Nucleic Acids Research Advance Access published January 25, 2011

Nucleic Acids Research, 2011, 1–9

doi:10.1093/nar/gkq1357

A high-throughput DNA methylation analysis

of a single cell
Martin Kantlehner1, Roland Kirchner1, Petra Hartmann1, Joachim W. Ellwart2,
Marianna Alunni-Fabbroni1 and Axel Schumacher3,*
1
Beckman Coulter Biomedical GmbH, Advalytix Products, 2Helmholtz Zentrum München, German Research
Center for Environmental Health - Institute of Molecular Immunology, 81377 Munich, Germany and
3
Neuroscience Department, Centre for Addiction & Mental Health, Toronto, M5T1R8, Canada

Received December 23, 2009; Revised December 24, 2010; Accepted December 28, 2010

ABSTRACT schizophrenia, bipolar disorder, diabetes, Alzheimer’s

disease, Parkinson’s disease and various cancers remains
In recent years, the field of epigenetics has grown

Downloaded from [Link] by guest on February 9, 2011

a major challenge. In recent years, it became evident that
dramatically and has become one of the most the study of epigenetic mechanisms, which are consistent
dynamic and fast-growing branches of molecular with various non-Mendelian irregularities of complex
biology. The amount of diseases suspected of diseases, may hold the key to the understanding of some
being influenced by DNA methylation is rising of the characteristics of these disorders (1). One of the
steadily and includes common diseases such as central obstacles hampering progress in the burgeoning
schizophrenia, bipolar disorder, Alzheimer’s field of epigenetics related to human disease is the
disease, diabetes, atherosclerosis, cancer, major inherent cell heterogeneity of a given tissue. Standard
psychosis, lupus and Parkinson’s disease. Due to population measurement techniques merely describe
cellular heterogeneity of methylation patterns, epi- average behavior and are insufficient to investigate vari-
ability among cells (2). Epigenetic analyses traditionally
genetic analyses of single cells become a necessity.
probe cell ensembles, thereby completely averaging over
One rationale is that DNA methylation profiles are relevant individual cell responses, such as differences in
highly variable across individual cells, even in the cell proliferation, lack of synchrony of cells in a culture,
same organ, dependent on the function of the responses to external stimuli and disease onset or stochas-
gene, disease state, exposure to environmental tic events (3). As a consequence, up to now it is impossible
factors (e.g. radiation, drugs or nutrition), stochastic to understand whether a small increase in the methylation
fluctuations and various other causes. Using a poly- level measured in the ensemble results from a small,
merase chain reaction (PCR)-slide microreaction homogeneous increase across all cells or a large increase
system, we present here a methylation-sensitive in a subset of cells. In cancer studies, methylation patterns
PCR analysis, the restriction enzyme-based in CpG islands that are important in gene regulation could
single-cell methylation assay (RSMA), in the be different from cell to cell even in a single tumor tissue.
Cells with distinct epigenetic profiles eventually display
analysis of DNA methylation patterns in single
distinct phenotypic behavior and drug response (4,5).
cells. This method addresses the problems of cell For example, variability in adipogenesis activity among
heterogeneity in epigenetics research; it is compar- preadipocytes presents significant challenges to drug inter-
ably affordable, avoids complicated microfluidic vention (6). To improve the efficacy of drug treatment, the
systems and offers the opportunity for high- source of cell-to-cell variability should be identified and
throughput screening, as many single cells can be targeted. Thus, the ability to detect methylation patterns
screened in parallel. In addition to this study, critical of regulatory sequences from a single cell is an essential
principles and caveats of single cell methylation factor for understanding the mechanism of tumor initi-
analyses are discussed. ation, the screening of molecular markers and early diag-
nosis of cancers. This holds true for the molecular
characteristics of most diseased tissues (especially highly
INTRODUCTION compartmented tissues such as brain or pancreas) where
current analyses are extremely limited by examination of
Despite significant effort, understanding the causes and pooled cell lysates. However, in contrast to one-cell tran-
mechanisms of complex non-Mendelian diseases such as scriptome analyses, the interrogation of DNA methylation

*To whom correspondence should be addressed. Tel: +1 416 5358501 6831; Fax: +1 416 979 4666; Email: axel_schumacher@[Link]

ß The Author(s) 2011. Published by Oxford University Press.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License ([Link]
by-nc/2.5), which permits unrestricted non-commercial use, distribution, and reproduction in any medium, provided the original work is properly cited.
2 Nucleic Acids Research, 2011

patterns in single cells is lagging behind, primarily because Bisbenzimide H33342 (Fluka) in a 10 mg/ml final concen-
of more complex methodologies. As a consequence, epi- tration and propidium iodide (Sigma) with a 50 mg/ml final
genetic cause of cell-to-cell variability cannot be readily concentration. Living single cells (propidium iodide
investigated in an average laboratory setup. For negative) in 1  PBS were deposited on the reaction sites
example, the study of methylation patterns in single cells of AmpliGrid slides (Beckman Coulter Biomedical
using standard equipment such as polymerase chain GmbH, Advalytix Products, Munich/Germany, BCB) ac-
reaction (PCR) reaction tubes with 0.2–1.5 ml volume is cording to their physical properties (size, granularity) with
usually unsuccessful due to numerous reasons, including a fluorescence-activated cell sorter (FACS), fitted with an
partial recovery of cell nuclei, loss of DNA during various AmpliGrid slide holder (BCG). Cell deposition was
experimental steps, incomplete chemical or enzymatic validated with a fluorescence microscope fitted with a
modification and insufficient amplification of the DNA, DAPI filter. Slides with deposited cells were stored at 4 C.
among other causes. Indeed, previous attempts to estab-
lish protocols for single cell analyses of methylation DNA preparation
pattern suffered from low success rate and low throughput A volume of 0.5 ml of a 2 % solution of CEK enzyme
(7). In contrast to gene expression analyses, successful (BCG) in 1  reaction buffer was pipetted to the reaction
methylation studies using microfluidic devices are still site of an AmpliGrid slide containing a single cell and
lacking, presumably because microfluidic protocols are ex- covered immediately with 5 ml of sealing solution (BCB).
pensive, usually have limited multiplexing capabilities, are After complete loading, the AmpliGrid slide was
laborious or cannot handle the necessary larger number of

