• Login
  • |
  • Register

NGS Leaders Blog

Guest Post: Perspectives on Sequencing Cancer - Part 1

 Permanent link

 Editor’s Note: In this 2-part guest blog series, Keith Robison, PhD (Lead Senior Scientist, Informatics, Infinity Pharmaceuticals) gives a reacap of the session he chaired at last month's Applying Next-Gen Sequencing conference. Keith also maintains the popular Omics! Omics! blog. 

October 11, 2011 

Keith Robison : (Part 1 of 2) I had an opportunity recently to chair a session at CHI’s Applying Next-Generation Sequencing meeting in Providence RI which was a nice microcosm of cancer genomics, covering in three talks one important corner of that vast space. I’ll also steal a bit from a talk in the joint session with Next-Generation Data Management which followed, as if I had been an omniscient and all-powerful chair, I would have yanked it into my session.

Whole Genome SequencingGiulia Fabbri of Columbia University led off by describing a successful scan of chronic lymphocytic leukemia (CLL) patient samples to identify recurrent causative mutations.  Using exome sequencing on five CLL samples, a set of 48 candidate mutations were identified, which were then validated on a larger panel.  The big fish netted by this campaign are activating mutations in NOTCH1.  The most common type of NOTCH1 mutation seen is interesting, as they are frameshifts near the C-terminus, which is the location of a destabilizing domain that targets NOTCH1 protein to the proteasome.  Hence, the mutations remove a negative influence, enabling a higher degree of oncogenic signaling. Other NOTCH1 mutations seen in CLL apparently have a similar effect, as Fabbri reported that so far there seems to be no difference in the clinical course of the truncating mutations versus other mutations.  Any NOTCH1 mutation appears to be a serious negative prognostic for the patient.  One striking bit of luck is that the frequency of NOTCH1 in CLL is about 1%, which means a sample of 5 samples could well have not included this mutation.  Further studies will use larger cohorts to scan for other mutations, as well as cohorts focused on more aggressive forms of the disease.

Mike Makrigiorgos of the Dana Farber followed with a discussion of his COLD-PCR technique and its extension to second-generation sample preparation. COLD-PCR is a twist on standard PCR in which each cycle contains two extra steps which result in perfectly matched sequences staying double-stranded and encouraging mutation-bearing fragments to be single-stranded. This is much like the reverse of consensus methods in synthetic biology for reducing errors; whereas in gene synthesis the common, correct matches are desired, in hunting rare mutations from clinical cancer samples there is a desire to prejudice the amplification against the wild-type. Makrigiorgos showed data on the Illumina platform in which samples with a minor allele frequency of a few percent were buried in noise with standard PCR, but became very strong signals with COLD-PCR sample preparation. Many of the questions after the talk were aimed at better understanding the degree to which each amplicon must have a custom COLD-PCR protocol; the early experiments benefitted from a thermocycler capable of running each well on a different program. But, Makrigiorgos gave brief sketches of extending COLD-PCR to emulsion PCR and hinted at changes which will allow each sample to be run in a common thermocycling scheme.

The final talk of the session was given by me describing my work recently at Infinity Pharmaceuticals on using PCR and the Ion Torrent PGM platform to search for mutations.  I described some of the successes and challenges I’ve discovered, some of which will be familiar to readers of my Omics! Omics! Blog. An attraction of the platform is the ability to quickly design and validate assays, We’ve had success detecting mutations spiked into samples at less than 1% frequency, can detect many alleles simultaneously and generally observed good linearity between the predicted number of sequencing reads and the observed number of reads. The sensitivity we’ve observed on PGM is much better than Makrigiorgos was reporting for Illumina, though other groups have achieved very high sensitivity on the Illumina as well with standard PCR. On the other hand, we’ve also had great variability in the number of reads coming off the instrument and sometimes the reads decay badly before they get to the region bearing the interesting alleles. Attempts to barcode samples were successful, but different barcodes sometimes amplified at wildly different frequencies and a first attempt to correct this did not yield much improvement.  Primer designs for the same amplicon also behaved wildly different, suggesting that empirical validation will be necessary. We also observed some unexpected KRAS alleles in a mixed cell line experiment; some of these appear to be noise that may result in different sensitivities for different alleles, whereas another unexpected allele appears to be the result of an unusual compound heterozygote in KRAS.

I also displayed one case of trying to find a 5-bp deletion with the PGM which was complicated by the homopolymer-calling errors on that platform. While the most commonly called allele was the correct one, several other deletion alleles were called in the neighborhood which could be rationalized as results of homopolymer miscalls. Also, inconsistency in the alignment of some nucleotides contributed to confusion in this case.  For example, the COSMIC-annotated deletion and the most common call were not identical, but given the sequence were equivalent. This points the need for better indel-calling software which can collapse such results.  One general advantage of the amplicon approach is it is not wedded to a platform; once MiSeq is available it should be straightforward to convert these amplicons to ones suitable for that instrument.

To summarize so far, Fabbri led us off with finding mutations by exome sequencing and then being faced with the challenge of scanning a larger sample cohort to assess frequency and relation to clinical characteristics.  Performing that scanning, and later routine testing in the clinic, is most commonly via PCR. Makrigiorgos proposed COLD-PCR as an approach to enrich that signal, since tumor specimens are always heterogeneous. I described a more head-on assault using Ion Torrent, though COLD-PCR might be an interesting twist, except when quantitative results are required (any allele-biased enrichment, by definition, distorts the allele frequency).

Read Part 2 

I-Study: Genomic Interpretation - Who Will Pay?
During this webinar, members of the study review team present preliminary findings of the I-Study, conducted at the Harvard Medical School's 2011 Personalized Medicine Conference.
Twitter Feed
Privacy Policy|Terms of Use