Introduction

LINE-1 (long interspersed element-1) or L1-mediated retrotransposition has significantly impacted upon human genome evolution (for recent reviews, see Deininger et al. 2003; Kazazian 2004; Han and Boeke 2005; Hedges and Batzer 2005) but has also given rise to human genetic disease (Chen et al. 2005, 2006). Intriguingly, L1 elements have driven the generation of some 10% of the human genome mass by mobilising Alu sequences (Lander et al. 2001; Batzer and Deininger 2002). Although there is no doubt that Alu insertion is initiated by L1 endonuclease and reverse transcriptase (RT)-dependent target site-primed reverse transcription (TPRT; Dewannieux et al. 2003; Hagan et al. 2003), the mechanism by which the newly synthesised 3′ end of a given Alu cDNA attaches to the target genomic DNA is less well understood. In this regard, the integration of full-length L1 elements has recently been proposed to occur via a template-jumping model whereas the integration of 5′-truncated L1 elements is thought to result predominantly from a microhomology-mediated end-joining (MMEJ) model (Zingler et al. 2005; Babushok et al. 2006). The integration of full-length Alu elements can also be explained, at least in principle, by the template-jumping model. However, unlike 5′-truncated L1 elements, 5′-truncated Alu elements appear by and large not to be integrated via the MMEJ model (Zingler et al. 2005).

Recently, we have identified two pathological simple Alu insertions (termed #1 and #2, respectively) in the CFTR gene (manuscript submitted). Interestingly, #1 represents the shortest (starting position at 236) of the 28 currently known pathological simple Alu insertions (i.e. no loss of target gene sequence) that are informative with respect to the starting position of the Alu insert (Fig. 1). More interestingly, of the six 5′-truncated simple Alu insertions, #1 represents the only example of the occurrence of a 2 bp microhomology between the 3′ end of the top strand’s 5′ overhang in the target sequence and the 3′ end of the nascent Alu cDNA (Supplementary Table S1). In addition, the second shortest pathological simple Alu insertion (starting position at 47) exhibited a one bp microhomology (Supplementary Table S1). In sharp contrast, none of the remaining four 5′-truncated simple Alu insertions (starting positions at 16, 39, 39, and 41, respectively) exhibited microhomology (Fig. 1; Supplementary Table S1). We were intrigued by this phenomenon and wondered whether microhomology could have played a role in the integration of shorter Alu sequences into the human genome. To test this idea, we performed a meta-analysis of the Alu insertion polymorphisms deposited in the Database of Retrotransposon Insertion Polymorphisms in Humans (dbRIP; http://falcon.roswellpark.org:9090/searchRIP.html; Wang et al. 2006). This analysis, when considered together with a re-evaluation of the mechanism underlying how the three previously annotated large deletion-associated short pathological Alu inserts (Chen et al. 2005) were generated, has enabled us to present a unifying model for Alu insertion in the human genome.

Fig. 1
figure 1

Alignment of the consensus sequences of five Alu sub-families. Dashes indicate gaps introduced so as to maximise alignment. Nucleotides identical between all sequences are indicated by asterisks. Pathological Alu insertions (including 28 simple ones and three associated with large genomic deletions) that are informative with respect to starting position in their respective Alu sub-family consensus sequences, are positioned accordingly in the aligned sequences. Note that the sub-family of the shortest Alu insert, which comprises CGTCTC plus A40and is associated with the Δ1444 bp in the SERPINC1 gene (Beauchamp et al. 2000; Chen et al. 2005), could not be assigned. Shaded arrows indicate either entries (underlined) that can be alternatively annotated as full-length Alu inserts or those that are not informative with respect to the ‘microhomology’ question (refer to Supplementary Table S1 for details). Note that (i) microhomology existing between the top strand’s 5′ overhang and the sequence that lies 5′ to the truncation position in the Alu consensus sequence was identified in the same way as for the Alu insertion polymorphisms (see second section of the text) and (ii) only Alu inserts with starting position 6 or greater were regarded as 5′-truncated in accordance with Zingler et al. (2005)

Full size image

Identification of microhomology existing between the top strand’s 5′ overhang and the sequence that lies 5′ to the truncation position in the Alu consensus sequence

