Thursday, 21 August 2014

FFPE: the bane of next-generation sequencing? Maybe not for long...

FFPE makes DNA extraction difficult; DNA yields are generally low, quality can be affected by fixation artefacts and the number of amplifiable copies of DNA are reduced by strand-breaks and other DNA damage. Add on top of this almost no standardisation in the protocols used for fixation and widley different ages of samples and it's not suprising FFPE causes a headache for people that want to sequence genomes and exomes. In this post I'm going to look at alternative fixatives to formalin, QC methods for FFPE samples to assess their suitability in NGS methods, some recent papers and new methods to fix FFPE damage.
 
Why do we use formalin-fixation: The ideal material to work with for molecular studies is fresh-frozen (FFZN) tumour tissue, as nucleic acids are of high-quality. But many cancer samples are fixed in formalin for pathological analysis and stored as Formalin-Fixed Parrafin-Embeded (FFPE) blocks, preserving tissue morphology but damaging nucleic acids. The most common artifacts are, C>T base substitutions caused by deamination of cytosine bases converting them to uracil and generating thymines during PCR amplification, and strand-breaks. Both of these reduce the amount of correctly amplifiable template DNA in a sample and this must be considered when designing NGS experiments.
 
Molecular fixatives: Our Histopathology core recently published a paper in Methods: Tissue fixation and the effect of molecular fixatives on downstream staining procedures. In this they demonstrated that overall, molecular fixatives preserved tissue morphology of tissue as well as formaldehyde for most histological purposes. They presented a table, listing the molecular-friendly fixatives and reporting the average read-lengths achievable from DNA & RNA (median read-lengths 725 & 655 respectively). All the fixatives reviewed have been shown to preserve nucleic acid quality, by assessment of qPCR Ct values or through RNA analysis (RIN, rRNA ratio, etc). But no-one has performed a comparison of these at the genome level, and the costs of sequencing probably keep these kind of basic tests beyond the limits of most individual labs.

The paper also presents a tissue-microarray of differently fixed samples, which is a unique resource that allowed them to investigate the effects of molecular fixatives on histopathology. All methods preserved morphology, but there was a wide variation in the results from staining. This highlights the importance of performing rigourous comparisons, even for the most basic procedures in a paper (sorry to any histpathologists reading this, but I am writing from an NGS perspective).

The first paper describing molecular a fixative (UMFIX) appeared back in 2003, in it the authors describe the comparison of FFZN to UMFIX tissue for DNA and RNA extraction, with no significant differences between UMFIX and FFZN tissues on PCR, RT-PCR, qPCR, or expression microarrays. Figure B from their paper shows how similar RNA bioanalyser profiles were from UMFIX and FFZN.

UMFIX (top) and FFZN (bottom)

 

Recent FFPE papers: A very recent and really well written paper in May 2014 by Hedegaard et al compared FFPE and FFZN tissues to evaluate their use in exome and RNA-seq. They used two extraction methods for DNA and three for RNA with different effects on quality and quantity.  Only 30% of exome libraries worked, but with 70% concordance (FFZN:FFPE). They made RNA-seq libraries from 20 year old samples with 90% concordance, and found a set of 1500 genes that appear to be due to fixation. Their results certainly make NGS analysis of FFPE samples seem to be much more possible than previous work. Interestingly they made almost no changes to the TruSeq exome protocol, so some fiddling with library prep, perhaps adding more DNA to reduce the impact of strand-breaks for instance would help a lot (or fixing FFPE damage - see below). The RNA-seq libraries were made using RiboZero and ScriptSeq. Figure 2 from their paper shows the exome variants with percentages of common (grey), FFZN-only (white) and FFP-only (red), there are clear sample issues due to age (11, 7, 3 & 2 years storage) but the overall results were good.


