[koadman]

Eh? Why did the first index work then?
I mean does that make sense to you? NaOH sits around for days in the instrument during read1. Then it successfully denatures the first read to allow annealing of the index1 primer. But after the additional several hours it takes to do 8 cycles then it doesn't work?

--
Phillip

To be honest very little of it ever made sense to me. I don't run the instruments, the core facility does, and I don't communicate with Illumina tech support, the core facility does. I'm an informatics guy with only a cartoon view of how sequencing works! I am merely trying to communicate more broadly with others what Illumina claimed the problem was. It would be super interesting and extremely valuable for future runs to learn that the problem is something more temperamental and the fact that it worked when we (e.g. the core people here) loaded fresh NaOH was just a fluke.

[pmiguel]

That's good to know - unfortunately, we can't do rapid chemistry as yet due to only having a HiScanSQ. We are looking at upgrading in the next few months to a 2500, but the user is a bit more desperate for the data than that.

Yes, we also used to run a HiScanSQ. Man, that machine was a headache compare to the HiSeq.

The "hassle" comes at the bioinformatics stage - because the sample sheets only give you the option of TruSeq OR Nextera (not both) and are applied across the whole flow-cell, I was told that we would essentially have to create two sample sheets (one TruSeq, one Nextera) and run the BaseCalltoFASTQ script twice in order to pull out the indices correctly.

I know they say this. It is untrue and bone-headed though. Single index libraries generate legitimate dual 8 base index sequences. Just stick those in your sample sheet and you can just run CASAVA once. Works on the MiSeq too. So, I don't see any impediment for mixing dual and single index libaries in the same lane, should the need arise.

That said, you also are essentially "wasting" cycles doing a second index read on the TruSeq samples when they only have a single index (unless using the new HT kits). Splitting SBS reagents for 101|8|8|101 indexing is a bit of a chore as well (a 200-cycle kit doesn't have enough spare reagents for the Nextera dual indexing, so you have to pool four 50-cycle kits). I can see larger facilities with high workloads not wanting the extra time to deal with this.

Yeah, this I am dreading. But hopefully we'll have 2 flowcells and we'll order two 200-cycle kits and one 50-cycle kit as well. Then just split the 50 cycle kit between the two 200-cycle kits.

Or maybe Illumina will turn on the 3rd swath for Rapid flowcells and then we don't need to use the slow chemistry any more.

--
Phillip

[matth431]

Well, maybe for certain definitions of the word "efficiently".
The MiSeq produces a higher cluster density for a given concentration of library. However, consider the absolute amount of library consumed by the MiSeq per cluster generated:

600 ul of 10 pM library = 6 fmol of library to generate 15 million clusters on a MiSeq v2 run.

120 ul of 15 pM library = 1.8 fmol of library to generate >200 million clusters on the cBot running v3 cluster chemistry.

MiSeq 2.5 million clusters/fmol
cBot/HiSeq 111 million clusters/fmol




How about using Rapid chemistry? Doesn't that come with the Nextera primers already in the mix? Also does on instrument clustering.

--
Phillip

That's good to know - unfortunately, we can't do rapid chemistry as yet due to only having a HiScanSQ. We are looking at upgrading in the next few months to a 2500, but the user is a bit more desperate for the data than that.

As far as Docbio's core facility not offering mixed runs, I suspect this is more down to the hassle rather than the actual ability to do it. Swapping out the cluster generation and indexing primers from the standard kit for those supplied in the Nextera Sequencing Primer kit is not an issue - these worked well enough on the TruSeq samples we had on that run so there should be no issue having some lanes TruSeq and some Nextera.

The "hassle" comes at the bioinformatics stage - because the sample sheets only give you the option of TruSeq OR Nextera (not both) and are applied across the whole flow-cell, I was told that we would essentially have to create two sample sheets (one TruSeq, one Nextera) and run the BaseCalltoFASTQ script twice in order to pull out the indices correctly.

