Serial number tagging reveals a prominent sequence preference of retrotransposon integration.

Transposable elements (TE) have both negative and positive impact on the biology of their host. As a result, a balance is struck between the host and the TE that relies on directing integration to specific genome territories. The extraordinary capacity of DNA sequencing can create ultra dense maps of integration that are being used to study the mechanisms that position integration. Unfortunately, the great increase in the numbers of insertion sites detected comes with the cost of not knowing which positions are rare targets and which sustain high numbers of insertions. To address this problem we developed the serial number system, a TE tagging method that measures the frequency of integration at single nucleotide positions. We sequenced 1 million insertions of retrotransposon Tf1 in the genome of Schizosaccharomyces pombe and obtained the first profile of integration with frequencies for each individual position. Integration levels at individual nucleotides varied over two orders of magnitude and revealed that sequence recognition plays a key role in positioning integration. The serial number system is a general method that can be applied to determine precise integration maps for retroviruses and gene therapy vectors. Published by Oxford University Press on behalf of Nucleic Acids Research 2014. This work is written by (a) US Government employee(s) and is in the public domain in the US.

INTRODUCTION

Transposable elements and retroviruses have evolved a variety of strategies to populate the genomes of their host. One critical feature of their propagation is the position in the host genome where they integrate. Many transposable elements avoid the disruption of host genes by integrating into non-coding regions, such as heterochromatin or intergenic sequences (1). The long terminal repeat (LTR) retrotransposon Tf1 of Schizosaccharomyces pombe integrates specifically into the promoters of pol II transcribed genes (2–5). One interesting feature of this integration pattern as revealed by high-throughput sequencing of 73 125 integration sites is that the promoters of stress response genes are preferred targets (4).

Retroviruses, such as murine leukemia virus (MLV) and human immunodeficiency virus type one (HIV-1), integrate near the transcription start sites of genes transcribed by pol II or into transcription units, respectively (6–8). Integration of HIV-1 is known to be directed by the host factor LEDGF (9–12) and MLV integration appears to be positioned by BET proteins (13,14). Despite the knowledge that HIV-1 and MLV integrate into specific features of genes and the observation that Tf1 integrates into promoters, it has not been possible to quantify integration at specific regions or sites because there are no methods for measuring insertion frequencies at individual nucleotide positions.

High-throughput sequencing has greatly increased the number of positions with integration that can be detected. However, independent insertions with the same orientation and genome position result in duplicate sequence reads that can not be distinguished from duplicates produced by the polymerase chain reaction (PCR) method used to detect the insertions. As a result, high-throughput sequencing greatly increases the number of positions detected but cannot measure the amount of integration that occurs at individual positions. Importantly, to understand the molecular details of integration mechanisms it is critical to distinguish the positions that are rare sites of integration from the sites that have high levels of integration. To quantify integration at specific nucleotide positions we have developed a serial number system that generates a unique sequence tag during each integration event. Although this method could be adopted to measure the integration of any retrovirus or retrotransposon we describe here the use of the serial number system with Tf1 as a proof of principle.

Integration of Tf1 into specific promoters was initially measured in plasmid-encoded targets and the results showed key nucleotide positions sustained large numbers of independent insertions (2,3,5). A deletion of a C-terminal region of integrase (IN) called the chromodomain (Tf1-CHD) disrupted both the recognition of promoters and the repeated integration at specific nucleotides (3).

The serial number system allowed us to measure overall levels of integration at each promoter and to determine the genome-wide contribution of the chromodomain to the pattern of integration. This unique method relies on random sequence tags within Tf1 to record the numbers of independent integration events at single nucleotide positions throughout the genome of S. pombe. By inducing the expression of Tf1 from a library of plasmids containing serial number tags we identified integration at 133 743 positions in the S. pombe genome. By tabulating the total number of serial numbers sequenced at each position we recorded a total of 1 061 068 independent insertions. We observed integration numbers at individual positions varied over two orders of magnitude and these frequencies provided a uniquely precise profile of transposition. We found that the positions with the highest numbers of insertions possessed a strong signature of flanking sequence indicating that sequence recognition plays a much greater role in targeting integration than previously known.

