ABSTRACT
Ordered shotgun sequencing (OSS) has been successfully carried out with an Xq25 YAC substrate. yWXD703 DNA was subcloned into
[lambda]
phage and sequences of insert ends of the
[lambda]
subclones were used to generate a map to select a minimum tiling path of clones
to be completely sequenced. The sequence of 135 038 nt contains the entire ANT2 cDNA as well as four other candidates suggested
by computer-assisted analyses. One of the putative genes is homologous to a gene
implicated in Graves' disease and it,
ANT2 and two others are confirmed by EST matches. The results suggest that OSS
can be applied to YACs in accord with earlier simulations and further indicate
that the sequence of the YAC accurately reflects the sequence of uncloned human
DNA.
Mapping of the human genome has achieved long-range contiguity in cloned DNA primarily by the use of yeast artificial chromosomes (YACs;
1
). The X chromosome, for example, has reached >95% coverage in YACs formatted
with PCR-based markers (sequence-tagged sites or STSs;
2
) and for most regions, YACs provide the only current coverage. If individual
YAC clones could provide sequencing substrates, their large size and ready availability would reduce the number of sequencing subprojects to be done, with a
corresponding decrease in logistical complexity.
Large insert bacterial clones are attractive alternative substrates for long-range sequencing (see Discussion), but sequencing of YACs or other large
insert clones has a number of prerequisites. For example, computer-assisted assembly of sequences of subclones from a very large clone could
be hampered by the number of repetitive elements that often occur in clusters
and can be highly homologous. Also, in the case of YACs, human DNA represents
only ~1% of the DNA content in a yeast cell and consequently sequencing
substrates subcloned from YACs must be recovered against the high yeast
background.
We sought to address these issues by deploying an `ordered shotgun sequencing'
strategy (OSS;
3
) on a typical YAC. OSS attempts to facilitate sequencing a large clone by
ordering a set of subclones based on paired end sequences (see below). Simulations (
3
-
5
) had suggested that such an approach could have several advantages. Here we
report the first use of an OSS approach to obtain the full sequence of a YAC,
which contains a 135 038 bp segment of Xq25. The results show that OSS is
practicable for large scale sequencing, even by a small working group.
The strategy and implementation of OSS and the assembly of the map are outlined
in Results, leading to Figures
1
and
2
. Here we focus on the preparation of DNA substrates and the methods to analyze
the resultant sequence.
YAC yWXD703 has been mapped in a region of Xq25 where STS content is concordant
in a number of YACs (
17
; Nagaraja
et al
., manuscript in preparation). Furthermore, STSs made from the ends of the YAC
map in the contig and the YAC had tested positive for an STS made from the ANT2
gene, for which a cDNA sequence has been published (
18
). Thus, several preliminary criteria of quality and verifiability were satisfied.
The success of an OSS approach depends on the ability to obtain sequence
information efficiently from both ends of subclones studied. Using optimized
conditions for PCR amplification we were able to amplify all of the human DNA
inserts from archived [lambda] subclones. In the current work, >90% of the amplified inserts yielded
usable end sequence tracts. Two thirds of those are longer than 450 bases and
even the much shorter tracts are useful for OSS map construction (see below).
To construct a partial map (Fig.
1
), the initial set of insert end sequences are searched: (i) against Genbank as
a first screen to identify possible genes which may provide further ordering information; (ii) against consensus Alu and L1 sequence elements. Inferred clone overlaps involving Alu or L1 sequences are flagged to watch out for similar but non-identical repetitive elements, but the scoring of any dubious overlaps
becomes definitive during the full sequencing of overlapping [lambda] subclones, since essentially every clone that ends in a repetitive element also includes unique sequences.
The project began with 76 series A subclones of 6-9 kb. In general, the recursive generation of OSS maps revealed problem
areas early on, permitting focused attention while the rest of the region was
systematically finished. In particular, when it became clear that the A clones
were not random enough to cover the entire YAC extent (see Discussion), the end sequences were supplemented from 27 of the larger B subclones. The amplification efficiency for B
clone inserts is only slightly lower (>90% in recent experiments) and the
distribution of B end sequences tended to complement the A series, connecting
many small contigs and thus carrying the OSS map toward closure (Fig.
2
). For this project, finishing required the filling of one gap (Fig.
2
) by sequencing a 2 kb fragment amplified from the YAC with primers at the edges
of the two remaining contigs. In addition, fragment A152 contained a poorly
resolved sequence associated with apparent polymerase slippage at a poly(T) or
poly(A) tract; it was clarified using a (T)16V primer (as in a comparable case;
14
).
