© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
TECHNICAL REPORTS
Replacing PCR with COLD-PCR enriches variant
DNA sequences and redefines the sensitivity of
genetic testing
Jin Li1, Lilin Wang1, Harvey Mamon1, Matthew H Kulke2, Ross Berbeco1 & G Mike Makrigiorgos1
PCR is widely employed as the initial DNA amplification step
for genetic testing. However, a key limitation of PCR-based
methods is the inability to selectively amplify low levels of
mutations in a wild-type background. As a result, downstream
assays are limited in their ability to identify subtle genetic
changes that can have a profound impact in clinical decisionmaking and outcome. Here we describe co-amplification at
lower denaturation temperature PCR (COLD-PCR), a novel form
of PCR that amplifies minority alleles selectively from mixtures
of wild-type and mutation-containing sequences irrespective of
the mutation type or position on the sequence. We replaced
regular PCR with COLD-PCR before sequencing or genotyping
assays to improve mutation detection sensitivity by up to
100-fold and identified new mutations in the genes encoding
p53, KRAS and epidermal growth factor in heterogeneous
cancer samples that had been missed by the currently used
methods. For clinically relevant microdeletions, COLD-PCR
enabled exclusive amplification and isolation of the mutants.
COLD-PCR will transform the capabilities of PCR-based genetic
testing, including applications in cancer, infectious diseases
and prenatal identification of fetal alleles in maternal blood.
PCR has become the cornerstone of molecular diagnosis, with almost
every genetic test that aims to identify DNA sequence variation
incorporating PCR. As commonly applied, PCR does not contain an
inherent selectivity toward variant (mutant) alleles, thus both variant
and nonvariant alleles are amplified with approximately equal efficiency. The burden of identifying and sequencing a mutation in a PCR
product falls on downstream assays such as Sanger sequencing,
pyrosequencing, matrix-assisted laser desorption/ionization time-offlight (MALDI-TOF) mass spectrometry, restriction fragment length
polymorphism (RFLP) assay, denaturing HPLC and others. Although
it is reliable for screening germline or prevalent somatic mutations,
sequencing of unknown low-prevalence mutations using these otherwise powerful technologies is still problematic. The importance of
identifying these mutations, however, is crucial in several fields
of medicine, including cancer, prenatal diagnosis and infectious
diseases1–3. For example, in cancer, low-level mutations in the gene
encoding epidermal growth factor (EGFR) that cannot be sequenced
by regular means can confer either positive therapeutic response to
tyrosine kinase inhibitors4 or drug resistance5,6. Mutations in plasmacirculating DNA that are useful as biomarkers for early detection7 or
tumor response to treatment8 cannot be sequenced using conventional
methods. Mutations in tumors with frequent stromal contamination,
such as pancreatic, lung or prostate tumors, can be ‘masked’ by the
presence of wild-type alleles5,9, thus requiring microdissection, or the
mutations can be missed altogether.
COLD-PCR is a new form of PCR that preferentially enriches
‘minority alleles’ from mixtures of wild-type and mutation-containing
sequences, irrespective of where an unknown mutation lies. Consequently, COLD-PCR amplification of genomic DNA yields PCR
products containing high percentages of variant alleles, thus permitting their detection. Because PCR is a common initial step in almost
all forms of genetic analysis, COLD-PCR provides a general
platform to improve the sensitivity of essentially all DNA variation–
detection technologies.
A single-nucleotide mismatch anywhere along a double-stranded
DNA sequence generates a small but predictable change to the melting
temperature (Tm) for that sequence10,11. Depending on the sequence
context and position of the mismatch, melting Tm changes of
0.2–1.5 1C are common for sequences up to 200 base pairs (bp) long
or longer10,11. We observed that for each DNA sequence, there is a
critical denaturation temperature (Tc) lower than the Tm below which
PCR efficiency drops abruptly and that the Tc is strongly dependent on
DNA sequence. DNA amplicons differing by a single nucleotide
reproducibly have different amplification efficiencies when PCR denaturation temperature is set to Tc. This observation can be exploited
during PCR amplification for selective enrichment of minority alleles
differing by one or more nucleotides at any position of a sequence. In
COLD-PCR, an intermediate annealing temperature is used during
PCR cycling to allow cross-hybridization of mutant and wild-type
alleles; heteroduplexes, which melt at lower temperatures than homoduplexes, are then selectively denatured and amplified at Tc, whereas
homoduplexes remain double-stranded and do not amplify efficiently.
1Department of Radiation Oncology, Divisions of Genomic Stability and DNA Repair, Physics and Radiation Therapy and 2Department of Medical Oncology, Dana Farber
Cancer Institute, Harvard Medical School, Room JF514, 44 Binney Street, Boston, Massachusetts 02115, USA. Correspondence should be addressed to G.M.M.
(mmakrigiorgos@partners.org).
