Precision oncology using a limited number of cells: optimization of whole genome amplification products for sequencing applications

Pancreatic cancer cell line HPAF-II was obtained from American Type Culture Collection (ATCC, Virginia, USA), and grown using EMEM medium (ATCC) supplemented with 10% fetal bovine serum (ATCC) and 100 U/mL penicillin-streptomycin (ATCC). All cell lines were grown at 37 °C with 5% CO₂ and were routinely passaged at 80% confluence using an iso-osmotic sodium citrate solution for cell release (Thermo, Massachusetts, USA).

In preparation for laser microdissection, cells were released from the culture plates using the iso-osmotic sodium citrate solution (Thermo). Following a wash with the culture medium, each cell line was diluted to a density of 1000 cells per 100 μL. Approximately 1000 cells (100 μL) were smeared on PEN membrane slides (Leica, Wetzlar, Germany), air-dried for 10 min, and fixed with 100 μL of 100% ethanol. Cells were then isolated using the PALM MicroBeam laser microdissection system (Zeiss, Oberkochen, Germany). 1, 5, 10 or 20 cells were laser microdissected and collected into 200 μL opaque tube caps (Zeiss) using the laser pressure catapult function. Cell transfer to the tube cap was confirmed by imaging the cap prior to cap closure using the cap-check function.

Isolated cells were subjected to genomic DNA isolation and WGA using one of the three commercially available single-cell WGA kits according to the manufacturer’s protocol: REPLI-g Single-Cell Kit (Qiagen, California, USA), Multiple Annealing and Looping Based Amplification Cycles (MALBAC) Single-cell WGA Kit (Yikon Genomics, Beijing, China) and PicoPlex Single-Cell WGA Kit (Rubicon, Michigan). Reactions were performed a minimum of 3 times for all conditions tested. WGA products were purified using the QIAquick PCR Purification Kit (Qiagen) and quantified with NanoDrop 2000 (Thermo), in keeping with a previously described CTC molecular analysis methodology [15].

Modified MDA protocol was performed using the same reagents available from the REPLI-g Single Cell Kit (Qiagen), with modifications made in the cell lysis step and the final amplification step. Cell lysis was performed over a course of 30 min to ensure complete lysis of genomic material from isolated cells (as opposed to 10 min recommended by the manufacturer). Prior to the final amplification step, the 50-ul MDA reaction mix was mixed for 30 s by pipetting up and down and then partitioned into 16 individual reactions (approximately 3ul each) and the amplification occurred at 30 °C for 8 h. This resulted in 16 individual MDA reactions with reduced individual reaction volumes (3ul instead of 50ul), all taking place in parallel. Following the amplification step, contents within the 16-wells were collected into one tube, followed by the purification and quantification step as described above.

Prior literatures have described the role of multiplex PCR in evaluating DNA quality isolated from formalin-fixed paraffin embedded (FFPE) tissues prior to downstream analysis using aCGH [20]. Based on this concept, we developed an 8-gene multiplex PCR/quantitative PCR (qPCR) based QC assay for evaluation of WGA-DNA quality. We selectively chose for 8 cancer-related genes that are considered molecular targets for targeted cancer therapy and therefore highly implicated in guiding therapeutic decisions. This way, the QC assay identifies WGA products suitable for precision oncology applications by evaluating for the coverage and “accessibility” of these important genomic locations within the WGA-DNA. Genes evaluated by the QC assay included BRAF, EGFR, KIT, KRAS, NRAS, PIK3CA, PTEN and P53.

Score of 0–8 was assigned based on the number of genes successfully amplified and detected using multiplex PCR and qPCR. Failure to detect one or more of the 8-gene from the WGA product signifies lack of coverage or potential ADO during the WGA process, both of which indicate a poor quality WGA product and jeopardizes accurate representation of the original starting genome. Thus, we only gave WGA-DNA products with 8 out of 8 score a “pass” and deemed them fitting for further downstream analysis.