Table
1
summarizes features of the shotgun sequencing of 21 [lambda] inserts from 18 A and three B clones. For two reasons, an overall 5-fold redundancy of sequence coverage (96 sequencing reactions; Table
1
) was attempted for each [lambda] subclone across the YAC. First, this number of sequences was sufficient
to lower the average number of gaps/lambda to 0 or 1, a level that made it
straightforward to finish the sequence by primer walking or other methods.
Second, ~1 in 3000 bases were idiosyncratically different in individual reads, a
variation that we attribute to misincorporations that occurred during PCR
amplification of the [lambda] inserts and were preserved in individual M13 subclones. About five
readings of the same sequence were sufficient to identify the odd readings and
infer a consensus sequence.
Table 1
Sequencing statistics for YAC 703
The precision (reproducibility) of the sequencing results was earlier assessed
at >99.9% for the 219 kb DNA in Xq28 (
14
). With multiple subclone coverage and higher sample purity, a comparable or
greater level was reached here. Consistent with high precision, sequence comparisons of several kilobases of regions of overlap between [lambda] subclones or between [lambda] and PCR-amplified DNA segments have detected no differences.
Comparable concordance was also found in comparisons of genomic sequence with
encoded cDNA segments.
Initial analysis with the CENSOR program was done with default settings, but in
general with a threshhold score of 35. This eliminated the identification as
`repetitive elements' of exonic sequences that are paucirepetitive in the genome. With these settings a total of 210 repetitive sequences were detected. They included 129 Alus,
nearly all of them full sequences (i.e. `dimers' rather than `monomers'),
accounting for 24.9% of the total sequence. They are roughly equally
distributed throughout the region and are indicated in Figure
3
(middle rows) in blue. The complete tabulation also included nine L1 matches of
at least 400 bp and 17 shorter sequences that were judged to be fragments of
L1s, comprising 5.6% of the total sequence (orange in Fig.
3
). The other 55 repetitive sequences included 21 classes of MERs, totalling 9.3%
of the sequence content (green in Fig.
3
). Thus, overall the region was composed of 40.9% repetitive sequences.
In the first step to detect candidates for genes, CpG islands (
19
) were indicated by telltale concentrations of GC and CpG dinucleotides. In Figure
3
, three strong CpG islands, coinciding with at least two restriction sites for enzymes with CpG in their recognition sites,
are indicated by the numbers 2-4. Another weaker possible island, numbered 1, coincided with only a
single rare-cutter restriction site and is not further considered here. CpG island
association and other evidence for each of five gene candidates, ANT2 and 703-1 to 703-4 (Fig.
2
), is summarized as follows.
703-1.
Two exons are predicted by GRAIL at nt 5754-5838 and 9072-9326, encoded on the top strand. The latter exon overlaps rat EST
L105369, with 80% identity over 255 bp. Like the predicted exons, the EST is
transcribed from the top strand, left to right in Figure
3
. Thus, the 5'-end of the gene presumably lies more centromeric in Xq25. Neither
the EST nor the genomic sequence resemble any known gene/protein, so that the
gene remains anonymous.
703-2.
This gene, initiated at CpG island 2, includes a first exon that is predicted
by GRAIL at nt 28 802-29 076 and is also highly homologous to a reported mitochondrial gene
(M31659;
20
; see also Discussion). Furthermore, although it is not predicted by GRAIL, a
second exon in an open reading frame 7000 bp toward the right arm (nt 35 858-36 081) shows a putative correct splice junction that would continue
precisely with further sequence homologous to the same mitochondrial gene (Fig.
3
). Four additional exons predicted by GRAIL and extending through nt 65 907 may represent additional coding sequence of the 703-2 gene.
703-3.
Three exons transcribed toward the left arm are predicted by GRAIL and could be
associated with CpG island 3. In this case, there is as yet no verification of
the exons by detection of cDNA, transcribed RNA or a gene of known function.
ANT2.
This gene, associated with CpG island 4, contains four exons that are predicted by GRAIL (nt 108 408-108 831, 109 212-100 352, 109 578-110 066 and 111 101-111 281). All are confirmed by comparison with the published cDNA
sequence.
We note that the YAC sequence deviates from the published cDNA sequence for ANT2
at several residues, but sequence tracts identical to those in the YAC occur in
many ESTs in dbEST that are homologous to the 3'-end of ANT2, including 18 in the Merck-Washington University database.
703-4.
A sequence not suggested by GRAIL as a putative exon nevertheless matched both
ends of a clone arising from the 3'-end of a cDNA. The EST, again from the Washington University/Merck
consortium project (clone ID158302) is essentially identical to nt 125 258-125 914 in the yWXD703 sequence. The clone is polyadenylated, so that it
most likely represents a cDNA rather than a contaminating genomic DNA or
unspliced message sequence, but the extent of the corresponding gene is
unknown.