Downloaded from [Link] by guest on February 9, 2011

incubated for 2 h at 60 C on an AmpliSpeed slide cycler
chemical or enzymatic steps involved. One of the main (BCB). Subsequently the sealing solution was removed by
restrictions of microfluidic devices is the inefficient dipping the slide into a staining jar filled with hexane for
transfer of the analyte (i.e. DNA) between each processing 30 s, followed by drying of the remaining aqueous phase at
step and elimination of waste products (8,9). To circum- 37 C on the AmpliSpeed slide cycler.
vent these limitations, we developed a protocol that is
comparably affordable and contains all the necessary Methylation-sensitive enzymatic cleavage of DNA
elements for single-cell DNA methylation profiling of
CpG sites integrated in a single device. The protocol A double digestion solution was prepared containing
consists of cell placement and cell lysis, enzymatic DNA 0.25 U/ml NlaIV (BspLI) and 0.25 U/ml MspI, HpaII or
Hin6I in 0.5  Tango buffer. All restriction enzymes
fragmentation, cleavage of DNA with methylation-
were purchased from Fermentas, St. Leon-Rot/
sensitive restriction enzymes (MSREs) and parallelized
Germany. For the PCR positive control a single digestion
multi-well DNA amplification using 48-well slides.
solution was prepared containing 0.25 U/ml NlaIV in
To demonstrate the usefulness of the restriction
0.5  Tango buffer. For each digestion, 0.5 ml of reaction
enzyme-based single-cell methylation assay (RSMA) tech-
mix was pipetted to an AmpliGrid reaction site containing
nology in a clinical setup, we chose several CpG island
a single cell and immediately covered with 5 ml of sealing
sequences of genes such as CDKN2A/INK4a (p16) and
solution. The slide was incubated for 6 h at 37 C on an
COL1A2, which are known to be aberrantly methylated in
AmpliSpeed slide cycler followed by a 30-min heat de-
various cancers. Interrogation of DNA methylation
naturation of the enzymes at 65 C. Subsequently, the
patterns in regulatory regions such as CpG islands has
sealing solution was removed by dipping the slide into a
become an important tool for diagnostic purposes and
staining jar filled with hexane for 30 s followed by drying
understanding of tissue-specific gene regulation in both
of the remaining aqueous phase at 45 C on the
normal land pathological conditions. Ultimately, the
AmpliSpeed slide cycler.
basic methodology described in this report may be
adapted to include real-time methylation protocols and Single-cell PCR
even whole-genome ‘epigenetic’ microarrays (1,10).
Primers were designed by the Beckman Coulter
Biomedical GmbH primer design service (Advalytix
MATERIALS AND METHODS Products, Munich, Germany) and purchased from
MetaBion, Martinsried/Germany as 100 mM solutions in
Cell lines
water. Primer mixes for the COL1A2 promoter: 0.8 mM
As an epigenetically well-characterized test system, the long forward (50 -TTCGGCTAAGTTGGAGGT
human SW480 cancer cell line derived from a colorectal ACTG-30 ), 0.3 mM short forward (50 -CTAGACATGCT
adenocarcinoma, grade III–IV, was used (11). To study CAGCTTTGTGG-30 ) and 0.3 mM reverse (50 -CTTACA
tissue-specific effects, human lymphocytes (Pachmann TTGGCATGTTGCTAGG-30 ). CDKN2A gene: each
SIMFO, Dr. Med. Ulrich Pachmann, Bayreuth/ 0.5 mM for long forward (50 -CTCTGGAGGACGAAGT
Germany) were analyzed. SW480 cells were cultured TTGC-30 ), short forward (50 -CTTCCTGGGGAGTTTTC
under standard conditions (humidified incubator, at AGA-30 ) and reverse (50 -ATTCCTCTTCCTTGGCTT
37 C and 5% CO2) in Dulbecco’s modified eagle CC-30 ). One microliter of each primer mix was pipetted
medium (DMEM) containing 10% fetal bovine serum to a reaction site with restriction-enzyme-treated cells and
(FBS), 50 mg/l gentamycin (Biochrom, Berlin/Germany) dried at 45 C on an AmpliSpeed slide cycler.
and 100 mg/l kanamycin (Biochrom). Human lympho- Subsequently, 1.65 ml of PCR mix containing
cytes were used as delivered. Cells were stained with 1  Multiplex Master Mix (Qiagen, Hilden/Germany)
 Nucleic Acids Research, 2011 3

that contains HotStarTaq DNA Polymerase and 0.9x in Table S1. Bisulfite conversion of genomic DNA and
Q-solution were pipetted onto AmpliGrid’s reaction sites pyrosequencing were performed as described previously
and covered immediately with 5 ml sealing solution. PCR (13). All PCR reactions were checked on a 1.0% agarose
program: initial denaturation at 95 C for 10 min, followed gel to ensure successful amplification and specificity before
by 60 cycles of (30 s at 94 C, 1:15 min annealing at 60 C proceeding with pyrosequencing.
and 1:15 min elongation at 72 C), followed by a final
10-min elongation step at 72 C. After the amplification,
4 ml of loading dye (1.25) was added to each reaction site RESULTS AND DISCUSSION
of the AmpliGrid slide. After joining of aqueous phase
and loading dye, 4 ml of each PCR product was transferred The proposed RSMA method uses a classical profiling
to an 8% PAGE gel (Anamed, Gross-Bieberau/Germany) approach with MSREs to interrogate methylation
and separated for 30 min at 250 V. Silver staining was per- patterns at specific CpG dinucleotides. RSMA comprises
formed with 0.1% AgNO3-solution for 5 min, followed by four main steps (Figure 1): (i) deposition of single cells
a 10 s washing step in DI water and development in 0.4 M onto the microreaction sites; (ii) cell lysis and removal of
NaOH/0.1% formaldehyde for 10 min. DNA-bound proteins; (iii) methylation-sensitive restric-
tion of DNA; and (iv) amplification of the interrogated
Bisulfite sequencing DNA sequences. The final amplification is performed with
two forward primers and one reverse primer (Figure 1B).
DNA from SW480 colorectal cancer cells and from whole The restriction sites for the MSREs are located between

