channel (Fig. S2B). Although the geometry was simplified for the purpose of the model, the relationship is expected to be maintained when scaled to larger volumes. This result matches our experimental observation that curved channels experience slower flow rates.

Next, we simulated the aqueous sample loading into the chambers at both perpendicular and inclined inlet geometries (Fig. S2C and D). According to the model, sample loading from the channel into the chamber took 0.27 seconds and 0.17 seconds for 90° and 45° inlets, respectively. The relative loading speeds match our observation that an angled inlet loaded faster than a perpendicular inlet. However, in the on-chip experiment we observed that loading could take as long as 5–15 seconds. This deviation can be attributed to the driving force variation in the sample digitization process. Specifically, absolute pressure within the air phase of the chip increases as the sample loads, resulting in a reduction in the pressure differential driving the liquid as it moves from the sample inlet to outlet. Thus, it takes a shorter time to fill a chamber close to the inlet when compared to those located near the outlet. Nonetheless, the essential features of the sample loading and digitization presented by the simulation were in good agreement with our experimental observations.

Loading efficiency in 2D vs. 3D device

Experimentally, we sought to quantitatively assess the loading efficiency of the multilayer curved channel device (hw = 2 × hc) as compared to the identical pattern in a single layer (hw = hc). Images were acquired of each device during loading. A single, fixed well was chosen for the time course study, beginning at the point the liquid first passes the well inlet. At each subsequent time point, the filled area of the well was compared to the volume of fluid that had continued along the channel, which we termed the “waste volume” (Fig. 3A). The single layer device had a waste volume of 0.93 μL, 12% of the input volume (8 μL), whereas the multilayer produced only 0.34 μL, 4% (Fig. 3B). Furthermore, the multilayer device filled each well ∼3 times faster than the single layer, resulting in significantly higher loading efficiency of the sample.
image file: c8lc01189c-f3.tif
Fig. 3 (A) Continuous images were acquired during loading of dye into a single layer and multilayer device. A timer starts when the dye first passes the single observation well, and ends when the well is fully filled. (B) The fill rate, time to fill a single 750 pL well, is much faster for the multilayer device. The waste volume, volume of liquid that is in the channels past the observation well at the time of fully filled, is lower for the multilayer device than the single layer.

The multilayer architecture also offers several advantages for partitioning and assay performance. The device utilizes a vertical compartmentalization strategy that complements the densities of the entrapped materials, namely the reaction fluid (ρfl ≈ 1 g mL−1), the partitioning oil (100 cst silicone oil [Sigma-Aldrich], ρoil = 0.967 g mL−1), and air (ρair = 0.001 g mL−1). Since ρfl > ρoilρair, the partitioning oil drives any remaining reaction fluid from the channels down into the wells. Air, which notoriously prevents successful high-temperature reactions in enclosed devices, rises up to the channels, and is ejected by the partitioning oil (Fig. S3). The environment is thus robust to amplification via high temperature assays such as PCR.

Reducing PDMS distortion and misalignment

Despite the advantages of the multilayer device, achieving reliable and repeatable fabrication was nontrivial. PDMS distortion with respect to its mold is a well-known challenge in multilayer soft lithography. Several attempts have been made to address the issue, namely by characterizing the distortion ratio,41 developing intricate fabrication techniques,42 and including tolerances into the design.43 The presented multilayer design requires strict alignment of two peeled PDMS layers, which eliminates simple scaling as an option. In order to maintain high efficiency in loading and a high-density array footprint, large error tolerances were also not a viable option. Therefore, we investigated the relationship between distortion and PDMS height.

PDMS patterns of various heights were aligned and compared to the pattern mask using a laser microscope (Keyence VK-X100) (Fig. 4A). For both mold heights, we observed a clear trend that smaller PDMS heights led to more predictable and less overall distortion (Fig. 4B). By reducing the PDMS height to 26 μm, the distortion could be reduced to ≈0.32%, with a standard deviation of 0.015%. To achieve consistent results across layers and thereby minimize misalignment, an ultra-thin soft lithography fabrication technique was implemented (Fig. S1). This ultra-thin technique is simple and obviates the need for any additional materials during fabrication. The evaporation potential is minimized by reducing the volume of porous hydrogel above the wells. Although alignment of the two layers requires a trained user, misalignment can be effectively reduced by minimizing the PDMS distortion from the mold.


image file: c8lc01189c-f4.tif
Fig. 4 (A) The PDMS pattern was aligned against the mask to measure the percent distortion. (B) The distortion was compared at different PDMS thicknesses for molds of two different resist heights.