Received 4 September 2007; accepted 11 December 2007; published online 13 April 2008; doi:10.1038/nm1708
NATURE MEDICINE
VOLUME 14
[
NUMBER 5
[
MAY 2008
579
TECHNICAL REPORTS
Figure 1 Description of COLD-PCR for an
Selectively denature
Denature DNA
example 167-bp TP53 sequence. Two forms of
the mismatched sequences
Reduce temperature:
at critical temperature (Tc ~86.5 °C)
cross-hybridize
COLD-PCR are described; full COLD-PCR and
94 °C
mutant with
Mutant
fast COLD-PCR, which are used depending on
wild-type sequences
Tc = 86.5 °C
whether it is important to identify all possible
mutations or to identify the Tm-reducing
mutations that are frequent in cancer. (a) In full
70 °C
Reduce temperature
Wild-type
for primer annealing
COLD-PCR (for enrichment of all mutations),
after a number of regular PCR cycles that enable
55 °C
Return to
an initial buildup of target amplicon(s), the PCR
Extend at 72 °C
94 °C
program is switched to the one shown. After
Preferentially copy the mutated sequences
denaturation at 94 1C, the PCR amplicon(s) are
allowed to cross-hybridize at an intermediate
Selectively denature
Denature DNA
mutant sequences
temperature (for example, 70 1C for 2–8 min).
at critical temperature (Tc ~86.5 °C)
94 °C
Because mutant alleles are in the minority, most
Mutant
mutant alleles end up in a mismatch-containing
Tc = 86.5 °C
structure (heteroduplex) that has a lower melting
temperature than the fully matched structure
at 72 °C
Extend
(homoduplex). Next, the PCR temperature is
Wild-type
55 °C
Reduce temperature
raised to the Tc to denature the mismatchfor primer annealing
containing sequences preferentially over the fully
matched sequences. Finally, the temperature is
reduced to 55 1C to allow primers to bind and
Preferentially copy the mutated sequences
prime replication of the preferentially denatured
sequences. Because this crucial denaturation is performed at every PCR cycle, the differential enrichment of mutation-containing alleles is compounded
exponentially, and results in a large difference in overall amplification efficiency between mutant and wild-type alleles at the end of the cycling. (b) In fast
COLD-PCR (for enrichment of Tm-reducing mutations), preferential amplification of mutations via COLD-PCR is so pronounced that, for the majority of point
mutations, mutant enrichment occurs even without performing the intermediate cross-hybridization step at 70 1C. Thus, rapid PCR amplification performed
at the Tc instead of at 94 1C discriminates strongly toward the lower-Tm allele. Fast COLD-PCR is rapid and results in higher enrichments than full COLDPCR. However, to enrich for all possible mutations, including deletions or insertions, the full COLD-PCR program described in Figure 1a is necessary.
a
© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
b
By fixing the denaturation temperature to Tc, mutations at any
position along the sequence are enriched during COLD-PCR amplification. COLD-PCR can be applied in two formats, ‘full’ COLD-PCR
and ‘fast’ COLD-PCR, depending on whether it is more important to
identify all possible mutations or achieve the highest mutation
enrichment. The operation of full and fast COLD-PCR for an example
167-bp sequence is described below. Enrichment of mutations via
COLD-PCR is most efficient for relatively short (o200-bp) sequences,
presumably because single-base mismatches along the sequence of a
small DNA fragment have a larger impact on Tm. Below we show
enrichment of mutation-containing DNA of approximately one order
of magnitude for sequences of up to 210 bp and enrichment of up to
two orders of magnitude for o100-bp sequences.
RESULTS
Improvement of enzymatic mutation detection via COLD-PCR
To validate COLD-PCR, we used serial dilutions of DNA from tumorderived cell lines containing mutations or microdeletions at different
positions along TP53 exon 8 and KRAS exon 1 (nucleotides in codons
12 and 13) or samples with microdeletions in EGFR exon
19 (Supplementary Fig. 1 online). In addition, genomic DNA from
a series of colon and lung cancer surgical specimens and plasmacirculating DNA was used for validation. COLD-PCR can be applied
directly from genomic DNA. However, for convenience, in most
experiments we adopted a nested PCR format. First, we amplified a
larger DNA sequence using a high-fidelity polymerase, and then we
applied nested COLD-PCR in separate, smaller fragments to enrich
for mutations. Thereby, fragments of 87 bp, 98 bp, 129 bp, 167 bp and
210 bp from the TP53, KRAS and EGFR genes were tested by COLDPCR. For each amplicon, a single Tc was used to enrich for mutations
at any position.
For an evaluation of the mutation enrichment obtained via COLDPCR, mutation-containing DNA was mixed with wild-type DNA and
580
amplified using the full COLD-PCR program (Fig. 1a). The PCR
products were digested with restriction enzymes that selectively
recognize either the mutation-containing DNA or the wild-type
DNA. When no restriction enzyme that distinguishes between mutant
and wild-type DNA was available, the Surveyor nuclease that digests
mismatch-containing DNA was used instead9,12. The digested products were then examined via denaturing HPLC. For comparison to
COLD-PCR, identical experiments were conducted by regular PCR.