MATERIALS AND METHODS

Serial number library construction

Plasmids pHL2763 and pHL2770 contain WT and chromodomain deleted versions of Tf1-neo, respectively. They both contain a unique nucleotide tag in the U5 of the upstream LTR of Tf1-neo that distinguishes newly integrated copies from the pre-existing elements (4). The serial number plasmid libraries pHL2944 (WT) and pHL2943 (CHD) were constructed by ligating a PCR-amplified fragment of 250 bp of the upstream LTR into the unique XhoI and SpeI sites of pHL2763 and pHL2770. The 250 bp fragment was amplified from pHL2763 using the oligos HL2827 and HL2828 (See Supplementary Figure S1 for sequences of oligos). HL2827 included the unique SpeI sequence and a stretch of 8 nt randomized sequence and HL2828 had the unique XhoI sequence. The ligated products were electroporated into ElectroMAX DH5a cells (Invitrogen). Approximately 250 000 and 100 000 cfu(s) were pooled together and constituted the WT Tf1s-neo and the CHD Tf1s-CHD-neo libraries, respectively. Plasmid DNA was extracted using the Mega kit (Qiagen).

Transposition assay

S. pombe diploid strain YHL5661 was transformed with the serial number plasmid library of Tf1s-neo and Tf1s-CHD-neo. Approximately 55 000 and 37 000 independent yeast colonies were pooled together from Edinburgh minimal 2 medium (EMM) plates lacking uracil for the Tf1s-neo and Tf1s-CHD-neo libraries, respectively. Thiamine at a concentration of 10 μM was added to the plate media to repress the nmt1 promoter fused to the neo-marked Tf1. The pooled cells were mixed at RT for 1 h and then washed four times with 225 ml of EMM media lacking uracil to remove thiamine. Transposition was induced by growing the Tf1s-neo and Tf1s-CHD-neo yeast serial number libraries at 32ºC in EMM media in the absence of thiamine to induce the nmt1 promoter. The Tf1s-neo and Tf1s-CHD-neo cultures were passaged with repeated dilutions to optical density (O.D.) 0.05 until they reached 30 and 50 generations, respectively. Next, the cultures were diluted to O.D. 0.25 with 5-FOA (5-fluoroorotic acid)-containing EMM media and grown to O.D. 5.0. This selected against cells retaining the Tf1-containing plasmids as they contained the URA3 gene. For the final step the cultures were diluted 10-fold to O.D. 0.5 with 5-FOA-and G418-containing YES media and grown to O.D. 5.0 to isolate cells with copies of Tf1s-neo and Tf1s-CHD-neo that had transposed into the genome.

Quantitative transposition assay

Quantitative transposition assays were performed to monitor the frequencies of transposition as the cultures of Tf1s-neo and Tf1s-CHD-neo were passaged. The method used was a modified version of a previously described procedure (15). In brief, each sequential culture in EMM media lacking thiamine was assayed to determine the increase in transposition frequency relative to generation number. Starting with the undiluted culture a series of five 10-fold dilutions were made. 0.1 ml of cells from the three lowest dilutions was then spread onto EMM FOA plates. The same volume from the three highest dilutions was spread onto EMM FOA and YES G418 plates. The transposition frequency is the percentage of FOAr cells that were also G418r.

DNA preparation and sequencing

The cells from the final YES cultures containing 5- FOA and G418 were harvested, genomic DNA was isolated and samples were prepared for Illumina sequencing as described previously with some modifications (4). In brief, the genomic DNA was purified from 200 O.D. units of cells using zymolyase 100T treatment and spheroblast extraction (4). MseI was used to fragment the DNA because previous data indicated this enzyme did not introduce a bias in detection of insertion sites (4) and because in our lab restriction enzyme cleaved ends are more efficiently ligated to linkers than sonicated DNA fragments. For each library six 2 μg samples of genomic DNA were digested in 100 μl volumes with MseI for 16 h. The digests were purified with the Qiagen PCR purification kit. The digested DNA for each library was eluted in a 50 μl volume and used in 10 duplicate linker ligations with Invitrogen T4 DNA ligase for 1 h at 25°C (See Supplementary Figure S1 for sequences of linker oligos). After heat inactivation at 65°C for 10 min, 10 units of SpeI was added to separate the 5′ LTR from the 3′ LTR which is used in the amplification of the insertion sites. All the SpeI cut DNA was used directly as template in 80 PCR reactions, 20 μl per well, with Titanium Taq from Clontech. The primer that recognizes the linker end is HL2216 and the LTR amplification primers with barcodes are described in Supplementary Figure S1B. The PCR program used was:

94°C 4 min

94°C 15 s

65°C 30 s

72°C 45 s

goto step 2 for a total of six cycles.