Highly efficient detection of DNA methylation

To demonstrate the overall detection efficiency of the device, fixed amounts of synthetic DNA representative of fully methylated CDKN2A (p14ARF), a well-established methylation biomarker for various cancer types,44 were digitized on the device. 8 μL of reaction mix were prepared containing a serial dilution of methylated p14ARF from ≈1000 to 100 copies. Fluorescent images were acquired after PCR amplification to quantify the number of detected methylated targets on the multilayer device (Fig. 5A). The absolute copy number detected closely matched the expected amount loaded into the chip for each dilution.
image file: c8lc01189c-f5.tif
Fig. 5 (A) Fluorescent images were captured of the positive wells for a dilution of methylated p14ARF. The reference dye was used to detect loaded wells, and a threshold was applied to calculate the total number of amplified DNA copies. (B) After accounting for the unloaded wells, the calculated copies per μL was compared to the expected. The relationship is linear with a slope of 0.93. (C) On average, 73% of the template molecules are detected. 7% of the sample is lost during preparation or storage, resulting in an estimated loading efficiency of 80%.

The results demonstrate an ultimate detection efficiency of 70–80%, and detection of as few as 100 copies in 8 μL (Fig. 5A). The reference channel was used to calculate the DNA concentration from the filled wells. The results demonstrate the high accuracy of the digital system at 93% (Fig. 5B). The 7% loss is considered inherent to the system, and may occur due to sample preparation or the material properties of the device. Therefore, the average loading efficiency of the multilayer device is 80% (Fig. 5C), which is 7 times higher than our previous work.

The high detection efficiency of this system qualifies it as a suitable method for detection of rare molecules in liquid biopsies. The reported average detection efficiency of 73% includes all potential sources of loss or variability. No system can be expected to be 100% efficient, as single copies of DNA can be lost in preparation steps, pipetting, freeze–thaw cycle, etc. Nevertheless, these improvements on loading and detection efficiency are important steps in advancing the clinical utility of this digital melt platform.

Epiallelic discrimination by digital melt

A key benefit of the HYPER-melt platform is the ability to not only detect and enumerate target molecules, but also to enable genotyping of a target locus. As shown previously,37 in the case of DNA methylation, HRM can be used to readily determine the methylation density of each detected epiallele. We used this paradigm to demonstrate epiallelic discrimination in the present device by loading and digitizing samples containing synthetic oligonucleotide targets representative of bisulfite-converted sequences of heterogeneously-methylated p14ARF epialleles at methylation densities of 33%, 67% and 100%, as well as an unmethylated (0%) population representing background DNA from ostensibly healthy, noncancerous cells.

Following amplification, the device was taken to the digital melt platform for parallelized melt curve acquisition (Fig. S4). The negative derivative of the curve was taken to determine the melt temperature (“peak”) of each amplicon. Amplification and HRM of the unmethylated control DNA at a concentration of 500 copies per nL, or 4 million copies resulted in an even distribution of positive wells containing a homogenous population of amplicons exhibiting melt centered at 83 °C (Fig. 6A). Next, 400 copies of each methylated epiallele were digitized amongst the 8 million copy background. The melt temperature was calculated for each amplicon, and the corresponding methylation density was distinguished by thresholding. A methylation heatmap was then generated to provide a quantitative metric of the methylation heterogeneity of the sample (Fig. 6B). Representative traces of each epiallelic population demonstrate the high sensitivity of this digital melt platform, and a sequence resolution of ≈4 CpG sites.


image file: c8lc01189c-f6.tif
Fig. 6 (A) Unmethylated sequences of p14ARF were amplified and discriminated by melt. (B) Epialleles at three defined methylation densities were loaded into the device along with the unmethylated background. All patterns were then amplified and discriminated by melt temperature.

To demonstrate the versatility of this approach for epiallelic discrimination, we utilized the HYPER-melt to detect and discriminate synthetic molecules representative of three epialleles (40%, 60%, and 100% methylated) of the BRCA1 locus in a background of healthy male genomic DNA from ostensibly healthy donors (Fig. S5). The robust detection and discrimination of epialleles with genomic DNA validates the potential clinical utility of the platform. The CpG-by-CpG resolution of the DREAMing assay can be tuned by locus selection and primer design, as previously described.37 For example, designing primers around a shorter sequence with fewer CpGs will produce higher CpG resolution than a longer, CpG-dense sequence. For further analysis, a linear regression model can be performed between control points and an in silico model (such as uMelt45), which will provide confidence intervals for methylation density at each melt temperature and identifies differentiable patterns.