Supplementary Figure 2 online depicts representative results from
this study. A TP53 exon 8 fragment from SW480 cell line DNA that
forms an NlaIII digestion site when a G-A mutation is present at
nucleotide position 14487 in codon 273 was enriched more than
tenfold from a mixture of mutated and wild-type DNA by full COLDPCR (Fig. 1a). The enrichment enables clear detection of the mutation
after NlaIII digestion. In contrast, when regular PCR was applied and
the digestion repeated, the mutation was barely detectable at the
dilution used (Supplementary Fig. 2a). Wild-type DNA showed no
mutation in either method. A wide range of additional TP53 and
KRAS mutations, including mutations that increase (A-G, T-G),
retain (G-C) or decrease (G-A, C-T, G-T) the amplicon Tm,
were similarly enriched, indicating that all nucleotide substitutions can
be enriched by full COLD-PCR. The fast COLD-PCR protocol
(Fig. 1b) was similarly evaluated for nucleotide substitutions that
decrease amplicon Tm. Ten- to fifteenfold mutation enrichments were
obtained with fast COLD-PCR (Supplementary Fig. 2f). When, we
applied COLD-PCR directly from genomic DNA instead of using a
nested PCR format, we obtained 11–22-fold mutation enrichment
(Supplementary Fig. 2g). A more pronounced mutation enrichment
was obtained by fast COLD-PCR than by full COLD-PCR for the
same (C-T) mutation. When a genomic DNA mixture with three
distinct mutations within a single amplicon was amplified with fast
COLD-PCR directly from the genomic DNA, the amplification
resulted in simultaneous enrichment of all 3 mutations, showing the
VOLUME 14
[
NUMBER 5
[
MAY 2008
NATURE MEDICINE
TECHNICAL REPORTS
Regular PCR 167 bp TP53 seq.
Dilution 1:10 SW480 (homozyg.) into wild-type
Reverse strand
Reverse strand
G→A mutation not visible
~10% mutation
Dilution 1:10 CT7 into wild-type
C→T mutation not visible
~5% mutation
G→A mutation not visible
5% mutation
Dilution 1:20 HCC2218 into wild-type
C→T mutation not visible
5% mutation
Dilution 1:10 MDA-MB231 into wild-type
G→A mutation visible
~30% mutation
Dilution 1:20 HCC2218 into wild-type
C→T mutation visible
~65% mutation
Forward strand
© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
Dilution 1:10 SW480 into wild-type
G→A mutation visible
(sample becomes heterozygous)
~70% mutation
Dilution 1:10 CT7 into wild-type
C→T mutation visible
~60% mutation
Dilution 1:10 MDA (heterozyg.) into wild-type
Reverse strand
COLD-PCR
Wild-type sample: no mutations
Wild-type sample: no mutations
Reverse strand
multiplex enrichment of mutations by COLD-PCR. Overall, the data
indicate that an improvement of one order of magnitude (range,
6–22-fold) in the sensitivity of RFLP (using restriction enzymes for
specific mutations) and unknown mutation scanning (using Surveyor
nuclease) is achieved by replacing regular PCR with COLD-PCR
(Supplementary Fig. 2).
Improvement of sequencing technologies via COLD-PCR
To examine the impact of mutation enrichment on the sensitivity of
sequencing technologies, PCR products obtained by COLD-PCR or
regular PCR were processed for Sanger dideoxy-terminator sequencing
or, alternatively, for pyrosequencing. When fast COLD-PCR at a Tc of
86.5 1C was applied to the 167-bp exon 8 fragment of TP53, the
enrichment was evident for all TP53 mutations tested (Fig. 2). For
example, the C-T mutation in HCC2218 cells that is initially diluted
down to a 5% mutant-to-wild-type ratio in wild-type DNA becomes
enriched to a B65% mutant-to-wild-type ratio after COLD-PCR,
as estimated by observation of the sequencing chromatograms
(that is, an enrichment of the mutation by B13-fold) (Fig. 2). The
wild-type TP53 sample amplified via COLD-PCR showed no
mutation (Fig. 2).
Overall, for all of the TP53 mutations studied for this 167-bp
fragment (Supplementary Fig. 1), the mutation enrichment varied by
B5–13-fold depending on the mutation position. The mutation
enrichment increased further upon amplification of a shorter
(87-bp) COLD-PCR amplicon from the same TP53 segment. Thus,
addition of 1% homozygous G-A mutant DNA to 99% wild-type
sample results in a ‘heterozygous’ sequencing chromatograph, indicating a B50-fold enrichment of the mutation during COLD-PCR
(Supplementary Fig. 3a online). The mutation enrichment is relatively smaller (B5-fold) when a longer (210-bp) amplicon is examined, although it is still clearly evident (Supplementary Fig. 3b). In
addition to examining mutations spanning TP53 exon 8, we studied
KRAS mutations representing commonly encountered nucleotide
changes in codons 12 and 13 by comparing sequencing chromatographs obtained after COLD-PCR or regular PCR (Supplementary
Fig. 4 online). Similarly to the results obtained with TP53, enrichments of 5–12-fold were obtained, depending on the mutation
studied, and the sequencing chromatographs revealed low-prevalence
mutations that were undetectable with regular PCR. It is noteworthy
that, for the conditions and sequences applied, polymerase substitution errors were not evident after COLD-PCR enrichment.
NATURE MEDICINE
VOLUME 14
[
NUMBER 5
[
MAY 2008
Figure 2 COLD-PCR improves the sensitivity of Sanger dideoxy-terminator
sequencing. DNA containing mutations in TP53 exon 8 (SW480 cell line,
genome nucleotide position 14487 G-A; colon cancer genomic DNA 7,
14486 C-T; MDA-MB231 cell line, 14508 G-A; HCC2218 cell line;
14516 C-T) was diluted in wild-type DNA and amplified by COLD-PCR at
a single critical denaturation temperature (Tc ¼ 86.5 1C) to obtain a 167-bp
PCR product. Sanger dideoxy-terminator sequencing using the reverse or
forward primer was applied to the PCR products. Regular PCR–sequencing
was also performed on the same samples for comparison. Mutations that
were not detectable after regular PCR become clearly detectable after
COLD-PCR. Mutation enrichments of 7-fold, 12-fold, 6-fold and 13-fold for
SW480, CT7, MDA-MB231 and HCC2218, respectively, were obtained via
COLD-PCR compared to regular PCR. The wild-type sample did not contain
mutations, as detected by either method of amplification.
