While L1s carry adequate information for transcriptional termination and polyadenylation 42, the polyadenylation site is non-canonical, so that L1 transcripts often do not end exactly at the end of the element 43. This is manifest in the number of L1 elements carrying 3' transduced sequence from their progenitor locus: about 10% of all retrotransposition events 44454647. We predicted that a small proportion of all transcripts from expressed L1s would carry non-L1 sequence resulting from read-through of the transcript into the flanking genomic region. These sequences could then be used to identify the genomic location of the element. In some cases, the terminal few bases of the L1 3' UTR might be sufficient in themselves to locate the element uniquely in the human reference genome.

We primed first strand synthesis of cDNA using oligo(dT), followed by second strand synthesis with an oligonucleotide located at the end of the 3' UTR of the L1 (Figure 1a). Due to the LINE1-associated poly(A) tract, 3' end sequence amplicons tend to be of low complexity (Figure 1b). We have been unsuccessful in obtaining adequate sequence quality and length from these amplicons using next generation sequencing methodologies; below, we describe our results using manually curated sequence reads that were generated by the Sanger method.

We obtained 3' end transcript sequence from 2,152 cDNA clones from lymphoblastoid cell lines from a single European-American individual, GM10861, from the Centre d'Etude du Polymorphisme Humain (CEPH) population (Table 1). Nearly half of these expressed sequence tags had been primed from the polyadenylation site immediately downstream of the L1, and therefore aligned to multiple identical L1 3' UTR locations in the genome. However, 1,148 expression tags were unique in the reference genome; these represented 204 distinct sites, 54 of which corresponded to full-length L1 elements. Thirty-eight L1 expression tag clusters could not be mapped adjacent to an L1 in the reference genome.

Expression tags were typically short (Figure 1b, c; Additional data file 1), with a mean end position of 34 nucleotides from the end of the L1 (median = 30.5 nucleotides). Seventy-five percent (152) of tagged sites terminated transcription less than 40 nucleotides from the end of the L1, and 93% (190) terminated less than 60 nucleotides from the L1 (Figure 1c). The distribution of polyadenylation positions for expression tags corresponding only to full-length elements was similar (mean = 35 nucleotides; median = 29 nucleotides; 81% located less than 40 nucleotides from the L1). Many of these short transductions represent non-canonical L1 3' ends or atypical polyadenylation cleavage sites, rather than the use of novel polyadenylation signals downstream of the L1 itself.

Thirty-seven L1 elements were tagged five or more times (Additional data file 2); these include six full-length elements, two containing intact, putatively functional ORFs (Table 2, 4p15.32 and 7q31.1). The Chao2 ecological index, which estimates the number of types based on the rate of sampling singletons and doubletons 48, predicts a total of 363 expressed sites in this individual. As over half of the 204 sites we identified are represented by only one or two expression tags, it is likely that increased sequencing will yield few significantly expressed new sites.

We have also obtained 3' sequence tags from five additional individuals: GM17032 and GM17033 are African-Americans, GM17045 is of Middle Eastern origin, and GM11994 and GM11995 are European-American individuals who are the parents of GM10861 described above. In total, 3,828 3' expression tags were sequenced from all six individuals (Table 1; Additional data files 1, 2 and 3), encompassing 1,592 sequences corresponding to 271 unique sites. Of these sites, 228 corresponded to an L1 element in the reference genome. The remaining 43 clusters, while containing L1 3' UTR sequence at one end, do not map to any of the reference L1s, and, therefore, may represent private or polymorphic insertions. Due to the extremely short, homopolymeric nature of these tags, we cannot map the putative location of these 43 clusters in the reference genome or design PCR oligonucleotides to verify their presence in genomic DNA.

Forty-seven L1 sites were sampled five or more times, while 26 were sampled ten or more times. These relatively highly expressed sites include ten full-length elements (Additional data file 1). Expression tags corresponding to different elements were cloned from different lines, and no elements were cloned from all six lines (Additional data file 3). We focused our interest on full-length elements, which might be transcribed from the native promoter in the 5' UTR and could potentially produce active ORF1 and/or ORF2 protein. Sixty-nine full-length elements in the human reference genome were identified in our 3' expression tag analysis (Table 2), which is significantly greater than the proportion of full-length elements in the reference genome from the Pa7 family or younger (Fisher's exact test P = 1.0 × 10^-15). Of the full-length elements tagged, more are from the human specific subfamily (30) than their proportions in the genome (Fisher's exact test P = 7.8 × 10^-21); however, this is not surprising because the primers that were used contained a nucleotide at the 3' end that biased amplification towards the L1Hs human specific subfamily.

