REVIEWS

Clinical cancer genomic profiling

Debyani Chakravarty 1,2
and David B. Solit 1,3 ✉

Abstract | Technological innovation and rapid reduction in sequencing costs have enabled the
genomic profiling of hundreds of cancer-​associated genes as a component of routine cancer
care. Tumour genomic profiling can refine cancer subtype classification, identify which patients
are most likely to benefit from systemic therapies and screen for germline variants that influence
heritable cancer risk. Here, we discuss ongoing efforts to enhance the clinical utility of tumour
genomic profiling by integrating tumour and germline analyses, characterizing allelic context
and identifying mutational signatures that influence therapy response. We also discuss the
potential clinical utility of more comprehensive whole-​genome and whole-​transcriptome
sequencing and ultra-​sensitive cell-​free DNA profiling platforms, which allow for minimally
invasive, serial analyses of tumour-​derived DNA in blood.

Precision oncology
Cancer is a genetic disease resulting from the accumu­ the development of more effective and less toxic person­
The process of using molecular lation of mutations in genes that regulate cell division, alized treatment strategies for all patients with cancer12.
data from the analysis of a survival, invasion or other hallmarks of the transformed Although progress has been slower and more incremen­
patient’s tumour or healthy phenotype. Some cancers are indolent and remain latent tal than many predicted, lung cancer quickly emerged
cells to inform treatment
and localized for years, whereas others rapidly invade as a cancer type in which precision oncology has been
selection.
nearby organs or metastasize to distant sites. Researchers transformative (Box 1). A diversity of targetable molec­
Companion diagnostics have long sought to understand the biological basis of ular alterations, including EGFR and BRAF mutations
Within the context of precision this variability of clinical outcomes by subclassifying and ALK, ROS1 and RET fusions, have a central role in
oncology, companion cancers into increasingly small but more phenotypically lung cancer pathogenesis. As tumours harbouring these
diagnostics are medical devices
designed to identify the subset
uniform subtypes. Historically, tumour classification molecular drivers are often indistinguishable under light
of patients most likely to was based primarily on cell type or tissue of origin and microscopy13, clinical tumour genomic profiling is now
respond to and benefit from morphological characteristics, in particular, histological broadly viewed as necessary by oncologists to ensure
a targeted or other systemic appearance under light microscopy1. Greater insight into optimal therapy selection in patients with advanced
or local therapy.
the molecular pathophysiology of cancer has promp­ lung cancer.
Next-​generation sequencing ted the adoption of molecular classification schemes The expansion in the number of therapeutically
(NGS). Massively parallel that integrate genomic information with clinical char­ actionable genes in lung and other cancer types exposed
high-​throughput sequencing acteristics to better predict an individual patient’s risk the limitations of single-​analyte companion diagnostics,
methods designed to analyse of recurrence or cancer-​specific death2. As the likeli­ which were historically co-​developed with new targeted
DNA or RNA more rapidly and
at higher resolution than older
hood of response to cytotoxic3, immune4 and targeted therapies14–16. In the early days of precision oncology,
methods such as Sanger therapies5–7 often varies as a function of molecular most companion diagnostic tests could detect only a sin­
sequencing. tumour subtype, accurate tumour classification is crucial gle mutation, such as BRAF p.Val600Glu (BRAF-​V600E),
to ensure optimal treatment selection. or a single gene fusion, such as EML4–ALK. The prolif­
For some cancer subtypes, a pathognomonic eration of clinically validated and investigational drug
1
Kravis Center for Molecular
genomic alteration is both a driver of tumour initiation targets, and more recently tumour-​agnostic biomarkers
Oncology, Memorial Sloan
Kettering Cancer Center,
and a potential therapeutic vulnerability. For example, of drug response, and the limited tumour tissue availa­
New York, NY, USA. almost all chronic myelogenous leukaemias (CMLs) ble for analysis for many patients with cancer prompted
2
Department of Pathology, have a translocation (the Philadelphia chromosome) the development of multiplexed diagnostic assays (ini­
Memorial Sloan Kettering involving the ABL1 tyrosine kinase gene on chromo­ tially based on mass spectrometry or PCR technology)
Cancer Center, New York, some 9 and the breakpoint cluster region (BCR) on that could define the mutational status of several dozen
NY, USA.
chromosome 22 (refs8,9). The resulting BCR–ABL trans­ cancer-​associated genes in a single reaction17–19. The
3
Department of Medicine, location is constitutively active, and drugs such as imati­ more recent development of next-​generation sequencing
Memorial Sloan Kettering
Cancer Center, New York,
nib that selectively inhibit ABL1 are highly effective in (NGS)-​based platforms has made feasible the concurrent
NY, USA. patients with CML10. The success of imatinib in patients analysis of hundreds of genes or even the entire genome
✉e-​mail: solitd@mskcc.org with CML generated widespread hope — which some using small quantities of tumour tissue collected by
https://doi.org/10.1038/ would later characterize as hype11 — that adoption of needle biopsy20–23 or DNA shed from tumour cells into
s41576-021-00338-8 clinical tumour genomic profiling would quickly enable plasma, so-​called cell-​free DNA (cfDNA)24,25.

