990 volume 31 number 11 november 2013 nature biotechnology

Jason Y. Park is at the Department of Pathology, University of Texas Southwestern Medical Center and Childrens Medical Center, Dallas, Texas, USA, and the Eugene McDermott Center for Human Growth and Development, University of Texas Southwestern Medical Center, Dallas, Texas, USA; Larry J. Kricka is at the Department of Pathology and Laboratory Medicine, University of Pennsylvania Medical Center, Philadelphia, Pennsylvania, USA; and Paolo Fortina is at the Cancer Genomics Laboratory, Kimmel Cancer Center, Department of Cancer Biology, Thomas Jefferson University, Jefferson Medical College, Philadelphia, Pennsylvania, USA. e-mail: jaspar@childrens.com

The authors describe experimental protocols for handling surgically collected solid tumor specimens that have been fixed in formalin and embedded in paraffin. A key step in the assay is the use of custom-synthesized DNA capture probes to enrich a sample for specific regions of the genome. This allows the authors to sequence their target regions to >500 average depth of coverage, thereby providing enough informa-tion to confidently call mutations. In total, the capture probes target 4,557 exons from 287 cancer-related genes, 47 introns from 19 genes frequently rearranged in cancer and 3,549 single-nucleotide polymorphisms.

Frampton et al.1 also improve clinical NGS protocols by providing a robust analytical vali-dation strategy (Fig. 1). The authors use 53 cell lines to create three types of reference materi-als designed to assess the performance of their method for detecting base substitutions, indels or copy-number variations. The cell lines are both of non-tumor origin (HapMap) and cancer derived. They are pooled into various admixtures to examine detection both of genetic alterations and genetic alterations diluted into normal DNA. Sensitive detection of alterations in diluted material simulates a surgical specimen that may have rare tumor cells in a non-tumor tissue background.

Importantly, the validation approach is con-sistent with recommendations outlined by the Next-generation Sequencing: Standardization of Clinical Testing (Nex-StoCT) workgroup3 convened by the US Centers for Disease Control and Prevention. These recommenda-tions and others4,5 are intended to ensure that diagnostic tests based on NGS meet clinical laboratory regulatory requirements.

Identifying mutations is only the first step. All clinical NGS tests must also demonstrate usefulness in improving patient outcomes. The clinical significance of novel mutations is difficult to determine, even when they occur in well-known oncogenic pathways. Frampton et al.1 say that their test reveals

in leukemia) or to target therapy (e.g., HER2 amplification in breast carcinomas to deter-mine treatment with Herceptin (trastuzumab, Roche/Genentech)). But it is unusual to test for multiple mutations that have not been previously described in a specific tumor type. An oncologist in the United States may decide that a patient will benefit from a US Food and Drug Administration (FDA)-approved drug, such as trastuzumab, on a tumor type or mutation type not approved by the FDA. Such practice is referred to as off-label use. Of all chemotherapeutic prescriptions, 3347% are off-label2.

If a patient has not responded to conventional therapies, there may be a rationale to perform NGS testing in order to identify potential drug targets and then find an appropriate clinical trial or off-label drug. However, routine imple-mentation of clinical NGS in oncology is still in its infancy. Current testing methods for cancer mutations include point mutation assays, exon sequencing, fluorescence in situ hybridization and karyotyping. Such tests are available for only a handful of genetic alterations that can inform therapy, and the diagnostic yield of any single test is low. For several reasons, including cost, a shotgun approach that combines several meth-ods is impractical. The potential advantages of NGS in achieving high-throughput testing are clear. Indeed, at a few centers, NGS has been applied clinically for several years to detect base substitutions and small deletions in large panels of genes (tens to thousands), whole exomes and whole genomes. The study of Frampton et al.1 is important because it shows that NGS can pro-vide superior analytic performance for detect-ing mutations in oncogenic pathways. Such an approach may also be lower in cost than con-ducting multiple non-NGS tests.