94°C 15 s

60°C 30 s

72°C 45 s

goto step 6 for a total of 24 cycles.

68°C 10 min

4°C until sample is retrieved.

All PCR reactions were pooled and divided into 6 samples that were each purified on a separate Qiagen PCR purification column. Each set of 80 PCR reactions were purified on a single 10 cm 2% TBE agarose gel. The gels were run at 70 volts until the dye reached half the length of the gel. The DNA of size 150–500 bp was cut from the gel and purified with Qiagen gel extraction kits. The concentration of the purified DNA was determined with qPCR (KAPA SYBR FAST kit, Kapa Biosystems) and a fluorimeter using picogreen. All six libraries were combined and loaded onto two lanes of an Illumina Genome Analyzer IIx (GAIIx) device and primer HL2747 was used to sequence 100 nt single end reads. The sequencing was performed by the Illumina FastTrack Sequences Services (Hayward, CA, USA).

The description of Data Analysis and the use of Rate Distortion Theory to remove erroneous serial numbers generated by Illumina misreads are included in the supplement.

FUNCTIONAL ENRICHMENT ANALYSIS

The intergenic regions were ranked by decreasing number of integration events. The genes with promoters within the 250 tandem intergenic regions with the highest number of insertions were selected. The functional enrichment analysis was performed using the genes induced by environmental stress (16) as query. Significant enrichment was tested by comparison of the number of genes induced by environmental stress in the 250 highest verses all intergenic regions (17). The P-values were calculated using a hypergeometric test and corrected by false rate discovery. P < 0.01 served as the cut-off criterion.

Accession number of sequence data

The Illumina data from the Tf1s-neo and Tf1s-CHD-neo experiments was submitted to the Short Read Archive (SRA) at National Center for Biotechnology Information (NCBI) under the accession number SRA044888.1.

RESULTS

To study genome-wide integration, a copy of Tf1 marked with neo (Tf1-neo) is expressed in S. pombe and cells with insertions are selected on media containing G418 (18,19). High-throughput sequencing of Tf1-neo integration sites relies on amplifying insertions with ligation-mediated PCR (4). The previous study that demonstrated Tf1-neo integrates into pol II transcribed promoters measured the number of positions with integration. However, the number of insertions at each position could not be determined; the duplicate sequence reads were discarded because those generated by independent integration could not be distinguished from ones resulting from PCR or cell propagation (Figure 1A).

Figure 1.

The serial number system quantifies the number of independent insertions at single nucleotide positions. (A) Two hypothetical promoters each have insertions at three different positions. However, one promoter (right) could have many more insertions at each site than the other. Because the insertion libraries are created by PCR duplicate sequence reads are typically discarded. As a result, two promoters could have very different amounts of integration but be reported to have the same amount of integration. (B) (i) The serial number system was created in Tf1 by inserting 8 bp of random sequence in the U5 region of the 5′ LTR (red stripe). Tf1 mRNA was expressed from the nmt1 promoter within a library of plasmids. (ii) The library of expression plasmids with the serial number sequences were introduced into S. pombe (iii) and the expression of Tf1 was induced (iv). (v) Genomic DNA from cells with Tf1 was ligated to linkers (red) and the insertions were amplified by PCR. The products were sequenced with Illumina technology. The sequence primer read across the serial numbers and into the sequence of the insertion sites (blue).

To measure independent integration events at single nucleotide positions we created a library of Tf1-neo expression plasmids that contained an 8-nucleotide random sequence in the U5 region of the 5′ LTR (Figure 1B). This random sequence we call the serial number was positioned in the U5 section of the LTR at a location previously shown to be unimportant for self-priming of reverse transcription and for transposition (Supplementary Figure S2) (18–21). Once introduced into S. pombe the library of 65 536 possible Tf1s (Tf1s signifies the presence of the serial number system) elements was expressed and cells with integration were selected in liquid cultures with G418. As a result of reverse transcription, the serial numbers were transferred to the U5 of the 3′ LTR. Sequence reads starting in the integrated Tf1s-neo elements recorded the serial number of each insertion and continued into the insertion site (Figure 1B, v). The number of independent insertions at each position of integration was the total number of different serial number sequences detected per site. The raw numbers of independent insertions were corrected to remove the false serial numbers produced by Illumina sequence errors. In a representative stretch of 8 bp of LTR sequence ∼3% of the reads had at least one nucleotide substitution. In cases where PCR created high numbers of identical sequences containing the same serial number and insertion site, 3% of the reads were expected to produce erroneous serial numbers generated by the Illumina errors. These were identified as clusters of serial numbers closely related in sequence to a founder present in high copy. The method that identified the clusters of erroneous serial numbers was based on a rate distortion algorithm for estimating the true number of serial numbers (Supplement).

