optimization of whole genome amplification products for sequencing applicati

12 Pages·2017·1.52 MB·English

Shoetal.BMCCancer (2017) 17:457
DOI10.1186/s12885-017-3447-6
RESEARCH ARTICLE Open Access
Precision oncology using a limited number
of cells: optimization of whole genome
amplification products for sequencing
applications
Shonan Sho1,2* , Colin M. Court1,2, Paul Winograd1,2, Sangjun Lee3, Shuang Hou3, Thomas G. Graeber3,
Hsian-Rong Tseng3 and James S. Tomlinson1,2,4
Abstract
Background: Sequencing analysis of circulating tumor cells(CTCs) enables “liquid biopsy” to guide precision
oncology strategies. However, this requires low-template wholegenomeamplification(WGA) thatis prone to errors
and biases from uneven amplifications.Currently, quality control (QC) methods for WGAproducts, as well as the
number ofCTCs needed for reliable downstream sequencing, remain poorly defined. We sought to define
strategies for selecting and generating optimal WGA products from low-template input as it relates to their
potential applications inprecision oncology strategies.
Methods: Singlepancreatic cancer cells (HPAF-II) were isolated using laser microdissection. WGA was performed
using multiple displacement amplification (MDA), multiple annealing and looping based amplification (MALBAC)
and PicoPLEX. Qualityof amplified DNA products were assessed using a multiplex/RT-qPCR based method that
evaluates for 8-cancer related genesand QC-scores were assigned. We utilized this scoring system to assess the
impact of de novo modifications to the WGA protocol. WGA products were subjected to Sanger sequencing,array
comparative genomic hybridization (aCGH)and next generation sequencing (NGS) to evaluate their performances
inrespective downstream analysesproviding validation of theQC-score.
Results: Single-cell WGA products exhibited a significant sample-to-sample variability in amplified DNA qualityas
assessed by our 8-gene QCassay.Single-cellWGA products that passed thepre-analysis QC had lower amplification
biasand improved aCGH/NGS performance metrics when compared to single-cell WGA products that failedthe
QC. Increasing thenumberof cellular input resulted in improved QC-scores overall, buta resultant WGA product
that consistentlypassedthe QC step required a starting cellularinput of atleast 20-cells. Our modified-WGA
protocol effectively reduced this number, achieving reproducible high-quality WGAproducts from ≥5-cells as a
starting template. Astartingcellular input of 5to 10-cells amplified using the modified-WGA achieved aCGH and
NGSresults that closelymatched that ofunamplified, batch genomic DNA.
Conclusion: The modified-WGA protocol coupled with the8-gene QC serve as an effective strategyto enhance the
quality of low-template WGA reactions. Furthermore, a threshold numberof 5–10 cellsare likely needed for a
reliable WGA reactionand product withhighfidelity to theoriginal starting template.
Keywords: Precision oncology, Whole genome amplification, Single-cellsequencing, Next generation sequencing,
Multiple displacement amplification
*Correspondence:[email protected]
1DepartmentofSurgery,UniversityofCaliforniaLosAngeles,10833LeConte
Ave,California,LosAngeles90095,USA
2DepartmentofSurgery,GreaterLosAngelesVeteran’sAffairsAdministration,
11301WilshireBlvd,California,LosAngeles90073,USA
Fulllistofauthorinformationisavailableattheendofthearticle
©TheAuthor(s).2017OpenAccessThisarticleisdistributedunderthetermsoftheCreativeCommonsAttribution4.0
InternationalLicense(http://creativecommons.org/licenses/by/4.0/),whichpermitsunrestricteduse,distribution,and
reproductioninanymedium,providedyougiveappropriatecredittotheoriginalauthor(s)andthesource,providealinkto
theCreativeCommonslicense,andindicateifchangesweremade.TheCreativeCommonsPublicDomainDedicationwaiver
(http://creativecommons.org/publicdomain/zero/1.0/)appliestothedatamadeavailableinthisarticle,unlessotherwisestated.
Shoetal.BMCCancer (2017) 17:457 Page2of12
Background originalgenomictemplate.Therefore,onemustbeableto
“Liquidbiopsy”ofcirculatingtumorcells(CTCs)hasbeen accurately differentiate between high- and low- quality
suggested in many recent studies as an ideal biopsy tech- WGA products in order to ensure accurate sequencing
niqueforprecisiononcologyapplications[1–5].CTCsare resultsforguidingcancertherapy.
thoughttoarisefrombothprimaryandmetastaticlesions, Currently, however, quality control metrics and selection
allowing for a more comprehensive representation of the criteria for high-quality WGA products from single cells
tumor genomic make-up [6]. Furthermore, the need for have not been sufficiently defined. Much of the existing
only a simple peripheral blood draw in “liquid biopsy” literature utilizing whole genome amplified DNA lack ana-
makes it amenable to repeated samplings without incur- lysis of the quality of WGA samples being used. Given the
ring significant costs or risks to patients. Although known sample-to-sample variability in minimal template
successful CTC enrichment, capture and downstream WGA products, WGA quality must be defined in order to
molecular analysis has been described, major obstacles accuratelyinterpretandcompareWGA-DNAderiveddata.