To evaluate the clinical implications of improving the sensitivity of
Sanger sequencing, we examined DNA from 18 colon and 25 lung
cancer surgical specimens for TP53 or KRAS mutations with either
regular PCR or COLD-PCR. All mutations identified via regular PCR–
Sanger sequencing were also identified via COLD-PCR–Sanger
sequencing. However, COLD-PCR-sequencing also identified three
samples with low-prevalence mutations that were missed by regular
PCR–sequencing. In two clinical samples, TL64 and CT20,
low-prevalence G-A mutations were detected via COLD-PCR–
sequencing of TP53 exon 8 within codon 273 (a ‘hotspot’ for TP53
mutations; Fig. 3). We also examined DNA from the plasmacirculating DNA of individuals with cancer who were undergoing
radiation therapy. In a sample from a subject with colon cancer, a
TP53 exon 8 (G-A) mutation was detected by COLD-PCR–Sanger
sequencing (Supplementary Fig. 5 online) but not by regular PCR–
Sanger sequencing. Finally, TP53 (C-T) mutations that had been
missed by regular sequencing were also revealed by COLD-PCR using
DNA obtained from a formalin-fixed, paraffin-embedded (FFPE)
specimen obtained from a subject with non–small cell lung cancer
(Supplementary Fig. 5). The bottom chromatograph in Supplementary Figure 5 shows the detection of G-A mutations within KRAS
codon 12 at nucleotide position 35 in another FFPE sample obtained
from a subject with lung cancer that had been missed after regular
167 bp TP53 exon 8 sequence
REGULAR PCR
(94 °C denaturation)
Lung tumor 64
Reverse seq.
No mutation visible
Forward seq.
No mutation visible
Colon tumor 20
No mutation visible
Reverse seq.
COLD-PCR
(denaturation at Tc = 86.5 °C)
Lung tumor 64
Reverse seq.
G→A mutation visible
Independent verification
via RFLP
G→A mutation verified
Reverse seq.
Colon tumor 20
G→A mutation visible
Independent verification
via RFLP
G→A mutation verified
Figure 3 Examples of low-level mutations in solid tumor clinical samples,
previously ‘invisible’ in Sanger dideoxy sequencing, that now become
detectable via COLD-PCR. Sanger sequencing of a 167-bp fragment from
TP53 exon 8 from clinical tumor samples 64 and 20 was performed after
either COLD-PCR or regular PCR. Lung tumor 64 and colon tumor 20 were
shown to contain TP53 nucleotide 14487 (codon 273) mutations that
were not detectable with regular PCR. Independent verification of the
low-prevalence somatic mutations in these clinical samples was done via
RFLP-PCR as described9 and sequenced using the reverse primer.
581
TECHNICAL REPORTS
a
KRAS-A549:w.type = 1:33 (GGT
Mutation
not visible
AGT)
b
KRAS-FFPE-6 (GGT
Mutation
not visible
Regular PCR
c
GAT)
Mutation not visible
Regular PCR
Regular PCR
TP53 exon 8
Lung tumor 64
C
G
A
G
C
5
T
G
Mutation
clearly visible
A
T
C
10
C
G
A
G
C
5
T
G
Mutation
clearly visible
COLD-PCR
A
T
A C A G C T G A G T G C G A G T
C
10
Codon 273 TP53 mutation
Mutation clearly visible
COLD-PCR
COLD-PCR
© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
C
G
A
G
C
5
T
G
A
T
C
10
C
G
A
G
C
5
T
G
A
T
C
10
A C A G C T G A G T G C G A G T
Figure 4 COLD-PCR improves the sensitivity of pyrosequencing. (a) DNA from cell line A549 was diluted 33-fold into wild-type DNA, and a 98-bp KRAS exon
2 segment was amplified via COLD-PCR (Tc ¼ 80 1C) or regular PCR, followed by pyrosequencing. The G-A mutation of the A549 cell line was only visible
when COLD-PCR was applied. (b) DNA from FFPE tumor 6 was screened similarly to KRAS nucleotide 35 (within codon 12) mutations. The G-A mutation
was detectable after COLD-PCR–pyrosequencing but missed after regular PCR. (c) DNAs from lung tumor 64 were amplified with COLD-PCR (Tc ¼ 83.5 1C) or,
alternatively, with regular PCR to obtain an 87-bp TP53 exon 8 PCR product that was processed for pyrosequencing. Lung tumor 64 showed a G-A mutation
at TP53 nucleotide 14487 (within codon 273) that was only detectable when COLD-PCR was performed before pyrosequencing and missed via regular
PCR–pyrosequencing.
PCR. The mutations identified via COLD-PCR were subsequently
independently verified from genomic DNA via RFLP9.