The 1624 non-redundant Alu insertion polymorphisms deposited in dbRIP (as of December 6, 2006) were subjected to manual evaluation with respect to whether microhomology exists between the top strand’s 5′ overhang and the sequence lying 5′ to the truncation position in the Alu consensus sequence, in line with previously established principles (e.g. Zingler et al. 2005; Babushok et al. 2006). Where a microhomology (the longest match where applicable) was identified, the top strand cleavage site was assigned as 3′ to the matched nucleotide(s) in the target sequence whilst the starting position of the 5′ truncated Alu insert was designated as the nucleotide 3′ to the matched base(s) in the Alu consensus sequence. Two examples―one involving a full-length Alu insert and the other involving a 5′ truncated Alu insert―are illustrated in Fig. 2. In many cases, this treatment yielded a modification of the originally defined end positions of the target site duplications (TSDs) and the start positions of the Alu inserts. Although detailed sequence information for each entry is given in Supplementary Tables S2–S6, several issues warrant further clarification here. First, that many of the entries can be alternatively annotated with respect to the microhomology question is due to the lack of a strict consensus sequence for top strand cleavage, although a weak preference for the sequence 5′-TYTN/R-3′ has recently been proposed (Gilbert et al. 2005). Second, a substantial proportion of the Alu insertion polymorphisms from dbRIP were excluded from further analysis; these included (i) entries overlapping with the pathological Alu insertional mutations listed in Supplementary Table S1, (ii) entries for which the repeat sequences and/or TSDs are unknown, (iii) full-length Alu insertions with additional nucleotides at their 5′ ends and (iv) various other entries that were uninformative with respect to the question of microhomology (Supplementary Table S6). Lastly, as is evident from inspection of Supplementary Tables S3 and S4, a significant proportion of the Alu insertions with starting positions at 2, 3 and 4 can be alternatively interpreted as full-length inserts; this issue will be addressed further at the end of the following section.

Fig. 2
figure 2

Two examples of how the starting positions of Alu inserts were modified, taking into account the question of ‘microhomology’. Both examples (A and B) were taken from dbRIP, the Database of Retrotransposon Insertion Polymorphisms in Humans (http://falcon.roswellpark.org:9090/searchRIP.html). (a) Target site duplications (TSDs) are highlighted in bold and underlined; Alu sequence plus the poly(A) tail are italicised; the starting position of the Alu insert is indicated by an Arabic numeral. (b)Top sequence: ±10 bp flanking the top strand cleavage site (indicated by an arrow) deduced from a; lower sequence: whilst italicised sequence on the right side corresponds to the ten 5′-most nucleotides of the Alu insert illustrated in a, sequence not italicised on the left side was taken from the Alu insert’s respective consensus sequence at corresponding positions where applicable. Microhomology is shaded wherever applicable. Note that in A, re-assigning the first G of the originally annotated full-length Alu insert into the upstream TSD resulted in the generation of a one base-microhomology between the top strand’s 5′ overhang and the now 5′-truncated (1 bp) Alu insert. In B, re-assigning the 5′-most TGGCT of a 5′-truncated Alu insert into the upstream TSD resulted in the generation of more extensive microhomology

Full size image

The sub-family of each selected Alu insert was checked/annotated using RepeatMasker (http://www.repeatmasker.org/cgi-bin/WEBRepeatMasker; as of December 6, 2006). Although in some cases, annotations were different from those previously reported in Chen et al. (2005, 2006) and dbRIP, this did not affect the conclusions of the study in any way. Consensus sequences of AluYa5, AluYa8, AluYb8, AluYb9, AluY, AluSq, AluYg6, AluYd8 and AluSp sub-families were taken from Repbase (http://www.girinst.org/repbase/update/browse.php; Jurka et al. 2005). Sequence alignments were performed with ClustalW (http://www.ebi.ac.uk/clustalw/#).