That said, you also are essentially "wasting" cycles doing a second index read on the TruSeq samples when they only have a single index (unless using the new HT kits). Splitting SBS reagents for 101|8|8|101 indexing is a bit of a chore as well (a 200-cycle kit doesn't have enough spare reagents for the Nextera dual indexing, so you have to pool four 50-cycle kits). I can see larger facilities with high workloads not wanting the extra time to deal with this.

[pmiguel]

Matt,

Sorry I can't be of much help. I'm sending my samples to a core facility and they flat out refuse to mix chemistries on a flow cell. I had been led to believe that it wasn't even possible to do what you're describing due to limitations of the cBot, but it seems you've found a strategy Illumina endorses. Not sure what the problem would be, unless there is some inherent different in the hybridization affinities between the two adapter types or some difference in the insert sizes that resulted in differential clustering of the two library types due to different concentrations of ends at a given pM concentration. It's late so I'm not even sure if that makes sense...

I do know Illumina says the MiSeq clusters more efficiently than the HiSeq, but you probably know that already.

Well, maybe for certain definitions of the word "efficiently".
The MiSeq produces a higher cluster density for a given concentration of library. However, consider the absolute amount of library consumed by the MiSeq per cluster generated:

600 ul of 10 pM library = 6 fmol of library to generate 15 million clusters on a MiSeq v2 run.

120 ul of 15 pM library = 1.8 fmol of library to generate >200 million clusters on the cBot running v3 cluster chemistry.

MiSeq 2.5 million clusters/fmol
cBot/HiSeq 111 million clusters/fmol

Sorry I can't be of much help... if you figure out a solution post an update because I'm curious.

Best,
DocBio

How about using Rapid chemistry? Doesn't that come with the Nextera primers already in the mix? Also does on instrument clustering.

--
Phillip

[pmiguel]

After much discussion with Illumina tech support we discovered that the NaOH which is used to denature the DNA prior to the 2nd barcode read had apparently neutralized during the several days of the first read, and this resulted in failure of the 2nd barcode. We attempted another run, replacing the NaOH with fresh solution before the 2nd barcode read and indeed, this seems to have resolved the problem. Illumina tech support said this issue may be related to the particular environment at our facility (too much sunshine in Davis?) and that not all users will experience the problem.

So, for those who are attempting dual index reads, if the 2nd barcode read quality is suspect it might be as simple as adding fresh NaOH!

Eh? Why did the first index work then?
I mean does that make sense to you? NaOH sits around for days in the instrument during read1. Then it successfully denatures the first read to allow annealing of the index1 primer. But after the additional several hours it takes to do 8 cycles then it doesn't work?

--
Phillip

[koadman]

Following up on my earlier posting I want to warn others of some problems we encountered with dual-index sequencing on the hiseq and how we resolved them to ultimately arrive at a good result. Our first attempt with a single lane gave encouraging results, so we loaded up the better part of a flowcell with dual indexed samples. Unfortunately the 2nd index failed to read on that run (and yes it compromised much of the planned analysis). Most of the 2nd index reads were of extremely low quality and those that did read were inside the adapter downstream from the first barcode. After much discussion with Illumina tech support we discovered that the NaOH which is used to denature the DNA prior to the 2nd barcode read had apparently neutralized during the several days of the first read, and this resulted in failure of the 2nd barcode. We attempted another run, replacing the NaOH with fresh solution before the 2nd barcode read and indeed, this seems to have resolved the problem. Illumina tech support said this issue may be related to the particular environment at our facility (too much sunshine in Davis?) and that not all users will experience the problem.

So, for those who are attempting dual index reads, if the 2nd barcode read quality is suspect it might be as simple as adding fresh NaOH!