Pyrosequencing is another sequencing technology that could benefit
from incorporating COLD-PCR. The current detection limit of
pyrosequencing is B10% mutant/wild-type ratio13. We repeated the
comparison of regular PCR to COLD-PCR via pyrosequencing for
several of the mutations in TP53 exon 8 and KRAS codons 12 and 13
that we had also examined via Sanger sequencing. Serial dilutions of
DNA from mutant cell lines indicated that COLD-PCR–pyrosequencing can identify mutations down to a prevalence of 0.5–1% for the
mutations examined, and mutation enrichments of 5–35-fold were
obtained compared to regular PCR-pyrosequencing. For example,
screening of DNA from cell line A549, clinical samples TL64 and
FFPE lung tumor 6 showed that COLD-PCR–pyrosequencing reveals
TP53 position 14487 and KRAS mutations that had been missed by
regular PCR–pyrosequencing (Fig. 4). TP53 exon 8 mutations within
codon 273 have been associated with bad prognosis in lung cancer14,15
and can also serve as biomarkers in plasma-circulating DNA16–20.
Similarly, KRAS mutations within codon 12 have prognostic implications in lung adenocarcinoma21. One of the principal weaknesses of
conventional sequencing technologies has been that they may miss
clinically relevant somatic mutations in heterogeneous samples22. By
replacing regular PCR with COLD-PCR, this weakness is alleviated, as
both Sanger sequencing and pyrosequencing can now identify lowprevalence somatic mutations in cancer samples.
Improvement of genotyping technologies via COLD-PCR
High-throughput genotyping technologies such as MALDI-TOF have
been proposed for the detection of somatic mutations in clinical
samples23. The current sensitivity limit of MALDI-TOF for low-level
mutation detection is B5–10% (ref. 23). However, the prevalence of
mutations in heterogeneous tumor specimens or in DNA from plasma
can be on the order of 5% or lower9, which generates questions about
the reliability of MALDI-TOF in these cases. By applying the same
approach and samples used for Sanger sequencing to MALDI-TOF, we
obtained mutation enrichments of 10–100-fold by replacing regular
PCR with fast COLD-PCR before MALDI-TOF; somatic mutations
could be detected at a prevalence of 0.1–0.5% (Supplementary Fig. 6
online). The mutation enrichment caused by fast COLD-PCR was
higher for samples containing the lowest mutant/wild-type ratios.
When we screened clinical samples containing low-prevalence somatic
582
mutations with COLD-PCR-MALDI-TOF, TP53 exon 8 mutations in
sample TL64 and in plasma-circulating DNA were detectable (Supplementary Fig. 6). With regular PCR–MALDI-TOF, the samples in
these clinical samples were barely detected or completely missed
(Supplementary Fig. 6).
Selection of unknown deletions using COLD-PCR
Microdeletions (3–20 bp) of a known or variable sequence are an
important group of mutations to identify. For example, 5–15-bp
microdeletions occurring at EGFR exon 19 confer a positive response
to tyrosine kinase inhibitors in lung adenocarcinoma4. When we used
two consecutive rounds of full COLD-PCR to amplify a 129-bp EGFR
exon 19 segment from a lung adenocarcinoma cell line with an
established deletion, we obtained an overall deletion enrichment of
450-fold, and we could sequence the deletion/wild-type ratio of
1:300, as opposed to B1:6 by regular PCR sequencing (Supplementary Fig. 7a online).
We expected that application of fast COLD-PCR for detection of
microdeletions that reduce Tm would also lead to high enrichment of
deletions. For validation, we engineered TP53 exon 8 microdeletions
of 7 bp or 3 bp (present in human neuroblastoma cell line TGW) by
PCR mutagenesis of the GC-rich region of TP53 exon 8 (see Tm graph
in Supplementary Fig. 1) and performed fast COLD-PCR at the same
critical denaturation temperature used for single-base mismatches,
Tc ¼ 83.5 1C. The data indicate that, for both 3- and 7-bp deletions,
COLD-PCR amplifies the deletion-containing sequence exclusively,
essentially leading to isolation of deletion mutant sequences in a single
COLD-PCR reaction (Supplementary Fig. 7b and Supplementary
Fig. 8 online). Deletion enrichments of at least 300-fold were
obtained, and deletion/wild-type ratios of 1:3,000 could be directly
sequenced (Supplementary Fig. 7c).
DISCUSSION
Because clinical decisions will increasingly rely on molecular tumor
profiling in the forthcoming era of personalized molecular medicine,
the reliability of identifying somatic mutations in diverse clinical
specimens, including heterogeneous tumors and bodily fluids, must
be high. In agreement with earlier reports5,9,22, our data indicate the
deficiencies of Sanger sequencing, as well as of pyrosequencing and of
MALDI-TOF, in detecting low-prevalence mutations in mixed tumor
samples. Four of forty-three surgical samples and three of ten plasma
VOLUME 14
[
NUMBER 5
[
MAY 2008
NATURE MEDICINE
TECHNICAL REPORTS
Table 1 Mutation prevalence for various types of somatic mutations in human cancer and mutation enrichment anticipated via COLD-PCR
Prevalence of somatic mutation in cancer27
COLD-PCR
COLD-PCR
Enrichment
Mutation
© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
C:G-T:Aa
C:G-A:Ta
Lung
(%)
Colon
(%)
Breast
(%)
Melanoma
(%)
Glioma
(%)
Enrichment (full COLD-PCR)e,
all mutations
(fast COLD-PCR)e,
Tm-decreasing mutations
37
29
78
6
37
15
92
2
97
1
5–12-fold
10–100-fold
5–8-fold
None
3–5-fold
None
450-fold one or more rounds
4100-fold
T:A-A:Tb
4
2
1
2
0
C:G-G:Cb
15
4
36
1
0
T:A-G:Cc
3
2
4
B0
0
T:A-C:Gc
6
8
3
2
0
Microdeletions and insertionsd
6
B0
3
1
2
aThese
bThese
mutations generally reduce the Tm of a DNA sequence.