A trimodal length distribution of simple Alu inserts and the role of microhomology in generating shorter Alu inserts

Studies of recently inserted genomic L1 elements in the human genome (Myers et al. 2002; Pavlicek et al. 2002; Szak et al. 2002; Boissinot et al. 2004), pathological L1 direct insertions (Chen et al. 2005), and de novo L1 insertions in cultured human cells (Gilbert et al. 2002; 2005) as well as in a transgenic mouse model (Babushok et al. 2006) have consistently shown that simple L1 inserts display a bimodal length distribution with a large peak of short (<2 kb) and a smaller peak of longer (∼6 kb) integrations. Although the exact mechanism underlying this bimodal distribution remains controversial (e.g. Farley et al. 2004; Gilbert et al. 2005), the generation of the abundant short L1 inserts would appear to be facilitated by the presence of microhomologies frequently found between the top strand’s 5′ overhang in the target genomic sequence and the 3′ end of the nascent L1 RT-transcribed cDNA strand (Zingler et al. 2005; Babushok et al. 2006).

As shown in Fig. 3, a trimodal length distribution of the 1402 informative Alu insertion polymorphisms is apparent: a major peak of full-length or almost full-length inserts (starting positions at 1–5; termed Group I for ease of discussion) with a frequency of ∼85% (1198/1402), a smaller peak of 115 inserts initiating from positions 8–47 (frequency, ∼8%; termed Group II), and the remaining inserts beginning from after position 51 to the end (termed Group III). The major peak was not unanticipated since (i) a full-length Alu insert is <290 bp and (ii) the L1 RT is believed to be of high processivity, by analogy with the property of Bombyx mori R2Bm RT (Bibillo and Eickbush 2002; Gilbert et al. 2005). Here it is worth noting that the observed frequency of Group I inserts is consistent with the finding that genome-wide ∼90% of Alu insertions are full-length [with full-length being defined as those elements initiating within the first five nucleotides of the consensus sequence; Zingler et al. (2005)]. Thus, by contrast with the situation pertaining with L1 elements, for most Alu sequences the process of cDNA synthesis would have a high probability of completion before being counteracted by the host repair machinery.

Fig. 3
figure 3

Global survey of Alu insertion polymorphisms selected from dbRIP (Wang et al. 2006). The Figure is presented essentially in the same manner as Fig. 1. However, for full-length or near full-length entries (i.e. starting positions at 1–5), only the total number is provided, respectively. See Supplementary Tables S2–S5 for details of all entries

Full size image

The smaller peak constituting Group II is however intriguing. On the one hand, all 115 truncations occurred within a relatively short region of 40 bases that is well-conserved between different Alu sub-families (Fig. 3). On the other hand, microhomology was only evident in 34.8% of the 115 entries (Fig. 3; Table 1). With respect to the mechanism underlying the generation of these Group II Alu insertions, we currently envisage two possible models, one operating at the level of transcription (i.e. from DNA to RNA), the other at the level of reverse transcription (i.e. from the RNA to the nascent cDNA strand). Both models are predicated upon the assumption that the behaviour of L1 RT is similar to that of Bombyx mori R2 RT, which readily jumps from the 5′ terminal end of the R2 RNA but very inefficiently from internal positions (Bibillo and Eickbush 2004). The first of these models proposes that the truncations arise through the use of alternative transcriptional start sites, in the context of the internal RNA polymerase III promoter [see Fig. 1 in Murphy and Baralle (1983) and Fig. 1 in Shankar et al. (2004) for the RNA polymerase III promoter structure and location within the Alu element itself]. This proposition is based upon two observations. First, the Group II inserts are located entirely within the A- and B-box consensus sequences of the polymerase III promoter (Murphy and Baralle 1983; Shankar et al. 2004); this strongly implies the involvement of alternative transcription sites in the generation of these 5′ truncated Alu inserts. Second, the use of alternative transcription start sites is not infrequent in genes that are transcribed by RNA polymerase II, although this has not been empirically demonstrated for RNA polymerase III transcripts. Formation of Group II inserts would proceed in the same way as for full-length inserts: upon reaching the 5′ end of the truncated Alu RNA, the L1 RT would jump from the RNA template to the 3′ end of the top strand’s 5′ overhang [see Fig. 3A in Zingler et al. (2005) and Fig. 5D, 2 in Babushok et al. (2006)]. The alternative model proposes that the truncations result from the degradation of Alu RNA by cellular RNase H (Ostertag and Kazazian 2001a; Zingler et al. 2005), the clustering of truncation sites being due to the occurrence of a specific secondary structure that prevents further RNA degradation by binding to trans-stabilising factors. Under this model, the formation of these truncated insertions would be identical to that envisaged under the first model, given that L1 RT can process to the 5′ end of a 5′ degraded Alu RNA.