[docbio]

Matt,

Sorry I can't be of much help. I'm sending my samples to a core facility and they flat out refuse to mix chemistries on a flow cell. I had been led to believe that it wasn't even possible to do what you're describing due to limitations of the cBot, but it seems you've found a strategy Illumina endorses. Not sure what the problem would be, unless there is some inherent different in the hybridization affinities between the two adapter types or some difference in the insert sizes that resulted in differential clustering of the two library types due to different concentrations of ends at a given pM concentration. It's late so I'm not even sure if that makes sense...

I do know Illumina says the MiSeq clusters more efficiently than the HiSeq, but you probably know that already.

Sorry I can't be of much help... if you figure out a solution post an update because I'm curious.

Best,
DocBio

[matth431]

Sorry I didn't see this post earlier. We've been running NexteraXT libraries on the HiSeq with good results. Following the recommendations of Illumina, we stop after library cleanup (no bead based balancing), then we quantitate the libraries with a high-sensitivity Qubit assay.

Our HiSeq core requires minimum 10nM pools in 25uL volume. Using the approximation of 2nM = 1ng/ul for 1kb fragments, we figure that's 125,000 pg of DNA in 25uL, and divide that between however many samples we're pooling (48-96) to figure out the number of picograms of each library to pool. The pool volume winds up being quite high, but we run it through a 30k mwco amicon concentrator column to get our final 25uL pool. I was nervous about the whole ordeal but the balancing has been acceptable and the clustering has been very good.

Another caveat about the bead based balancing and the MiSeq, once you come out of the BBB you've got single stranded libraries. They claim only a 10-20% loss in a freeze-thaw cycle but in practice we've seen clustering drop from 50-70% with a single freezethaw between MiSeq runs.

So they work, if you pay attention to the caveats and avoid the bead based balancing step.

Hi DocBio,

Could I check whether you ever perform mixed flow-cells with some lanes TruSeq and some Nextera (not both in same lane)? We carried out a run last year with six lanes of TruSeq and 2 lanes Nextera XT, following the caveats about using the Nextera Sequencing Primer kit for the cBot/PE read, etc. However, despite good clustering of the TruSeq samples (we load at 10pM), we only obtained ~8 million clusters per lane for the Nextera samples. Illumina were at something of a loss to explain this. We didn't use the normalisation beads following library prep - the samples were individually diluted to 10pM (quantified using Qubit which works well for loading Nextera libraries on our MiSeq). I should note that we typically see equivalent cluster densities for both library types at 8pM loading on the MiSeq.

Any help appreciated as we have a user wishing to do a similar run in the next week or so!

Matt

[docbio]

Sorry I didn't see this post earlier. We've been running NexteraXT libraries on the HiSeq with good results. Following the recommendations of Illumina, we stop after library cleanup (no bead based balancing), then we quantitate the libraries with a high-sensitivity Qubit assay.

Our HiSeq core requires minimum 10nM pools in 25uL volume. Using the approximation of 2nM = 1ng/ul for 1kb fragments, we figure that's 125,000 pg of DNA in 25uL, and divide that between however many samples we're pooling (48-96) to figure out the number of picograms of each library to pool. The pool volume winds up being quite high, but we run it through a 30k mwco amicon concentrator column to get our final 25uL pool. I was nervous about the whole ordeal but the balancing has been acceptable and the clustering has been very good.

Another caveat about the bead based balancing and the MiSeq, once you come out of the BBB you've got single stranded libraries. They claim only a 10-20% loss in a freeze-thaw cycle but in practice we've seen clustering drop from 50-70% with a single freezethaw between MiSeq runs.

So they work, if you pay attention to the caveats and avoid the bead based balancing step.

[Zuuhl]

FWOS, just forget the NaOH to check Illumina did not for nothing changed the sealing on the NaOH row of the cBot plate...

Ahh, but the NaOH on the cBot plate it not used to make the initial library single-stranded.
It is used during the Denaturation-Hybridization step at to get the clusters single-stranded before the seq. primer annealing.

Kim

[pmiguel]

It is remarkable that the diversity between what people cluster is so big. We use 10 pM and some others 16. Something that I find hard to believe is machine dependent.