mutations generally retain the Tm of a DNA sequence.
mutations generally increase the Tm of a DNA sequence. dThese
mutations may increase, retain or decrease the Tm of a DNA sequence. eThe enrichment is defined as the fold-increase of the prevalence of a mutation relative to performing regular PCR.
samples tested contained clinically important TP53, KRAS or EGFR
mutations that were not detected by these methods when preceded by
regular PCR. Replacement of regular PCR with COLD-PCR provides a
‘universal boost’ to these widely used technologies and enables them
to be used with the required confidence in routine screening of cancer
specimens for somatic mutations, including low-level mutation
screening of surgical and FFPE tumor samples or bodily fluids.
Single-molecule ‘deep’ sequencing (SMS) promises to mitigate, in
part, the problem of heterogeneity in tumor specimens22. However,
when mutations at the 1–5% level or less need to be detected in these
PCR amplicons, the high-throughput capabilities of SMS decline
rapidly owing to limitations imposed by statistics22. Accordingly,
replacement of regular PCR with COLD-PCR, either before or during
SMS, would increase the efficiency of the technology in proportion to
the mutation enrichment obtained.
Enrichment of mutated sequences during PCR at a single-sequence
position via peptide nucleic acids or thermostable restriction enzymes
has been reported before24,25. However, COLD-PCR enriches mutations at all sequence positions simultaneously without requiring
addition of new reagents, hence it can be used to improve unknown
mutation–detection technologies, as well as genotyping at specific
positions. PCR is routinely performed before genetic testing. Therefore, replacing PCR with COLD-PCR that increases the mutation
detection limits does not entail substantial additional investment. The
large majority of mutations encountered in human tumors are
amenable to enrichment by fast COLD-PCR, whereas full COLDPCR applies to 100% of mutations (Table 1). However, the full
COLD-PCR protocol requires the buildup of substantial PCR product
for mutation enrichment to occur to achieve efficient cross-hybridization (Fig. 1a), which restricts the enrichment to the late stages of PCR.
In contrast, for fast COLD-PCR, there is no requirement for PCR
product buildup, hence the mutation enrichment starts at earlier PCR
cycles than in full COLD-PCR, and fast COLD-PCR results in higher
mutation enrichments (Table 1).
We found it convenient to generate longer amplicons via regular
PCR and then apply COLD-PCR to shorter fragments from these
amplicons, provided the total number of PCR cycles is not increased
excessively. COLD-PCR performed directly from genomic DNA
(Supplementary Fig. 2) is an alternative that reduces the possibility
of polymerase-introduced ‘noise’ by excessive PCR cycling. The
mutation enrichment via COLD-PCR depends on the sequence
NATURE MEDICINE
VOLUME 14
[
NUMBER 5
[
MAY 2008
cThese
context, thus certain mutations within a DNA sequence may be more
difficult to detect than others. Computational predictions of the
expected mutation enrichment as a function of the DNA sequence
can help mitigate these concerns.
Other than increasing the sensitivity and reliability of mutational
screening in clinical samples from subjects with cancer, COLD-PCR
should find application in all situations where minority alleles may be
present and increased detection sensitivity is needed. In prenatal diagnosis, fetal DNA sequences require sensitive detection in the presence
of high excess maternal DNA3. In infectious diseases, detection of low
copy numbers of mutated microorganisms is crucial, for example in
antiretroviral resistance2. In epigenetics, identification of traces of
differentially methylated or unmethylated bisulfite-treated DNA is
employed to serve as a tumor biomarker in bodily fluids26. Adaptation
of COLD-PCR to nanotechnology platforms is also anticipated.
METHODS
DNA and tumor samples. We extracted genomic DNA containing defined
mutations from cell lines (purchased from American Type Culture Collection,
listed in Supplementary Fig. 1). We obtained surgical cancer samples from the
Massachusetts General Hospital Tumor Bank. We obtained DNA from the EGFR
exon 19 deletion–positive cell line HCC827 from the Lowe Center for Thoracic
Oncology, Dana Farber Cancer Institute. We collected plasma-circulating DNA
from subjects undergoing radio-chemotherapy at Dana Farber–Brigham and
Women’s Cancer Center after Dana Farber Institutional Review Board approval.
All subjects signed informed consent forms after collection of blood tissue
specimens. We isolated genomic DNA with the DNeasy Tissue Kit (Qiagen).
Primers used for COLD-PCR. The primers used for amplification of TP53,
KRAS and EGFR gene segments via COLD-PCR are listed in Supplementary
Table 1 online, along with the amplicon size and Tc.