Other recent papers looking at FFPE include: Ma et al (Apr 2014): they developed a bioinformatics method fo gene fusion detection in FFPE RNA-seq. Li et al (Jan 2014): they investigated the effect of molecular fixatives on routine histpathology and molecular analysis. They achieved high-quality array results with as little as 50ng of RNA. Norton et al (Nov 2012): they manually degraded RNA in 9 pairs of matched FFZN/FFPE samples, and ran both Nanostring and RNA-seq. Both gave reliable gene expression results from degraded material. Sinicropi et al (Jul 2012): they developed and optimised RNA-seq library prep and informatics protocols. And most recently Cabanski et al published what looks like the first RNA-access paper (not open access and unavailable to me). RNA-access is Illumina's new kit for FFPE that combines RNA-seq prep from any RNA (degraded or not) with exome capture (we're about to test this, once we get samples).

QC of FFPE samples: It is relatively simple to extract nucleic acids from FFPE tissue and get quantification values to see how much DNA or RNA there is, but tolerating a high failure rate, due to low-quality, in subsequent library prep is likely to be too much of a headache for most labs. Fortunately several groups have been developing QC methods for FFPE nucleic acids. Here I'll focus mostly on those for DNA.

Van beers et al published an excellent paper in 2006 on a multiplex PCR QC for FFPE DNA. This was developed for CGH arrays and produces 100, 200, 300 and 400bp fragments from nonoverlapping target sites in the GAPDH gene from the template FFPE DNA. Figure 2 from their paper (reproduced below) demonstrate a good (top) and a bad (bottom) FFPE samples results.


Whilst the above method is very robust and generally predictive of how well an FFPE sample will work in downstream molecular applications, it is not high-throughput. Other methods generally use qPCR as the analytical method as it is quick and can be run in very high-throughput. Illumina sell an FFPE QC kit which uses comparison of a control template to test sampeples and a deltaCq method to determine if samples are suitable for arraya or NGS. LifeTech also sell a similar kit but for RNA, Arcturus sample QC, using two β-actin probes and assessing quality via their 3'/5' ratio.Perhaps the ideal approach would be a set of exonic probes multiplexed as 2, 3, or 4-colour TaqMan assays. This could be used on DNA and RNA and would bring the benefits of the Van beer and LifeTech methods to all sample types.

Fixing FFPE damage: Another option is to fix the damage caused by fomalin fixation. This is attractive as there are literally millions of FFPE blocks, and many have long-term follow up data. A paper in Oncotarget in 2012 reported the impact of using uracil-DNA glycosylase (UDG) to reduce C>T caused by cytosine deamination to uracil. They also showed that this can be incoporated into current methods as a step prior to PCR, something which we've been doing for qPCR for many years. There are not strong reasons to incorporate this as a step in any NGS workflow as there is little impact on high-quality templates.

NEB offer a cocktail of ezymes in their PreCR kit, which repairs damaged DNA templates. It is designed to work on: modified bases, nicks and gaps, and blocked 3' ends. They had a poster at AGBT demonstrating the utility of the method, showing increased library yields and success rates with no increase in bias in seqeuncing data.

Illumina also have an FFPE restoration kit; restoration is achieved through treatment with DNA polymerase, DNA repair enzyme, ligase, and modified Infinium WGA reaction, see here for more details.

These cocktails can almost certainly be added to: MUTYH works to fix 8-oxo-G damage, CEL1 is used in TILLING analysis to create strand-breaks in mismatched templates and could be included, lots of other DNA repair enzymes could be added to a mix to remove nearly all compromised bases. It may be possible to go a step further and fix compromised bases rather than just neutralise their effect.

Whatever the case it looks very much like FFPE samples are going to be processed in a more routine manner very soon.


Monday, 18 August 2014

$1000 genomes = 1000x coverage for just £20,000

It strikes me that if you can now sequence a genome for $1000, then you could buy 1000x coverage for not much more than a 30x genome cost a couple of years ago! Using a PCR-free approach I can imagine that this would be the most sensitive tool to determine tumour, or population, heterogeneity. I’m sure that sampling statistics might limit the ability to detect low-prevalence alleles but I’m amazed by the possibility none-the-less.
  • 1 X-Ten run costs $1,000
  • 1000x requires 33 X-Ten runs (30x each)
  • $33,000 = £20625
If you’re running a ridiculously high Human genome project on X-Ten do let me know!