Downloaded from [Link] by guest on February 9, 2011

blood was purified using the QIAamp mini kit and the two forward primers. In case the amplified DNA
QIAamp DNA blood mini kit (Qiagen), respectively, ac- fragment is methylated, the restriction enzymes will not
cording to the manufacturer’s standard protocols. cleave, resulting in two amplification products. If the
Bisulfite treatment was performed using a modified DNA is unmethylated, only the short amplification
standard protocol after Clark et al. (12). Briefly, 500 ng product is synthesized during the PCR. The resulting
genomic DNA was split into four reactions and separately DNA products are finally separated on a gel, where the
fragmented by EcoRI, PvuII and HindIII (Fermentas). In short band serves as positive PCR control. A large number
the fourth reaction, DNA was sheared briefly (1–2 s) with of MSREs can be used with RSMA, enabling the interro-
an ultrasound sonicator. After heat inactivation of the gation of nearly every gene or promoter in a mammalian
enzymes, all reactions were combined and then purified genome (1).
using a MiniElute reaction cleanup kit (Qiagen) and AmpliGrid slides, commonly used for single-cell PCR
eluted in 35 ml H2O. Bisulfite conversion was performed applications, were chosen as reaction platform (14). Each
using an EpiTect Bisulfite Kit (Qiagen) according to the slide contains 48 independent hydrophilic reaction sites
manufacturer’s manual and finally eluted in 20 ml EB surrounded by a hydrophobic circle which holds
buffer. Bisulfite-treated DNA was stored at 80 C until aqueous reaction solutions, such as enzyme buffers, in
needed. 10 ml of each sample was used for bisulfite place (Figure 1A). Cells were deposited on the slide
hot-start PCR in 25 ml volume reactions containing 1 mM surface by flow cytometry cell sorting. Before each experi-
primers, and a master mix containing TrueStart Taq poly- ment, the correct deposition of single cells on the reaction
merase (Fermentas). PCR program: Initial denaturation sites was verified with a microscope. The flatness of the
at 95 C for 2 min, followed by 35 cycles of (20 s at 94 C, glass substrate allows a simple optical control of the whole
1:10 min annealing at 61.5 C and 1 min elongation at workflow using a microscope or a slide scanner. Lysis of
72 C), followed by a final 10-min extension step at 72 C. single cells was performed with a protease solution (CEK)
CDKN2A bisulfite primer: F: 50 -GTAGTATGGAGTTT which digests proteins associated with DNA. A heat in-
TYGGTTGATTG-30 ; R: 50 -AAAAACTAAATAATCCC activation of CEK is not required as it is almost com-
AACACATCTT-30 (product size 498 bp). COL1A2 pletely inactive under the conditions where most MSREs
bisulfite primer: F: 50 -GGAGGTATTTTAGGGTTAGG are predominantly active (37 C).The relatively low CEK
GAAAT-30 ; R: 50 -CAAACAAACTAAAAACACTTAC reaction temperature (60 C) keeps the DNA double-
ATTAA-30 (product size 295 bp). All PCR reactions stranded, a necessary prerequisite for the subsequent en-
were checked on a 1.0% agarose gel to ensure successful zymatic cleavage. To prevent evaporation and cross-
amplification and specificity before proceeding with contamination, each reaction was covered with 5 ml of an
sequencing. The amplification product of the expected oil-based sealing solution (see ‘Materials and Methods’
size were cut from the gel and purified by QIAquick gel section). After the proteolysis, the sealing solution was
extraction kit (Qiagen). PCR products were diluted to removed and the aqueous droplets air-dried. The drying
2 ng/ml and subjected to direct sequencing (Eurofins step is required to prevent a surplus dilution of the
MWG Operon, Ebersberg/Germany). reaction mix in the methylation-sensitive cleavage and
DNA fragmentation. Restriction enzymes should be
Pyrosequencing added in 0.5 ml of 0.5  restriction enzyme buffer,
Primers were designed with Pyrosequencing Assay Design because higher buffer concentrations may result in inhib-
Software v1.0.6 (Biotage, Uppsala, Sweden). The ition of the following PCR reaction. Additionally, a
CDKN2A locus was designed to span the region contain- reduced amplification efficiency using some restriction
ing the same HpaII and Hin6I sites that were interrogated enzyme buffers was recognized (data not shown). Hence,
with RSMA. A full list of primer sequences can be found the final PCR amplification is performed in a relatively
4 Nucleic Acids Research, 2011

Downloaded from [Link] by guest on February 9, 2011

Figure 1. Principle of the RSMA test system for the analysis of DNA methylation patterns in single cells. (A) Single cells from various sources are
placed on high-throughput multi-well PCR slides (one cell per well), either by micromanipulation, laser capture microdissection or flow cytometry
cell sorting. The protocol was tested with AmpliGrid microreaction slides, which are composed of a standard microscope slide sized glass surface
with chemically structured, DNA free reaction centers. The reaction volume ranges from 500 nl up to 1.65 ml, small enough to sufficiently con-
centrate the reaction volume to a point where enzymatic steps proceed with high kinetics, but large enough to sufficiently dilute the cellular
components (1 pl) of each cell. It is possible to perform 48 methylation-profiling reactions on one 75 mm  25 mm slide. The structure of the
slides allows carrying out methylation-sensitive PCRs from single cells that can be lysed directly on the reaction centers, thereby generating
methylation profiles starting with DNA amounts as low as 6 pg. The hydrophilic and hydrophobic regions on the slide surface ensure that even
after inaccurate pipetting the droplets containing single cells will find their way to the correct position. An important feature of this technology is the
parallel analysis of multiple cells, a desirable feature that makes it possible to obtain statistically relevant methylation data on single cells in a
reasonable time frame. (B) DNA is fragmented by a restriction enzyme (here NlaIV), that cleaves outside of the target region. Primer mixes contain
two forward primers (F1+F2) and one reverse primer (R). The analyzed CpG dinucleotides (here HpaII and Hin6I sites) are located between the
two forward primers. If the restriction site is methylated, the restriction enzyme cannot cleave and both PCR products are synthesized, whereas for
unmethylated restriction sites only the short PCR product (internal PCR positive control) is produced.

