Next-generation sequencing of formalin-fixed, paraffin-embedded tumor biopsies: navigating the perils of old and new technology to advance cancer diagnosis

Formalin-fixed, paraffin-embedded (FFPE) tumor biopsies are a primary pathological specimen for oncology diagnostics. Hundreds of millions of such biopsies exist, and the method for creating them persists due to the ease and stability of sample fixation and the clarity of morphological features of disease that is revealed. These advantages, however, come at a cost that could not have been anticipated when formalin fixation was first described by Blum 120 years ago [1]. Formaldehyde readily crosslinks both proteins and nucleic acids, and it ignites a series of chemical modifications that compromise enzymatic treatments in downstream assays. The consequences can be dire for nucleic acid testing. Every step along the path from sample to signal – extraction, amplification, detection and interpretation – is influenced by the FFPE process. Most concerning is the lack of standardization for preparing FFPE blocks, leading to highly variable results [2]. Although a number of DNA- and RNA-friendly preservation methods have been proposed [3,4], the entrenched use of FFPE procedures, coupled with the vast repository of archived specimens, has ensured its place in molecular pathology.

By comparison, next-generation sequencing (NGS) technology emerged less than a decade ago. Yet, its trajectory has been meteoric: the price of sequencing per raw-base pair decreased by approximately 100,000-fold from 2000 to 2010 [5], and much of this gain was realized through massively parallel methods that have rapidly displaced Sanger sequencing. Like PCR, it only took about 5 years for NGS to find its way into the clinic. Now that it has, there can be little doubt of its revolutionary nature. With capabilities to accurately detect point mutations, structural variation, copy number changes, methylation status and gene expression, NGS is a multifaceted and versatile tool for molecular investigations.

Opportunities for improved cancer diagnosis

The promise of NGS to improve cancer diagnosis arguably exceeds any other nucleic acid-based technology today. First, NGS offers a unifying approach for molecular testing. With the advent of rapid benchtop sequencers such as the MiSeq®, the Ion Personal Genome Machine® (PGM) and the 454 GS Junior, as well as hybrid workhorses such as the HiSeq® 2500, DNA- and RNA-based analyses can be performed on a single instrument with turnaround times that are compatible with oncology testing needs. In fact, it was recently estimated that 85% of the more than 600 solid tumor clinical tests in use today are ‘immediately replaceable’ by NGS methods [6]. Second, the ability to retrieve profiles across different classes of cancer biomarkers provides an integrated approach that can fold multiple genetic aberrations into a handful of corrupt signaling pathways and inform the selection of targeted therapies [7]. Third, the combination of NGS and increasingly sophisticated bioinformatics reveal the dimensionality of cancer-associated cellular changes and expand treatment strategies from one mutation/one drug to combination therapies that can attack multiple nodes of dysregulated pathways and unveil incipient ‘driver’ mutations [8]. This information can guide more effective treatments based on a pathway rather than an anatomical view of cancer and begin to shift the tide of therapy from reactive to preventative medicine.

Today’s problems

The potential of NGS to reshape cancer diagnosis and treatment is not without its challenges. Although topics such as regulatory oversight and insurance reimbursement are vital to the broader conversation, our focus in this commentary is the technical barriers. We address three hurdles: the biopsy specimen, the NGS platform and associated reagents, and the bioinformatic data analysis and reporting.

FFPE tumors are heterogeneous, reflecting diseased cells and molecular subclones of variable composition, and the isolated DNA is often refractory to amplification that can limit the template diversity in NGS library preparations. More insidious is the observation that low-quality FFPE can spark both false negatives and positives [9,10]. For example, we recently identified a false-positive rate of up to 89% when the lowest quality FFPE DNA is processed using a popular commercial enrichment method [10].

The NGS instruments and reagent ecosystems are also problematic. Although benchtop sequencers such as the PGM have realized exponential gains in performance within the last 2 years, continual protocol and reagent changes have created an unstable architecture that is an anathema to clinical testing. As examples, there have been 14 updates to the PGM Torrent Suite since 2011, and MiSeq software has undergone major revisions approximately every 2 months since its introduction. In addition, the intricacy of NGS workflows, with requirements for up to hundreds of raw materials and auxiliary high complexity instrumentation (such as the RainDance ThunderStorm^™ and Fluidigm Access Array^™), compounds the difficulty in establishing a fixed assay configuration with unwavering performance.

Last, the analysis challenges are daunting. Publically available tools have been honed for standard genotyping applications [11], but these methods have not proven extensible to calling low abundance variants – especially small insertions/deletions (indels) – in FFPE tumor DNA. Consequently, algorithms have been heavily modified from existing code or built from scratch to reliably call mutations in low allele fractions [12]. NGS instrument manufacturers have begun to offer clinically focused analysis and annotation packages, either local or in the cloud, yet the implementation of these offerings in clinical laboratories remains uncertain. Indeed, recent publications focused on clinical NGS of FFPE tumor DNA have emphasized in-house analysis pipelines, visualization tools or annotations that query databases such as COSMIC and dbSNP to contextualize identified variants [13–15]. The creation of a full-fledged, in-house bioinformatic solution is impractical for many laboratories. In addition, tiered reporting that prioritizes ‘clinically actionable’ variants – at best, a slippery term – has emerged as a backstop for an incomplete knowledge base of disease-relevant genomic changes [16].