A secondary QC-score was generated to confirm the internal validity of the original QC-score based on the 8-cancer related genes. The secondary QC-score was generated from a distinct set of 8 primer pairs representing 8 housekeeping genes: NDUFA, UQCRC, ACTG, CYB5A, GABA-RAPL, MIF, MYC, PRPH.

WGA products were subjected to multiplex PCR pre-amplification followed by qPCR assay for detection of amplified targeted genes. Multiplex PCR was performed using primer sets representing genes mentioned above. Primer sets used for the primary QC assay included: BRAF (forward 5′ – TAC TGC TCT TTC TTC TCC AAC AC – 3′; reverse 5′ – CCT GAT TGT ATT TGA GAT CTA GTA GGG – 3′) EGFR (forward 5′-CAG CCT TCT CCG TAA TTA GCA T – 3′; reverse 5′ – TGA CAC AGA TAA TTG TCC CAC AG – 3′), KIT (forward 5′ – GGC ATT GAG GAG GGA TAG TAA AT – 3′; reverse 5′ – CTG AAC AAT TTG CTT GAA TGT TGG – 3′), KRAS (forward 5′ – GTG TTA CTT ACC TGT CTT GTC TTTG – 3′; reverse 5′ – GCC TTC TAG AAC AGT AGA CAC AA – 3′), NRAS (forward 5′ – AAT GGA ATC CCG TAA CTC TTG G – 3′; reverse 5′ – GAT GAT GTA CCT ATG GTG CTA GTG – 3′), PIK3CA (forward 5′ – AGG GCA AAT AAT AGT GGT GAT CT – 3′; reverse 5′ – CAG CAA TTA CTT GTT CTG GTA CAC – 3′), PTEN (forward 5′ – CTT TCT CTA GGT GAA GCT GTA CT – 3′; reverse 5′ – GGT TCA TTG TCA CTA ACA TCT GG – 3′) and P53 (forward 5′ – AAG AGA AGC AAG AGG CAG TAA G – 3′; reverse 5′ – CTT AGG CTC CAG AAA GGA CAA G – 3′). Primer sets used for the secondary QC assay included: NDUFA7 (forward 5′ – TGC TCT GGA TGT GAA GAT GCC A – 3′; reverse – 5′ – TTC CAG GTA AAT CCA GCC CAG G – 3′), UQCRC1 (forward 5′ – CAG CCA GTC AGC ATC ATC CAA C – 3′; reverse 5′ – GAA AGC CGG ATT GCG GTA ACA T - 3′), ACTG1 (forward 5′ - GCT CAA TGG GGT ACT TCA GGG T – 3′; reverse 5′ – GTG GAC GTT ACG TAA AAG GCC C – 3′), CYB5A (forward 5′ – GGC AAC GCT TAG ACT CTG TGT G – 3′; reverse 5′ – CTG CCC TTG GCC TAA CTA ACC T – 3′), GABARAPL2 (forward 5′ – CCA GCC AAT TCA TGA GTC GGT G – 3′; reverse 5′ – CCT GAC AAC TCG CAA GTA GCA C – 3′), MIF (forward 5′ – AGA AGT CAG GCA CGT AGC TCA G – 3′; reverse 5′ – GGC ACG TTG GTG TTT ACG ATG A – 3′), MYC (forward 5′ – GGA TAG CTC TGC AAG GGG AGA G – 3′; reverse 5′ –TCG TCG CAG TAG AAA TAC GGC T – 3′), PRPH (forward 5′ – GTT CCT CAA GAA GCT GCA CGA G – 3′; reverse 5′ – CGT TAG ACT CTG GAT CTG GCG T – 3′). Details of multiplex PCR pre-amplification is available in the Additional file 1: Supplementary Material. Briefly, PCR reactions were carried out on a C1000 Thermal Cycler (Bio-Rad) with the Multiplex PCR Plus Kit (Qiagen) using total volumes of 17 uL per reaction. The reaction conditions were as follows: 95 °C for 15 min, denaturation at 94 °C for 30 s, annealing at 64 °C for 30 s, and extension at 72 °C for 30 s for a total of 10 cycles, with a final step of 72 °C for 10 min.