Table 1 Correlation between the Presence of Microhomology (1–7 bp) and the length of the 5′ truncation of Alu insertion polymorphismsa
Full size table

As mentioned above, only 34.8% of the Group II Alu inserts were found to exhibit microhomology. By contrast, microhomology was found in some 50% (44/89) of the Group III Alu inserts. As a matter of fact, in the context of the 5′ truncated Alu insertion polymorphisms (i.e. starting positions, 8–271), there exists a positive correlation between the presence of microhomology and the length of the 5′ truncation (Table 1), thereby suggesting an important role of the MMEJ mechanism in generating shorter Alu inserts. Under this model, the generation of most of the shorter Alu inserts could have been promoted by the inadvertent annealing of the microhomology present between the 3′ end of the nascent Alu cDNA strand and the 3′ end of the top strand’s 5′ overhang. This would then be followed by the premature termination of nascent cDNA strand synthesis with concomitant initiation of second Alu cDNA strand synthesis by either a second L1 RT or a host DNA repair enzyme. In addition, we should point out that our finding differs from the recent genome-wide analysis that has concluded that 5′ truncated Alu elements exhibit no (or only a weak) tendency to exhibit microhomology (Zingler et al. 2005). The discrepancy may be due to one or more of the following reasons. Firstly, Zingler et al. (2005) did not address the microhomology issue in relation to the different lengths of 5′ truncation. Secondly, these authors used only computer-generated data with respect to the analysis of the 5′ truncated Alu insertions. In other words, they did not analyse the relevant data manually. As shown in Supplementary Tables S3–S6, our manual evaluation led to the re-annotation of a significant fraction of the dbRIP entries.

Finally, as in the case of the pathological Alu insertional mutations (Supplementary Table S1), most of the near full-length Alu insertion polymorphisms (i.e. starting positions at 2–5) can be alternatively interpreted as bona fide full-length insertions (Table 2). Assuming that L1 RT is of high processivity and given that a full-length Alu element is < 290 bp, we believe that most, if not all, of the above entries that can be alternatively interpreted are genuinely full-length insertions. Consequently, we propose that Alu insertions should be regarded as full-length whenever possible. Finally, it should be noted that all Alu insertions with starting positions beyond five, analysed in this study, cannot be alternatively interpreted to be full-length.

Table 2 Near Full-Length Alu insertion polymorphisms (i.e. starting positions 2–5 in accordance with their respective consensus sequences) that can be alternatively interpreted as full-length insertionsa
Full size table

Large deletion-associated short Alu inserts appear to be integrated through qualitatively different mechanisms

It is no longer in dispute that L1-mediated retrotransposition generates large genomic deletions, as evidenced by complementary observations made in the context of in vitro studies (Gilbert et al. 2002, 2005; Symer et al. 2002), identification of disease-causing mutations (Chen et al. 2005; Mine et al. 2007) and genome-wide analysis (Callinan et al 2005; Han et al. 2005). As we already pointed out in our previous meta-analytical study (Chen et al. 2005), the regions spanning the upstream deletion breakpoints in the target ABCD1, APC and SERPINC1 genes were annotated as Alu sequences by RepeatMasker and hence share significant similarity with the Alu inserts of interest (Fig. 4). Alu retrotransposition-mediated deletions have also been identified in the human genome in an evolutionary context (Callinan et al. 2005), but it is unclear whether these lesions share the same sequence features as noted in the three above-mentioned pathological mutations.

Fig. 4
figure 4

Pairwise alignment of the top strand sequences (from 5′ to 3′) overlapping the presumed upstream breakpoints of the ABCD1 (Kutsche et al. 2002), APC (Su et al. 2000) and SERPINC1 (Beauchamp et al. 2000) genes and their respective Alu inserts. Dashes indicate gaps introduced in order to maximise alignment. Identical nucleotides are identified by vertical bars. The putative upstream breakpoints are denoted by vertical arrows. Alu sequences contained within the inserts are shaded. Unshaded Alu sequences are derived from the consensus Alu Yb9 sequence at corresponding positions. For the sake of simplicity, the sub-family of the precursor sequence that generated the shortest Alu insert associated with the 1444 bp deletion in the SERPINC1 gene (Beauchamp et al. 2000) was also arbitrarily designated Yb9 (this does not affect the conclusions drawn owing to the high sequence identity manifested by the members of the Alu sub-families; see Fig. 1)