Frampton et al.1 improve on previously reported NGS cancer testing methods by increas-ing the number and types of genetic alterations detected as well as demonstrating an extensive validation with numerous reference materials.

As next-generation sequencing (NGS) of tumor cells becomes more sophisticated, it is likely to inform all aspects of cancer management, from diagnostic testing to treatment and drug dis-covery. In this issue, Frampton et al.1 describe a comprehensive NGS assay applicable to clinical samples that identifies single-base substitutions, copy-number variations and focal amplifica-tions in 287 cancer-related genes, fusion events involving 19 frequently rearranged genes and 3,549 single-nucleotide polymorphisms in other locations throughout the genomeall within a single sequencing run. Validating the analytic performance of such a test is chal-lenging because it assays so many nucleotide positions in the genome (~1.5 Mb in total). The authors therefore use 53 cell lines to create reference materials for assessing the sensitiv-ity and specificity of variant detection. Finally, they apply their test to >2,000 clinical cases. The overall approach of the studyincluding assay design, creation of reference materials and validationserves as a model for the develop-ment of future clinical NGS tests.

In current clinical practice, mutation analy-sis of cancer samples is performed to estab-lish diagnoses (e.g., specific translocations

Next-generation sequencing in the clinicJason Y Park, Larry J Kricka & Paolo Fortina

Pools of cell lines carrying a variety of known mutations are used to validate the performance of a cancer diagnostic test based on next-generation sequencing.

n e w s a n d v i e w snp

g

2013

Nat

ure A

mer

ica,

Inc.

All

right

s re

serv

ed.

nature biotechnology volume 31 number 11 november 2013 991

clinically actionable mutations in 76% of tested tumors. They define a clinically actionable muta-tion as one that has a clinically available targeted treatment option or a mechanism-driven clinical trial. But just having a drug targeted to a spe-cific mutation does not guarantee its efficacy. For example, included in the list of identified altera-tions are point mutations and indels in ERBB2

(HER2). Trastuzumab is labeled by the FDA for use in breast or metastatic gastric or metastatic gastroesophageal junction adenocarcinomas, only when HER2 is overexpressed; the label does not mention point mutations or indels. The authors note that robust clinical evidence for tar-geting these alterations must still be generated, and cite a recent in vitro study of targeted therapy

for nonamplification HER2 mutations in breast cancer6. Although a single in vitro study may guide potential off-label use, it is not sufficient evidence for a successful FDA review of a diag-nostic test kit intended to guide chemotherapy.

There are scenarios when off-label use of chemotherapeutics is considered best prac-tice for patient care. However, off-label use has well-known risks to patients, including the unknown risk-benefit ratio of administer-ing compounds that frequently have toxic side effects7. In general, the biology of most cancers is not understood well enough to confidently transfer a chemotherapeutic from one clini-cal setting to another in the absence of clinical trials. For example, the mutational status of KRAS in metastatic colon cancer predicts the efficacy of anti-EGFR therapeutics. A logical hypothesis would be that KRAS mutational sta-tus may also predict the efficacy of anti-EGFR therapeutics in earlier-stage colon cancer. However, a recent clinical trial of earlier-stage colon cancer found that KRAS mutational sta-tus does not predict the efficacy of adding anti-EGFR therapy8. This is one of many examples of how chemotherapeutic selection is not a simple logic equation but a task best guided by clinical trials.

The challenge of properly prescribing che-motherapeutics off-label will remain even as the use of NGS in clinical settings becomes more widespread9. Multiple oncology trials to evaluate NGS are currently under way10. In addition to examining the clinical benefits of sequencing panels of genes, similar to the panel used by Frampton et al.1, some of these studies are also exploring mRNA transcript profiling and sequencing of whole exomes and whole genomes. To maximize the utility of such studies, we would argue that patients who are treated with drugs for off-label indi-cations based on the findings of NGS test-ing should be encouraged to participate in registries that document their outcomes. Frampton et al.1 have established such a regis-try for patients (http://clinicaltrials.gov/show/NCT01851213).