Identification of the critical denaturation temperature. We determined Tc
experimentally for each amplicon. First, we obtained a real-time melting curve
in the presence of LC-Green dye in a Cepheid real-time PCR machine to define
the Tm in the PCR buffer used. A series of denaturation temperatures lower
than the Tm at steps of 0.5 1C were then applied for COLD-PCR until the
temperature was low enough that no specific PCR product was produced. We
then set the Tc as the lowest temperature that reproducibly yielded a substantial
PCR product. In all cases, to achieve a substantial differential in the amplification efficiency of wild-type and point mutation–containing sequences, we
selected a Tc below the Tm of both the wild-type sequence and mutant or
mismatched sequences using the scheme in Figure 1.
583
© 2008 Nature Publishing Group http://www.nature.com/naturemedicine
TECHNICAL REPORTS
Full-COLD-PCR. For experiments in which a nested PCR format was chosen,
we first amplified larger DNA fragments (TP53 exon 8, 247 bp; KRAS exon 1,
160 bp; EGFR exon 19, 201 bp) from 5–10 ng genomic DNA in a standard PCR
reaction with high fidelity polymerase, HF-2 (error rate, B1 10–5 misincorporations per nucleotide after 25 cycles of PCR, BD-Clontech), before
applying COLD-PCR.
For the 98-bp KRAS exon 1 fragment at Tc ¼ 80 1C, we used a
1:1,000 dilution of the first PCR product for a second PCR (either COLDPCR or regular PCR). COLD-PCR was performed and monitored in real time
in a Smart Cycler I machine (Cepheid). Full COLD-PCR reactions contained
final concentrations of reagents as follows: 1 JumpStart buffer (Sigma), 0.2
mM each dNTP, 0.2 mM forward and reverse primers listed in Supplementary
Table 1, 0.1 LC-Green dye (Idaho Technologies), 1 JumpStart Taq polymerase (Sigma) and DNA. Full COLD-PCR cycling conditions were as follows:
95 1C, 120 s; 10 cycles of (95 1C, 15 s; 55 1C fluorescence reading, 30 s; 72 1C,
1 min); then 30 cycles of (95 1C, 15 s; 70 1C, 8 min; 80 1C, 3 s; 55 1C
fluorescence reading, 30 s; 72 1C, 1 min).
For the 87-bp, 167-bp and 210-bp TP53 exon 8 and 12- bp EGFR
exon 19 fragments, we used the same program and reagents used for
KRAS at the respective Tc for each fragment, as listed in Supplementary
Table 1.
For full COLD-PCR of the 87-bp TP53 exon 8 directly from genomic DNA,
we added 5–10 ng of genomic DNA to each reaction, with all other reagents
remaining the same as for the nested format. PCR cycling was modified to the
following conditions: 95 1C, 120 s; 25 cycles of (95 1C, 15 s; 55 1C fluorescence
reading, 30 s; 72 1C, 1 min); then 30 cycles of (95 1C, 15 s; 70 1C, 8 min;
83.5 1C, 3 s; 55 1C fluorescence reading, 30 s; 72 1C, 1 min).
Fast-COLD-PCR. For nested-PCR of the 98-bp KRAS exon 1 fragment at
Tc ¼ 80 1C, we modified the program as follows: 95 1C, 120 s; 10 cycles of
(95 1C, 15 s; 55 1C fluorescence reading ON, 30 s; 72 1C, 1 min); 30 cycles of
(80 1C, 3 s; 55 1C fluorescence reading ON, 30 s; 72 1C, 1 min).
For the 87-bp, 167-bp and 210-bp TP53 exon 8 and 129-bp EGFR exon
19 fragments, we used the same program and reagents at the respective Tc for
each fragment, as listed in Supplementary Table 1.
For fast COLD-PCR for the 87-bp TP53 exon 8 directly from genomic DNA,
we added 5–10 ng of genomic DNA to each reaction, with all other reagents
remaining the same as for the nested format. PCR cycling was modified to the
following conditions: 95 1C, 120 s; (95 1C, 15 s; 55 1C fluorescence reading ON,
30 s; 72 1C, 1 min) for 25 cycles; (83.5 1C, 3 s; 55 1C fluorescence reading ON,
30 s; 72 1C, 1 min) for 30 cycles. COLD-PCR experiments were reproduced at
least five times, and the fold enrichment varied within a range of ±15%
of the average.
Downstream assays. MALDI-TOF genotyping of COLD-PCR and regular PCR
amplicons was performed at the Harvard Partners Center for Genetics and
Genomics High Throughput Genotyping Core Facility. Pyrosequencing of
COLD-PCR and regular PCR amplicons was performed by EpigenDx. Sanger
sequencing of COLD-PCR and regular PCR products was performed at the
Dana Farber sequencing Core facility.
Note: Supplementary information is available on the Nature Medicine website.
ACKNOWLEDGMENTS
We gratefully acknowledge the assistance of A. Brown at the Harvard Partners
Center for Genetics and Genomics High Throughput Genotyping Facility and of
M. Miri, F. Haluska and P. Janne in obtaining specimens from the Massachusetts
General Hospital Tumor Bank and Dana Farber Cancer Institute. We also
acknowledge B. Price and A. D’Andrea for valuable comments on the manuscript.
This work was supported by training grant 5 T32 CA09078 (J.L.) and US
National Institutes of Health grants CA111994-01 and CA115439-01.
584
AUTHOR CONTRIBUTIONS
J.L. and L.W., experimental design; H.M. and M.H.K., clinical considerations
and rationale; R.B., modeling; G.M.M., project setup, experimental design and
manuscript preparation.