high volume (1.65 ml) to dilute the reaction buffers of the essential since the amplification products are usually
preceding reactions as much as possible. Single-cell located within GC-rich regions that favor the formation
methylation-sensitive PCRs have to be highly efficient of stable intramolecular hairpin structures interfering with
and sensitive as only one to two copies of the target the annealing of the primers. This is generally not a
sequence is present; hence, specific care has to be taken problem for conventional PCR with a high number of
to avoid problems that may cause decreased amplification DNA templates, but can be crucial in single-cell PCR,
efficiency. One problem during amplification steps is the where the amount of starting material is low. The occur-
mixed hydrophobic–hydrophilic nature of biological rence of hairpin structures also requires special care in
macromolecules, which leads to their adsorption to a primer design for single-cell methylation profiling applica-
variety of surfaces; often the DNA molecules stick to tions. To avoid primer sequences that are located in
reaction vessels, which may interfere with the amplifica- regions susceptible to hairpin structures in the chosen
tion. This characteristic may result in a significant loss of PCR conditions, the target regions were analyzed for
sample or in the possible inaccessibility of certain DNA their folding properties using the SirGraph package
sequences for several initial PCR cycles and eventually (Supplementary Data) (15). In our experience, primer an-
becoming accessible only later during the amplification. nealing can also be enhanced by using a higher concentra-
Indeed, we found that an increase from a standard tion of primers, which is often important for improving
30-cycle amplification to 60 cycles can improve the the amplification efficiency of the long PCR product.
PCR performance significantly (data not shown). In ex- Another key component for a successful single-cell PCR
pression studies, such cycle numbers may introduce dis- is the complete denaturation of the DNA template at the
tortions in the measured gene expression levels. However, beginning of the PCR. Incomplete denaturation of the
when performing methylation analysis the high cycle DNA results in an inefficient utilization of template in
number is not a disadvantage per se as the cell contains the earliest amplification cycles leading to poor yield of
only one or two copies of each locus which is either PCR product or may interrupt amplification.
methylated or not and no intermediate methylation Consequently, the initial denaturation should be per-
levels exist. In theory, this information can be used to formed over an interval of at least 5–8 min at 95 C if the
transform the methylation status of each CpG dinucleo- GC-content of the desired DNA sequence is 50% or less.
tide into a binary code for subsequent biostatistical This interval should be extended up to 10 min for GC-rich
analyses. Moreover, a large number of PCR cycles are templates such as gene promoters. In our experiments, the
 Nucleic Acids Research, 2011 5

use of a hot-start DNA polymerase, i.e. HotStarTaq poly- for PCR efficiency, only a single digestion with NlaIV was
merase, which itself requires an initial denaturation step to performed.
be fully activated, resulted in good and sufficient In total, we analyzed 429 single cells for COL1A and
amplification. CDKN2A (Figure 2). In 1.5% of the amplification reac-
tions using the SW480 cells, and in 5% of reactions with
Single-cell profiling in human cancer cell lines and lymphocytes, the PCR yielded no product (on average 3%
lymphocytes of all amplifications).
On average, our analysis showed that in 91% of the
RSMA can be applied with a variety of enzymes and SW480 cells the analyzed HpaII restriction site within
protocols (e.g. real-time measurements and multiplexing, the COL1A2 promoter was methylated (Supplementary
among others); however, to illustrate the overall Figure S2). Comparable methylation percentages were
approach, we demonstrate a simple protocol, that obtained for the Hin6I site, where in 87% of the cells
requires two very commonly used MSREs in epigenetics CpG methylation was observed. The opposite pattern
research, HpaII and Hin6I. A comprehensive list of po- was found in lymphocytes, with most cells being
tential restriction enzymes with genomic coverage for unmethylated. Only 5% of the analyzed sites were
DNA methylation profiling can be found in Schumacher methylated. These numbers are in agreement with pub-
& Petronis 2006 (1). One rationale in using the classical lished data for SW480 cells, which were reported to be
profiling approach with MSREs is the potential to use the hypermethylated (>80% methylation) at the +7 site in
data and principles of single-cell methylation profiling the COL1A2 promoter compared to non-cancer tissues