The ability to perform HRM-based discrimination enables rare and heterogeneous population analysis in an all-in-one platform, for which this is no current commercially available alternative. The impact of methylation heterogeneity in cancer and development is only just beginning to be elucidated. Several models have predicted that differential methylation occurs very early in carcinogenesis.46,47 This platform provides a simple, low-cost tool for quantification of rare DNA methylation heterogeneity.

Conclusions

In conclusion, we presented a multilayer microfluidic device that achieves efficient and highly sensitive detection and discrimination of DNA methylation heterogeneity of rare molecular populations. The curved channels and multilayer architecture improved loading efficiency up to 7× more than our previous design. The digital melt platform provides a rapid and facile technique for parallelized sequence interrogation at the single-copy level, allowing a comprehensive analysis of DNA methylation heterogeneity, especially in challenging samples such as liquid biopsies. We hope this technology can be applied to potentially improve sensitivity in cancer detection and therapeutic monitoring, and to provide insight into molecular mechanisms of tumor evolution.

Author contributions

T.-H. W. conceived the concepts. T. R. P. designed and conceived the assay. C. M. O. and S. L. designed the computational model. S. L. performed the simulations. C. M. O. and D. G. fabricated the devices. C. M. O. designed the device and experiments. C. M. O. constructed the platform hardware and software. C. M. O. analyzed the data. C. M. O. wrote the paper.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

This work was supported by the NIH (U01CA214165, UG3CA211457, R01AI137272).