Regardless of whether clinical NGS is suc-cessful in identifying genetic alterations that can be treated with existing drugs, Frampton et al.1 report a notably high frequency of muta-tions in oncogenic pathways (on average 1.57 mutations per case, with 1,579 distinct altera-tions in 2,112 cases). If these mutations are all important in driving tumorigenesis, the high number will represent a challenge to current drug development models that target a single mutation with a single chemotherapeutic.

COMPETING FINANCIAL INTERESTSThe authors declare no competing financial interests.

Copy-number variations

Base substitutions Insertions & deletions

28 discrete tumor cell lines41 pools from tumor cell lines MAF tested at 20%

7 discrete tumor cell lines42 pools from combinations of tumor cell lines and normal cell linesEach tumor cell line tested from 0 to 80% dilution with normal cell lines

Variant type

Reference materialused for validation

Solution-based target capturewith 23,685 individually synthesized,

biotinylated, 120-bp DNA capture probes

Paired-end sequencing

Pathology review DNA extraction

Library preparation

Map reads to human genome

Variant calling

Tumor samplein paraffin block

20 discrete HapMap cell lines2 pools of 10 cell lines eachMAF tested from

992 volume 31 number 11 november 2013 nature biotechnology

6. Bose, R. et al. Cancer Discov. 3, 224237 (2013).

7. Krzyzanowska, M.K. J. Clin. Oncol. 31, 11251127 (2013).

8. Alberts, s.R. et al. J. Am. Med. Assoc. 307, 13831393 (2012).

9. Garber, K. J. Natl. Cancer Inst. 103, 8486 (2011).

10. simon, R. & Roychowdhury, s. Nat. Rev. Drug Discov. 12, 358369 (2013).

1. Frampton, G.M. et al. Nat. Biotechnol. 31, 10231031 (2013).

2. Conti, R.M. et al. J. Clin. Oncol. 31, 11341139 (2013).3. Gargis, A.s. et al. Nat. Biotechnol. 30, 10331036 (2012).4. Rehm, H.L. et al. Genet. Med. 15, 733747 (2013).5. CAP Laboratory Accreditation Program. Molecular

Pathology Checklist http://www.cap.org/apps/docs/laboratory_accreditation/checklists/new/molecular_ pathology_checklist.pdf (College of American Pathologists, 2013).

Two views on light sheetsCarl G ebeling & erik M Jorgensen

A dual-view light-sheet microscope combines isotropic spatial resolution with high speed and minimal phototoxicity.

Carl G. Ebeling is at the Department of Biology, Howard Hughes Medical Institute, University of Utah, Salt Lake City, Utah, USA, and the Department of Physics, University of Utah, Salt Lake City, Utah, USA, and Erik M. Jorgensen is at the Department of Biology, Howard Hughes Medical Institute, University of Utah, Salt Lake City, Utah, USA. e-mail: jorgensen@biology.utah.edu

excitation objectives illuminate the sample from opposite sides, and two detection objec-tives orthogonal to the excitation plane col-lect the fluorescence emission. However, the downside to this technique is increased light exposure and phototoxicity.

Rotating the sample has its own drawbacks, such as prolonging acquisition time, but the advantages of imaging the same volume from multiple vantage points is an increase in reso-lution. With the proper post-analysis software, the best spatial information from each image is preserved, whereby the poor axial resolution in one orientation is replaced with the higher lateral resolution from the second orientation. In short, multiview acquisition begins to solve the axial resolution problem. Previous imple-mentations of multiple viewpoint acquisi-tion systems35 require anywhere from 4 to 36 different acquisitions of the same volume, unfortunately offsetting the inherent advan-tages of the low photo-dosage of conventional light-sheet microscopy. Moreover, multiview methods have not yet demonstrated true iso-tropic resolution (i.e., the same resolution in all spatial directions).