still remain prior to its clinical translation [6, 7] (Fig. 1). Furthermore, although prior studies have shown improved
One major challenge is the limited number of CTCs WGAqualitywithincreasingamountofgenomictemplate
available for molecular analysis. The number of CTCs input [19], the number of CTCs needed to generate a
obtainable from a single peripheral blood remains highly WGA product suitable for sequence analysis to be utilized
limited, with most studies showing <5 CTC/mL from a inprecisiononcologystrategiesremainunknown.
single peripheral blood [3, 8]. For many GIcancers, espe- In the current study, we sought to define strategies for
cially pancreas ductal adenocarcinoma, CTCs are even selecting and generating optimal WGA products from
more limited [9, 10]. Thus, for molecular analysis to be samples ranging from one to twenty cells as it relates to
performed using CTCs, the limited amount of genomic their potential applications in precision oncology strat-
materials available from CTCs must undergo whole gen- egies. To this end, we developed a quality control (QC)
omeamplification(WGA)togenerateadequatequantities assay to help facilitate the selection of high quality
ofDNAfordownstreamsequencinganalysis. WGA product suitable for use in downstream sequen-
Despite recent advances in WGA techniques, amplifi- cing applications, including point mutation detection,
cation processes remain prone to uneven amplifications, array comparative genomic hybridization (aCGH) and
resulting in amplification bias [11–14]. For heterozygous next generation sequencing (NGS). In order to better
sites, this can result in a total loss of one allele, a understand the key determinants of WGA quality, we
phenomenon called allele drop out (ADO) [15, 16]. This compared various WGA methods in addition to the
is especially problematic when working with a small number of input cells to determine their influence upon
number of cells, as in CTC analysis, as the stochastic amplification reactions. We then used our findings to
variation in the WGA process is exacerbated when start- develop a modified multiple displacement amplification
ing with low copies of genomic input [17, 18]. This (MDA) protocol with a notable improvement in ampli-
results in considerable variability in sample-to-sample fied DNA quality over the conventional MDA proto-
quality when working with CTCs [15]. Low-quality col. Ultimately, we utilized these findings to determine
WGA products with significant degrees of amplification the “threshold” number of cells needed for reliable
bias and ADO are inappropriate for precision oncology molecular analysis that could be utilized in precision
applications as they fail to accurately represent the oncology strategies.
Fig.1WorkflowofCTCsequencingandassociatedchallenges.SeveralchallengesexistinCTCsequencinganalysis.Qualitycontrol(QC)forthe
amplifiedDNAproductisparamount,aslowqualityWGAproductsleadtofailed/inaccuratedownstreamsequencingapplications.AQCstep
priortocostlydownstreamsequencingapplicationshelpsreduceavoidablecostsassociatedwithfailedsequencing
Shoetal.BMCCancer (2017) 17:457 Page3of12
Methods reduced individual reaction volumes (3ul instead of
Celllines 50ul), all taking place in parallel. Following the amplifi-
Pancreatic cancer cell line HPAF-II was obtained from cation step, contents within the 16-wells were collected
American Type Culture Collection (ATCC, Virginia, into one tube, followed by the purification and quantifi-
USA), and grown using EMEM medium (ATCC) supple- cation stepasdescribed above.
mented with 10% fetal bovine serum (ATCC) and 100 U/
mL penicillin-streptomycin (ATCC). All cell lines were Developmentofqualitycontrol(QC)-scoreforWGA-DNA
grownat37°Cwith5%CO andwereroutinelypassaged Prior literatures have described the role of multiplex
2
at 80% confluence using an iso-osmotic sodium citrate PCR in evaluating DNA quality isolated from formalin-
solutionforcellrelease(Thermo,Massachusetts,USA). fixed paraffin embedded (FFPE) tissues prior to down-
stream analysis using aCGH [20]. Based on this concept,
Lasermicrodissection we developed an 8-gene multiplex PCR/quantitative
In preparation for laser microdissection, cells were PCR (qPCR) based QC assay for evaluation of WGA-
released from the culture plates using the iso-osmotic DNA quality. We selectively chose for 8 cancer-related
sodium citrate solution (Thermo).Followingawashwith genes that are considered molecular targets for targeted
the culture medium, each cellline wasdiluted to adens- cancer therapy and therefore highly implicated in guid-
ity of 1000 cells per 100 μL. Approximately 1000 cells ing therapeutic decisions. This way, the QC assay identi-
(100 μL) were smeared on PEN membrane slides (Leica, fies WGA products suitable for precision oncology
Wetzlar, Germany), air-dried for 10 min, and fixed with applications by evaluating for the coverage and “acces-
100 μL of 100% ethanol. Cells were then isolated using sibility” of these important genomic locations within
the PALM MicroBeam laser microdissection system the WGA-DNA. Genes evaluated by the QC assay in-
(Zeiss, Oberkochen, Germany). 1, 5, 10 or 20 cells were cluded BRAF, EGFR, KIT, KRAS, NRAS, PIK3CA,
laser microdissected and collected into 200 μL opaque PTEN and P53.
tube caps (Zeiss) using the laser pressure catapult Score of 0–8 was assigned based on the number of
function. Cell transfer to the tube cap was confirmed by genes successfully amplified and detected using multi-
imagingthe cap priortocapclosure usingthe cap-check plex PCR and qPCR. Failure to detect one or more of
function. the 8-gene from the WGA product signifies lack of
coverage or potential ADO during the WGA process,
Wholegenomeamplification both of which indicate a poor quality WGA product and
Isolated cells were subjected to genomic DNA isolation jeopardizes accurate representation of the original start-
and WGA using one of the three commercially available ing genome. Thus, we only gave WGA-DNA products
single-cell WGA kits according to the manufacturer’s with 8 out of 8 score a “pass” and deemed them fitting
protocol: REPLI-g Single-Cell Kit (Qiagen, California, forfurtherdownstreamanalysis.