Of the 69 expressed full-length elements, 30 are present in genes (Table 2), which is somewhat more than expected from the proportions in the genome (Fisher's exact test P = 0.0013). Of the elements present within genes, slightly more than would be expected by their distribution in the genome are in the same orientation as the gene (Fisher's exact test P = 0.0026; L1Hs only, Fisher's exact test P = 0.017; Table 2). This is in keeping with the possibility that some of these L1s may be expressed as a side effect of transcription of the host gene.

Seven expressed full-length elements contain intact ORF1 and ORF2 and might be competent for retrotransposition under certain conditions. Four additional elements contain potentially active ORF2 in the absence of ORF1 (Table 2). The proportions of expression tagged elements from the L1Hs subfamily containing intact ORF1, ORF2, both or neither are not significantly different from those present in the genome as a whole (χ² = 2.36, degrees of freedom = 3, P = 0.5).

To supplement our 3' end analysis, we also conducted L1 5' RACE on RNA from lymphoblastoid cell lines corresponding to a single European-American individual, GM11994, the father of GM10861 described above. Expression tags obtained using 5' RACE identify L1 transcription start sites, either from the native L1 promoter or from an upstream promoter (Figure 1a). As the 5' end of a full-length L1 is not homopolymeric, we were able to obtain high quality reads using high-throughput 454 pyrosequencing. We recovered 36,088 sequences, of which 14,488 corresponded to 427 locations in the reference genome (Table 1; Additional data file 4). The Chao2 index predicts 494 sites in total; therefore, these loci include the majority of the expressed sites within this particular individual, and likely include all the highly expressed sites.

Only six of the full-length 5' RACE expression-tagged L1 elements were also found by 3' expression tagging (Table 2). This lack of overlap is instructive, though not entirely surprising, as 3' tags would include both full-length and 5' truncated elements, the latter being the most common in the genome. In contrast, 5' RACE is biased towards full-length elements, as relatively few L1s are 3' truncated. Moreover, the oligonucleotide used to prime 3' amplification contained a nucleotide change that biased it towards amplification of the L1Hs subfamily, whereas the 5' RACE primer was unbiased and would identify all L1 5' UTR-derived sequence.

We identified 347 sites corresponding to full-length expressed elements by 5' RACE analysis, 89 of which were sampled 10 or more times (Additional data file 4). Of the remaining expressed sites, 76 corresponded to deleted or degenerated 3' truncated elements from the L1P1, L1P2 and L1P3 subfamilies (Additional data file 4, grey font). Four tagged sites did not correspond to an L1 element in the reference genome (Additional data file 4, blue font). We were able to verify by PCR that one of these four sites, which mapped to chr12: 33908761, identifies a non-reference L1 present in the GM10861/GM11994/GM11995 familial trio. The precise insertion breakpoint of this L1 was determined by sequencing of the PCR verification product (Additional data files 4 and 5).

L1 5' start sites mapped by 5' RACE can be subdivided into three groups: those that are located in the upstream flanking sequence, those that are internal to the element, and those that splice from far upstream. Four expression tags indicated usage of a promoter far upstream (>15 kb) that produced a transcript that spliced immediately adjacent to a full-length L1 (Additional data file 4, green font). Of the start sites mapping internally or within 1,000 bp upstream of a full-length L1, 50% (170) were located within ± 50 nucleotides of the consensus start of the L1 (Figure 1d), with the median start site at position -21 relative to the L1. These relatively close, though variant, start sites are typical of usage of the native L1 promoter 12. However, 124 5' expression tags to full-length elements begin greater than 100 nucleotides upstream of the L1 (Figure 1d), suggesting that a proportion of L1 transcripts from certain loci might also originate from upstream flanking promoters.

