Understanding how the data have been generated is necessary to choose the right analysis tools, and to properly interpret the results.
Here we only talk about data obtained from the Illumina sequencing of target DNA fragments captured using RNA baits, because this is what many people currently use at Kew Gardens, especially in the context of the Plant And Fungal Trees Of Life (PAFTOL) project.
We, Kew scientists using target capture sequencing, provide here a summary of target capture and sequencing as it is performed at Kew, and we also include points of discussion that have proven to be relevant to many of us.
This document does not pretend to be exhaustive, and another great resource is for instance the documentation provided here and here by Matt Johnson (Texas Tech University, USA), Eliott Gardner (Morton Arboretum, USA) and N. Wickett (Northwestern University, USA), who contributed to the development of the PAFTOL bait kit (see below) together with Kew.
Bait design
DNA preparation and target capture
Sequencing
Details
Bait design details
Library size and splash-zone
Adapters
We first have to design baits to capture the target DNA
-
we choose some genes that we want to study. These genes are the target genes.
-
we design RNA sequences (usually 120 bp long) that align to the target genes.
These sequences are called baits, because they will allow to capture the target genes.
The baits can be designed to overlap each-other, we talk about bait tiling. -
we synthesize the baits with biotinylated nucleotides: this is the so called bait kit
People usually rely on a company to design and to synthesize the baits.
Once the bait kit is ready, it can be used to capture the target regions from one or multiple DNA samples.
-
we extract the genomic DNA of an individual.
-
we prepare the DNA of this sample so that all its DNA fragments have a similar size (details here)
-
we put adapters containing primers and indexes at both extremities of all the DNA fragments (details here). The result is called a DNA library.
-
we add the baits and a few other reagents, and we leave it 16 to 24 hours at 65 degrees Celsius. This is the hybridization reaction: the baits bind preferentially to the DNA to which they are the most similar (i.e. to the targets), but also to other DNA regions (details below). We usually multiplex, i.e. we put multiple DNA libraries together in a single tube (because it is cheaper). This is the hybridization pool.
-
At the end of the hybridization, the baits + target DNA are captured using magnetic beads coated with streptavidin, which binds to the biotin of the baits. The beads+baits+target DNA cluster on the magnet while the reagents and the DNA that did not attach well to the baits are washed away.
-
Finally, the DNA that attached to the baits is eluted, and amplified by PCR. We thus end up with the captured DNA, containing mostly the target genes but also some non-specific = off target DNA that came along.
The ratio of specific/non-specific (= on/off target) captured DNA illustrates how successful was the enrichment of the libraries in target DNA: the higher the ratio is, the more successful the enrichment is, the lesser sequencing effort is needed.
We often put multiple pools of captured DNA together (again because it is cheaper) in a single tube to proceed with the sequencing. This is the sequencing pool.
The sequencing pool is loaded into the Illumina sequencing machine (MiSeq, HiSeq, etc.), which will read the DNA sequences (details here).
We usually perform paired-end sequencing: the DNA fragments are read twice, once from each extremity, producing a pair of reads that are known to come from a same fragment.
NB: Because the DNAs of multiple samples are mixed in the hybridization pool(s) and/or in the sequencing pool, it is crucial that the DNA libraries of these samples are made with adapters containing different indexes, so that we can tell what DNA sequence is from what sample. The easiest way to ensure this is to think about what samples should go in a same sequencing pool before starting to prepare the libraries.
The RNA hybridizes with the DNA, and the more similar the bait is to the target, the more stable the hybridization is. An ideal bait should thus be very similar to the target gene and much less similar to any other un-desired region. One can optimize this by choosing target regions that don’t have very similar un-desired copies in the genome.
This does not obligatory mean that one has to chose low-copy regions: you may be interested in a gene family, or in recovering all paralogs of a certain gene, in which case you will design baits in the regions that you expect to be the most conserved across the genes family / the paralogs. But these baits should still ideally be much more similar to these regions than to any other non-desired region of the genome.