USA), Multiple Annealing and Looping Based Amplifi- A secondary QC-score was generated to confirm the
cation Cycles (MALBAC) Single-cell WGA Kit (Yikon internalvalidity of the original QC-score based onthe 8-
Genomics, Beijing, China) and PicoPlex Single-Cell cancer related genes. The secondary QC-score was
WGA Kit (Rubicon, Michigan). Reactions were per- generatedfrom adistinct setof8primerpairsrepresent-
formed a minimum of 3 times for all conditions tested. ing 8 housekeeping genes: NDUFA, UQCRC, ACTG,
WGA products were purified using the QIAquick PCR CYB5A,GABA-RAPL,MIF,MYC,PRPH.
Purification Kit (Qiagen) and quantified with NanoDrop
2000 (Thermo), in keeping with a previously described MultiplexPCRandquantitativePCR(qPCR)
CTCmolecular analysismethodology[15]. WGA products were subjected to multiplex PCR pre-
Modified MDA protocol was performed using the amplification followed by qPCR assay for detection of
same reagents available from the REPLI-g Single Cell Kit amplified targeted genes. Multiplex PCR was performed
(Qiagen), with modifications made in the cell lysis step using primer sets representing genes mentioned above.
and the final amplification step. Cell lysis was performed Primer sets used for the primary QC assay included:
over a course of 30 min to ensure complete lysis of gen- BRAF (forward 5′ – TAC TGC TCT TTC TTC TCC
omic material from isolated cells (as opposed to 10 min AAC AC – 3′; reverse 5′ – CCT GAT TGT ATT TGA
recommended by the manufacturer). Prior to the final GAT CTA GTA GGG – 3′) EGFR (forward 5′-CAG
amplification step, the 50-ul MDA reaction mix was CCT TCT CCG TAA TTA GCA T – 3′; reverse 5′ –
mixed for 30 s by pipetting up and down and then TGA CAC AGA TAA TTG TCC CAC AG – 3′), KIT
partitioned into 16 individual reactions (approximately (forward 5′ – GGC ATT GAG GAG GGA TAG TAA
3ul each) and the amplification occurred at 30 °C for AT – 3′; reverse 5′ – CTG AAC AAT TTG CTT GAA
8 h. This resulted in 16 individual MDA reactions with TGT TGG – 3′), KRAS (forward 5′ – GTG TTA CTT
Shoetal.BMCCancer (2017) 17:457 Page4of12
ACC TGT CTT GTC TTTG – 3′; reverse 5′ – GCC and 95 °C at 0.5 °C intervals. Real-time PCR data were
TTC TAG AAC AGT AGA CAC AA – 3′), NRAS reviewed and analyzed using the CFX manager (BioRad).
(forward 5′ – AAT GGA ATC CCG TAA CTC TTG G Specificity of the PCR amplification product was deter-
–3′;reverse 5′–GATGATGTACCTATG GTG CTA mined using melting curve analysis. PCR products with
GTG – 3′), PIK3CA (forward 5′ – AGG GCA AAT melting-temperature (Tm) matching the expected value
AATAGT GGT GAT CT – 3′; reverse 5′ – CAG CAA based on primer sequence, and threshold cycle (Ct) <30
TTA CTT GTT CTG GTA CAC – 3′), PTEN (forward were counted asreliableamplificationanddetection.
5′ – CTT TCT CTA GGT GAA GCT GTA CT – 3′;
reverse 5′ – GGT TCATTG TCA CTA ACATCT GG Array-CGH
– 3′) and P53 (forward 5′ – AAG AGA AGC AAG Sample WGA-DNA and reference DNA were differen-
AGG CAG TAA G – 3′; reverse 5′ – CTT AGG CTC tially labeled with cyanine-3 (CY3) and cyanine-5 (Cy5)
CAG AAA GGA CAA G – 3′). Primer sets used for the dyes using the GenetiSure Amplification and Labeling
secondary QC assay included: NDUFA7 (forward 5′ – Kit (Agilent) according to the manufacturer’s protocol.
