Substrate preparation

Four 100 μl infill reactions were set up per substrate using 100 pmol template oligonucleotides (T1-T15 in S1 Method) and 200 pmol biotinylated infill primer (P1 in S1 Method) in 1x Klenow buffer (Thermo Scientific) supplemented to 200 μM dNTPs (Invitrogen). Hybridization was performed by heating to 95°C for 5 minutes, followed by cooling to 25°C. The reactions were then supplemented with 10 U Klenow fragment (Thermo Scientific) and incubated at 37°C for 30 min. Klenow fragment was heat inactivated at 65°C for 20 min. Each reaction was purified and concentrated to 40 μl using Qiaquick columns (Qiagen) and eluting in nuclease free water. Final concentrations were determined by Nanodrop (Thermo Scientific). Each pair of adapter oligonucleotide (A1–A4 in S1 Method) were annealed in 1x PNK buffer (NEB) by incubating 20 μM of each oligonucleotide at 94°C for 3 minutes and slowly ramping down to 25°C.

Preparing the library

Each prepared substrate was treated with corresponding RE and incubated for 1 hour as described in Table 1, followed by heat inactivation. Dynal M-280 streptavidin beads (Invitrogen), 100 μl/sample were washed in 2x bind and wash buffer (2x BW: 10 mM Tris, 1 mM EDTA, 2 M NaCl), added to each reaction and incubated at room temperature (RT) for 30 min. The beads were washed 3 times in 1x TE and resuspended in 100 μl 1x BW. Half of the beads were resuspended in 22 μl 1x T4 DNA Ligase buffer (Thermo Scientific), and supplemented with 20 pmol adapter corresponding to the product formed (A1–A4 in S1 Method) and 5 U T4 DNA ligase (Thermo Scientific), incubated 1 hour at RT with agitation. The supernatant was removed and beads were washed. ssDNA was extracted by adding 12 μl 0.15 M NaOH for 5 min at RT with agitation. The supernatant was removed and neutralized with 6 μl 0.3 M HCl and 3 μl 10x ABL (100 mM Tris, 1 mM MgCl₂). Ligation products were diluted 20×–500× as input to PCR. Four 50 μl PCR reactions per sample were set up with 1x HotStarTaq PCR Buffer (Qiagen), 200 nM dNTPs (Qiagen), 50 nM of primer (P2 and P3 in S1 Method), and 2.5 U HotStarTaq DNA Polymerase (Qiagen). The PCR was run at 95°C 15 min, [94°C 1 min, 55.5°C 1 min, 72°C 1 min] × 22 cycles, 72°C 10 min, 4°C hold. The four reactions of each sample were pooled and purified using MinElute spin columns (Qiagen) to 20 μl EB (Qiagen). 60 ng per sample were pooled and concentrated using a MinElute column to 13 μl, which was used for sequencing.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Reaction conditions for each enzyme used in the study.

https://doi.org/10.1371/journal.pone.0117059.t001

Sequencing

The pooled prepared samples were sent for sequencing to GATC Biotech AG Next Gen Lab (Germany) and sequenced using an Illumina Genome Analyzer II, 1×76 bp obtaining 11.8 million reads. The raw sequencing reads are available at the NCBI Sequence Read Archive under accession SRS627913.

Analysing the data

The sequence files were processed with custom made Python scripts. First, for each read in the raw data, the first six bases were matched to the indices on the adapters used for ligation. A hamming distance of two or less to the synthesized adapter was considered a match. A cleavage distance search was initially conducted on the entire reads, where perfect matches to 10 bp immediately before the randomized region of the designed constructs (digestion products in S1 Method) was used to determine cleavage distance. As the primary analysis the recognition sequence of each RE was matched in descending order from the enzyme with longest cleavage distance only matching at the preferred distance of cleavage for each RE +/− 2 bp. For each enzyme with a perfect match, at each distance (−2, −1, 0, +1, +2 bp), a sequence logo (Weblogo 3.4, [21] was made from all the unique sequences classified to that distance. Non-unique sequences were used for FauI and SmuI, as the shorter cleavage distance of these RE (6 bp for non-slipp cleavage) limited the available number of variants. To assay the impact of sequence quality, a very stringent filtering step where all reads were trimmed to 45 bp and any read containing a base with less then Q20 was discarded. The filtered file was then analysed in the same way as the non-filtered.

