Gene-specific FACS sorting method for target selection in high-throughput amplicon sequencing

Experimental overview

In efforts to perform ultra-deep amplicon sequencing of the hypervariable Dog Leukocyte Antigen (DLA) target using the Roche/454 Genome Sequencer we have developed a strategy based on dual-labelling of emulsion PCR (emPCR) beads and fluorescence activated cell sorting (FACS) to isolate beads carrying the target sequence on their surface as described below and illustrated in Figure 1.

The aim was to eliminate the large amounts of unspecific products present in many amplicon libraries, which are difficult to fully remove by gel or SPRI-based (eg AMPure) purification and take up precious sequencing capacity. We have achieved this by using SYBR Green to label all types of DNA-carrying beads and a target-specific probe facilitating Alexa647 detection and collection of beads carrying the target sequence.

FACS sample preparation

The analyzed sample consisted of an amplicon library prepared from 34 individuals. Amplification of the library resulted in significant formation of unspecific products, as shown in Figure 2A, so the library was purified by size separation with gel electrophoresis which, according to the manufacturer, offers best control of fragment size selection hence giving rise to the best sequencing result. The purified material, which appeared to be free from short unspecific products according to gel analysis (Figure 2B) was used for emulsion PCR and split into two batches for the comparative analysis. The control sample was processed according to the standardized Genome Sequencer FLX procedure, involving enrichment of the DNA-covered beads using Streptavidin-covered magnetic beads to capture the biotinylated protruding strands. The parallel emPCR sample was labelled as outlined above.

Flow cytometric analysis of the labelled sample showed a major peak representing naked beads that had failed to capture and amplify a DNA sequence while the DNA-covered, SYBR Green positive beads formed two minor populations with strong and weak positive SYBR Green fluorescence, respectively, as indicated by green arrows in Figure 1B. Beads with strong Alexa647 fluorescence, which thus carried an amplified target gene fragment on their surface, are indicated by purple brackets in Figure 1C. These beads are also depicted as purple dots in Figure 1D, which illustrates the SYBR Green fluorescence and SSC-A (Side Scatter Area) of the beads. Clearly, beads that yield positive Alexa647 signals also give rise to strong SYBR Green signals. In contrast, beads displaying weak positive SYBR Green signals show no Alexa647 fluorescence. Thus, the level of SYBR Green fluorescence is correlated to the amount of DNA on the bead surface and was used for guidance when setting the FACS sort gate on the Alexa647 parameter. The beads that yielded Alexa647 fluorescence signals (5.5% of the beads in total), and thus carried an amplified target gene fragment on their surface were sorted out for subsequent sequencing (see the square gate in Figure 1D). An additional experiment using FACS probe labelling was conducted confirming the previous results (see below).

Sequence analysis

The average sequence length was greatly improved by FACS enrichment (as shown by the read-length distributions from the Genome Sequencer runs for the control and FACS-sorted samples in Figure 3). Interestingly, despite thorough purification by gel electrophoresis, short unspecific fragments still accounted for a large proportion of the standard-enriched control sample library (Figure 3A). Just 41% of the reads in the control sample were longer than 200 bp, and of these 63% (25% of the total) contained the target gene sequence. In contrast, in the FACS-enriched sample many of the short fragments had been removed (Figure 3B); 69% of the reads were longer than 200 bp and 97% of these (67% of the total) contained the target gene sequence. The additional FACS sorted sample demonstrated that 75% out of all reads exceeded 200 bp and 72% out of all reads contained the target sequence. Thus, overall our FACS sorting protocol gave a nearly three-fold increase in the fraction of target sequence reads obtained.

To assess if any sequence bias was introduced by the FACS enrichment approach we calculated the change in frequencies of sequence counts for each individual (as indicated by their respective tags) in the two FACS-enriched samples as compared to the standard-enriched control sample (Figure 4A). Only minor differences were observed; the largest variations in frequency was for individual 17 and 21 where the frequencies were 2 and 3 percentage points higher in the FACS-sorted sample than in the control sample, respectively. However, for the corresponding individual in the other FACS sorted sample this change was merely 0,1 and 0,2 percentage points. For all other individuals the corresponding changes were approximately 1 percentage point or lower. Hence, the slight differences in frequency observed appear to be of stochastic nature.