TGC TCT GGA TGT GAA GAT GCC A – 3′; reverse Briefly, a 15.5 ul reaction mixture containing sample or
– 5′ – TTC CAG GTA AAT CCA GCC CAG G – 3′), reference DNA, Random Primer Mix and water was
UQCRC1 (forward 5′ – CAG CCA GTC AGC ATC placed in 98 °C for 3 min for DNA denaturation and
ATC CAA C – 3′; reverse 5′ – GAA AGC CGG ATT transferred to 4 °C for 3 min. Next, 9.5 ul of Labeling
GCG GTA ACA T - 3′), ACTG1 (forward 5′ - GCT Master Mix containing 5ul of 5× Reaction Buffer, 2.5ul
CAA TGG GGT ACT TCA GGG T – 3′; reverse 5′ – of 10× dNTP Mix, 1.5 ul of Cy5-dUTP/Cy3-dUTP and
GTG GAC GTTACG TAA AAG GCC C – 3′), CYB5A 0.5 Exo (−) Klenow were added to each tube containing
(forward 5′ – GGC AAC GCT TAG ACT CTG TGT G the sample/reference. DNA labeling reaction was per-
– 3′; reverse 5′ – CTG CCC TTG GCC TAA CTA formed in 37 °C for 45 min followed by inactivation step
ACC T – 3′), GABARAPL2 (forward 5′ – CCA GCC in 65C for 10 min. Labeled DNA was purified using the
AAT TCA TGA GTC GGT G – 3′; reverse 5′ – CCT Post Labeling Purification Columns (Agilent) according
GAC AAC TCG CAA GTA GCAC – 3′), MIF (forward to the manufacturer’s protocol. Purified labeled DNA
5′ – AGA AGT CAG GCA CGT AGC TCA G – 3′; samples were prepared for hybridization, which took
reverse 5′ – GGC ACGTTG GTG TTTACGATG A – place on Agilent 8x60K CGH microarray slides at 67 °C
3′), MYC (forward 5′ – GGA TAG CTC TGC AAG for 6 h. Following the hybridization, the slides were
GGG AGA G – 3′; reverse 5′ –TCG TCG CAG TAG washed per manufacturer protocol, and prepared for
AAA TAC GGC T – 3′), PRPH (forward 5′ – GTT scanningusingtheAgilentSureScanMicroarrayScanner
CCT CAA GAA GCT GCA CGA G – 3′; reverse 5′ – (Agilent). Microarray image was prepared and analyzed
CGT TAG ACT CTG GAT CTG GCG T – 3′). Details usingtheAgilentCytoGenomicssoftware(Agilent).
of multiplex PCR pre-amplification is available in the
Additional file 1: Supplementary Material. Briefly, PCR Sequencinglibrarypreparationandsequencing
reactions were carried out on a C1000 Thermal Cycler Purified WGA products were sheared to generate DNA
(Bio-Rad) with the Multiplex PCR Plus Kit (Qiagen) fragments of 350 bps using the Covaris sonicator
using total volumes of 17 uL per reaction. The reaction (Covaris). Followingcleanupof the sonicated DNA, end-
conditions were as follows: 95 °C for 15 min, denatur- repair and ligation were performed using the KAPA
ation at 94 °C for 30 s, annealing at 64 °C for 30 s, and DNA Library Preparation Kit (KAPA Biosystems) ac-
extension at 72 °C for 30 s for a total of 10cycles, with a cording to the manufacturer’s protocol, followed by
final stepof72°Cfor 10min. library amplification by PCR. Sequencing was performed
Following the pre-amplification step by multiplex on an Illumina HiSeq 2000 using random primers and
PCR, the resulting amplified products were analyzed and pair-endreads of75bps(2X75bps).
detected using qPCR. Reactions took place on BioRad
CFX-96 real time system (BioRad) using the QuantiTect Sequencinganalysisandvisualization
SYBR Green PCR Kit (Qiagen). A 25-ul reaction mix- The sequencing read data was analyzed using Ginkgo
tures were prepared, which contained 12.5 ul of the (http://qb.cshl.edu/ginkgo), an open-source platform for
SYBR Green PCR Master Mix (Qiagen), 9.5 ul of RNA- the analysis of single-cell copy-number variations (CNV).
grade water, 1 ul of individual primers sets (10uM) and This analysis platform was developed specifically to
2ul of the multiplex PCR product. The reaction condi- address the unique challenges of single-cell sequen-
tion were as follows: 95 °C for 15 min, followed by 35 - cing data, and incorporates built-in computational
cycles of 94 °C for 15 s, 64 °C for 20 s, and 72 °C for tools for data optimization. Specifically, limitations as-
20 s. The plate was read following the extension step at sociated with low depth of sequencing coverage, amp-
72 °C. Melting curve analysis was performed between 70 lification bias and inflated read counts from poorly
Shoetal.BMCCancer (2017) 17:457 Page5of12
assembled genomic regions were addressed in Ginkgo’s in amplified DNA quality was noted for MALBAC, with
analysis. Detailed description of the computational meth- QC-scores ranging between 2 and 6. Although Picoplex
odologyandcorrectiontoolsareprovidedbyGarvinetal. resulted in less quality disparity, majority of samples
[21]. Mapped reads were binned using variable length- achieved scores of only 3. QC-score profiles for the
binning and underwent GC bias normalization prior to MDA and MALBAC methods were overall similar, but
segmentation using circular binary segmentation. Quality the WGA product that passed our QC criteria (QC-
metrics data, including Lorenz curve, histogram of read score = 8) was only found in the single-cell sample
count distribution and index of dispersion were obtained amplifiedwith usingtheMDAmethod.