Serial number analysis of six independent cultures identified a total of 1 061 068 integration events

To create dense profiles of integration sites and evaluate the reliability of the serial number system, insertions of wild-type Tf1s-neo were generated in three independent cultures. The strains used were all diploid so that integration events would not likely result in strong loss of function and bias the pattern. In addition, the role of the chromodomain in positioning integration was tested by creating insertions of Tf1s-CHD-neo in three additional cultures. While the overall frequency of integration of Tf1s-CHD is 14-fold reduced (3), we were able to obtain complex libraries of insertions. Integration sites from each of the six cultures were positioned by sequencing libraries resulting from ligation-mediated PCR (4). Basic Local Alignment Search Tool analysis identified a total of 133 743 unique positions in the S. pombe genome that had insertions (Materials and Methods). These are defined as strand specific positions (SSPs) that consider the two orientations of integration independently. Unless otherwise indicated, all references to sites or positions refer to SSPs. Using the number of serial numbers at each unique position and the correction for sequencing errors, we recorded a total of 1 061 068 independent integration events throughout the three chromosomes of S. pombe (Table 1). A wide variation in integration levels was observed throughout all three chromosomes (Figure 2). The levels of integration at individual positions varied from 0 to 622 (Supplementary Table S1). Comparing all 6 integration data sets, the percent of positions that had more than one insertion varied from 40.08% to 27.4%. The percentage of positions that had greater than 100 insertions varied from 0.57% to 0.01% (Supplementary Tables S1 and S2). To visualize the number of integrations per site we ordered the positions by insertion number and displayed this as a graph (Figure 3). To determine whether the deletion of the chromodomain altered the distribution of insertions per site, Figure 3 compares pooled sets of data with equivalent numbers of insertion positions. By comparing similar numbers of positions we found the distribution for WT2 plus WT3 (128 386 positions) was very similar to CHD1 plus CHD2 plus CHD3 (123 881).

Table 1.

Summary of integration for each of 6 independent experiments

	WT1	WT2	WT3	CHD1	CHD2	CHD3	TOTAL
Integration positions	92 264	68 069	60 317	49 014	41 263	33 604
Independent integration events	455 140	190 478	155 105	110 715	86 391	63 239	1 061 068
% events upstream of ORF	80.2	80.3	80.0	78.4	78.1	78.0
% events in ORF	3.5	3.9	4.0	5.1	5.4	5.6
% events downstream of ORF	16.3	15.8	16.0	16.5	16.5	16.4

Figure 2.

The levels of integration throughout the three chromosomes of S. pombe varied widely. The integration from WT Tf1s-neo (WT1, WT2 and WT3) was mapped on the three chromosomes of S. pombe. The Y-axis is the total number of insertions per nucleotide position. Only insertions that mapped to unique sites are shown. Integration in centromeres, telomeres and repeated regions of the chromosomes is not shown.

Figure 3.

The number of insertions per position varied widely. Each of the 128 386 insertion positions of WT2 and WT3 (blue) were ordered on the X-axis from the positions with the highest number of insertions to those with the lowest insertions. The Y-axis is the number of independent insertions at each position. The 123 881 integration positions of CHD1, CHD2 and CHD3 (red) were also ordered and graphed.