We further analyzed the genotypes and found that all genotypes were represented in both FACS-enriched and control samples. Up to five mismatches between FACS probe and target sequences was observed. An example is shown in Figure 4B, where the two consensus sequences from the FACS-sorted and control samples for individual 27 are depicted. The same genotypes, with three or five mismatches against the target-specific probe, were detected in both the FACS-enriched and control sample.

To investigate whether similarity to FACS probe sequence had an effect on the enrichment efficiency of particular genotypes, the frequency of each probe region sequence present for every individual in the FACS sorted sample and the control sample, was calculated (see Additional file 1). The average differences in genotype frequencies between the FACS-enriched sample and the control sample was 10%. Many samples had changes below 5% and investigation of individuals with changes above 5% showed that there was no trend of genotypes with a higher sequence identity to the probe having a higher frequency in the FACS-enriched sample, than in the control sample. Thus, the FACS enrichment protocol does not affect what amplicon sequences are present in the amplicon library after FACS sorting.

Sample preparation

An amplicon library of 34 individuals was prepared by PCR of the Dog Leukocyte Antigen gene DLA-DRB1 exon 2 (target size: 303 bp) [13], introducing identifying tags and the common Roche/454 A/B FLX sequences. The individuals had previously been shown to be heterozygous for this gene by traditional Sanger sequencing. The library was analyzed by agarose gel electrophoresis and ethidium bromide staining. Approximately 400-bp long fragments were extracted by gel purification and the product was then analysed by agarose gel electrophoresis and GelRed (Biotium, Inc., Hayward, CA, USA) staining. EmPCR and strand-separation were carried out according to instructions of the manufacturer of the sequencing system (Roche). Briefly, an emulsion was formed by shaking "mock buffer" with mineral oil, then PCR reagents, the DNA library and beads were shaken into the emulsion. Following clonal amplification of library fragments on beads, the emulsion was broken and beads were collected by centrifugation. For labelling and subsequent FACS enrichment, the protruding DNA strands were removed by washing the collected beads in 0.1 M NaOH according to the manufacturer's protocol. For comparison, a control emPCR reaction mixture was prepared from the same library, and unspecific enrichment of DNA-covered beads was carried out, using streptavidin-covered beads, again according to the manufacturer's instructions.

emPCR product labelling

For detecting beads carrying the amplified gene fragment by FACS, a biotinylated 50 mer was designed to match the central part of the amplicon product and still allow for some mismatches in the hypervariable DLA target sequence. To investigate whether the amount of DNA on the beads could be visualised by FACS, beads were also labelled with random hexamers and SYBR Green.

Dual labelling was carried out by adding 20 pmol random hexamers and 20 pmol biotinylated gene-specific 50 mer probe, both from Eurofins MWG Operon (Ebersberg, Germany) in a total volume of 100 μl 1× Annealing Buffer (1×AB, Roche, Basel, Switzerland) and applying the following conditions: 1 min at 95°C, then decreasing to 25°C over 10 min. 1 μl 50 × SYBR Green I Nucleic Acid Gel Stain (SYBR Green, Molecular Probes Inc., Eugene, OR, USA) and 4 μl streptavidin-conjugated Alexa647 dye (0.2 mg/ml, Invitrogen, Carlsbad, CA, USA) were then added to each sample, and the resulting mixture was incubated for 10 min at room temperature. Samples were then transferred to 5 ml Polystyrene Round-Bottom FACS Tubes (BD Biosciences, Franklin Lakes, NJ, USA) and diluted by adding 500 μl phosphate-buffered saline (PBS) containing 0.1% Pluronic F108 NF surfactant (PBSP) (BASF Corporation, Mount Olive, NJ, USA). As negative FACS control, unreacted beads were labelled in the same manner.

FACS enrichment

FACS enrichment of beads carrying an amplified gene fragment on their surface was achieved by sorting out beads that were Alexa647 positive using a FACSAria instrument (Beckton Dickinson, San Jose, California). Beads were transferred to Ultrafree^®-MC Centrifugal Filter Units (Millipore, Billerica, MA, 0.65 mm pore size) and washed twice in 140 μl 0.1 M NaOH for 40 s and twice in 250 μl 95°C 1×AB for 40 s, followed by re-suspension in 1×AB.

Sequencing

FACS-enriched and standard-enriched control samples were sequenced in separate 1/16-lanes of a picotiter plate according to the instructions of the sequencing system.

Sequence analysis

Sequences were sorted by individual using an in-house developed Perl script, aligned using Muscle [14], Jalview version 2 [15] was employed for visualisation and analysis of the alignments.