Downloaded from [Link] by guest on February 9, 2011

derived from this study to develop further protocols (22). This increased methylation of the collagen gene in
with higher genomic coverage. Indeed, the interest in colorectal cancer cells was inversely correlated with
MSREs is now resurging as these enzymes are the key collagen messenger RNA (mRNA) steady-state levels.
tools for large-scale epigenomic profiling using micro- Similarly, in our experiment, the CDKN2A promoter
arrays (1) and next-generation sequencing (NGS) was found to be almost completely methylated in SW480
approaches (16), which ultimately may lead to the devel- cells (99%), but relatively hypomethylated in lymphocytes
opment of single-cell whole methylome technologies. (21%). Again, our results are in agreement with previous
To validate the RSMA method, restriction sites located studies, which reported methylation levels of the
in the promoter region of two genes (COL1A2 and CDKN2A promoter between 92% and 100% in SW480
CDKN2A) were analyzed in the human colorectal cells and no significant methylation in blood cells (20,23–
cancer cell line SW480 and in human lymphocytes. In 25). Additionally, using a methylation-specific polymerase
initial screenings, the FGF19 and RARB gene promoters chain reaction followed by analysis of the methylated
were also analyzed; however, we detected only cytosine content of the product by thermal denaturation,
unmethylated CpGs in the interrogated cells (Figure 1B). a recent study by Kovatsi and co-workers (26) also
Hence, we focused our analysis on the well-characterized described a partial methylation of the CDKN2A
collagen type I alpha 2 (COL1A2) and cyclin-dependent promoter in peripheral blood samples from healthy
kinase inhibitor 2A (CDKN2A) genes, which are known individuals.
to exhibit cancer- and tissue-specific methylation patterns. Overall, the single-cell data show the expected signifi-
A number of reports have demonstrated the silencing of cant difference between the SW480 cancer cells and
COL1A2, a candidate tumor suppressor gene, due to lymphocytes. It is important to note that relatively
aberrant methylation within the promoter region in similar methylation levels were obtained for the closely
primary cancer tissues and in several cancer cell lines spaced HpaII and Hin6I restriction sites within the same
(17,18). Similarly, CDKN2A, a known tumor suppressor cell type and the same promoter region. This finding indi-
gene, is either mutated or deleted in a wide range of cates that the methylation status of each CpG dinucleotide
cancers and in addition has been found to be silenced by within the same promoter may be representative for the
CpG methylation in many tumor types (19–21). methylation status of surrounding CpG sites. This finding
To assess the restriction enzyme efficiency in the chosen is significant as only a subset of CpGs in CpG islands is
model system, DNA from single cells was also cleaved recognized by specific methylation sensitive restriction
with MspI, an isoschizomer of HpaII, which cuts inde- enzymes and therefore suitable for methylation profiling.
pendent of the CpGs methylation status. If a digestion To validate our RSMA results, we performed direct
was successful, amplification of the interrogated regions bisulfite sequencing on the CDKN2A and COL1A2
should only produce the short amplification product promoter regions as well as quantitative pyrosequencing
(Figure 1B). All MSRE cleavage reactions were performed of the downstream CDKN2A HpaII and Hin6I sites. As
as double digests together with NlaIV (50 -GGN^NCC-30 ). exemplified for the CDKN2A promoter, direct sequencing
The main purpose of NlaIV, which cleaves independent of of PCR products from bisulfite-treated DNA revealed
methylation status, is the fragmentation of the genomic hypermethylation (>98%) of both gene promoters in
DNA. Without fragmentation, we observed a significantly SW480 cells, whereas both CpG islands revealed to be
decreased processivity for the long PCR product of several hypomethylated in lymphocytes (Figure 3). Additionally,
different genes (data not shown). In our experimental pyrosequencing of the 30 border of the CDKN2A
setup, NlaIV does not cleave within the amplified DNA promoter in additional lymphocyte DNA samples
sequence, but close to the primer annealing sites outside revealed residual DNA methylation level (up to 18%),
the long PCR products (Figure 1B). As positive control confirming our RSMA data (Supplementary Figure S3).
6 Nucleic Acids Research, 2011

Figure 2. Polyacrylamide gel electrophoresis of products from multiple parallel single-cell PCRs as exemplified for the CDKN2A promoter. Each
lane represents a PCR product from a single cell in SW480 cells (top row) and lymphocytes (bottom row). All of the SW480 cells in the presented
slide contained only fully methylated HpaII and Hin6I sites, indicated by the presence of two bands, the uncleaved (methylated) long band (191 bp,
F1/R) and the short control band (157 bp, F2/R). In contrast, most of the lymphocytes contained only the short band, indicating that the analyzed
CpG dinucleotides were unmethylated. Some cells were treated with MspI/NlaIV as control for the enzymatic reaction and should produce only the
short band. In parallel, some cells were treated with NlaIV alone, which serves as positive control for the long PCR product. If the chosen PCR

Downloaded from [Link] by guest on February 9, 2011

conditions are appropriate, both amplification products should be visible. Typically, the short control band is noticeable weaker compared to the
long band, which may be explained by the fact that in later PCR cycles the short band may also serve to a significant extent as primer for the long
PCR product. In addition, due to better hybridization efficiency, the longer PCR product may be amplified preferentially. M, methylated; U,
unmethylated.

These results underscore the power of the RSMA more than 3 months increased the rate of failed reactions
approach, as more detailed methylation data can be (e.g. incomplete digestion or repressed amplification) sig-
produced compared to direct sequencing. The observed nificantly. For example, control digestion with MspI,
variation of methylation patterns between individual which should always produce only the small PCR
cells of the same cell type illustrates that a certain stochas- product, failed in 7% of experiments when using old
tic variation exists, which would be masked by conven- cells. Similarly, NlaIV digests failed in about 9% of old
tional bulk measurements. Bisulfite sequencing may not cells, whereas incomplete digestion was nearly absent in
be the first choice for single-cell methylation profiling freshly prepared cells. In general, DNA methylation
because treatment with bisulfite significantly degenerates patterns are known to be stable in post-mortem tissues
the four nucleotide code (1), which results in the loss of (29) and storage of cells on the slides should not affect
specificity of a large portion of the genome and increases the DNA methylation patterns as the cells are dried and
the chance of cross hybridization of PCR primers. no enzymatic activity is to be expected.
Additionally, due to the harsh conditions, a certain per- The main reason for the lack of technologies to study
centage of DNA degrades during bisulfite conversion, a single-cell methylation patterns can be attributed to the
decisive disadvantage when working with a limited poor yields encountered when subjecting mammalian
amount of DNA. As a consequence, conventional bisulfite cells to multiple enzymatic or chemical processes. In our
sequencing requires several hundred or even thousands of experience, to circumvent these problems, it is advanta-
cells in order to obtain a sufficient amount of DNA for geous to perform all experimental steps in a single
analysis. Obtaining this number of cells of a single type is reaction vessel, without transfer of the components.
challenging, particularly when attempting to study rare Ideally, to avoid loss of DNA, pumping (circulation) of
stem cells such as tumor-forming cancer stem cells solutions should be avoided. So far, researchers focusing
(CSCs) (27,28). Therapeutic innovations may emerge on epigenetic profiling of a limited number of cells rely on
from a better understanding of the epigenetic mechanisms microplates or standard reaction tubes for their assays. As
of CSCs. The high-throughput RSMA technology has the a result, large dead volumes, template adsorption and no
potential to transform such single-cell methylation optical quality control are responsible for a considerable
analyses from a major technological challenge to a uncertainty and low reproducibility. We found that using
routine procedure for research and clinical diagnostics. microreaction slides could overcome all of these problems,
whereas using standard reaction vessels, such as 0.5 ml
PCR tubes resulted in inconsistent or completely failed
Critical parameters for single-cell methylation profiling enzymatic reactions (data not shown). For kinetic
In our experiments, we studied several other important reasons, it is advantageous to perform enzymatic reactions
parameters that may prove crucial in single-cell analyses. in very small volumes, but, on the other hand,
For example, in rare cases, we observed incomplete diges- large-enough dimensions (>0.5 ml) to allow for reaction
tion of the DNA, most likely due to aged cells that were components to be easily transferred or added between
stored on the microreaction sites for longer periods. reaction steps in a standard laboratory setup.
Usually, cells can be stored on the reaction sites for Another crucial step is the complete lysis of the cell and
several months; however, we found that cells stored for removal of unwanted proteins before enzymatic reactions
 Nucleic Acids Research, 2011 7