as a part of the Ginkgo analysis pipeline. Sequencing
data visualization was performed using Nexus soft- Determinationofnumberofcellsneededtoachieve
ware (Biodiscovery, inc.). reproducibleWGAproductquality
We noted significant random variations in WGA prod-
KRASPCRandsangersequencing uct quality when using single-cells as starting template
PCR amplification of KRAS exon 2 was performed using (Fig. 2). Increasing the amount of starting genomic
the following primer (Integrated DNA Technologies): template has been shown to improve WGA product
Forward 5′ – AAG GTA CTG GTG GAG TAT TTG – quality [19, 22]. Thus, we hypothesized that there would
3′ and Reverse 5′ – GTA CTC ATG AAA ATG GTC be a certain “threshold” number of cellular input above
AGA G – 3′, with expected amplicon length of 295 bps. which a reliable and reproducible WGA process is
PCR reactions were carried out on a C1000 Thermal possible, i.e. a number of cells which will achieve a
Cycler (Bio-Rad) with Platinum PCR SuperMix High “pass”inourQC steponaconsistentbasis.
Fidelity Kit (Invitrogen) using total volumes of 50 μLper In order to test this hypothesis, 5-cell, 10-cell and 20-
reaction according to the manufacturer’s protocol. The cell samples ofHPAF-II cells were cut and isolated using
reaction conditions were as follows: denaturation at a laser micro-dissector. WGA was performed using the
94 °C for 30 s, annealing at 55 °C for 30 s, and exten- MDAmethod, followed bythe QC assay and assignment
sion at 68 °C for 45 s for a total of 40 cycles. The PCR of the QC-scores (Fig. 3a). Less variability in amplified
products were purified using the QIAquick PCR Purifi- DNA quality was noted when 5 or more cells were used
cation Kit (Qiagen) and eluted into 50 uL of nuclease- as starting genomic template. All samples persistently
free water (Qiagen). DNA was diluted to a concentra- achieved a QC-score of ≥5 (5-cell) or ≥6 (10-cell), as
tion of 10 ng/uL based on Nanodrop quantification of opposed to single-cell WGA reactions with scores ran-
the PCR product. Automated dideoxy terminator se- ging from 1 to 8. Furthermore, at least one of the tripli-
quencing was performed by capillary electrophoresis by cates in 5-cell and 10-cell group passed the QC-step
the UCLA GenoSeq Core on an ABI 3730 DNA (QC-score =8). However, ahighly reliable WGA process
analyzer using Big Dye Terminator chemistry (Applied withalloftheWGAproductspassingtheQCstepcould
Biosystems). All sequences wereanalyzed by manual in- not be achieved until 20 cells were used as starting gen-
spection of the individual trace files using Four Peaks omictemplate input.
(Nucleobytes).
ModifiedMDAreactiontoimprovequalityofWGA
Results productfromlimitedtemplatesamples.
WGAproductvariabilityinsingle-cells With the existing MDA protocol, multiple single-cell
We applied our 8-gene QC assay to single-cell WGA- WGA reactions must be performed in order to obtain
DNA samples. We tested three different WGA reaction one high-quality WGA product suitable for downstream
methods: MDA, MALBAC and Picoplex. Total of 18 analysis (Fig. 2). Even when more than a single-cell was
individual HPAF-II cells were isolated using laser micro- used, a reliable WGA reaction with its amplified DNA
dissection and used in the respective WGA methods. product passing the QC step on a consistent basis could
The experiment was repeated 6 times in order to ac- not be achieved until 20 cells were used (Fig. 3a). For
count for the variability expected with the single-cell many GI cancers that are known to generate only few
WGA process. QC assay was performed and the QC- CTCs (i.e. pancreas ductal adenocarcinoma), obtaining
score wasassignedforeachWGA product. 20 CTCs from a single peripheral blood draw may be
Despite using a clonally expanded cell line and per- unrealistic.
forming WGA reactions in a parallel fashion under the Given these problems, we sought to develop a mo-
same condition, we noted significant variability in dified WGA protocol that would reduce the sample-
sample-to-sample WGA-DNA quality as assessed by the to-sample variability in DNA quality and lower the
8-gene QC assay (Fig. 2). QC-scores for MDA amplified number of starting cells needed to achieve a reliable
single-cells ranged between 1 and 8. Similar variability WGA process. Multiple prior reports have described
Shoetal.BMCCancer (2017) 17:457 Page6of12
Fig.2QC-scoresforsingle-cellWGAproducts.Single-cellWGAwasperformedusingMDA,MALBACandPicoPLEX.Significantsample-to-sample
variabilityinamplifiedDNAqualitywasnoted,asmeasuredbyour8-geneQCscore
the benefit of performing MDA by partitioning the reac- For single-cells, 2 of 3 WGA reactions resulted in a
tion into parallel smaller volume reactions [16, 18, 22]. product with the perfect QC-score of 8, compared to 1
The small reaction volumes and template DNA partition- in6WGA reactionsusing the conventional MDAproto-
ing restricts the degree of aberrant preferential amplifica- col. Importantly, for 5-cell and 10-cell samples, all WGA
tion, leading to a more uniform WGA process overall. reactions generated productsthat passed our QC criteria
Based on this principle, we developed a modified MDA (Fig. 3b). The modified MDA protocol in our hands
protocol with key changes in two aspects: (1) increasing effectively reduced the “threshold” number of cells
the cell lies step from 10 min to a minimum of neededforareliableWGAreactiondownto5cellsfrom
30 min to ensure adequate release of genomic mate- 20cells,wellwithinthenumberofCTCsattainablefrom
rials from cells, and (2) partitioning the final MDA a singleperipheral blooddraw.