Of the full-length elements identified, 24 are from the L1Hs human-specific subfamily, which is not significantly greater than what would be expected based upon the proportions found in the genome (Fisher's exact test P = 0.26; Table 3). However, elements from the next youngest L1Pa2 (Fisher's exact test P = 9.9 × 10^-13) and L1Pa3 (Fisher's exact test P = 1.7 × 10^-10) subfamilies are overrepresented, while the older L1Pa5 (Fisher's exact test P = 2.6 × 10^-5) and L1Pa6 (Fisher's exact test P = 5.6 × 10^-15) elements are underrepresented. This is consistent with the hypothesis that more evolutionarily recent elements are more likely to have retained sequences that would be permissible for transcription. Of the 24 full-length L1Hs elements, eight contain intact ORF1 and ORF2, two contain an intact ORF2 only, and nine contain an intact ORF1 only (Table 3). Relative to the proportions in the genome, the distribution of elements containing intact ORFs is not significant (χ² = 0.7, degrees of freedom = 3, P = 0.9).

We have characterized the nature of transcription from four full-length elements identified by 3' expression tags. The most frequent 3' expression tag (Table 2; Additional data file 1) we identified corresponds to an element from the L1Hs subfamily located on band 4p15.32 at coordinates chr4:15452168-15458393 (Figure 2a). We isolated 263 sequence tags from this element from lymphoblastoid cells of GM10861 (Additional data file 2), corresponding to 24% of all mapped tags from that individual. An additional 86 tags to this element were isolated from four more individuals (parents GM11994 and GM11995, and the unrelated individuals GM17032 and GM17033; Additional data file 3), indicating that the element at 4p15.32 is highly expressed in lymphoblastoid cell lines. A previous study found that the 4p15.32 element is nearly fixed in four human populations (heterozygosity ≤ 0.05) 49.

The majority of expression tags to this locus end 42 nucleotides downstream of the element (Figure 2b, chr4 short tag), just upstream of a polyadenylation stretch in the genomic DNA. However, two expression tags extend to 182 nucleotides downstream (Figure 2b, chr4 long tag), suggesting that at least some of the transcripts might continue further into the flanking DNA. Directed 3' RACE using a primer located just downstream of the L1 amplified a single product terminating at this same position in both individuals GM11994 and GM11995 [dbEST:64858885]. These 182 nucleotides are also found downstream of another 5' truncated L1 located at chr6: 66316760-66318742 (Figure 2b, chr6 transduction), which was previously described as a member of a transduction family 47. The chromosome 6 insertion, which is polymorphic in different ethnic populations 495051, is therefore likely the descendent of the full-length element on chromosome 4. These lines of evidence all point to at least some fraction of the L1 transcript at 4p15.32 terminating 182 nucleotides downstream of the element (Figure 2b, chr4 3' long tag).

The L1 at 4p15.32 contains an intact ORF1 gene; however, ORF2 is truncated 96 amino acids early, downstream of the known functional domains. The presence of the transduced polymorphic descendent element on chromosome 6 suggests that the 4p15.32 element has been active in the recent past. Nine expression tags from two different individuals were also isolated from a similar sequence (U35) to 4p15.32 that does not occur in the human reference genome (Figure 2b, U35 tag; Additional data files 1 and 3). U35 may represent an allele or an additional non-reference L1 insertion related to the element at 4p15.32.

The L1 at 4p15.32 is located in intron 7 of the CD38 gene, in the same orientation as the gene (Figure 2a). CD38 (cluster of differentiation 38) is a cell-surface glycoprotein involved in lymphocyte cell adhesion and signaling 52. We examined steady-state RNA levels of the L1 at 4p15.32 in CEPH familial lymphoblastoid cell lines using a TaqMan quantitative RT-PCR assay specific for the L1 transcript (Figure 2c). Note that expression tags were cloned at high frequency from both GM10861 and GM11994 (Additional data files 1, 2 and 3). There are significant differences in expression among the different individuals, with GM11992 showing little to no expression (Figure 2d), and individual GM10861 showing relatively high expression. We compared the expression of the L1 element to that of the surrounding CD38 gene. We found that, while the abundance of the CD38 transcript is several orders of magnitude higher, the pattern of expression of the L1 element follows that of expression of the gene (Figure 2e).