Following the pre-amplification step by multiplex PCR, the resulting amplified products were analyzed and detected using qPCR. Reactions took place on BioRad CFX-96 real time system (BioRad) using the QuantiTect SYBR Green PCR Kit (Qiagen). A 25-ul reaction mixtures were prepared, which contained 12.5 ul of the SYBR Green PCR Master Mix (Qiagen), 9.5 ul of RNA-grade water, 1 ul of individual primers sets (10uM) and 2ul of the multiplex PCR product. The reaction condition were as follows: 95 °C for 15 min, followed by 35 cycles of 94 °C for 15 s, 64 °C for 20 s, and 72 °C for 20 s. The plate was read following the extension step at 72 °C. Melting curve analysis was performed between 70 and 95 °C at 0.5 °C intervals. Real-time PCR data were reviewed and analyzed using the CFX manager (BioRad). Specificity of the PCR amplification product was determined using melting curve analysis. PCR products with melting-temperature (Tm) matching the expected value based on primer sequence, and threshold cycle (Ct) <30 were counted as reliable amplification and detection.

Sample WGA-DNA and reference DNA were differentially labeled with cyanine-3 (CY3) and cyanine-5 (Cy5) dyes using the GenetiSure Amplification and Labeling Kit (Agilent) according to the manufacturer’s protocol. Briefly, a 15.5 ul reaction mixture containing sample or reference DNA, Random Primer Mix and water was placed in 98 °C for 3 min for DNA denaturation and transferred to 4 °C for 3 min. Next, 9.5 ul of Labeling Master Mix containing 5ul of 5× Reaction Buffer, 2.5ul of 10× dNTP Mix, 1.5 ul of Cy5-dUTP/Cy3-dUTP and 0.5 Exo (−) Klenow were added to each tube containing the sample/reference. DNA labeling reaction was performed in 37 °C for 45 min followed by inactivation step in 65C for 10 min. Labeled DNA was purified using the Post Labeling Purification Columns (Agilent) according to the manufacturer’s protocol. Purified labeled DNA samples were prepared for hybridization, which took place on Agilent 8x60K CGH microarray slides at 67 °C for 6 h. Following the hybridization, the slides were washed per manufacturer protocol, and prepared for scanning using the Agilent SureScan Microarray Scanner (Agilent). Microarray image was prepared and analyzed using the Agilent CytoGenomics software (Agilent).

Purified WGA products were sheared to generate DNA fragments of 350 bps using the Covaris sonicator (Covaris). Following cleanup of the sonicated DNA, end-repair and ligation were performed using the KAPA DNA Library Preparation Kit (KAPA Biosystems) according to the manufacturer’s protocol, followed by library amplification by PCR. Sequencing was performed on an Illumina HiSeq 2000 using random primers and pair-end reads of 75 bps (2X75bps).

The sequencing read data was analyzed using Ginkgo (http://​qb.​cshl.​edu/​ginkgo), an open-source platform for the analysis of single-cell copy-number variations (CNV). This analysis platform was developed specifically to address the unique challenges of single-cell sequencing data, and incorporates built-in computational tools for data optimization. Specifically, limitations associated with low depth of sequencing coverage, amplification bias and inflated read counts from poorly assembled genomic regions were addressed in Ginkgo’s analysis. Detailed description of the computational methodology and correction tools are provided by Garvin et al. [21]. Mapped reads were binned using variable length-binning and underwent GC bias normalization prior to segmentation using circular binary segmentation. Quality metrics data, including Lorenz curve, histogram of read count distribution and index of dispersion were obtained as a part of the Ginkgo analysis pipeline. Sequencing data visualization was performed using Nexus software (Biodiscovery, inc.).