Wu et al.1 offer a solution to many of the issues plaguing light-sheet microscopy. They show that their dual-view, inverted SPIM (diSPIM) setup achieves isotropic resolution with minimal phototoxicity while keeping the sample immobile and mounted by conven-tional methods. The enabling innovation is the use of a duty cycle whereby the two objectives alternate in rapid sequence between excitation and detection (Fig. 1c). Each volume is there-fore imaged only twice, and from orthogonal vantage points. The two volumes are then com-bined computationally into a single, isotropic image by a fast, joint-deconvolution algorithm (Fig. 1d). Together with fast, scientific-grade, complementary metal oxide semiconductor cameras, an imaging readout speed of 200 Hz is achieved, yielding a volumetric acqui-sition time of 2 Hz, which is about 10 times faster than what has been demonstrated with the fastest comparable methods. The method also makes selective plane illumination more practical because standard sample prepara-tions, namely a specimen on a coverslip, can be used, as opposed to the time-consuming agar-immersion methods of sample prepara-tion that SPIM instruments generally require. This has the further advantage of limiting the bench-to-instrument time, which is critical in studies looking at rapid dynamic processes.

The authors show that the high acquisition speed, low phototoxicity and isotropic resolu-tion of their diSPIM setup allow the imaging of biological processes that would have been difficult to observe with more conventional

and photobleaching. Confocal microscopy eliminates out-of-focus light by masking it with a pinhole before detection, greatly improving the axial resolution of the image. However, laser light is absorbed throughout the sample, causing phototoxicity and photobleaching. The basic problem is that the excitation beam overlaps the detection path. In addition, the slow acquisition rates preclude imaging of fast events in living specimens.

In SPIM, by contrast, the specimen is illu-minated by an orthogonal light sheet, thereby eliminating out-of-focus light in the detection path (Fig. 1b)2. The resulting image is sharp, and the specimen does not absorb light out-side the focal plane. But light-sheet micros-copy is hampered by poor axial resolution. Creating the orthogonal light path requires multiple objectives in close proximity to the sample and hence to each other. High numeri-cal aperture objectives would generate thinner light sheets on the excitation side and better resolution on the detection side, but because of their bulky diameters and short working distances, it is not feasible to use them. Lower numerical aperture objectives with longer working distances are used instead. This weakness has led to a search for improved SPIM microscope designs and the creation of many SPIM spin-offs.

A common approach to overcome poor axial resolution is to image the sample from multiple viewpoints and then computationally fuse the data into a single composite image. This concept was first implemented by rotat-ing the sample so that the z axis becomes the new x axis3. To mitigate scattering effects and to increase simultaneous coverage of the sam-ple, a second strategy has been to add another set of excitation and detection pathways4,5. While still rotating the sample, opposing

The confocal microscope, long the dominant imaging technique in cell biology, is approach-ing obsolescence, soon to be superseded by a new generation of optical microscopes. The approach most likely to take its place is selec-tive plane illumination microscopy (SPIM), a technique that delivers lower doses of light to the sample and achieves faster image acqui-sition compared with confocal systems. The major drawback of conventional SPIM is its poor axial resolution: whereas optical sections in confocal microscopes are 800-nm thick, those in light-sheet microscopes are 26 m. In this issue, Wu et al.1 describe a new dual-view SPIM instrument capable of providing 330-nm resolution, not only along the lateral axes (x and y planes) but also along the opti-cal axis (z plane). The key insight is to use two orthogonal objectives that alternate in a duty cycle between excitation and detection as they scan through the sample. Light-sheet microscopy now stands to deliver upon its ini-tial promiserapid imaging of living organ-isms with low photo-dosage and high spatial resolution in three dimensions.

In traditional epifluorescence microscopy, a column of light excites fluorescent molecules above and below the focal plane in the speci-men (Fig. 1a). The out-of-focus light generates a blurry image without detail, and absorption of light by biological tissue leads to phototoxicity

News AND V iewsnp

g

2013

Nat

ure A

mer

ica,

Inc.

All

right

s re

serv

ed.