Slippage

For the set of enzymes studied the RE target templates digested by 1 or 2 bp off the defined distance varied from 1–54% (average 13.1%, median 7%) (Table 2). For this set of enzymes, roughly three classes of slippage frequency was observed, defined as low slippage (0–2%), intermediate slippage (2–15%), and high slippage (>15%). Low slippage enzymes were BseRI, AcuI, BbvI, BpmI and FokI, intermediate slippage enzymes were GsuI, BsgI, Eco57I, Eco57MI, SmuI and FauI, and high slippage enzymes were BpuEI, MmeI and EcoP15I (Fig. 2, Table 2). The enzyme least prone to slip was BseRI with 1.1% products +/− 2 bp of its defined distance, where +1 was the most common slippage product (0.9%). The enzyme most prone to slip was MmeI with a total slippage of 54% where +1 was almost the only product (Table 2, S1 Table). The second enzyme most prone to slip was BpuEI with 41% of the product formed missing one base in length. EcoP15I, the only Type III enzyme in this study, slipped in 26% of the cases. For Eco57MI, a modified version of Eco57I to recognize CTGRAG, the alteration between A to G slightly influenced the enzyme’s slippage, and Eco57MI displayed a slight increase in slippage as compared to the original enzyme (Eco57I). The detected slippage for Eco57I was 7%, and 11% for CTGAAG vs. 13% for CTGGAG (Table 2). The isoschizomers also varied. AcuI displayed 1.1% slippage and Eco57I 7.2%, BpmI 1.5% and GsuI 5.1%, and SmuI 12.4% and FauI 15.2%. Slipping is clearly not a random event of +/−1 base, but different enzymes show differing propensity for a particular direction. We found that a 2 bp variation is very rare. FauI showed the highest general tendency to slip +2 bp, at 0.31% and BpuEI had the highest tendency for −2 at 0.26%.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Slippage distributions for the Type IIS enzymes tested.

Bar plots showing the percentage of reads obtained at the expected distance (0) and +/− 2 bp away for different Type IIS enzymes. The enzymes display a wide range of length specificity but usually not more than 1 bp away from the preferred distance.

https://doi.org/10.1371/journal.pone.0117059.g002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Distribution of total slippage detected of the restriction enzymes.

https://doi.org/10.1371/journal.pone.0117059.t002

Sequence dependence

The experimental design also enabled us to investigate the effect on sequence context between the recognition and cleavage sites. This sequence dependency showed a common preference for adenines between the site of recognition and the site for cleavage. A general preference for cytosines proximal to the adapter ligation site is also observed, but not universal (S2–S9 Figs.). As the case for slippage, variation of sequence context is not limited to different REs but also between isoschizomers. This difference is perhaps most striking between FauI and SmuI, where FauI displays a clear preference for adenine-containing substrates while SmuI does not (Fig. 3). The most common slippage of SmuI (+1 bp in 12% of the sequences) also seems to be promoted by a thymine right before the cleavage position where this is not the case for FauI (Fig. 3). The T is even more conserved in the few reads (0.08%) detected with +2 slippage (Fig. 3). For isoschizomers AcuI and Eco57I the sequence dependence is similar, but a slight preference for adenines can be seen for Eco57I and is not present for AcuI (S2 Fig.). Isoschizomers BpmI and GsuI both show similar behaviour. BpmI shows a faint preference for A, whereas GsuI shows no sequence dependence at the preferred length of cleavage. There is a slight bias for cytosine at the site of ligation shared for both of the enzymes (S3 Fig.). BpmI seems to have a stronger dependence on base composition for +1 slippage, with an adenine potentially influencing this behaviour 4 bp away from cleavage (S3 Fig.). This could potentially explain the generally lower frequency of slipped products for BpmI than GsuI. The same enzyme, Eco57MI, showed differing sequence context depending on which recognition sequence was used, where CTGAAG was less dependent on sequence content, and where CTGGAG was slightly inclined to interact with A rich constructs (S4 Fig.). The sequence dependence assay revealed that the +2 slippage of Ecop15I, actually was a −1 bp slippage of a randomly occurring repetition of the recognition sequence (CTGCTGCTG instead of CTGCTG, see S5 Fig.).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Slippage and sequence dependence of isoschizomers FauI and SmuI.