You mean in the other thread? Yes. Actually looks like all the results are fairly linear for pM to Kclusters/mm^2, except for one outlier.

Anyway, each strand is a separate amplicon. One initially anneals through the p5, the other through p7, or do I misunderstand you Phillip?

Thanks, that is what I meant.

--
Phillip

[josdegraaf]

It is remarkable that the diversity between what people cluster is so big. We use 10 pM and some others 16. Something that I find hard to believe is machine dependent.

Anyway, each strand is a separate amplicon. One initially anneals through the p5, the other through p7, or do I misunderstand you Phillip?

FWOS, just forget the NaOH to check Illumina did not for nothing changed the sealing on the NaOH row of the cBot plate...

[pmiguel]

I am pretty sure that the cbot relies more on heat denaturation than on the NaOH base denaturation. So the base is just a failsafe. The first step in cbot cluster gen is an incubation around 98C. So the material that actually clusters is always single stranded. The single strands should cluster without much bias as long as they have both p5 and 7 adapters.

What oligos are on the flow cell? A double stranded amplicon would be structured like:

5'->>>>P5>>>>>>>>>>>>>---------insert----->>>>P7-complement>>>>-3'
3'-<<<P5complement<<<---------insert-----<<<<P7<<<<<<<<<<<<<<<<-5'

The flow cells have to have both P5 and P7 oligos on them to allow cluster PCR to occur. Does that mean that each strand is a separate amplicon, or is one non-functional for some reason?

--
Phillip

[FWOS]

Cbot denaturation

I am pretty sure that the cbot relies more on heat denaturation than on the NaOH base denaturation. So the base is just a failsafe. The first step in cbot cluster gen is an incubation around 98C. So the material that actually clusters is always single stranded. The single strands should cluster without much bias as long as they have both p5 and 7 adapters.

[pmiguel]

Hi all, I've been looking at this as well. Reading through the Illumina protocol pdf, I noticed the following:



Which suggests that one might be able to play with the relative amounts of hybridization buffer and bead-normalized sample material. Would anybody care to speculate on whether this would or would not work for getting a library at the right loading concentration for the HiSeq? My understanding is that HiSeq loading concentration should be about 1.5x the miseq. The protocol calls for 24uL library in 600uL. What about 36uL library in 600uL?

cBot wants 120 ul of library/lane. Note that HiSeq lanes have a much larger surface area than a MiSeq lane.

A few labs reported results from clustering at a given concentration on their HiSeq in another thread. We had results consistent with those obtained by most of the labs posting in that thread. That is 16 pM resulted in about 800-850 Kclusters/mm^2. 16 pM is about 10 million amplicons/ul. So 1.2 billion amplicons were consumed per lane to yield about 240 million raw clusters. 20 % efficiency -- not bad. (But can either strand of an amplicon anneal to the flow cell oligos? That would be a 2x factor to consider.)

To get about the same cluster density you would need 600 ul of ~11 pM. Is that right, 4 billion amplicons? What does that get you using v2 hardware? 12 million clusters? 0.3% efficiency?

Please check my logic/math. Seems pretty outlandish that you would use more amplicons to yield orders of magnitude lower numbers of reads on the MiSeq.

If true, that suggests, in principle, the normalization beads should yield enough amplicons to cluster a HiSeq lane with.

A couple of cautions, though:

(1) Not sure about the flow cell oligos on a HiSeq flow cell being able to capture either strand of an amplicon. Since the bead-based normalization of NexteraXT presumably removes one of the strands, it would be critical that there be oligos to capture the strand that is retained.

(2) NexteraXT calls for a heat denaturation prior to clustering. Whereas all HiSeq protocols call for base denaturation. Ignoring dilution issues, will heat denatured templates even work on the cBot? NexteraXT calls for the run to be started as soon as possible after heat denaturation of templates. Could be the cBot has a pace to leisurely for this to work?

--
Phillip