Published online at http://www.nature.com/naturemedicine
Reprints and permissions information is available online at http://npg.nature.com/
reprintsandpermissions
1. Kobayashi, S. et al. EGFR mutation and resistance of non–small-cell lung cancer to
gefitinib. N. Engl. J. Med. 352, 786–792 (2005).
2. Hoffmann, C. et al. DNA bar coding and pyrosequencing to identify rare HIV drug
resistance mutations. Nucleic Acids Res. 35, e91 (2007).
3. Lo, Y.M. et al. Presence of fetal DNA in maternal plasma and serum. Lancet 350,
485–487 (1997).
4. Paez, J.G. et al. EGFR mutations in lung cancer: correlation with clinical response to
gefitinib therapy. Science 304, 1497–1500 (2004).
5. Janne, P.A. et al. A rapid and sensitive enzymatic method for epidermal growth factor
receptor mutation screening. Clin. Cancer Res. 12, 751–758 (2006).
6. Engelman, J.A. et al. Allelic dilution obscures detection of a biologically significant
resistance mutation in EGFR-amplified lung cancer. J. Clin. Invest. 116, 2695–2706
(2006).
7. Diehl, F. et al. Detection and quantification of mutations in the plasma of patients with
colorectal tumors. Proc. Natl. Acad. Sci. USA 102, 16368–16373 (2005).
8. Kimura, T. et al. Mutant DNA in plasma of lung cancer patients: potential for
monitoring response to therapy. Ann. NY Acad. Sci. 1022, 55–60 (2004).
9. Li, J. et al. s-RT-MELT for rapid mutation scanning using enzymatic selection and real
time DNA-melting: new potential for multiplex genetic analysis. Nucleic Acids Res. 35,
e84 (2007).
10. Lipsky, R.H. et al. DNA melting analysis for detection of single nucleotide polymorphisms. Clin. Chem. 47, 635–644 (2001).
11. Liew, M. et al. Genotyping of single-nucleotide polymorphisms by high-resolution
melting of small amplicons. Clin. Chem. 50, 1156–1164 (2004).
12. Yeung, A.T., Hattangadi, D., Blakesley, L. & Nicolas, E. Enzymatic mutation detection
technologies. Biotechniques 38, 749–758 (2005).
13. Ogino, S. et al. Sensitive sequencing method for KRAS mutation detection by
pyrosequencing. J. Mol. Diagn. 7, 413–421 (2005).
14. Huang, C. et al. Mutations in exon 7 and 8 of TP53 as poor prognostic factors in
patients with non-small cell lung cancer. Oncogene 16, 2469–2477 (1998).
15. Huang, C.L. et al. Mutations of TP53 and K-ras genes as prognostic factors for nonsmall cell lung cancer. Int. J. Oncol. 12, 553–563 (1998).
16. Jackson, P.E. et al. Specific TP53 mutations detected in plasma and tumors of
hepatocellular carcinoma patients by electrospray ionization mass spectrometry.
Cancer Res. 61, 33–35 (2001).
17. Shao, Z.M., Wu, J., Shen, Z.Z. & Nguyen, M. TP53 mutation in plasma DNA and its
prognostic value in breast cancer patients. Clin. Cancer Res. 7, 2222–2227 (2001).
18. Mayall, F., Jacobson, G., Wilkins, R. & Chang, B. Mutations of TP53 gene can be
detected in the plasma of patients with large bowel carcinoma. J. Clin. Pathol. 51,
611–613 (1998).
19. Silva, J.M. et al. Tumor DNA in plasma at diagnosis of breast cancer patients is a
valuable predictor of disease-free survival. Clin. Cancer Res. 8, 3761–3766
(2002).
20. Gonzalez, R. et al. Microsatellite alterations and TP53 mutations in plasma DNA of
small-cell lung cancer patients: follow-up study and prognostic significance. Ann.
Oncol. 11, 1097–1104 (2000).
21. Eberhard, D.A. et al. Mutations in the epidermal growth factor receptor and in KRAS
are predictive and prognostic indicators in patients with non-small-cell lung cancer
treated with chemotherapy alone and in combination with erlotinib. J. Clin. Oncol. 23,
5900–5909 (2005).
22. Thomas, R.K. et al. Sensitive mutation detection in heterogeneous cancer specimens
by massively parallel picoliter reactor sequencing. Nat. Med. 12, 852–855 (2006).
23. Thomas, R.K. et al. High-throughput oncogene mutation profiling in human cancer.
Nat. Genet. 39, 347–351 (2007).
24. Sun, X., Hung, K., Wu, L., Sidransky, D. & Guo, B. Detection of tumor mutations in the
presence of excess amounts of normal DNA. Nat. Biotechnol. 20, 186–189
(2002).
25. Fuery, C.J. et al. Detection of rare mutant alleles by restriction endonuclease-mediated
selective-PCR: assay design and optimization. Clin. Chem. 46, 620–624 (2000).
26. Belinsky, S.A. et al. Gene promoter methylation in plasma and sputum increases with
lung cancer risk. Clin. Cancer Res. 11, 6505–6511 (2005).
27. Greenman, C. Patterns of somatic mutation in human cancer genomes. Nature 446,
153–158 (2007).
VOLUME 14
[
NUMBER 5
[
MAY 2008
NATURE MEDICINE