Agilent SureSelect - coverage of high GC regions

[NGSfan]

We have successfully run a targeted enrichment with SureSelect and were able to achieve similar results to the Tewhey et al (2009 Genome Biol) for our own targeted subset. As shown in their paper, we also noticed that regions of high GC content were difficult to capture - we see lower read coverage in these areas. Does anyone have any experience trying to increase the coverage of these more difficult regions? Say, for example, by increasing the number of baits overlapping a high GC region?

We are wondering if this is a worthwhile approach and if by chance anyone has tried it already with useful results. We have some extra design space on our SureSelect and are considering "piling on" the baits in these regions for a few important genes.

[krobison]

I would be very interested in how you fare with this.

One possible explanation for dropout of extremes of %GC is not so much the SureSelect hybridization but the various PCR steps. Do you think you could significantly shave the total number of PCR cycles the library is exposed to?

[Xi Wang]

Originally Posted by NGSfan

We have successfully run a targeted enrichment with SureSelect and were able to achieve similar results to the Tewhey et al (2009 Genome Biol) for our own targeted subset. As shown in their paper, we also noticed that regions of high GC content were difficult to capture - we see lower read coverage in these areas. Does anyone have any experience trying to increase the coverage of these more difficult regions? Say, for example, by increasing the number of baits overlapping a high GC region?

We are wondering if this is a worthwhile approach and if by chance anyone has tried it already with useful results. We have some extra design space on our SureSelect and are considering "piling on" the baits in these regions for a few important genes.

How do you define the sequence coverage? If you take log values of the coverage, what is the results?

[upenn_ngs]

The drop in high-GC content is largely from secondary structure formation. Adding formamide and increasing the temperature might tilt the table toward hybridization with RNA oligo.

[NGSfan]

Originally Posted by krobison

I would be very interested in how you fare with this.

One possible explanation for dropout of extremes of %GC is not so much the SureSelect hybridization but the various PCR steps. Do you think you could significantly shave the total number of PCR cycles the library is exposed to?

For the PCR steps, I mostly worry over "PCR duplicates". And if I remember, wouldn't the PCR bias the coverage in favor of high GC?

http://www.ncbi.nlm.nih.gov/pubmed/18660515

[NGSfan]

Originally Posted by Xi Wang

How do you define the sequence coverage? If you take log values of the coverage, what is the results?

Good question - I am just doing this "by eye" so to speak. So for example, the average bp coverage of a target region is 20X and then drops to 2 or 0 in high GC regions.

[NGSfan]

Originally Posted by upenn_ngs

The drop in high-GC content is largely from secondary structure formation. Adding formamide and increasing the temperature might tilt the table toward hybridization with RNA oligo.

The secondary structure issue would be my guess as well. My only concern with adding formamide and/or increasing the temp is the effect on lower GC targets. I like the idea, but shifting the binding energies might cause as many problems as it solves.

The idea of adding more baits was to help increase coverage of a subset of targets without affecting the enrichment of other targets.

[upenn_ngs]

Another factor, many GC rich regions are dropped from both whole genome sequencing as well as the exome capture. This image from the Broad.

http://www.postimage.org/image.php?v=aV6cBnA

[Xi Wang]

Originally Posted by NGSfan

Good question - I am just doing this "by eye" so to speak. So for example, the average bp coverage of a target region is 20X and then drops to 2 or 0 in high GC regions.

I guess if you take log value of the bp coverage for each region, and then take the average, the phenomenon will be different. I am just wondering the amplification is exponentially increased the DNA fragments.

[dottomarco]

Originally Posted by NGSfan

We have successfully run a targeted enrichment with SureSelect and were able to achieve similar results to the Tewhey et al (2009 Genome Biol) for our own targeted subset.

@ NGSfan : What sequencer did you use? Illumina GA II?
I am interested in using Agilent's SureSelect for sequence enrichment to get the targets to sequence with a 454 FLX. Do you think using a long fragmented 454 library with SureSelect can create any problem with the hybridization? Agilent do not provide any ufficial protocol for 454 libraries, but I assume that their long baits could work well with our ~400-500 bp fragments.

[krobison]

The main platform-customization of the SureSelect as I understand it is there are blocking oligos to prevent daisy-chaining of products -- without these sometimes a correctly hybridized fragment will hybridize to an off-target fragment via the adapter regions.

[NGSfan]

Originally Posted by Xi Wang

I guess if you take log value of the bp coverage for each region, and then take the average, the phenomenon will be different. I am just wondering the amplification is exponentially increased the DNA fragments.

Yes, that is a good point. Amplification is a concern and will certainly bias things. However, I am not seeing PCR duplicates to be a big issue in my data set.

I have talked to some big sequencing centers about the GC issue and they also have encountered it, however their approach is to simply bump up the sequencing - to 70X coverage (we are at 30-40X).