Earlier studies of Tf1 integration revealed that 96.5% of the insertions occur in intergenic sequences and the majority of the sites cluster upstream of the open reading frames (ORFs) (4). The integration of the serial number elements exhibited the same preferences for intergenic sequences (an average of 96.2% for Tf1s-neo and 94.6% for Tf1s-CHD-neo) and for regions upstream of ORFs (Table 1 and Figure 4). Previous analyses revealed that the integration levels in individual intergenic regions varied with ∼20% of the regions being favored (above random) for integration (4). The levels of Tf1s-neo and Tf1s-CHD-neo integration within each intergenic region were tabulated and again ∼20% of the intergenic sequences had higher integration levels than the control set of random insertions called the matched random control (MRC) (Figure 5 and Supplementary Figure S3). The intergenic regions with high integration of Tf1s-neo correlated well with the intergenic regions that had high levels of Tf1 integration in our previous study (4). A linear regression of integration events of Tf1s-neo compared to the number of insertion sites reported previously (4) for each intergenic sequence produced an R2 = 0.76. These intergenic sequences with high levels of integration were previously found to contain significantly higher numbers of stress response promoters than predicted by the proportion of stress response genes in the genome (4). This observation was made using the intergenic sequences between two tandem oriented genes so that we knew which promoters were targeted by Tf1. Using the data from Tf1s-neo we identified the 250 intergenic sequences with the highest number of integration events and found the downstream genes were significantly enriched (adjusted P-value = 7.00E-05, see Materials and Methods) with genes that are induced by stress as defined by Chen et al. (16).

Figure 4.

Integration clustered upstream of ORFs. The integration events produced by WT (Tf1s-neo) and CHD (Tf1s-CHD-neo) versions of Tf1were mapped relative to ORFs. The ORFs of S. pombe were divided into 15 equal segments (red) and the percentage of the integration within each segment is displayed. Inserts outside of ORFs are displayed upstream or downstream depending on whether they were closer to the 5′ or 3′ end of the ORF. The integration outside of the ORFs was displayed in 100 bp intervals.

Figure 5.

Approximately 1000 intergenic regions had higher numbers of insertions than observed for random integration. All intergenic regions of S. pombe were ordered on the X-axis based on numbers of integration events and the amount of integration in each intergenic region is plotted on the Y-axis. (A) The integration produced by WT1 (Tf1s-neo) is displayed along with an equal number of insertions positioned randomly (MRC WT1). (B) The integration of CHD1 (Tf1s-CHD-neo) is compared to an equal number of insertions positioned randomly (MRC CHD1). The centromeres and telomeres were excluded from the intergenic sequences because they are significantly longer than other intergenic regions.

To determine how reproducible the integration levels were in each intergenic region of the 6 integration experiments we used linear regression (Table 2 and Supplementary Table S3). The results showed strong reproducibility of integration in intergenic regions for Tf1s-neo with R2 values all greater than 0.93. The R2 values comparing integration of the three Tf1s-CHD-neo patterns also revealed very high reproducibility (R2 > 0.95). We also found that there were strong correlations in the amounts of intergenic integration regardless of the deletion of the chromodomain (Tf1s-neo versus Tf1s-CHD-neo, R2 > 0.75).

Table 2.

The integration levels within the intergenic regions from 6 independent experiments were compared by linear regression. R2 values demonstrated that the integration amounts were highly reproducible.

	WT2	WT3	CHD1	CHD2	CHD3
WT1	0.942	0.930	0.801	0.758	0.780
WT2		0.986	0.885	0.858	0.857
WT3			0.891	0.870	0.865
CHD1				0.957	0.963
CHD2					0.955

The serial number system identified unique features of highly targeted sites

Previous studies of integration were capable only of recording which genomic positions were selected as insertion sites. The unique contribution of the serial number system is that it made it possible to tabulate numbers of independent insertions at single nucleotide positions even when they occurred in the same orientation. On a total chromosome scale the full value of this information is not apparent. But when individual intergenic regions are examined unique patterns of integration were visualized (Figure 6A and B). The specific patterns of integration were highly similar in each of the 6 independent experiments demonstrating that the position and levels of integration were reproducible. Some promoter regions possessed peak-like patterns with center coordinates that have high numbers of independent inserts flanked by sequences with integration that declined with distance from the peak. The width of the peaks was promoter specific. Other regions exhibited stretches of low level integration or no insertions at all (Figure 6B, center). What was surprising was that some single nucleotide positions possessed hundreds of independent insertions while the flanking positions had little or no integration (Figure 6A, center).

Figure 6.

The serial number system revealed unique patterns of integration with measures of independent integration at individual nucleotide positions. (A) and (B) depict two regions of the S. pombe genome in chromosome 2. The number of insertions at each nucleotide position is shown for all 6 experiments.