We also examined three full-length elements that were represented less frequently by 3' expression tags. The L1 element on chromosome band 13q14.2, located at coordinates chr13:47937193-47943803, was represented by six expression tags total cloned from each member of the GM11994/GM11995/GM10861 familial trio (Table 2; Additional data files 2 and 3). The 3' end tags terminate 20 nucleotides downstream of the end of the element, within a poly(A) rich region (Figure 3a). The associated L1, while classified in the human-specific L1Hs subfamily, does not contain intact ORFs and is not expected to be competent for retrotransposition.

The 13q14.2 L1 is located inside intron 22 of the Retinoblastoma 1 (RB1) tumor suppressor gene 53, in the same orientation (Figure 3a). 5' RACE analysis specific to this locus amplified a single product from individual GM11994 [dbEST:64858883], designating a transcriptional start site 457 bp upstream of the start of the element that encompasses part of intron 21 and all of exon 22 of RB1 (Figure 3a). This position is over 161 kb downstream of the start of RB1. Real-time RT-PCR analysis in familial cell lines indicates differential expression of the L1 amongst related individuals (Figure 3b), with individuals GM10861 and GM11994, from which expression tags were cloned, showing relatively high expression, and individual GM11992 showing an almost complete absence of expression. Expression of the RB1 exons flanking the L1 correlated with expression of the L1 transcript (Figure 3c), although the RB1 gene was expressed overall at a higher level (approximately 20 to 50 times more abundant).

L1s are typically cytosine methylated, a modification that is associated with suppression of expression (for example, 29). We hypothesized that methylation might be lost in expressed elements; therefore, we examined cytosine methylation, both at the start of transcription of the L1 and within the L1 5' UTR, in two cell lines (GM10860 and GM11992; Figure 3b) that showed stark differences in expression levels. In both cases, the sequences were densely methylated (see Materials and methods), indicating a lack of an obvious role for cytosine methylation in quantitative expression differences at this site.

The L1Hs element at 6p22.2, genome coordinates chr6:24919786-24926013, was represented by a single 3' expression tag from individual GM17032 (Table 2; Additional data file 3). However, this element had been previously found to be one of the most retrotranspositionally active in a cell culture assay, as well as polymorphic in natural populations 1954. As expected from an element that is known to be active, both ORFs are intact. The 3' expression tag terminates 119 nucleotides downstream of the element (Figure 4a). 3' RACE analysis in two unrelated individuals (GM11994 and GM11995) using a primer immediately next to the L1 identified a single product terminating 442 nucleotides downstream of the L1 [dbEST:64858886]. Therefore, the transcript from this L1 can transduce at least 442 bp of genomic sequence.

The L1 at 6p22.2 is located within intron 21 of the FAM65B gene (Figure 4a), which encodes a factor involved in trophoblast differentiation 55. Expression of the element was confirmed in lymphoblastoid cell lines from individuals GM10861 and GM11994, both of which are heterozygous for this insertion. The element in GM10861 was expressed greater than two-fold more than in GM11994 (Figure 4b). Transcripts corresponding to the FAM65B gene, while more abundant than those from the L1 (approximately 30 times), appear to correlate with expression of the L1 in these two individuals (Figure 4b).

The L1 at 1p22.3, genomic coordinates chr1:86917252-86923882, was identified by a single 3' expression tag cloned from individual GM10861 (Table 2; Additional data file 2). This element, a member of the L1Hs subfamily, encodes two intact ORFs, and therefore might be retrotranspositionally competent. In contrast to the three elements characterized above, the L1 at 1p22.3 is intergenic, located 19 kb and 23.6 kb from its nearest neighboring genes (Figure 5a).

The 3' expression tag corresponding to this locus ends 27 nucleotides downstream of the end of the L1 (Figure 5a). Directed 5' RACE from cell line GM11994 identified a single transcriptional start site 125 bp upstream of the element [dbEST:64858884] (Figure 5a). The L1 transcript shows some variant expression within lymphoblastoid lines from two CEPH families (GM10860/GM11992/GM11993 and GM10861/GM11994; Figure 5b). Bisulfite sequencing analysis on this trio indicates differential methylation levels, with GM11992 showing the least CpG methylation (Figure 5c). Because this is also the individual expressing the least transcript, cytosine methylation does not appear to correlate with transcriptional repression at this site.