Equimolar amounts of two long-range PCR products which together encompass the complete mitochondrial genome, which is a double-stranded circular molecule of 16,6 kb, were pooled and fragmented by sonication, ligated to a biotinylated DNA adapter, denatured, and immobilized on streptavidin-coated magnetic beads (, top left). The immobilization prevents self-hybridization of the bait molecules that occur if they are free in solution. DNA extracted from blood or saliva from 46 individuals 
were used to produce indexed Solexa libraries 
, which were pooled in equimolar amounts, denatured (, top right) and incubated with the beads for 48 hours. The beads were then washed and the captured molecules were heat-eluted, amplified and sequenced (, bottom) on one lane of a Solexa Genome Analyzer II.
Overview of the capture-on-beads method.
The number of reads per individual varied between 237,763 and 801,556 (). On average, 16% of the reads in each sample mapped 
to the reference mtDNA sequence (NC_012920) () and the average mtDNA coverage varied between 43- and 151-fold (). The minimum coverage at any base in any sample was 8-fold (). The coverage across the mitochondrial genome and samples was fairly uniform, with a 6-fold difference between the positions of highest and lowest coverage ().
Number of reads sequenced (green bar) and aligned to the mitochondrial genome (red bar) for each sample.
Average (red squares) and minimum coverage (green squares) of the mitochondrial genome for each sample.
Coverage of each position across the whole mitochondrial genome, considering all the samples together.
To validate the method, we compared the sequences determined by us to sequences for parts (hypervariable region I) of the same mtDNAs produced by a traditional approach where PCR products were sequenced by the Sanger method 
. After the exclusion of a homopolymeric C-stretch which can vary in length due to PCR-induced nucleotide misincorporations, a total of 17,134 bases (approximately 372 per individual) could be compared. They agreed except at seven positions in single individuals, where Ns were called by the capture/Solexa method. These Ns most probably arise due to rare recombination events during the amplification of the pool of indexed libraries and can be avoided by omitting this step 
. One N was called both in the PCR/Sanger and the capture/Solexa in one individual. This is probably due to heteroplasmy, i.e. the presence of two different mtDNA sequences in this individual.
are insertions of parts of mitochondrial genome into the nuclear genome 
. Because of their similarity to the mitochondrial genome numts
can potentially hybridize to the mitochondrial DNA-derived baits and lead to ambiguities in mtDNA sequences (represented as Ns) or even to incorrect sequence determination. To test for the potential presence of numts
we mapped all the reads overlapping ambiguous positions (Ns) against the human genome with blat 
. Only 0.08% of the reads had a higher score to the nuclear genome then to the organellar mtDNA and are thus potentially numts
. Additionally, we translated all protein-coding sequences in silico
(13 per mitochondrial sequence) and found no premature stop codons. This demonstrates that the capture method is reasonably insensitive to human numts
The method described allows the efficient capture of any unique sequence for which a PCR product can be generated. It is cost efficient in that it requires only standard laboratory equipment and reagents and fast in that the capture can be performed immediately when the PCR products are at hand. A similar method for capturing mtDNAs was recently described 
. The authors performed 100 PCR reactions to produce biotinylated baits covering the mtDNA and performed two consecutive hybridizations in solution. The approach presented here is different in that the bait is immobilized on the beads during capture. This prevents the bait molecules from self-hybridizing making both strands accessible for the target capture and the production of the bait simpler (e.g. only two PCR reactions are needed). Additionally, we have shown that our approach can be multiplexed, allowing for efficient analysis of many samples in parallel. In our research group it has been used to capture complete mitochondrial genomes from complex samples such as saliva and ancient hominin bones. Although the efficiency of capture is slightly lower when the human DNA is contaminated by one or two orders of magnitude greater amounts of microbial DNA, it is possible to retrieve complete mitochondrial genomes from most such samples using this method.