Four exons are predicted by GRAIL as transcribed toward the left arm distal to
703-4, but none of them is homologous to any known expressed sequence or gene.
In the absence of any additional evidence at this time, they are not suggested
as an additional gene.
The YAC thus contains two regions (Fig.
3
), each extending about half of its length, with the first half transcribed
toward the right arm and the rest transcribed toward the left arm.
This study was designed: (i) to test the OSS formulation, with some comparison
with the current standard of random shotgun sequencing; (ii) to assess the
feasibility of the YAC as a sequencing substrate; (iii) to analyze the sequence
and begin to assess the fidelity of the YAC compared with uncloned DNA.
OSS is designed with a small workgroup in mind, with the goal of sequencing a
large target in an easily manageable way. However, implementation of OSS
imposes requirements that include the subcloning of a large clone into 7-10 kb fragments and the efficient sequencing of the ends of the cloned
fragments. The requirements have been successfully met in this test.
As a method to generate fragments from a large target, partial restriction
digestion with
Sau
3A1 has the advantages that it generates cohesive ends which can be ligated more
efficiently than blunt ends and the ends can be partially filled in to prevent
self-ligation and improve cloning efficiency. The disadvantages are that there
is preferential cleavage at some sites and some regions may lack
Sau
3A1 sites.
One way to compensate for local deviations in
Sau
3A1 site distribution uses a mixture of insert sizes. As an example in this
study, the B subclone (average ~10 kb) library complemented coverage by the A library (average 7 kb). A
strategy using a mixture of clones of more than one size range can help achieve
overall coverage and also can counter the problem of non-random distribution of subclones. In fact, simulations indicate that even
a small fraction of larger inserts can provide such benefits (
4
). The fraction here (1:6 or 14%) is in the range recommended from simulations,
but because large clones are harder to amplify, the practical case may deviate
from models and much more work will be required to optimize the approach.
PCR amplification easily provides increasingly reliable clone inserts of pure
human DNA as substrates for the generation of end sequences. As an alternative to clone DNA preparation, amplification of subclone inserts of up to 10 kb is currently robust and straightforward with
long-range PCR kits. Essentially all the inserts are single bands on agarose
gels and after treatment with ExoI and sAP, to purify the products (see
Materials and Methods), direct sequencing of the ends of amplified inserts can be efficiently performed with large numbers of samples.
The OSS approach has another requirement, recovery of sequence from both ends of the majority, if not all, of the subclones. If there are appreciable numbers of subclones for which sequence is available
for only one end, the efficiency of OSS map construction declines sharply,
since connections to other clones cannot be established. In this project the
fraction of clones for which both ends yielded sequence was >65% (which was
improved to >80% in later experiments), high enough to sustain map
construction. Furthermore, we have routinely recovered sequence from an
apparently poor run even when it was as short as 30 bases. Sequences that short can nevertheless help to make a map (
1
) and are an important aid in mapping gaps and confirming overlaps.
Long read lengths are critical, however, at the level of random shotgun
sequencing of the [lambda] intermediate clones and thus determine their optimum size. Current read
lengths of the order of 550 bases can easily sustain the analysis of [lambda] inserts of 10-20 kb. Ten kilobases is now the upper limit for high efficiency long-range PCR, but as long-range PCR continues to improve, a smaller number of
larger [lambda] substrates should suffice, simplifying the project.
The rate of a project depends on the rate of sequencing of subclones. A single
technician can now sequence up to two 10 kb [lambda] clones/week. Each 150 kb YAC or other large insert clone thus requires ~2 person-months of effort to reach the gap closure phase. The rate of
an individual project can be increased by analyzing more [lambda] subclones simultaneously, but this increases the chance that a region
will unintentionally be sequenced more than once. For example, when several [lambda] subclones from yWXD703 were simultaneously chosen for minimal overlaps
and sequenced completely, the group containing [lambda] A055 (Fig.
2
) merged with that containing A080, with a substantial overlap in sequenced
area. Therefore, to avoid duplicated effort, each project can be assigned to
only one person at a time and the rate of very large projects can rather be
increased by expanding the number of large insert clones being analyzed.
In critical respects ordered and random shotgun sequencing are equivalent. For example, comparable precision requires comparable sequencing redundancy and both methods require finishing efforts to fill gaps.
But each approach has advantages.