PCR amplification of KRAS exon 2 was performed using the following primer (Integrated DNA Technologies): Forward 5′ – AAG GTA CTG GTG GAG TAT TTG – 3′ and Reverse 5′ – GTA CTC ATG AAA ATG GTC AGA G – 3′, with expected amplicon length of 295 bps. PCR reactions were carried out on a C1000 Thermal Cycler (Bio-Rad) with Platinum PCR SuperMix High Fidelity Kit (Invitrogen) using total volumes of 50 μL per reaction according to the manufacturer’s protocol. The reaction conditions were as follows: denaturation at 94 °C for 30 s, annealing at 55 °C for 30 s, and extension at 68 °C for 45 s for a total of 40 cycles. The PCR products were purified using the QIAquick PCR Purification Kit (Qiagen) and eluted into 50 uL of nuclease-free water (Qiagen). DNA was diluted to a concentration of 10 ng/uL based on Nanodrop quantification of the PCR product. Automated dideoxy terminator sequencing was performed by capillary electrophoresis by the UCLA GenoSeq Core on an ABI 3730 DNA analyzer using Big Dye Terminator chemistry (Applied Biosystems). All sequences were analyzed by manual inspection of the individual trace files using Four Peaks (Nucleobytes).

We applied our 8-gene QC assay to single-cell WGA-DNA samples. We tested three different WGA reaction methods: MDA, MALBAC and Picoplex. Total of 18 individual HPAF-II cells were isolated using laser micro-dissection and used in the respective WGA methods. The experiment was repeated 6 times in order to account for the variability expected with the single-cell WGA process. QC assay was performed and the QC-score was assigned for each WGA product.

Despite using a clonally expanded cell line and performing WGA reactions in a parallel fashion under the same condition, we noted significant variability in sample-to-sample WGA-DNA quality as assessed by the 8-gene QC assay (Fig. 2). QC-scores for MDA amplified single-cells ranged between 1 and 8. Similar variability in amplified DNA quality was noted for MALBAC, with QC-scores ranging between 2 and 6. Although Picoplex resulted in less quality disparity, majority of samples achieved scores of only 3. QC-score profiles for the MDA and MALBAC methods were overall similar, but the WGA product that passed our QC criteria (QC-score = 8) was only found in the single-cell sample amplified with using the MDA method.

We noted significant random variations in WGA product quality when using single-cells as starting template (Fig. 2). Increasing the amount of starting genomic template has been shown to improve WGA product quality [19, 22]. Thus, we hypothesized that there would be a certain “threshold” number of cellular input above which a reliable and reproducible WGA process is possible, i.e. a number of cells which will achieve a “pass” in our QC step on a consistent basis.

In order to test this hypothesis, 5-cell, 10-cell and 20-cell samples of HPAF-II cells were cut and isolated using a laser micro-dissector. WGA was performed using the MDA method, followed by the QC assay and assignment of the QC-scores (Fig. 3a). Less variability in amplified DNA quality was noted when 5 or more cells were used as starting genomic template. All samples persistently achieved a QC-score of ≥5 (5-cell) or ≥6 (10-cell), as opposed to single-cell WGA reactions with scores ranging from 1 to 8. Furthermore, at least one of the triplicates in 5-cell and 10-cell group passed the QC-step (QC-score = 8). However, a highly reliable WGA process with all of the WGA products passing the QC step could not be achieved until 20 cells were used as starting genomic template input.

With the existing MDA protocol, multiple single-cell WGA reactions must be performed in order to obtain one high-quality WGA product suitable for downstream analysis (Fig. 2). Even when more than a single-cell was used, a reliable WGA reaction with its amplified DNA product passing the QC step on a consistent basis could not be achieved until 20 cells were used (Fig. 3a). For many GI cancers that are known to generate only few CTCs (i.e. pancreas ductal adenocarcinoma), obtaining 20 CTCs from a single peripheral blood draw may be unrealistic.