The large number of insertion events sequenced with the serial number system allowed us to detect patterns of nucleotide preferences at the insertion sites. We aligned 128 386 integration positions of Tf1s-neo combined from the WT2 and WT3 experiments and examined the nucleotide frequencies of flanking sequence using the logo algorithm (22). The insertion sites possessed a pattern of preferred nucleotides with modest strength that extended 50 bp in either direction (Figure 7A). The preferred nucleotides flanking Tf1s-neo inserts formed a palindrome, a pattern observed with other integrating elements that is thought to be due to stabilizing contacts between the IN and target DNA (4,8,23–26). To determine whether the chromodomain contributed to the overall pattern of nucleotide preference we analyzed the 123 881 insertion sites produced by all three of the Tf1s-CHD-neo experiments. The resulting logo was very similar to that produced by Tf1s-neo indicating that the chromodomain contributed little to the average pattern of nucleotide preference (Figure 7B).

Figure 7.

The integration of WT (Tf1s-neo) and CHD (Tf1s-CHD-neo) have sequence preferences. (A) The 128 386 insertion sites produced by WT2 and WT3 were aligned and the logo pattern of nucleotide preferences was determined. The positions of the 5 nucleotides at the target sites that are duplicated during integration are indicated by target site duplication (TSD). The height of the nucleotide in the logo represents the bit score associated with the bias and the proportional height of a nucleotide relative to the others at a position constitutes the percent of the insertions that have that specific nucleotide. The A at the center of the palindrome is 35.4% of the height indicating that 35.4% of the insertions had an A at that position. (B) The logo pattern was determined for the 123 881 insertion positions generated by CHD1, CHD2 and CHD3.

The bulk of integration sites had modest to low levels of sequence specificity (Figure 7, bit scores <0.1) suggesting that the overall pattern of integration positions was not the result of nucleotide preferences. However, we wondered whether the high numbers of independent insertions found at the ‘hottest’ positions might result from the recognition of specific nucleotides. To test this possibility we aligned the 50 insertion sites from each collection of Tf1s-neo that had the highest number of independent insertions. These 150 positions had numbers of independent insertions ranging between 71 and 622. The logo pattern from these top positions possessed a marked increase in nucleotide specificity with bit scores that in some positions were five times higher than the scores of the complete set of insertions (Figure 8A versus Figure 7A). The nucleotide preferences of Tf1 lacking the chromodomain (Tf1s-CHD-neo), at the 150 positions with the highest number of insertions also had greatly increased nucleotide specificity compared to all Tf1s-CHD-neo insertions (Figure 8B versus Figure 7B). However, the logo pattern of the top Tf1s-CHD-neo sites had nucleotide specificities higher even than the top sites of wild-type Tf1 (Figure 8B versus Figure 8A). For example, at position 18 of the top Tf1s-CHD-neo sites, the bit score was nearly 1 because 62% of the sites had a C at this location (Figure 8B and Table 3). The preference for C at position 28 was also higher in the top Tf1s-CHD-neo sites than in the top Tf1s-neo sites (Table 3, 58% versus 43%). In addition to its heightened level of specificity, Tf1 lacking the chromodomain integrated at its top 150 sites with a unique asymmetry (Figure 8B). The strongest positions of nucleotide preference only occurred downstream of the insertion sites. This surprising absence of palindromic symmetry indicates that the chromodomain influences the orientation of integration events and the recognition of nucleotides at the insertion sites with the highest number or repeated events.

Figure 8.

The integration positions with the highest number of independent insertions had strong nucleotide preferences. (A) The 50 integration positions of WT Tf1s-neo with the greatest number of independent insertions were aligned from the WT1, WT2 and WT3 experiments. A logo was generated from these top 150 sequences. The positions of the 5 nucleotides at the target sites that are duplicated during integration are indicated by TSD. The height of the C at position 18 of the palindrome indicates that 52% of the insertions had a C at this position. (B) A logo was produced by combining the 50 integration positions of CHD1, CHD2 and CHD3 (Tf1s-CHD-neo) with the highest number of independent insertions. The resulting logo was generated from 150 sequences. 62% of the insertions had a C at position 18. (C) Fifty insertion sites from each of the WT1, WT2 and WT3 experiments with a single insertion event were chosen at random and used to create a logo. (D) Fifty insertion sites from each of the CHD1, CHD2 and CHD3 experiments with a single insertion event were chosen at random and used to create a logo.