In current large scale sequencing projects, random shotgun sequencing is much
further along the learning curve, so that costs have been progressively reduced by incremental improvements and a stable and increasing level of efficiency has been demonstrated. Also, random
shotgun methods employ only a single subcloning step from the large insert
clone to sequencing substrates, whereas in the approach to OSS used here, two
successive cloning steps are used. The bridge to smaller clones at each step
is, however, made more efficient by replacing clone DNA preparation with
automatable PCR amplification, and the OSS approach affords several possible
compensating advantages.
First, OSS is relatively forgiving of substantial levels of contamination. With
BACs or PACs as substrates, contamination levels from bacterial DNA can be as
low as <5% or as high as >25% in different preparations (work in progress), but even higher levels require little effort to identify and discard contaminating subclones at the end sequencing stage of OSS. For example, single-pass pulsed-field gel purification had enriched yWXD703 DNA ~50-fold over the starting DNA. With these preparations,
there is still close to 50% contamination with yeast, but most of those clones
could be detected by probing with yeast DNA. Furthermore, extraneous clones can
easily be identified by sequencing one end of each 7-12 kb subclone and finding yeast sequences by comparison with the yeast
sequence in GenBank. In the instant case, <10% of the sequencing reactions in the project (180 end sequences out of 2400
total reactions for this project) would be required to discard all vector and
yeast subclones without prior screening. In contrast, in random shotgun sequencing, every contaminating clone receives the same attention as each authentic clone,
so that half the reactions would be required to eliminate 50% contamination.
Second, a similar consideration applies to sequencing of the vector component of
subclones. BACs or PACs contain 8-16 kb of vector sequence in random subclones, whereas those segments are
again largely discarded at an early stage in OSS.
Third, and related to the other potential advantages, OSS offers added
flexibility in the choice of subclones or regions for discretionary sequencing.
For example, regions of a YAC or other large substrate that overlap a
neighboring YAC are again quickly found by end sequencing and need not be
completely redone. If one estimates that the two neighboring clones would each
overlap a sequencing substrate by the order of 20% of its length, then a
corresponding fraction of the total sequencing effort might be saved using OSS
rather than a random shotgun approach. One can even imagine approaches in which
the early identification and mapping of large L1 elements (or other repetitive
sequenes in tandem) could activate the option of less thorough sequencing there
than in regions of unique DNA (
21
). For the human genome project, precise end-to-end sequencing may preclude such selectivity, but sequencing of
other organisms might make it more attractive, especially when funding is
limited.
As a subsidiary feature, gaps in the overall map are localized at an early stage
in such a project, permitting focused attention. Similarly, within each
subfragment, the average number of gaps (0.4) is relatively simple to fill and,
because repetitive elements are segregated into subclones, they are easier to
discriminate during sequence assembly. Perhaps because of the added information
provided by the compartmentalization, the overall redundancy of sequencing was ~6-fold (see Table
1
), compared with 8- to 10-fold in the sequencing of a comparable amount of DNA by random
shotgun methods. This comparison, however, is contingent, since various
approaches are becoming increasingly efficient and approaches have not been
compared with the same clones.
It is also possible that OSS will have advantages for a small group. In ongoing
work a group of three staff with some informatics support has been able to
reach a level of ~1 Mb of sequence throughput per year. This level would be enough to manage
many large scale sequencing projects and compares favorably with the per capita
throughput of large scale random shotgun sequencing by large groups. It remains
untested, however, whether random shotgun sequencing on a comparable scale can
also be efficiently adapted to a small group.
Based on this study and simulations, subcloning of a single YAC can provide
material for projects across at least 150 kb and likely for much larger
stretches of DNA.
In ongoing work, OSS has also been applied to bacterial clones like BACs or PACs
with ease. Bacterial clones show much lower levels of co-cloning and are easier to prepare free of contaminating host DNA than are YACs, though as we comment above, the sequencing
required to discard adventitious yeast clones is a very small fraction of total
effort and could be reduced further, for example by a second pulsed-field gel electrophoresis (
22
) or by passage of the YAC through a `window' strain (
23
).
As an additional consideration, there are many regions that are only available
in YACs. (Even in the model eukaryotic organism
Caenorhabditis elegans
, 10-20% of the genome has thus far been cloned only in YACs;
24
.) YACs may thus be the substrate of choice (or necessity) for long-range sequencing of certain portions of genomes. It is therefore
encouraging that OSS can already be applied to YACs, with an efficiency that
can obviously be improved. If subclones are randomly derived from a starting
target clone, sequences from their ends indeed produce a map (Fig.
2
), as had been anticipated (
3
), and sequencing effort would be expected to scale linearly with the size of
the clone.
Repetitive sequences.