Downloaded from [Link] by guest on February 9, 2011

Figure 3. Bisulfite sequencing of larger cell populations to verify the results from single-cell measurements. (A) Direct sequencing of the CDKN2A
core promoter region revealed hypermethylation of all of the analyzed CpG dinucleotides within the promoter CpG island in colorectal cancer cells,
whereas no significant methylation in lymphocytes could be detected. The small methylation level in the HpaII and Hin6I sites can be explained by
their location at the CpG island downstream border, after which methylation levels increase. The black areas of the lollipops represent the average
degree of methylation for each CpG dinucleotide, determined by semiquantitative measurement of peak height for each nucleotide from multiple
bisulfite sequence electropherograms. Gray lollipops = no sequencing data. (B) Structure and analyzed sequence of the CDKN2A gene. Sample
sequencing traces from different tissues are shown on the bottom. The bisulfite sequencing results confirm the methylation data derived from
single-cell methylation profiling by the RSMA technology.

take place, primarily to avoid protein interference. This is methylation patterns. For example, embryonic stem (ES)
important because an individual cell’s total cellular cells are easily affected by culture conditions and in vitro
protein content is quite high, averaging 8  109 molecules manipulation, which could in turn result in increased vari-
per cell, which is about equal to 700 pg (3). Many of the ability in the methylation and transcription patterns of the
proteins, especially histone and transcription factors, are ES cell genome. Indeed, previous work on ES cells has
bound to the DNA and hence can interfere with analysis. shown that stem cell-derived tissues and embryos often
The small absolute amount of DNA and the low DNA fail to maintain stable epigenetic states (30,31). Often, epi-
concentration compared to cellular proteins presents sig- genetic instability is reported in differentially methylated
nificant challenges for detection. Efficient proteolytic steps regions of mostly growth-related imprinted genes, but less
have to be applied to ensure that the minute amount of often neuronal genes (32). The tendency for living cells to
DNA can be enzymatically processed and amplified by be perturbed by in vitro manipulation imposes stringent
PCR. In our experience, a 2-h treatment with CEK requirements in performing a successful biological experi-
protease (see ‘Material and Methods’ section) was suffi- ment. Changes in the cell’s environment such as
cient to lyse the cell and free the DNA from proteins for nutrients, pH, ionic strength and temperature can lead
subsequent enzymatic treatment. to variation in the intracellular concentrations of many
Another very crucial step in single-cell methylation molecular species involved in methylation homeostasis.
profiling is the very quick transfer of the freshly Furthermore, epigenetic patterns are in a constant state
prepared cell to the reaction site, followed by immediate of flux, e.g. cyclical methylation and demethylation of
drying or lysis of the cell. The act of placing a cultured CpG dinucleotides at specific loci may occur within
(adherent) or microdissected cell into suspension for minutes (33). Hence, in some single-cell studies, careful
transfer into the analytical devices may damage the cell consideration of cell-cycle status and transfer time to the
membrane and activates various signaling or apoptotic reaction site for quick drying are warranted.
pathways. This in turn may disturb the epigenetic machin- In contrast to expression studies, DNA methylation
ery of the cell, ultimately resulting in technically biased analysis of single cells are even more sensitive to
8 Nucleic Acids Research, 2011

contamination, as a single foreign DNA molecule could screened in parallel. This method may prove to be espe-
significantly alter the result, whereas a single RNA cially useful for clinical studies, as the procedure is
molecule would—in most cases—not dramatically alter straightforward and does not require complex
the measured expression levels. Given the potential for microfluidics devices. By coupling RSMA with
contamination from even very small amounts of DNA, a micromanipulation, laser capture microdissection, or
useful microreaction system will likely need to be single flow cytometry cell sorting, the heterogeneous nature of
use. tumors, neurodegeneration and other complex disorders
Another difficulty when studying single diploid cells can be investigated at the single-cell level. Overall, it is
comes from heterozygous DNA methylation that can be argued that single-cell epigenetic strategies such as
found, for instance, in imprinted genes where each allele RSMA, when applied in parallel with the traditional
can be differentially methylated depending on its parental genetic ones, may significantly advance the discovery of
origin. Depending on the applied protocol, assessing the etiopathogenic mechanisms of complex diseases and sto-
exact methylation pattern can be challenging. For chastic epigenetic variance.
example, in our protocol, if only one allele is methylated,
the resulting banding patterns in the gel will look similar
to fully methylated sequences (two PCR bands), only with SUPPLEMENTARY DATA
a higher intensity of the low-molecular PCR product. Supplementary Data are available at NAR Online.
Consequently, to confirm heterozygous methylation, the