References

  1. L. A. Diaz and A. Bardelli, J. Clin. Oncol., 2014, 32, 579–586 CrossRef PubMed.
  2. C. D. M. Campos, J. M. Jackson, M. A. Witek and S. A. Soper, Cancer J., 2018, 24, 93–103 CrossRef CAS PubMed.
  3. D. A. Haber and V. E. Velculescu, Cancer Discovery, 2014, 4, 650–661 CrossRef CAS PubMed.
  4. P. Podlesniy and R. Trullas, in Methods in Molecular Biology, Humana Press, New York, NY, 2018, vol. 1768, pp. 111–126 Search PubMed.
  5. L. Gorgannezhad, M. Umer, M. N. Islam, N.-T. Nguyen and M. J. A. Shiddiky, Lab Chip, 2018, 18, 1174 RSC.
  6. A. Agarwal, M. Balic, D. El-Ashry and R. J. Cote, Cancer J., 2018, 24, 70–77 CrossRef CAS PubMed.
  7. M. Chimonidou, A. Strati, A. Tzitzira, G. Sotiropoulou, N. Malamos, V. Georgoulias and E. S. Lianidou, Clin. Chem., 2011, 57, 1169–1177 CrossRef CAS PubMed.
  8. T. Forshew, M. Murtaza, C. Parkinson, D. Gale, D. W. Y. Tsui, F. Kaper, S.-J. Dawson, A. M. Piskorz, M. Jimenez-Linan, D. Bentley, J. Hadfield, A. P. May, C. Caldas, J. D. Brenton and N. Rosenfeld, Sci. Transl. Med., 2012, 4, 136ra68 Search PubMed.
  9. D. T. Miyamoto, D. T. Ting, M. Toner, S. Maheswaran and D. A. Haber, Cold Spring Harbor Symp. Quant. Biol., 2016, 81, 269–274 CrossRef PubMed.
  10. E. Crowley, F. Di Nicolantonio, F. Loupakis and A. Bardelli, Nat. Rev. Clin. Oncol., 2013, 10, 472–484 CrossRef CAS PubMed.
  11. X. Han, J. Wang and Y. Sun, Genomics, Proteomics Bioinf., 2017, 15, 59–72 CrossRef PubMed.
  12. M. Ignatiadis, M. Lee and S. S. Jeffrey, Clin. Cancer Res., 2015, 21, 4786–4800 CrossRef CAS PubMed.
  13. H. Schwarzenbach, D. S. B. Hoon and K. Pantel, Nat. Rev. Cancer, 2011, 11, 426–437 CrossRef CAS PubMed.
  14. G. Sharma, S. Mirza, C. P. Prasad, A. Srivastava, S. D. Gupta and R. Ralhan, Life Sci., 2007, 80, 1873–1881 CrossRef CAS PubMed.
  15. T. Xu, X. Kang, X. You, L. Dai, D. Tian, W. Yan, Y. Yang, H. Xiong, Z. Liang, G. Q. Zhao, S. Lin, K.-N. Chen and G. Xu, Theranostics, 2017, 7, 1437–1446 CrossRef CAS PubMed.
  16. P. Ulz, E. Heitzer, J. B. Geigl and M. R. Speicher, Int. J. Cancer, 2017, 141, 887–896 CrossRef CAS PubMed.
  17. M. Szyf, Drug Resist. Updates, 2003, 6, 341–353 CrossRef CAS PubMed.
  18. V. Santini, H. M. Kantarjian and J.-P. Issa, Ann. Intern. Med., 2001, 134, 573 CrossRef CAS PubMed.
  19. J. G. Herman and S. B. Baylin, N. Engl. J. Med., 2003, 34921349, 2042–2054 CrossRef PubMed.
  20. S. A. Belinsky, Nat. Rev. Cancer, 2004, 4, 707–717 CrossRef CAS PubMed.
  21. P. A. Jones and S. B. Baylin, Nat. Rev. Genet., 2002, 3, 415–428 CrossRef CAS PubMed.
  22. B. Vogelstein, N. Papadopoulos, V. E. Velculescu, S. Zhou, L. A. Diaz and K. W. Kinzler, Science, 2013, 339, 1546–1558 CrossRef CAS PubMed.
  23. P. B. Gupta, C. M. Fillmore, G. Jiang, S. D. Shapira, K. Tao, C. Kuperwasser, E. S. Lander, T. Huang, T. R. Golub, E. S. Lander and J. P. Mesirov, Cell, 2011, 146, 633–644 CrossRef CAS PubMed.
  24. D. A. Landau, K. Clement, M. J. Ziller, P. Boyle, J. Fan, H. Gu, K. Stevenson, C. Sougnez, L. Wang, S. Li, D. Kotliar, W. Zhang, M. Ghandi, L. Garraway, S. M. Fernandes, K. J. Livak, S. Gabriel, A. Gnirke, E. S. Lander, J. R. Brown, D. Neuberg, P. V. Kharchenko, N. Hacohen, G. Getz, A. Meissner and C. J. Wu, Cancer Cell, 2014, 26, 813–825 CrossRef CAS PubMed.
  25. N. C. Sheffield, G. Pierron, J. Klughammer, P. Datlinger, A. Schönegger, M. Schuster, J. Hadler, D. Surdez, D. Guillemot, E. Lapouble, P. Freneaux, J. Champigneulle, R. Bouvier, D. Walder, I. M. Ambros, C. Hutter, E. Sorz, A. T. Amaral, E. de Álava, K. Schallmoser, D. Strunk, B. Rinner, B. Liegl-Atzwanger, B. Huppertz, A. Leithner, G. de Pinieux, P. Terrier, V. Laurence, J. Michon, R. Ladenstein, W. Holter, R. Windhager, U. Dirksen, P. F. Ambros, O. Delattre, H. Kovar, C. Bock and E. M. Tomazou, Nat. Med., 2017, 23, 386–395 CrossRef CAS PubMed.