The suite of programs assembled for the analysis of an earlier region of 220 kb
(
14
) has been useful in this case as well. In the earlier case, in a region of very
high GC (57.2%), the repetitive sequences were heavily weighted toward Alu,
with only a marginal contribution of L1 and MER. Here again, in a region of
moderate GC content (45.5%), Alus dominate. The corresponding percentages of
sequence that are Alu, L1 and MERs were 20.4, 1.6 and 3% in the high GC region
and 24.9, 5.7 and 9.3% here. Thus, as first indicated in the comparison of the
Xq28 segment with other reports of long-range sequencing (see Discussion;
14
), the results show no trend in the proportions, for example, of Alu:L1. Rather,
they are in agreement with the earlier analysis of Alu- and L1-containing restriction fragments in YACs from Xq24-qter, which indicated that both Alus and L1s were widely
distributed with variable densities across nearly all of the 50 Mb region (
25
).
Gene number and characteristics.
Gene content, however, continues to follow the trend observed for large scale
sequence tracts, that coding capacity is proportional to overall GC content in
`isochores' (
26
), zones of up to 1 Mb length with characteristic roughly constant GC. Like
other regions with 42-45.5% GC (
27
,
28
), yWXD703 contains about four genes in 140 kb [including three candidate genes
(703-2, 703-3 and ANT2) and two partial genes (703-1 and 703-4)], or about 1/32 kb.
It is of interest that of the five gene candidates in the YAC, four have already
been supported by EST content (703-1, 703-2, ANT2 and 703-4) or homology to known genes (ANT2 and 703-2). (Furthermore, the EST homologous to 703-4 derives from fetal kidney and the sequence has
also now been detected in fetal kidney RNA in Northern analysis; work in
progress.)
Gene 703-2 is especially interesting, since, as Figure
3
shows, it is highly homologous to `GDC', a carrier protein that has been
associated with Graves' disease (
29
). GDC is a mitochondrial protein and one can wonder whether the X-linked gene detected here is a form that is functionally related. It is
unlikely that the gene product of 703-2 is also a constituent mitochondrial protein, since it would then be expressed in every cell, but no corresponding cDNA was present in any of 12 cDNA libraries assayed by PCR-based tests.
The increasing number of ESTs available from systematic cDNA projects and the
consituent ANT2 gene provide alternative determinations of sequence and
particularly of the gene segments that are usually considered the most
important part of genomic sequence. Comparisons permit some inferences about
the fidelity of the YAC sequence to Xq25 genomic DNA.
The simplest criterion for fidelity is co-linearity. Genomic DNA should show no deletion of entire exons or within
exons of cDNA. The orders of exons and their sequences are co-linear in yWXD703 and all expressed sequences available thus far,
including every EST assigned to the region, a portion of a gene similar to one
associated with Grave's disease and all of the ANT2 gene. In addition, the
primer sequences for three other STSs developed from genomic DNA [registered in
the WU Genome Center database as sWXD1832 (2736L; DXS7340), sWXD774 (TH4) and
sWXD770 (TH10)] are all contained in yWXD703 at the proper interprimer
distances.
A more stringent requirement for fidelity is based on the detailed comparison of
sequence. That YAC sequences can indeed be faithful to genomic DNA has been
indicated in another case (
9
), in which G6PD sequence was derived with 99.99% precision and complete
agreement for both a YAC and a cDNA. For segments of yWXD703, disregarding a
low level of polymorphic variation in sequence, the maximum agreement cannot
exceed the accuracy of the determination of cDNA sequences, or ~96% (EST sequences in GenBank are also less accurate from 3'-ends of cDNAs and toward the end of long sequence tracts).
Since the genomic sequences are ~95-97% identical to the independently determined cDNA sequences, the
results are consistent with a YAC sequence that is likely to be faithful to
uncloned DNA. However, the comparisons are of course still limited. More
stringent evidence could be obtained by comparative sequencing of tracts
sampled from uncloned DNA by PCR-based methods at statistically chosen intervals or by the comparison of
Southern hybridization patterns of genomic and YAC DNA when probed with labeled
YAC DNA.
In conclusion, this 135 kb segment of Xq25 has been analyzed in a single YAC
clone using OSS. The approach worked relatively smoothly, though there is still
room for improvement. The sequence analysis programs also worked to give
accurate gene predictions and comparisons to ESTs, indicating that the human
DNA in the YAC likely retains the sequence of genomic DNA.
We thank Warren Regala, Sandra MacMillan and Patricia Taillon-Miller for technical help with sequencing, YAC preparations and subcloning. Support for this project came from the NIH (GESTEC grant
HG00201).
REFERENCES
Return