Thursday, 14 August 2014

How many MiSeq owners are using the bleach protocol to minimise carryover?

A comment popped up on a post I'd written in April last year "MiSeq (and 2500) owners better read this and beware"that made me think I'd ask readers the question in the title: "how many of you are using the MiSeq bleach wash protocol?

The carryover issue led to a small residue of the last library to be run being sequenced in the subsequent run. This caused a potential problem to MiSeq users, particularly those interested in low frequency mutations. My post was prompted after some discussions with other MiSeq owners and a thread on SEQanswers, which Illumina posted to describing their experiences with reducing this carryover, and that it was seen typically at 0.1%.

The comment on my post was about a more aggressive bleach protocol which reduced carryover to almost undetectable levels (thanks Illumina), but that appears to have not been communicated to all users. It was impossible for me to find on the Illumina website but it's not the most easy to navigate site in the world so I thought I'd put the document up here for you to see (click the image for the TechNote).

https://drive.google.com/file/d/0B383TG7oJh2CTWp6MzZkMHBDd28/edit?usp=sharing

You need to request this through your techsupport team as it needs a new recipie on your MiSeq. And you really must follow the instructions to the letter, too much bleach and you'll probably kill your MiSeq!

Ilumina demonstrated that this protocol can reduce carry-over to less than 0.001% or one read per 100,000. We've been using this as the default wash for many months and reports of carry-over are nearly non-existent.

Sunday, 27 July 2014

When will BGI stop using Illumina sequencers

With the BGI aiming to get their own diagnostic sequencing tests on the market, and the purchase and development of Complete Genomics technology - Omega, a question that could be asked is whether BGI will ever stop using Illumina technology?

BGI are still the largest install of HiSeq's but they have not purchased an X-Ten and it's not clear if they've switched over to v4 chemistry on the updated 2500. The cost of upgrades or replacement on a scale on 128 machines would be high, but BGI have deep pockets. So is this the start of the end for Illumina in China?

If BGI stops using Illumina will Illumina notice? I'm sure they will and the markets will read lots into any announcement, but in the long run it's difficult to see China without an Ilumina presence. The Chinese science community is booming, their research spend is second only to the US and is likely to climb much more quickly, and they have a massive health-care market that NGS can make a big impact on.

Once we hear what BGI can do with the CGI technology (exomes for instance) we might find out Illumina has a strong competitor and with LifeTech/Thermo effectively putting Ion Torrent on-hold competition in the NGS market is something the whole community, including Illumina needs.

PS: This is my last post for a couple of weeks while I'm off on holiday in Spain. Hasta luego!

Monday, 21 July 2014

1st Altmetric conference - Sept 25/26th in London

I've been a user of Altmetric for a while now and very much like what they are doing with article metrics. I'm sure many Core Genomics readers will also have seen the Altmetric badge on their own papers. Now Altmetric are hosting their first conference.


The meeting aims to demonstrate how users are integrating Altmetric tools into their processes. Hopefully they'll cover lots of interesting topics and spend some time talking about how the community can keep tools like Altmetric from becoming devalued by gaming.

Might see you there...

Thursday, 17 July 2014

A hint at the genomes impact on our social lives

GWAS is still in the news and still finding hits, the number of GWAS hits has increased rapidly since the first publication for AMD in 2005. Watch the movie to see the last decade of work!




A recent paper in PNAS seems to have got people talking: in Friendship and natural selection Nicholas Christakis and James Fowler describe their analysis of the Framingham Heart Study (FHS) data; specifically the data of people recorded as friends by participants. The FHS recorded lots of information about relatives (parents, spouses, siblings, children), but also asked participants “please tell us the name of a close friend". Some of those friends were also participants and it is this data the paper used to determine a kinship coefficient, higher values indicate that two individuals share a greater number of genotypes (homophily.