Tomorrow’s solutions

Although these challenges are significant, there is reason for optimism. We are now at an inflection point for NGS-based cancer diagnostics where technological capability can be contained within methodological proficiency to serve the specific needs of the clinical laboratory. Consistent with this interpretation, more than a half dozen clinical NGS guidelines have been published across international professional organizations and diagnostic workgroups since 2012 [17–20]. Achieving accurate NGS in clinical settings requires insight into where the process can go wrong and how to build robust and reliable procedures around that knowledge.

The biopsy specimen

Any NGS diagnostic assay can fail even before it begins. Although FFPE biopsies can be prepared using standardized procedures that lessen the insult of chemical modification to clinical DNA and RNA, these procedures are surprisingly uncommon outside of carefully defined translational medicine projects. In fact, only about 10% of FFPE DNA templates from representative specimens are amplifiable, even when relatively short targets are selected for enrichment [9,10].

How can this problem be addressed? First, the nucleic acid extraction method should be optimized and rigorously scrutinized. Bulk DNA yield should not be a singular pursuit. In fact, reduced total DNA yields may be preferable if the recovery of ‘functional’ DNA is improved. Second, and further to this point, the objective of DNA quantification should be to enumerate templates that can be readily manipulated in downstream enrichment procedures. Inputs should be based on the FFPE DNA that will actually be interrogated by NGS rather than the larger fraction that would not. For this reason, we recently described a preanalytical DNA copy number assay (QFI^™-PCR) that predicts NGS metrics and guides FFPE DNA inputs to help assure reliable interpretations [10]. Last, there is a compelling need for inexhaustible standards that mirror both the quality and mutational heterogeneity of FFPE clinical specimens and can be used for test validation and proficiency testing [17,21]. As a first step, National Institute of Standards and Technology (NIST) reference materials, such as NA12878, are planned for an early 2014 release, and commercial controls with FFPE-like properties are emerging.

The NGS platforms & reagents

Although shifting sands of reagent formulations, instrument upgrades and protocol updates have been the norm, manufacturers are now beginning to ‘lock down’ diagnostically facing NGS systems. Illumina, for example, recently announced a CE mark on their MiSeqDx^™ Cystic Fibrosis System. Furthermore, both Illumina and Life Technologies are actively pursuing FDA-cleared NGS instruments and assays, which will require rigorous design control. NGS diagnostic tests will certainly benefit from these durable system configurations, particularly if they integrate FFPE-relevant reference materials, predictive sample qualification assays (such as QFI^™-PCR) and in-process QC assays and metrics that safeguard reliable sequencing results.

The data analysis & reporting

The growing clinical value of low-level somatic mutations in FFPE tumors places heavy demands on NGS variant calling algorithms. Several groups have described approaches developed specifically for this purpose [13,22,23], including methods that accommodate FFPE DNA ‘noise’ such as an elevation in transition mutations [13]. Additional improvements might include the integration of sample-level information, including preanalytical QC measures, to inform adaptive models for bioinformatic analysis. Furthermore, methods to assess clinically important copy number changes and indels continue to develop. But the single most efficient strategy for enabling diagnostically and operationally robust NGS processes is simply to target content that maximizes test value. For example, targeting oncogene mutational ‘hotspots’ that are recommended by practice guidelines will not only produce a high diagnostic yield, but also reduce the analysis and reporting burden associated with lower impact regions. Given the criticality of secondary confirmation from the primary NGS result [17,21], and the complexity of variants of unknown significance that are collateral damage from casting a broad net [21], thoughtfully selected and focused content offers a compelling solution for the first generation of NGS diagnostic assays. We favor deep sequencing (e.g., >1000× read depth) of focused gene panels with high analytical sensitivity for established, actionable mutations over relatively shallow exome scale or whole-genome sequencing that offers diminishing returns beyond the most diagnostically informative gene regions.

Summary & outlook

The key advances needed to realize the promise of NGS in cancer diagnostics are within reach:

• • Focused gene-enrichment assays that provide accurate and sensitive detection of clinically actionable sequences;

• • Preanalytical assays and metrics that quantify sample-specific molecular characteristics;

• • Validated reference materials that mimic real-world FFPE specimens;

• • In-process QC assays and predictive metrics;

• • Stabilized workflows, reagents and platforms;

• • Optimized, traceable and validated bioinformatic analyses and reporting across multiple categories of somatic genetic variation.

Targeted NGS assays are poised to revolutionize cancer diagnostics and drug development in ways that we can anticipate and others we cannot. The full value of these innovations will require integration across multiple domains of technology and methods old and new. The payoff, however, will be yet another critical step forward in individualizing both diagnosis and treatment.

Financial & competing interests disclosure

Some of the research described in this editorial was supported by a grant from the Cancer Prevention and Research Institute of Texas (CP120017, PI: GJL). The author is Vice President of Research and Technology Development and is both an employee of and holds stock options in Asuragen, Inc. The author has no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.

No writing assistance was utilized in the production of this manuscript.