I should have mentioned we did single end reads. We should be getting paired end reads soon, and I hope this might help a little, since we'll be able to sequence a GC-rich region which was partially bound at the other end with an average GC content. Maybe?

[NGSfan]

Originally Posted by dottomarco

@ NGSfan : What sequencer did you use? Illumina GA II?
I am interested in using Agilent's SureSelect for sequence enrichment to get the targets to sequence with a 454 FLX. Do you think using a long fragmented 454 library with SureSelect can create any problem with the hybridization? Agilent do not provide any ufficial protocol for 454 libraries, but I assume that their long baits could work well with our ~400-500 bp fragments.

Their long baits ~120bp are quite reasonable - but I have no clue on the behavior of hybridization when having longer fragments (400-500). I suspect you might run into self-hybridizing issues more often, but who really knows!

We generally followed the Agilent protocol - fragmenting ~200bp . Something to note: if you are after exons, then you don't want too long a fragment because you'll be sequencing at the ends and your aligned reads will be more often "off target" so to speak - in the sense that they will be around the exon, rather than on the exon.

[Xi Wang]

Originally Posted by NGSfan

Yes, that is a good point. Amplification is a concern and will certainly bias things. However, I am not seeing PCR duplicates to be a big issue in my data set.

I have talked to some big sequencing centers about the GC issue and they also have encountered it, however their approach is to simply bump up the sequencing - to 70X coverage (we are at 30-40X).

I should have mentioned we did single end reads. We should be getting paired end reads soon, and I hope this might help a little, since we'll be able to sequence a GC-rich region which was partially bound at the other end with an average GC content. Maybe?

Oh. But if you have the data, you can try what just I mentioned.

And for PE reads, I don't think it can improve a lot. Because it is the DNA fragments that amplified. So the coverage should have some relationship with the GC-content of the DNA fragments. On the other hand, the read GC-content and the DNA fragment GC-content have a high correlation. As a result, the relationship between the read GC-content and the coverage reflects a lot the reality.

[bioinfosm]

Another point which I did not notice here is, # of reads actually sequenced to get 30x exome coverage for the agilent capture stuff.

We notice that only 20% of reads map on-target! Is that a common thing? (Illumina 75bp PE)

[bioinfosm]

Originally Posted by upenn_ngs

Another factor, many GC rich regions are dropped from both whole genome sequencing as well as the exome capture. This image from the Broad.

http://www.postimage.org/image.php?v=aV6cBnA

Do you know what the WGS track is, and from where one can obtain it for the IGV?

[NGSfan]

Originally Posted by Xi Wang

Oh. But if you have the data, you can try what just I mentioned.

And for PE reads, I don't think it can improve a lot. Because it is the DNA fragments that amplified. So the coverage should have some relationship with the GC-content of the DNA fragments. On the other hand, the read GC-content and the DNA fragment GC-content have a high correlation. As a result, the relationship between the read GC-content and the coverage reflects a lot the reality.

I'm not clear on why converting the read coverage to a log scale would help understand distribution better. Simply visualizing coverage on a log scale will simply change the scale you're looking at, no?

Maybe I'm not understanding the advantage. Could you show me an example?

[if1]

Originally Posted by bioinfosm

...
We notice that only 20% of reads map on-target! Is that a common thing? (Illumina 75bp PE)

Hi,
we have also run a TE using SureSelect with Illumina 76bp reads Single End and we obtained similar results to the Tewhey et al (Genome Biol 2009): 50% of uniquely aligned reads were on target with a uniformity of capture similar to what reported in the paper.
I am wondering if someone else has results on Illumina 76 Paired-End, as it seems from Agilent website that the % on target should increase from 50% to 70% using PE protocol.

Thanks

[NGSfan]

Originally Posted by bioinfosm

Another point which I did not notice here is, # of reads actually sequenced to get 30x exome coverage for the agilent capture stuff.

We notice that only 20% of reads map on-target! Is that a common thing? (Illumina 75bp PE)

Becareful with Tewhey's enrichment calculation, they calculated 48% "on or near target" meaning +/- 150bp around the target! If you look in the text, they actually only got 37% "on target" , so while your 20% is low, it's not terrible in comparison. How much of the genome are you targeting?

In our case with 76-bp single end reads, we targeted 0.09% of the genome and enriched to 35% of the sequences being "on target" , which is a ~390-fold enrichment. If you were to actually convert Tewhey's numbers to solely "on target" (from 0.12% to 37%), then their claim of "about 400 fold enrichment" is actually ~290-fold. Just a small criticism.

We have just completed a 76-bp paired end run with 4 samples multiplexed - I will let you know what we get with our alignment results

[bioinfosm]

Thanks NGSfan.

The on or near number, taking +/- 200bp goes to 15%, still pretty low I would guess.

It would be interesting to check, where all the rest 85% reads went!