reaction into 16 individual reactions containing ~3ul
each on a 96-well plate, prior to the final isothermal WGAproductQCassayusing8-vs.16-genemultiplex
amplification step at 30 °C for 8 h. PCR
WGA products obtained using this modified MDA Our proposed 8-gene QC assay evaluates 8 genomic
protocol resulted in improved reproducibility and higher locations within a WGA product. Evaluating more loci in
QC-scores overall (Fig. 3b). Amplification reaction gain theory provides more comprehensive evaluation of the
was overall lower for the modified MDA products com- amplifiedDNAproduct.Wetestedwhethertherewasany
pared to the conventional MDA, with reduction of ap- benefit to evaluating more genetic loci beyond the 8-
proximately 50% on average for each sample undergoing cancer genes during the QC process. To test for this, we
the modified MDAprotocol (Additional file 2: Table S1). performed a secondary QC assay using a different set of
Fig.3NumberofcellsusedforgenomictemplateinputandWGA-DNAquality.aConventionalMDA.OverallQC-scoresimproveasthenumber
ofcellsusedforstartingcellularinputincreases.20-cellsarerequiredinordertoattainapassingQC-scoreof8reproducibly.bModifiedMDA.A
passingQC-scoreof8isachievedreproduciblyfrom5-cells(asopposedto20-cells)usingthemodifiedMDAprotocol
Shoetal.BMCCancer (2017) 17:457 Page7of12
8-housekeeping genes on the same single-cell conven- The same two single-cell WGA products were also
tional MDA products, as well as the 1/5/10-cell modified analyzed by massive multiplex short read sequencing.
MDA products (Additional file 2: Table S2A and B). The Figure 5 shows the quality metrics of sequencing data
original (8-cancer gene) and the secondary (8-housekeep- associatedwitheachWGA-DNA sampleand unamplified
ing genes) QC assays generated highly concordant QC- batch gDNA. When we compare the two single-cell
scoresforallofthesamplestested.Allofthesamplesthat WGA products(WGA-QCpass and WGA-QCfail), Lorenz
passed the QC-step based on the original QC assay also curves, histograms of read count frequency and indexes of
passed using the secondary QC assay. For the sole modi- dispersion all indicated a superior NGS data quality with
fied MDA sample with the low QC-score (sample 1-C), highercoverageuniformityforWGA-QCpasscomparedto
the secondary QC also resulted in a similarly low QC- WGA-QCfail. The Lorenz curve provides information on
score indicating poor quality WGA-DNA. Thus, evalu- theuniformityofthesequencingreadsdistribution.Perfect
ation of more genetic loci beyond the original 8-cancer coverage results in a straight line with slope of 1 (y = x).
genes did not provide any additional QC information and The wider the curve below the line of y = x,the lower the
didnotchangeQCresultsforanyofthesamplestested. coverage uniformity and greater the amplification bias. As
canbeseeninFig.5a,WGA-QCpassresultedinmoreuni-
Using8-cancergeneQCassaytoselectsamplesfor formdistributionofreaddepthcomparedtoWGA-QCfail.
downstreamaCGHandNGSapplications Thehistogramofreadcountfrequencyalsoprovidesinfor-
Weperformedpointmutationdetection,aCGHandNGS mationoncoveragedispersion(Fig.5b).Histogramswitha
using single-cell WGA products that either passed wide range of distribution, as seen in the sequencing data
(WGA-QCpass) or failed (WGA-QCfail) the QC-step. obtained using WGA-QCfail, indicates a greater degree of
Figures 4 and 5 illustrate the relationship between amplification bias. Genomic profiles generated from NGS
QC results and performances in various downstream sequencing data are as shown in Fig. 5c. Comparison be-
applications. tweenWGA-QCpassandWGA-QCfailrevealssignificantly
Point mutation detection within the KRAS gene was more“noise”inthegenomicprofilederivedfromthelatter,
successful in both QCpass and QCfail WGA products withagreaterindexofdispersioncomparedtothatderived
(Fig. 4a). However, while the QCpass WGA sample fromWGA-QCpass.