Sequence logos were generated showing the most frequent base per position (in bits) in the region between recognition and cleavage. The detected slippage from −2 to 2 bp are shown independently (row 1 through 5). Recognition site is not included in the plot but was identified immediately left to the plots shown. In the corner the number of sequences identified (and used for the plot) and the relative frequency for that enzyme are shown.

https://doi.org/10.1371/journal.pone.0117059.g003

Slippage influencing factors

In this experiment enzyme-to-substrate concentration ratio was kept constant (2U of enzyme per pmol of substrate, except for BsgI, where 1U were used per pmol substrate). Enzyme-to-substrate ratios could potentially influence cleavage effectiveness, e.g. a FokI monomer has been shown to be inactive until dimerising with a second monomer [22, 23]. The varying enzyme concentrations from the suppliers lead to variations in glycerol contents in the reactions used for this experiment (S2 Table). Glycerol concentration has long been known to promote star activity [24, 25], however, no correlation was found between glycerol concentration in the reaction and slippage activity (S11 Fig.) or clear trend for sequence dependence (S2–S9 Figs.). Yet, to further investigate the glycerol factor a subset of five enzymes were selected for a new experiment were the glycerol levels during restriction were kept at 1% (S2 Method, S3–S4 Tables). We can report that a lower glycerol level has a minor non-systematic effect on the slippage levels (S5 Table). We observe both a slippage increase and a slippage decrease and the categorization into low, intermediate and high mode did not change except for BpmI (from low to intermediate) (S5 Table). To validate that the result was not an effect of biased starting material, a follow-up study was conducted where the digestion substrates for the enzymes used to generate the additional dataset were sequenced. Analysis of the randomized region spanning from the recognition to two bases downstream of the digestion site revealed no sequence biases in the starting material (S2 Method, S10 Fig.).

By looking at the 15 enzymes as a group we searched for simple correlations in the dataset, i.e. reaction temperature (S12 Fig.), preferred S-adenosylmethionine (SAM) concentration (S13 Fig.), cleavage distance (S14 Fig.), number of sequence reads obtained (S15 Fig.), and organism source (Table 2), but could not establish any patterns.

Quality filtering

When all reads were trimmed to 45 bp and any that contained a base with quality score below Q20 (1/100 probability of wrong call) were filtered out, nearly 2.1 million reads remained, and nearly 1.9 million (90.5%) of these were classified as an expected construct and used for slippage analysis. On average the stringent quality filtering resulted in a 0.99 pp shift in slippage. E.g. going from 1.10% to 0.82% total slippage for BseRI (0.28 pp reduction), and from 53.6% to 52.2% for MmeI (1.4 pp reduction) (S6 Table). This indicates a slight inflation of reported slippage due to sequencing errors. Interestingly AcuI exhibited a higher slippage in the quality filtered data (3.35% vs. 1.10% slippage). 3.35% slippage is however more in line with the result gained in the initial approach of analysing the entire read in a search for 10 bp recognition sequence right before the randomized region (total slippage 3.86%) (S6 Table).

Previous studies

The enzymes with previous slippage data available (AcuI, MmeI and EcoP15I) agreed well with our findings (Fig. 2)(compare S4 Fig. in Drmanac et al [8] for AcuI and EcoP15I; and S2 Fig. in Shendure et. al [6] for MmeI). The sequence dependence for the endonuclease activity has previously been reported for MmeI [6] and is to our knowledge the only enzyme for which this has been published using an extensive dataset. MmeI showed very little sequence dependence in general in our study, which is concordant with the previous report. The previous study reported on an adenine preference close to the recognition sequence, and reasoned that this could have been introduced during A-tailing when constructing the library. We do not find any preference proximal to the recognition sequence, which supports that conclusion. The previous study was based on mapping reads to a reference genome and found a slight bias downstream of the cleavage site. In our system this potential preference is cleaved away, but the slight bias for cytosine immediately upstream the site of cleavage is supported in our findings, albeit to a higher degree. It is reasonable that a C or G will result in increased ligation efficiency, and is also in general (but not always) the preferred base in the immediate proximity to cleavage in our data.

Accuracy

The Illumina system is well suited for the task of assaying this type of libraries as (A) it is cost-effective and high throughput, and (B) it has a low insertion-deletion (indel) error frequency. Both of these features contribute to making a sensitive protocol for slippage analysis. The indel rate has been determined to 4 ppm for mapping 100 bp reads to the phiX genome [26]. The impact of sequence quality was investigated by reanalysing the data with a stringent raw data-filtering step. Except for AcuI, this step resulted in a reduction of slippage as compared to the non-filtered analysis, indicating a slightly inflated result, but in general low contribution from sequencing errors.