The study has generated a lot of interest and news (GenomeWeb, BBC, Altmetric) but also some negative comments, mainly about how difficult this is to prove in a study where you cannot rule out genetic relationships individuals themselves don't know exist (i.e. I don't know who my third cousins are and might make friends by chance).

The data in supplemental files from PNAS paper show Manhattan plot (top) for the identified loci, its not as stunning an example as you'd see in other fields. Compare it to a well characterised GWAS hit from a replicated study in Ovarian Cancer (bottom).








Monday, 14 July 2014

Sequencing exomes: what sort of read to use?

What's the right way to sequence an exome? We've been looking at Illumina's v4 chemistry for HiSeq 2500 and wondering whether we should jump to PE125bp or not, or should we try to reconfigure our exome capture for shorter or longer fragments.

Exome-seq: Exomes have been a big hit, there are currently over 3000 publications in PubMed with the search term "exome". Given that the first in-solution exome paper was only published in 2009 that's pretty amazing, but then again the exome is an amazing research tool.

Note to readers: This post started out as a writing down my thoughts about whether we should move to longer reads for exomes. But it has become a bit more rambling as I started to find out I need to do some mroe digging. I may well come back to this post with an update or version two...

There are many ways to prepare an exome for sequencing and in my lab we're currently using Illumina's rapid exome kit. We're also about to compare this to Agilent's new SureSelect QXT kit which is a direct competitor to Illumina's Nextera-based offering. But we've never tried Nimblegen or AmpliSeq, however this post is more about how to sequence the exome than prepare it so enough of kit comparisons.

The standard exome: Their are two things you need to consider when sequencing exomes: read depth and read-length. I'm not going to worry about depth in this post, and instead I'm going to focus on read-length. Today most labs appear to be running exomes at PE75bp, a standard which I am not sure has ever been agreed by anyone, but it has been accepted as being good-enough for most projects (Illumina recommend PE75-100). I know of some groups that moved over to PE100 to simplify lab logistics as much as anything else, but I am not clear that there are significant benefits to increasing length so we've stuck at PE75 for the time being.

Are longer reads better: With the advent of v4 chemistry on HiSeq 2500 we should be able to generate high-quality paired-end 125bp reads, albeit with a slightly higher error rate at the end of the read. At first glance this additional data seems too good to ignore, especially when Illumina do not sell a 150cycle SBS kit, and 3x50cycle SBS would not be that much cheaper (and more hassle for my lab staff!) By my reckoning PE75 costs £900 per lane whilst PE125 is £1200, or £300 for an extra 100bp of coverage. So if cost does not prevent us using PE125, should we simply switch?

Insert size vs read-length: As you can see below the average distribution of exome fragments size spans the read-length of the sequencer. The solid black line indicates 150bp (PE75): everything to the left of this will be fragments sequenced with an overlapping reads (opes), whilst everything to the right is sequenced with non-overlapping reads (nopes). As read-length increases the percentage of fragments sequenced with an overlap also increases, at PE100 (dashed line)  this is over 50% of reads, and at PE125 (dotted line) it's about 75% of all fragments. An overlapping read creates some issues as the two reads are not independent, tools need to take the overlap into account when calculating on-target coverage, etc; but it also offers the opportunity to increase variant calling quality by increasing Q-scores in the overlap region.



Exome libraries may not be the best size for sequencing: If a non-overlapping read is the best kind to generate then we may need to reconfigure library prep in the light of v4 chemistry. An interesting comparison can be made to the Agilent Bioanalyser trace below the computed insert size distribution. If you overlay and rescale the two images, then the Agilent trace appears to be peak-shifted to larger fragments, and the right-hand fragment distribution is much broader. This appears to demonstrate the preference of clustering:sequencing for shorter fragments.