exhibited a mutant to wild-type allelic ratio closely
matching that of the batch gDNA (HPAF-II cells are Determinationofthresholdnumberofcellsneededfor
known to have 4 to 1 mutant:wild-type allelic ratio), the optimalperformancesinaCGHandNGSanalysis
QCfail WGA sample exhibited evidence of amplification Although the single-cell WGA product passing our QC
bias,demonstratedbyanalteredallelic ratio. criteria (WGA-QCpass) performed well in point muta-
When we compared aCGH performances between the tion detection, aCGH and NGS, room for improvement
two single-cell WGA products, WGA-QCpass resulted still existed when compared to the performance metrics
in a notably better derivative log2 ratio spread (DLRS) of unamplified batch gDNA (Figs. 4 and 5). Increasing
and signal-to-noise ratio compared to WGA-QCfail the number of starting cells has been shown to improve
(Fig. 4d). DLRS is a key quality metric for aCGH data WGA product quality [15, 19, 22]. However, the number
measuringthepoint-to-pointconsistencyor“noisiness”in ofcellsneededforgeneratingaWGAproductthatfaith-
data,withhighvaluesindicatingpoorsignal-to-noiserela- fully represents the unamplified, original DNA in down-
tionshipanddifficultyinassessingtruecopynumbervari- stream molecular analysisremainstobedefined.
ation (CNV) status. The poor quality metrics (DLRS: 1.3, Using our modified MDA protocol, we lowered the
signal-to-noise ratio: 3.9) associated with WGA-QCfail numbers of cells needed to reliably pass our QC-step
renders its aCGH data unfit for meaningful analysis and down to 5 cells from 20 cells. However, it remained to
interpretation. On the other hand, DLRS of 0.72 and be answered how closely 5-cell and 10-cell WGA prod-
signal-to-noise ratio of 53.6 associated with WGA- ucts approximated the unamplified DNA in aCGH and
QCpassmeetsthequalitythresholdforsingle-cellderived NGS analysis, and whether obtaining 5 to 10 CTCs from
aCGH set forth by Agilent Technologies [23]. Gain and a single liquid biopsy is truly sufficient for sequencing
loss profiles generated from aCGH data are as shown in applications within precision oncology strategies. Figure
Fig. 4c. Comparison between WGA-QCpass and WGA- 4b and 5 illustrate the performances of 5-cell and 10-cell
QCfail reveals disparities in single-cell aCGH profiles, modified MDA products (QC score = 8 for both) in
even though clonally expanded HPAF-II cells were used aCGH and NGS. As the number of cells used for starting
for both. When compared to the unamplified batch genomic template increased from 1 to 5 to 10 cells, pro-
gDNA, we found multiple areas of alterations that were gressive improvement in all quality metrics of aCGH and
detected with WGA-QCpass, but not with WGA-QCfail NGS were noted. Notably, the Lorenz curves for WGA-
(Fig.4a,redarrows). DNA and unamplified batch gDNA nearly overlapped by
Shoetal.BMCCancer (2017) 17:457 Page8of12
Fig.4PointmutationdetectionandaCGHanalysisusingasingle-cellWGAproductsthatpassed(QC-pass)orfailed(QC-fail)thequalitycontrol
step,b)5and10-cellWGAproductswithpassingQC,andcunamplified,batchgenomicDNA.Single-cellWGAproducts(QC-failvs.QC-pass)
exhibitsignificantlydifferentdownstreamanalysesresultsdespiteusingcellsisolatedfromaclonallyexpandedcellline.Thesingle-cellWGA
productwiththepassingQC(QC-pass)generatedSangersequencingandaCGHresultsthatmorecloselyresembledthatoftheunamplified
batchgDNAcomparedtothesingle-cellWGAproductfailingtheQC(QC-fail).Redarrowssignifytheareasofalterationthatweredetectedin
QC-passandbatchgDNA,butnotinQC-fail.daCGHqualitymetrics(DLRSandsignal-to-noiseratio)for1,5and10-cellWGAproductsand
batchgDNA.ImprovedDLRSvalueswereassociatedwithWGAproductsthatpassedtheQCstepaswellasusinganincreasingnumberof
startingcellularinput
10-cells (Fig. 5a). Furthermore, by 10-cells, WGA-DNA dispersion as the number of cells used increased (Figs. 4b
derivedaCGHandNGSgenomicprofilesappearedhighly and 5c). Lastly, the resolution limits of 1, 5 and 10-cell
concordantcomparedtothatofunamplifiedbatchgDNA, amplified products applied to the NGS analysis were
with progressive reduction in signal “noise” and index of assessed.Byevaluating for the accuracyofCNVcallingat
Shoetal.BMCCancer (2017) 17:457 Page9of12
Fig.5NGS application using 1, 5 and 10-cell WGA products and batch gDNA. a Lorenz curve illustrating the amplification bias in read
coverage. Lorenz curve provides information on the uniformity of the sequencing reads distribution. Perfect coverage results in a straight
line with slope of 1 (y = x) as shown by the dotted line. The wider the curve below the line of y = x, the lower the coverage uniformity
and greater the amplification bias. Single-cell WGA product passing the QC-step (QC-pass) is observed to have less amplification bias
compared to the single-cell WGA product that failed the QC-step (QC-fail). As the number of starting cellular input increases, the degree
of amplification bias becomes even less pronounced, approaching that of unamplified batch gDNA by 10-cells. b Histogram of read count
distribution. Wider range of distribution without a distinct peak (as seen for single-cell WGA: QC-fail) signifies worse coverage dispersion.