Downloaded from [Link] by guest on February 9, 2011

resulting amount of each PCR product would have to be
carefully measured, such as by quantitative real-time PCR ACKNOWLEDGEMENTS
(qPCR). We thank Gabriel Oh and Catherine Ng for the help with
An additional caveat for single-cell approaches is that Pyrosequencing.
stochastic biological events or even the chance analysis of
a very rare cell can create noise which may confound ob-
servations. Hence, single-cell methylation profiling must FUNDING
retain the capacity to perform population statistics This work was supported by A.S. and by Beckman, which
(therefore a need for high-throughput). Although the had no role in study design, data collection and analysis,
general level of cell-to-cell epigenetic variation in decision to publish, or preparation of the manuscript.
mammals is still largely unknown, it can be expected Funding for open access charge: XXX
that the cell’s epigenetic machinery is inherently noisy. It
could be shown that significant differences in gene expres- Conflict of interest statement. M.A.-F., P.H., M.K. and
sion levels exist between phenotypically identical cells R.K. are employees of Beckman Coulter Biomedical
in vivo, and that these differences exceed any noise contri- GmbH.
bution from global mRNA amplification (34). Due to the
highly parallelized methodology, RSMA also offers the
REFERENCES
opportunity to study stochastic influences in methylation
dynamics, an important component of epigenetic drift 1. Schumacher,A. and Petronis,A. (2006) Epigenetics of complex
(1,35,36). Indeed, it was suggested that a substantial diseases: from general theory to laboratory experiments.
Curr. Top. Microbiol. Immunol., 310, 81–115.
portion of phenotypic variance in disease and aging, trad- 2. Le,T.T., Harlepp,S., Guet,C.C., Dittmar,K., Emonet,T., Pan,T.
itionally attributed to environmental effects, may in fact and Cluzel,P. (2005) Real-time RNA profiling within a single
result from stochastic epigenetic events in the cell (1). By bacterium. Proc. Natl Acad. Sci. USA, 102, 9160–9164.
studying methylation patterns in multiple single cells (up 3. Sims,C.E. and Allbritton,N.L. (2007) Analysis of single
to 48 cells per slide) under controlled environmental con- mammalian cells on-chip. Lab Chip, 7, 423–440.
4. Cluzel,P., Surette,M. and Leibler,S. (2000) An ultrasensitive
ditions, quantification of stochastic events becomes bacterial motor revealed by monitoring signaling proteins in
possible. The typical channel dimensions found in single cells. Science, 287, 1652–1655.
microfluidic devices (10–100 mm) and the ability to ma- 5. Wilkinson,G.R. (2005) Drug metabolism and variability among
nipulate nanoliters of reagents on-chip have made these patients in drug response. N. Engl. J. Med., 352, 2211–2221.
6. Lee,Y.H., Chen,S.Y., Wiesner,R.J. and Huang,Y.F. (2004) Simple
devices encouraging platforms for the analysis of single flow cytometric method used to assess lipid accumulation in fat
cells. However, most of these systems do not possess a cells. J. Lipid Res., 45, 1162–1167.
high-throughput capacity or they are too expensive for 7. Kerjean,A., Vieillefond,A., Thiounn,N., Sibony,M., Jeanpierre,M.
the daily use in most laboratories. With our proposed and Jouannet,P. (2001) Bisulfite genomic sequencing of
microdissected cells. Nucleic Acids Res., 29, E106–E106.
system, high-throughput as well as cost efficiency can be 8. Easley,C.J., Karlinsey,J.M., Bienvenue,J.M., Legendre,L.A.,
ensured. Roper,M.G., Feldman,S.H., Hughes,M.A., Hewlett,E.L.,
Merkel,T.J., Ferrance,J.P. et al. (2006) A fully integrated
microfluidic genetic analysis system with sample-in-answer-out
capability. Proc. Natl Acad. Sci. USA, 103, 19272–19277.
CONCLUSIONS 9. Toriello,N.M., Douglas,E.S., Thaitrong,N., Hsiao,S.C.,
In summary, we demonstrated an affordable technology Francis,M.B., Bertozzi,C.R. and Mathies,R.A. (2008) Integrated
microfluidic bioprocessor for single-cell gene expression analysis.
to assess the methylation patterns of CpG dinucleotides Proc. Natl Acad. Sci. USA, 105, 20173–20178.
in single cells. RSMA offers the opportunity for 10. Schumacher,A., Kapranov,P., Kaminsky,Z., Flanagan,J.,
high-throughput screening, as many single cells can be Assadzadeh,A., Yau,P., Virtanen,C., Winegarden,N., Cheng,J.,
 Nucleic Acids Research, 2011 9