Exome libraries are probably the best size for capturing exons: The average exon length in the Human genome is 170 bp with 80–85% exons less than 200bp (Zhu et al & Sakharkar et al) so the 185bp average fragment length seems almost ideal.

http://iospress.metapress.com/content/x40k2kuge74488kf/fulltext.pdf
Table reproduced from Shkharkar et al 2004

So what's the sweetspot for Exome capture and sequencing: The simple answer is I don't know, and several factors are likely to affect this. As we increase read-length we'll get more fragments with overlapping reads that could be wasteful; the same happens if we decrease fragment size so longer reads give us more and more overlap with higher quality. But unless there are tools to make use of this the data are redundant. So fragments should not be longer than reads.

But fragments are captured by probes of 95bp so we should probably not make fragments shorter than probes.

Exome capture kits contain blocking oligos to prevent adapter:adapter hybridisation and off-target pull-down. As fragment length increases then the amount of near-target sequence captured may increase meaning we should not make fragments too long. A long fragment risks too much off-target enrichment by the secondary capture of off-target fragments.

Lastly (for now) we'd like to be able to use independent fragments for our analysis so read-pairs might be better replaced with longer single-reads, but twice as many. So perhaps the answer is probes that efficiently capture exons with little or no fragment:fragment hybridisation, coupled to single-end 185bp sequencing with low error-rate across the reads.

Monday, 7 July 2014

Anatomy of a NextSeq flowcell


Personally I'm thinking that the aluminium plate might make a pretty nifty bottle opener!

Thursday, 3 July 2014

How to find the best papers to read is tough

We've all been there: PubMed returns over 2500 RNA-seq papers, and there's still 800 left when you only search the title! How do you find the best papers to read? PubMed can help a little more with your quest to find out more about RNA-seq as there are just 19 reviews, but it's often primary papers you need to dig into to truly understand what's going on in a field. There are other ways to find out what's a hot paper and I've just started using a relatively new one: the Altmetric explorer.

Before I go any further I will say this is a demo account (thanks Altmetric) and their pricing plans are squarely aimed at institutions. Hopefully they'll find a way to make tools for individuals with perhaps more limited search functionality.

What does Altmetric Explorer do: The search tool allows you to filter the vast amount of data Altmetric has collected, you can even enter a PubMed search directly. The first thing I did was to look at was my own publication record and see who's talking about the papers I've co-authored, turns out it is often just me (as far as Altmetric is concerned)!

I'd originally been in touch with the Altmetric team about using data from ORCID (I wrote about this last year) and seeing if it were possible to pull out more complex relationships between authors. The aim was to make creation of something like the Circos plot below easy to do for any group of individuals ro even institutes. I'm still a long way from doing this but if anyone can offer some help that would be great!



The searches I presented below simply used a PubMed search and list papers in the order of most activity, as recorded by Altmetric. You can filter on lots of other metrics including; keyword, date, journal, etc. Take a look and get in tocuh with the Altmetric team if you'd like to do more.
RNA-seq Altmetric activity:


ChIP-seq Altmetric activity:

My Altmetric activity:
PubMed = Hadfield J[author]

Tuesday, 1 July 2014

Single cell extravaganza

AGBT had lots of presentations on both clinical and single-cell work last February and on Wednesday, February 12th Aviv Regev, of the Broad Institute described her groups experience with the Fluidigm C1 system, and their early work in understanding cell-to-cell communication in the immune system. The results were published in Nature a couple of days ago. A more exciting story for me scientifically was published by Aviv Regev and Bradley Bernstein in Science awhere they describe the use of single-cell sequencing to understand tumour heterogeneity.

The big headlines for me were that just 1M RNA-seq reads were enough to get high-quality gene expression estimates, and that single cells are great: but we’re going to need lots of them to delve deeply into biological systems. Both papers showed a massive loss of data at QC stages, around 1/3rd of cells were lost and only 30% of reads mapped to the transcriptome. Hopefuly both of these are things we can improve in the next few years to make single-cell mRNA-seq even more powerful.