Table 3.

Ratio of the nucleotide frequencies at positions identified in logos of the top 150 positions depicted in Figure 8.

Nucleotide (position)	Frequency in WT	Frequency in CHD	Ratio considered	Ratio in WT	Ratio in CHD
C(18)	52	62
G(-18)	44.66	36.66	C(18) / G(-18)	1.164	1.691
C(28)	43.33	58
G(-28)	36.66	34	C(28) / G(-28)	1.182	1.706
T(31)	57.33	62
A(-31)	52.66	52.66	T(31) / A(-31)	1.088	1.177

The strong preferences for specific nucleotides by either wild-type Tf1s or Tf1s lacking the chromodomain were observed at the sites with highest numbers of independent insertions. A selection of 150 positions with single insertions chosen at random had little or no nucleotide preferences (Figure 8C and D). We therefore asked how much impact sequence preferences had overall on the selection of insertion sites. For this question we ranked the 345 583 insertion events produced by Tf1s-neo in the WT2 and WT3 experiments based on the number of independent insertion events that occurred at their position. The events that occurred at the sites with the greatest number of insertions had the highest ranking while the events that occurred at sites with single insertions had low rankings. These ranked events were put into 20 bins; each included 17 279 events or 5% of the total insertions. Then for the insertions in each bin we determined the average sequence similarity to the sequence preferences of the top 150 sites shown in Figure 8A. The resulting histogram showed an important trend, that the sites with increasing numbers of events also had increasing average sequence similarity to the top 150 positions (Figure 9A). The trend of increasing sequence similarity started at the 75% bin indicating that as much as 75% of the integration events exhibited sequence preference. Importantly, this 75% of the integration events occurred at just a third of the positions detected. Without the serial number system it wound have been impossible to identify which sites accounted for the bulk of integration. Another measure of how much sequence of the target site contributed to integration efficiency was to compare the first two bins on the left in Figure 9A. While each includes 5% of the events, the positions in the first bin had 4-fold more integration. This 4-fold increase in integration efficiency corresponds with a 3.1% increase in sequence similarity to the logo of the top sites.

Figure 9.

Sequence preference was observed in a substantial proportion of integration events. (A) The 345 583 insertion events produced by Tf1s-neo in the WT2 and WT3 experiments were ranked based on the number of independent insertion events that occurred at their position. The events that occurred at the sites with the greatest number of insertions had the highest ranking while the events that occurred at sites with single insertions had low rankings. These ranked events were put into 20 bins; each included 17 279 events or 5% of the total insertions. Then for the insertion events in each bin we determined the average sequence similarity to the sequence preferences of the top 150 sites shown in Figure 8A. (B) 345 583 insertion events (MRC) positioned in silico at random sites were ranked and organized into 20 bins as in (A).

To test what sequence preference would occur if integration were random, an equivalent number of randomly positioned events in an MRC was used to compare to the logo of the top sites. Each bin had equal percentages of average similarity (Figure 9B). These results indicate that sequence preference played a significant role in the number of insertion events that occurred at individual integration sites. Using the same ranking and binning system we tested the 260 345 insertion events of Tf1s-CHD-neo to determine whether the chromodomain contributed to the increase in sequence preferences associated with insertion positions that had higher numbers of independent integration events. The 20 bins of Tf1s-CHD-neo insertions had increasing sequence similarity to the top 150 sites with a pattern very similar to that of by Tf1s-neo, indicating that the chromodomain did not contribute significantly to the overall contribution of sequence preferences (Supplementary Figure S4 versus Figure 9A).

The results in Figure 9 reveal that sequence recognition is an important determinant in integration frequency. We tested whether sequence recognition was sufficient to predict which genome positions were selected for integration. Find Individual Motif Occurrences (FIMO) was used to scan the genome for sequences that matched the logo of the strongest integration sites (Figure 8A) with a P-value < 3.32E-05 (27). Of 98 857 matches we analyzed the top 5000 sequences. While this set had 20 positions with greater than 200 insertions, these hot spots were a minority. Only a small fraction of the 5000 matches had more than 10 insertion events (Supplementary Figure S5). In addition, the positions with over 200 inserts had match scores to the logo that distributed evenly throughout the set of 5000 positions sorted by match score. These 5000 matches to the logo of strong integration sites showed no preference for sequences upstream of ORFs (Supplementary Figure S6). We conclude that matching the recognition sequence in the logo is just one determinant in the integration process. The mechanism that targets promoters is another determinant.