Gingeras,T. et al. (2006) Microarray-based DNA methylation 23. Herman,J.G., Merlo,A., Mao,L., Lapidus,R.G., Issa,J.P.,
profiling: technology and applications. Nucleic Acids Res., 34, Davidson,N.E., Sidransky,D. and Baylin,S.B. (1995) Inactivation
528–542. of the CDKN2/p16/MTS1 gene is frequently associated with
11. Leibovitz,A., Stinson,J.C., McCombs,W.B. III, McCoy,C.E., aberrant DNA methylation in all common human cancers.
Mazur,K.C. and Mabry,N.D. (1976) Classification of human Cancer Res., 55, 4525–4530.
colorectal adenocarcinoma cell lines. Cancer Res., 36, 4562–4569. 24. Chim,C.S., Fung,T.K., Wong,K.F., Lau,J.S., Law,M. and
12. Clark,S.J., Harrison,J., Paul,C.L. and Frommer,M. (1994) High Liang,R. (2006) Methylation of INK4 and CIP/KIP families of
sensitivity mapping of methylated cytosines. Nucleic Acids Res., cyclin-dependent kinase inhibitor in chronic lymphocytic
22, 2990–2997. leukaemia in Chinese patients. J. Clin. Pathol., 59, 921–926.
13. Mill,J., Tang,T., Kaminsky,Z., Khare,T., Yazdanpanah,S., 25. Nagy,E., Veress,G., Szarka,K., Csoma,E. and Beck,Z. (2005)
Bouchard,L., Jia,P., Assadzadeh,A., Flanagan,J., Schumacher,A. Frequent methylation of p16INK4A/p14ARF promoters in
et al. (2008) Epigenomic profiling reveals DNA-methylation tumorigenesis of Epstein–Barr virus transformed lymphoblastoid
changes associated with major psychosis. Am. J. Hum. Genet., 82, cell lines. Anticancer Res., 25, 2153–2160.
696–711. 26. Kovatsi,L., Georgiou,E., Ioannou,A., Haitoglou,C.,
14. May,A., Kirchner,R., Muller,H., Hartmann,P., El Hajj,N., Tzimagiorgis,G., Tsoukali,H. and Kouidou,S. (2010) p16
Tresch,A., Zechner,U., Mann,W. and Haaf,T. (2009) Multiplex promoter methylation in Pb2+-exposed individuals. Clin. Toxicol.
rt-PCR expression analysis of developmentally important genes (Phila), 48, 124–128.
in individual mouse preimplantation embryos and blastomeres. 27. Gupta,P.B., Chaffer,C.L. and Weinberg,R.A. (2009) Cancer stem
Biol. Reprod., 80, 194–202. cells: mirage or reality? Nat. Med., 15, 1010–1012.
15. Zuker,M. (2003) Mfold web server for nucleic acid folding and 28. Rosen,J.M. and Jordan,C.T. (2009) The increasing complexity of
hybridization prediction. Nucleic Acids Res., 31, 3406–3415. the cancer stem cell paradigm. Science, 324, 1670–1673.
16. Lister,R. and Ecker,J.R. (2009) Finding the fifth base: 29. Ernst,C., McGowan,P.O., Deleva,V., Meaney,M.J., Szyf,M. and

Downloaded from [Link] by guest on February 9, 2011

genome-wide sequencing of cytosine methylation. Genome Res., Turecki,G. (2008) The effects of pH on DNA methylation state:
19, 959–966. in vitro and post-mortem brain studies. J. Neurosci. Methods, 174,
17. Chiba,T., Yokosuka,O., Fukai,K., Hirasawa,Y., Tada,M., 123–125.
Mikata,R., Imazeki,F., Taniguchi,H., Iwama,A., Miyazaki,M. 30. Humpherys,D., Eggan,K., Akutsu,H., Hochedlinger,K.,
et al. (2005) Identification and investigation of methylated Rideout,W.M. III, Biniszkiewicz,D., Yanagimachi,R. and
genes in hepatoma. Eur. J. Cancer, 41, 1185–1194. Jaenisch,R. (2001) Epigenetic instability in ES cells and cloned
18. Mori,K., Enokida,H., Kagara,I., Kawakami,K., Chiyomaru,T., mice. Science, 293, 95–97.
Tatarano,S., Kawahara,K., Nishiyama,K., Seki,N. and 31. Mann,M.R., Chung,Y.G., Nolen,L.D., Verona,R.I., Latham,K.E.
Nakagawa,M. (2009) CpG hypermethylation of collagen type I and Bartolomei,M.S. (2003) Disruption of imprinted gene
alpha 2 contributes to proliferation and migration activity of methylation and expression in cloned preimplantation stage
human bladder cancer. Int. J. Oncol., 34, 1593–1602. mouse embryos. Biol. Reprod., 69, 902–914.
19. Merlo,A., Herman,J.G., Mao,L., Lee,D.J., Gabrielson,E., 32. Schumacher,A. and Doerfler,W. (2004) Influence of in vitro
Burger,P.C., Baylin,S.B. and Sidransky,D. (1995) 50 CpG island manipulation on the stability of methylation patterns in the
methylation is associated with transcriptional silencing of the Snurf/Snrpn-imprinting region in mouse embryonic stem cells.
tumour suppressor p16/CDKN2/MTS1 in human cancers. Nat. Nucleic Acids Res., 32, 1566–1576.
Med., 1, 686–692. 33. Kangaspeska,S., Stride,B., Metivier,R., Polycarpou-Schwarz,M.,
20. Mund,C., Beier,V., Bewerunge,P., Dahms,M., Lyko,F. and Ibberson,D., Carmouche,R.P., Benes,V., Gannon,F. and Reid,G.
Hoheisel,J.D. (2005) Array-based analysis of genomic DNA (2008) Transient cyclical methylation of promoter DNA. Nature,
methylation patterns of the tumour suppressor gene p16INK4A 452, 112–115.
promoter in colon carcinoma cell lines. Nucleic Acids Res., 33, 34. Subkhankulova,T., Gilchrist,M.J. and Livesey,F.J. (2008)
e73. Modelling and measuring single cell RNA expression levels find
21. Zheng,S., Chen,P., McMillan,A., Lafuente,A., Lafuente,M.J., considerable transcriptional differences among phenotypically
Ballesta,A., Trias,M. and Wiencke,J.K. (2000) Correlations of identical cells. BMC Genomics, 9, 268.
partial and extensive methylation at the p14(ARF) locus with 35. Fraga,M.F., Ballestar,E., Paz,M.F., Ropero,S., Setien,F.,
reduced mRNA expression in colorectal cancer cell lines and Ballestar,M.L., Heine-Suner,D., Cigudosa,J.C., Urioste,M.,
clinicopathological features in primary tumors. Carcinogenesis, 21, Benitez,J. et al. (2005) Epigenetic differences arise during the
2057–2064. lifetime of monozygotic twins. Proc. Natl Acad. Sci. USA, 102,
22. Sengupta,P.K., Smith,E.M., Kim,K., Murnane,M.J. and 10604–10609.
Smith,B.D. (2003) DNA hypermethylation near the transcription 36. Wang,S.-C., Oelze,B. and Schumacher,A. (2008) Age-specific
start site of collagen alpha2(I) gene occurs in both cancer cell epigenetic drift in late-onset Alzheimer’s disease. PLoS ONE, 3,
lines and primary colorectal cancers. Cancer Res., 63, 1789–1797. e2698.