When we want to be able to capture genetically divergent taxa with a same bait, one way is to have baits targeting regions that we expect to be conserved between the taxa, typically exons. But the problem then is that the targeted regions may be too conserved to provide phylogenetic signal. There are multiple ways around this, for instance:
-
One can target regions that are conserved enough to be captured by a same bait but divergent enough so that they provide phylogenetic signal. This is possible to a certain extent because baits can hybridize to divergent DNA, but at some level of divergence this becomes inefficient because the baits will hybridize a lot to non-desired regions.
In addition, since the efficiency of hybridization is proportional to the bait-target similarity, if, in a same hybridization pool, some taxa have DNA more similar to the baits than other taxa, the captured DNA may come mostly from the taxa with the DNA the most similar to the baits. -
Another way is to design multiple baits targeting the same DNA region, with some baits more similar to some taxa than to others. This is the approach chosen for the PAFTOL project, but they refined it, as explained in their article.
-
Another, compatible, way is to target regions that are flanked by more variable regions. The flanking regions will come along during the capture if there are DNA fragments containing part of the target region (enough for a bait to hybridize) and part of the flanking region. The larger the library size is, the longer the flanking region that can be captured is, but the higher the effort required to sequence the complete flanking region will be, as shown in Figure 3.
The sequenced flanking regions form what some people call the splash zone.
At the inter-generic level, it may be difficult to find conserved regions flanked by regions that are variable but still conserved enough to be aligned between taxa, so the last approach is most beneficial at the intra-generic level.
This is why the PAFTOL project relies mostly on having multiple baits targeting the same DNA regions in multiple plant families, and on the fact that inside a family these regions are conserved enough to be targeted by a same bait but also variable enough to provide phylogenetic resolution at the genus level.
However, the PAFTOL bait kit can also be used to capture the flanking regions of the PAFTOL targets, which can then be used to answer intra-generic phylogenetic questions and/or build species-level barcoding libraries. Researchers at Kew are currently testing the utility of the PAFTOL bait kit at the infra-generic and infra-specific levels in numerous angiosperm families such as Araceae, Arecaceae, Combretaceae, Fabaceae, Iridaceae, Liliaceae, Magnoliaceae, Nepenthaceae, Nymphaeaceae, Saxifragaceae, and Solanaceae.
Because the Illumina machines do not work with DNA fragments longer than ca. 450 bp (Hiseq X), or ca. 800 bp (other Hiseq). (This is not up-to-date, more precise information welcome!)
Because Illumina gives much more data at a lower cost, and with a lower error rate.
Ideally, to get the most of each sample we collect, we should sequence its full genome using both short and long-read sequencing. However, long-read technologies are still too expensive, and it is usually not possible to get long reads from the herbarium specimens because their DNA is already fragmented.
Using the short read approach is thus not only cost-efficient and sufficient to achieve phylogenetic inference, it is also the only way to use DNA from our herbarium collections.
There are many things to consider:
-
If you sequence together DNAs of different fragment sizes, the smaller fragments will be sequenced preferentially, so in general you may want to fragment your DNAs so that all your libraries have a similar size. Or to sequence libraries of different sizes separately (which increases sequencing costs).
Some size variation is ok, and one can adjust the molarity of individual libraries to correct for it, but usually the more variable the libraries are, the less homogenous the results will be. -
The longer the fragment size is, the longer the splash zone may be (at the condition of a sufficient sequencing effort, see Figure 3)
-
If a non-negligible proportion of the DNA library is smaller than the read length you chose, the sequencing will fail.
Adapters are important because:
-
They contain two indexes (in our case index 5 and index 7), the combination of which is unique for each sample and thus allows to sequence multiple individuals together and still recognize which sequence comes from which sample.
-
They contain primers, which are used to amplify the DNA by PCR during the library preparation and after the hybridization, and to replicate the DNA during the sequencing process.
-
They allow the DNA fragment to be fixed on the flow cell during the sequencing process.
Details about how the adapters are put on the DNA, how the DNA is sequenced, what data result from this, and why we end up with pieces of adapters in the reads are given here (see the last page for a summary of what adapters are in the data).