DISCUSSION

The serial number system described here provides a solution to the significant problem that deep sequencing of integration sites cannot determine the frequencies of insertion at individual nucleotide positions. Previous profiles of integration supplied a yes or no answer to whether integration occurred at any given position. By tagging each insertion with an independent serial number we generated saturating profiles of integration sites that measured the frequency of integration at each position. With this technology we identified a combined total of 1.1 million insertion events of Tf1s-neo and Tf1s-CHD-neo at sites that had between 1 and 622 independent insertions in the genome of S. pombe. Insertion sites had frequencies of integration that varied over two orders of magnitude.

The highly dense and reproducible profile of integration reported here allowed us to test the role of the IN chromodomain in positioning Tf1 insertions. Although previous studies of integration in target plasmids indicated the chromodomain was required for positioning insertions in promoters (3), we found here that deletion of the chromodomain resulted in no substantial change in the genome-wide distribution of integration. It appears that the previous studies of integration sites in plasmids, were subject to alternative constraints that influenced integration. These alternative interactions or constrains did not occur when the insertions were positioned in chromosomal targets.

Sequence preferences contributed significantly to integration frequency at insertion sites

The wide range of integration frequencies at insertion sites raises the question what accounts for the variation in integration efficiency. In particular, why did some positions have very high numbers of independent events while nucleotides on either side had little or no integration. Logo analysis of 150 highest positions of integration clearly showed a sequence signature that was substantially stronger than that of the total set of 128 386 Tf1 insertion positions. This observation indicated that the sequence of the target sites played a significant role in causing high frequency of insertions. This finding led us to ask how much influence insertion site sequence had on the genome-wide profile of integration. By ranking all integration events by the frequency of their repeated insertion we found that sequence preference contributed to the efficiency of integration for 75% of the events. Importantly, we found that these events, the 75% of the insertion events, occurred at just 33% of the insertion positions. Without the serial number system we would have been unable to account for the positions of this 75% of the activity. And as a result of the precision provided by the serial number system, we discovered that the bulk of integration activity occurred at sites with a sequence signature.

Although the chromodomain did not contribute significantly to the overall pattern of integration it did play a role in sequence recognition at insertion sites with high numbers of events. Interestingly, integration at the top 150 positions of Tf1 lacking the chromodomain exhibited a unique asymmetric pattern of preferred nucleotides (Figure 8B). The asymmetry in this logo indicates that at the top insertion sites IN lacking the chromodomain was defective in mediating integration in one of the two orientations. Despite this reduction in one orientation of integration the remaining insertion events retained the same sequence preference downstream of the insertion sites. As there was a reduction in one orientation of integration the sequence preferences downstream were more pronounced. This is another important observation that was only made possible with the frequency data provided by the serial number system.

Understanding the contribution of specific nucleotides to integration efficiency will lead to important structural and mechanistic information about the complex of IN, donor DNA and the target DNA. For example, x-ray crystal structures of the prototype foamy retrovirus (PFV) IN identified contacts between amino acids and specific nucleotides in target DNA(25,28). Arg329 of IN is hydrogen bonded to guanine 3, guanine 21 and thymine 22 in the target DNA. Serial number data of PFV integration would be instrumental in measuring the frequency of integration at insertion sites and in testing how mutations in Arg329 change the frequency of integration at specific positions. Since the design of the serial number system can be readily incorporated into any retroviral or retrotransposon expression vector, experiments testing the role of individual amino acid residues in target site preference can now be conducted.

Retrovirus vectors have been adapted for introducing therapeutic genes into cells of patients in clinical trials. However, depending on the positions of integration they have caused leukemia by activating oncogenes (29–33). To gage the safety of new retrovirus vectors for gene therapy it is necessary to develop saturating profiles of integration sites. The extraordinary capacity of new deep sequencing technology now makes it possible to achieve these highly dense profiles. However, this tremendous increase in capacity also diminishes the distinction between highly targeted positions and rare insertion sites. Retrovirus vectors can be readily modified in U5 sequence in the 3′ LTR and this could be used to generate serial numbers that tag unique insertions (34). The application of the serial number design for retroviruses would reveal the frequency of integration at individual sites and as a result would provide true integration levels adjacent to cancer genes.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.