Given these problems, we sought to develop a modified WGA protocol that would reduce the sample-to-sample variability in DNA quality and lower the number of starting cells needed to achieve a reliable WGA process. Multiple prior reports have described the benefit of performing MDA by partitioning the reaction into parallel smaller volume reactions [16, 18, 22]. The small reaction volumes and template DNA partitioning restricts the degree of aberrant preferential amplification, leading to a more uniform WGA process overall. Based on this principle, we developed a modified MDA protocol with key changes in two aspects: (1) increasing the cell lies step from 10 min to a minimum of 30 min to ensure adequate release of genomic materials from cells, and (2) partitioning the final MDA reaction into 16 individual reactions containing ~3ul each on a 96-well plate, prior to the final isothermal amplification step at 30 °C for 8 h.

WGA products obtained using this modified MDA protocol resulted in improved reproducibility and higher QC-scores overall (Fig. 3b). Amplification reaction gain was overall lower for the modified MDA products compared to the conventional MDA, with reduction of approximately 50% on average for each sample undergoing the modified MDA protocol (Additional file 2: Table S1). For single-cells, 2 of 3 WGA reactions resulted in a product with the perfect QC-score of 8, compared to 1 in 6 WGA reactions using the conventional MDA protocol. Importantly, for 5-cell and 10-cell samples, all WGA reactions generated products that passed our QC criteria (Fig. 3b). The modified MDA protocol in our hands effectively reduced the “threshold” number of cells needed for a reliable WGA reaction down to 5 cells from 20 cells, well within the number of CTCs attainable from a single peripheral blood draw.

Our proposed 8-gene QC assay evaluates 8 genomic locations within a WGA product. Evaluating more loci in theory provides more comprehensive evaluation of the amplified DNA product. We tested whether there was any benefit to evaluating more genetic loci beyond the 8-cancer genes during the QC process. To test for this, we performed a secondary QC assay using a different set of 8-housekeeping genes on the same single-cell conventional MDA products, as well as the 1/5/10-cell modified MDA products (Additional file 2: Table S2A and B). The original (8-cancer gene) and the secondary (8-housekeeping genes) QC assays generated highly concordant QC-scores for all of the samples tested. All of the samples that passed the QC-step based on the original QC assay also passed using the secondary QC assay. For the sole modified MDA sample with the low QC-score (sample 1-C), the secondary QC also resulted in a similarly low QC-score indicating poor quality WGA-DNA. Thus, evaluation of more genetic loci beyond the original 8-cancer genes did not provide any additional QC information and did not change QC results for any of the samples tested.

We performed point mutation detection, aCGH and NGS using single-cell WGA products that either passed (WGA-QCpass) or failed (WGA-QCfail) the QC-step. Figures 4 and 5 illustrate the relationship between QC results and performances in various downstream applications.

Point mutation detection within the KRAS gene was successful in both QCpass and QCfail WGA products (Fig. 4a). However, while the QCpass WGA sample exhibited a mutant to wild-type allelic ratio closely matching that of the batch gDNA (HPAF-II cells are known to have 4 to 1 mutant:wild-type allelic ratio), the QCfail WGA sample exhibited evidence of amplification bias, demonstrated by an altered allelic ratio.