Full size image

The generation of the three disease-causing large genomic deletions associated with Alu insertions can in principle be accounted for by the model illustrated in Fig. 6B from Gilbert et al. (2002): each event was putatively initiated by L1 endonuclease cleavage on the bottom strand but, unlike the typical process of TPRT leading to the generation of a simple insertional event, the L1 RT-transcribed Alu cDNA strand appears to have invaded a double strand break located far upstream of the bottom strand nick/break (Chen et al. 2005). This model can be further refined in the light of new developments in the field. Thus, in a genome-wide analysis of both human and chimpanzee data sets, Han et al. (2005) observed a significant positive correlation between the size of the L1 direct insertion and the size of the associated deletions. Han et al. (2005) surmised that the longer the newly synthesised L1 cDNA strand was, the higher would be the probability of forming sufficient complementarity between the end of the L1 cDNA and the region flanking the 5′ end of the L1 insertion in the ancestral sequence. This is indeed a plausible explanation for the generation of large genomic deletions created upon L1 insertion. This model cannot however be readily extrapolated to cases of large genomic deletions caused by insertions of Alu elements, simply because the Alu inserts in the three disease-causing events are significantly 5′ truncated (see Fig. 1). This notwithstanding, the model of Han et al. (2005) stimulated us to propose a refined model for the generation of large genomic deletions caused by Alu insertions: the significant sequence similarity existing between the regions spanning the top strand’s upstream deletion breakpoints and the newly synthesised Alu cDNA strands in all three cases (Fig. 4) suggests that the longer the stretch of complementarity, the higher the likelihood of a newly synthesised Alu cDNA strand annealing to a double strand break-containing far-upstream region. In this refined model, the position of the Alu truncation would be specified by the position of the double strand break in the top strand whereas the synthesis of the Alu cDNA strand might not necessarily need to be completed in order to obtain sufficient complementarity for strand annealing/invasion.

One further point warrants further discussion. It is possible that the top strand’s upstream double strand break may be attributable to the activity of L1 endonuclease (Gasior et al. 2006). Were this to be the case, this could predict an active role for L1-mediated retrotransposition in creating large genomic deletions. It should however be emphasised that the L1 endonuclease used to generate the top strand’s upstream double strand break may not necessarily be the same as that used to create the bottom strand’s first nick (Mine et al. 2007), by analogy to the proposition that two different L1 RT molecules may be used for twin-priming, leading to L1 inversion (Ostertag and Kazazian 2001b). It is equally possible that the top strand’s upstream double strand break was created independently of L1 endonuclease. Were this to be the case, “a fascinating scenario would present itself: the organism could have ‘hijacked’ the L1 machinery to repair an existing double strand break through a mechanism akin to single strand annealing.” (Chen et al. 2005). In this particular context, L1 integration may represent a ‘host/parasite battleground’ as it has been termed by Gilbert et al. (2005), in which L1 integration finds itself in a ‘race’ to complete cDNA synthesis before being ‘hijacked’ to patch an upstream double strand break.

A unified model for Alu insertion into the human genome

Based upon the above observations, we propose a unified model for Alu insertion in the human genome. Since Alu elements are comparatively short, L1 RT is usually able to complete nascent Alu cDNA strand synthesis before jumping to the 3′ end of the top strand’s 5′ overhang, resulting in the generation of either full-length (i.e. Group I events) or 5′ truncated (i.e. Group II events) Alu inserts. Alternatively, the synthesis of the nascent Alu cDNA strand may be terminated prematurely if its 3′ end anneals to the 3′ terminal of the top strand’s 5′ overhang by means of microhomology-mediated mispairing, an event which would often lead to the formation of significantly truncated (Group III) Alu inserts. Furthermore, the nascent Alu cDNA strand may be ‘hijacked’ to patch existing double strand breaks located in the top-strand’s upstream regions (which should usually comprise Alu-rich sequences), leading to the generation of large genomic deletions. Clearly, the unified model proposed here is likely to be subjected to further modification/revision by new studies as they emerge.