c Genomic profiles generated from NGS sequencing data. Significantly less “noise” is observed with single-cell QC-pass when compared
to single-cell QC-fail. Increasing the number of starting cellular input further improves the data quality, as reflected in the decreasing
index of dispersion
Shoetal.BMCCancer (2017) 17:457 Page10of12
various bin sizes (250 kb, 500 kb, 1 Mb and 2.5 Mb), we cancer therapy. The eight genes included in our QC
found 5 and 10-cells to require a minimum bin size of assay are all considered important molecular targets for
approximately250kbtoreproducethebatchDNAresult. cancer therapy, and under active investigation in the
For single-cell, bin size of approximately 2.5 Mb was National Cancer Institute – Molecular Analysis for
needed(Additionalfile3:Fig.S1).Insummary,using5–10 Therapy CHoice (NCI-MATCH) trial [28–31]. Thus,
cells as the starting template for our modified MDA reac- evaluation for WGA-DNA quality using this QC assay
tion generated WGA-DNA that closely approximated the helps identify samples suitable for potential use in preci-
unamplifiedgDNAinbothaCGHandNGSperformances. siononcology strategies.
Along with the 8-cancer gene QC-assay, we have also
Discussion developed a modified MDA protocol in order to help
CTC analysis offers unprecedented potentials for fur- reduce the number of cells needed for a reliable WGA
thering precision oncology. Realization of a “liquid bi- reaction. Our modified MDA protocol based on the
opsy” through CTC sequencing helps avoid invasive and principle of small volume MDA required only a 96-well
costly traditional biopsy procedures. Moreover, it allows plate and negated the need for labor-intensive protocols
for a dynamic monitoring of evolution in tumor genome and costly special equipment described in previous
in response to cancer therapy [24]. However, most works. Although this meant that we could not achieve
downstream applications (i.e. aCGH, NGS) necessitate a the nanoliter reaction volumes described in prior stud-
whole genome amplification (WGA) step prior to the ies, we still observed a meaningful improvement in
analysis of single/few-cell, which is known to introduce sample-to-sample reproducibility and overall improve-
errors and biases [11–14, 25]. In the current study, we ment in our defined QC-scores. Importantly, we noted
describedour strategic approach for selectingand gener- reduced amplification reaction gains associated with our
ating optimal WGA products for analysis using aCGH modified MDA protocol, consistent with observations
and NGS. We presented and validated a quality control made inpriorliteratures onsmallvolumeMDA[18, 22].
assay for WGA product quality, and introduced a modi- As previously described, volume restriction limits ampli-
fied MDA protocol that helped improve the reproduci- fication reaction and decreases the overall reaction gain.
bility and reliability of the existing WGA process. However, this process can also restrict the degree of
Finally, we showed that by combining our QC-criteria, aberrant amplification of certain preferred sequences,
modified MDA protocol and using as little as 5–10 cells resulting in a more uniform product overall. Although
as a starting template, a WGA product with high fidelity an excessively low DNA yield may signify either a poor
totheunamplifiedtemplate DNAcouldbeobtained. starting DNA quality or an ineffective amplification
Multiple groups have published on performance com- process, an exceedingly high amplification gain is also
parison of different WGA methods over the recent associated with an increased amplification bias [22]. Par-
years, with significant differences in results and conclu- ticularly in MDA, reaction gain >107 has been shown to
sions existing between these studies [14, 16, 26]. How- correlate with poor amplification quality [16]. These
ever, the majority of these studies lack analysis of the findings suggest that the degree of amplification gain, or
quality of WGA samples being used. Given the known DNA yield, is likely an important parameter of amplified
high sample-to-sample variability in single-cell WGA DNAquality.
quality, no meaningful interpretation and comparison of In addition to improving the WGA process itself,
WGA-DNA derived data can be made without a well bioinformatic computational tools are also available to
defined quality standard. Our proposed 8-cancer gene improve the quality of sequencing data. A number of
QC assay has the potential to fulfill this gap, as it suc- algorithms have been described to date, each with estab-
cessfully predicts the performances of WGA-DNA in lished efficacy in optimizing the sequencing data from
downstream analysis by aCGH and NGS, in addition to single/few-cell amplified DNA [21, 32–35]. We utilized
providinglimitedsequencinginformation. the Ginkgo (http://qb.cshl.edu/ginkgo) platform in our
Prior reports have described multiplex PCR based QC current analysis, which incorporates built-in algorithms
assay for formalin-fixed paraffin embedded (FFPE) DNA designed specifically to address the challenges associated
and Ampli1 WGA-DNA [20, 27]. The Ampli1 WGA is a with single/few-cell amplified DNA [21]. Notably, des-
PCR-based amplification process using non-random pite the data optimization achieved by the Ginkgo algo-
primers, which is fundamentally dissimilar to the MDA rithm, the failed QC sample remained highly biased and
method of non-PCR based isothermal amplification was unsuitable for accurate interpretation. Although
process using random primers. To date, a QC assay statistical correction algorithms remain highly useful in
specific to the MDA method remains to be defined. Fur- single/few-cell amplified DNA analysis, generation and
thermore, no study to date has described a QC assay selection of optimal amplified DNA likely remains a key
designed specifically to evaluate for genes implicated in factor insuccessfuldownstream sequencing analysis.