  26. S. Li, F. Garrett-Bakelman, S. Chung, M. Sanders, T. Hricik, F. Rapaport and C. Mason, Nat. Med., 2016, 22, 792–799 CrossRef CAS PubMed.
  27. J. G. Herman, J. R. Graff, S. Myohanen, B. D. Nelkin and S. B. Baylin, Proc. Natl. Acad. Sci. U. S. A., 1996, 93, 9821–9826 CrossRef CAS.
  28. L. S. Kristensen, T. Mikeska, M. Krypuy and A. Dobrovic, Nucleic Acids Res., 2008, 36, e42 CrossRef PubMed.
  29. C. A. Eads, K. D. Danenberg, K. Kawakami, L. B. Saltz, C. Blake, D. Shibata, P. V. Danenberg and P. Laird, Nucleic Acids Res., 2000, 28, e32 CrossRef CAS PubMed.
  30. Y. M. Lo, L. Y. Chan, K. W. Lo, S. F. Leung, J. Zhang, A. T. Chan, J. C. Lee, N. M. Hjelm, P. J. Johnson and D. P. Huang, Cancer Res., 1999, 59, 1188–1191 CAS.
  31. C. D. Ahrberg, A. Manz and B. G. Chung, Lab Chip, 2016, 16, 3866–3884 RSC.
  32. M. Baker, Nat. Methods, 2012, 9, 541–544 CrossRef CAS.
  33. B. J. Hindson, K. D. Ness, D. A. Masquelier, P. Belgrader, N. J. Heredia, A. J. Makarewicz, I. J. Bright, M. Y. Lucero, A. L. Hiddessen, T. C. Legler, T. K. Kitano, M. R. Hodel, J. F. Petersen, P. W. Wyatt, E. R. Steenblock, P. H. Shah, L. J. Bousse, C. B. Troup, J. C. Mellen, D. K. Wittmann, N. G. Erndt, T. H. Cauley, R. T. Koehler, A. P. So, S. Dube, K. A. Rose, L. Montesclaros, S. Wang, D. P. Stumbo, S. P. Hodges, S. Romine, F. P. Milanovich, H. E. White, J. F. Regan, G. A. Karlin-Neumann, C. M. Hindson, S. Saxonov and B. W. Colston, Anal. Chem., 2011, 83, 8604–8610 CrossRef CAS PubMed.
  34. T. Mikeska, C. Bock, H. Do and A. Dobrovic, Expert Rev. Mol. Diagn., 2012, 12, 473–487 CrossRef CAS PubMed.
  35. K. C. A. Chan, P. Jiang, C. W. M. Chan, K. Sun, J. Wong, E. P. Hui, S. L. Chan, W. C. Chan, D. S. C. Hui, S. S. M. Ng, H. L. Y. Chan, C. S. C. Wong, B. B. Y. Ma, A. T. C. Chan, P. B. S. Lai, H. Sun, R. W. K. Chiu and Y. M. D. Lo, Proc. Natl. Acad. Sci. U. S. A., 2013, 110, 18761–18768 CrossRef CAS PubMed.
  36. C. M. O'Keefe, T. R. Pisanic, H. Zec, M. J. Overman, J. G. Herman and T. Wang, Sci. Adv., 2018, 4, eaat6459 CrossRef PubMed.
  37. T. R. Pisanic, P. Athamanolap, W. Poh, C. Chen, A. Hulbert, M. V. Brock, J. G. Herman, T.-H. Wang and T.-H. Wang, Nucleic Acids Res., 2015, 43, e154 CrossRef PubMed.
  38. Y. Fu, H. Zhou, C. Jia, F. Jing, Q. Jin, J. Zhao and G. Li, Sens. Actuators, B, 2017, 245, 414–422 CrossRef CAS.
  39. A. M. Thompson, J. L. Smith, L. D. Monroe, J. E. Kreutz, T. Schneider, B. S. Fujimoto, D. T. Chiu, J. P. Radich and A. L. Paguirigan, PLoS One, 2018, 13, e0196801 CrossRef PubMed.
  40. Q. Zhu, L. Qiu, B. Yu, Y. Xu, Y. Gao, T. Pan, Q. Tian, Q. Song, W. Jin, Q. Jin and Y. Mu, Lab Chip, 2014, 14, 1176–1185 RSC.
  41. S. W. Lee and S. S. Lee, Microsyst. Technol., 2007, 14, 205–208 CrossRef.
  42. C. Moraes, Y. Sun, C. A. Simmons and J. Micromechanics, Microengineering, 2009, 19, 6 Search PubMed.
  43. J. S. Marcus, W. F. Anderson and S. R. Quake, Anal. Chem., 2006, 78, 3084–3089 CrossRef CAS PubMed.
  44. M. Esteller, S. Tortola, M. Toyota, G. Capella, M. A. Peinado, S. B. Baylin and J. G. Herman, Cancer Res., 2000, 60, 129–133 CAS.
  45. Z. Dwight, R. Palais and C. T. Wittwer, Bioinformatics, 2011, 27, 1019–1020 CrossRef CAS PubMed.
  46. T. R. Pisanic, P. Athamanolap and T. H. Wang, Semin. Cell Dev. Biol., 2017, 64, 5–17 CrossRef CAS PubMed.
  47. W. Timp and A. P. Feinberg, Nat. Rev. Cancer, 2013, 13, 497–510 CrossRef CAS PubMed.

Footnote

Electronic supplementary information (ESI) available. See DOI: 10.1039/c8lc01189c

This journal is © The Royal Society of Chemistry 2019