When we compared aCGH performances between the two single-cell WGA products, WGA-QCpass resulted in a notably better derivative log2 ratio spread (DLRS) and signal-to-noise ratio compared to WGA-QCfail (Fig. 4d). DLRS is a key quality metric for aCGH data measuring the point-to-point consistency or “noisiness” in data, with high values indicating poor signal-to-noise relationship and difficulty in assessing true copy number variation (CNV) status. The poor quality metrics (DLRS: 1.3, signal-to-noise ratio: 3.9) associated with WGA-QCfail renders its aCGH data unfit for meaningful analysis and interpretation. On the other hand, DLRS of 0.72 and signal-to-noise ratio of 53.6 associated with WGA-QCpass meets the quality threshold for single-cell derived aCGH set forth by Agilent Technologies [23]. Gain and loss profiles generated from aCGH data are as shown in Fig. 4c. Comparison between WGA-QCpass and WGA-QCfail reveals disparities in single-cell aCGH profiles, even though clonally expanded HPAF-II cells were used for both. When compared to the unamplified batch gDNA, we found multiple areas of alterations that were detected with WGA-QCpass, but not with WGA-QCfail (Fig. 4a, red arrows).

The same two single-cell WGA products were also analyzed by massive multiplex short read sequencing. Figure 5 shows the quality metrics of sequencing data associated with each WGA-DNA sample and unamplified batch gDNA. When we compare the two single-cell WGA products (WGA-QCpass and WGA-QCfail), Lorenz curves, histograms of read count frequency and indexes of dispersion all indicated a superior NGS data quality with higher coverage uniformity for WGA-QCpass compared to WGA-QCfail. The 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). The wider the curve below the line of y = x, the lower the coverage uniformity and greater the amplification bias. As can be seen in Fig. 5a, WGA-QCpass resulted in more uniform distribution of read depth compared to WGA-QCfail. The histogram of read count frequency also provides information on coverage dispersion (Fig. 5b). Histograms with a wide range of distribution, as seen in the sequencing data obtained using WGA-QCfail, indicates a greater degree of amplification bias. Genomic profiles generated from NGS sequencing data are as shown in Fig. 5c. Comparison between WGA-QCpass and WGA-QCfail reveals significantly more “noise” in the genomic profile derived from the latter, with a greater index of dispersion compared to that derived from WGA-QCpass.

Although the single-cell WGA product passing our QC criteria (WGA-QCpass) performed well in point mutation detection, aCGH and NGS, room for improvement still existed when compared to the performance metrics of unamplified batch gDNA (Figs. 4 and 5). Increasing the number of starting cells has been shown to improve WGA product quality [15, 19, 22]. However, the number of cells needed for generating a WGA product that faithfully represents the unamplified, original DNA in downstream molecular analysis remains to be defined.

Using our modified MDA protocol, we lowered the numbers of cells needed to reliably pass our QC-step down to 5 cells from 20 cells. However, it remained to be answered how closely 5-cell and 10-cell WGA products approximated the unamplified DNA in aCGH and NGS analysis, and whether obtaining 5 to 10 CTCs from a single liquid biopsy is truly sufficient for sequencing applications within precision oncology strategies. Figure 4b and 5 illustrate the performances of 5-cell and 10-cell modified MDA products (QC score = 8 for both) in aCGH and NGS. As the number of cells used for starting genomic template increased from 1 to 5 to 10 cells, progressive improvement in all quality metrics of aCGH and NGS were noted. Notably, the Lorenz curves for WGA-DNA and unamplified batch gDNA nearly overlapped by 10-cells (Fig. 5a). Furthermore, by 10-cells, WGA-DNA derived aCGH and NGS genomic profiles appeared highly concordant compared to that of unamplified batch gDNA, with progressive reduction in signal “noise” and index of dispersion as the number of cells used increased (Figs. 4b and 5c). Lastly, the resolution limits of 1, 5 and 10-cell amplified products applied to the NGS analysis were assessed. By evaluating for the accuracy of CNV calling at various bin sizes (250 kb, 500 kb, 1 Mb and 2.5 Mb), we found 5 and 10-cells to require a minimum bin size of approximately 250 kb to reproduce the batch DNA result. For single-cell, bin size of approximately 2.5 Mb was needed (Additional file 3: Fig. S1). In summary, using 5–10 cells as the starting template for our modified MDA reaction generated WGA-DNA that closely approximated the unamplified gDNA in both aCGH and NGS performances.