Nature Reviews | Genetics

0123456789();:
Reviews

Box 1 | Lung cancer as a model for precision oncology In this Review, we discuss the current landscape
of clinical actionability in precision oncology and the
the epidermal growth factor receptor (eGFr) tyrosine kinase inhibitors gefitinib and challenges of variant interpretation that have arisen as
erlotinib were the first targeted therapies to receive us Food and Drug administration a result of the rapid and widespread adoption of larger
(FDa) authorization for the treatment of patients with lung cancer187,188. eGFr inhibitors panel NGS-​based diagnostic platforms. The complemen­
were initially tested in a tumour- and mutation-​agnostic manner, and, while limited
tary use of clinical tumour NGS to assist with not only
clinical activity was noted in most cancer types, rapid symptomatic relief and dramatic
tumour regressions were observed in a minority (15–20%) of patients with lung cancer. therapy selection but also cancer subtype diagnosis and
responses were more common in never-​smokers and in asian women, suggesting that the assessment of heritable cancer risk are summarized
underlying differences in disease pathogenesis were the likely basis for the variable in Fig. 1. Many of the technical challenges posed by the
responses to eGFr inhibitors among patients with lung cancer189,190. retrospective use of clinical samples for NGS-​based tumour genomic
studies later revealed that the vast majority of responders had somatic gain-​of-​function profiling have been discussed elsewhere31, thus we focus
mutations in the EGFR gene that induce constitutive eGFr activation and oncogene here on ongoing efforts to derive greater clinical value
dependence191–193. from tumour genomic profiling data through more
as the preclinical development of eGFr tyrosine kinase inhibitors pre-​dated large-​ robust evidence-​based variant classification, integration
scale sequencing initiatives such as the Cancer Genome atlas (tCGa), which have of tumour and germline sequencing and the identifica­
since defined the genomic landscape of most cancer subtypes, it was not known before
tion of allelic configurations and mutational signatures
the initial clinical studies of gefitinib and erlotinib that activating mutations in EGFR
were common in lung cancer. indeed, eGFr inhibitors initially received FDa approval predictive of drug response. Finally, we review the
for the treatment of all patients with non-​small-​cell lung cancer (NsCLC) irrespective potential future role for more comprehensive diagnostic
of EGFR mutational status, and for several years thereafter the clinical utility of EGFR platforms such as whole-​genome and RNA sequencing
mutational testing was an area of active debate and disagreement194. Given the limited and ultra-​sensitive cfDNA profiling platforms that can
treatment options available at the time for patients with metastatic lung cancer and the noninvasively monitor tumour recurrence and identify
possibility that eGFr overexpression or ligand-​driven eGFr kinase activation could also adaptive changes that mediate drug resistance.
confer eGFr dependence and drug sensitivity, many clinicians advocated treatment of
all patients with metastatic NsCLC with eGFr tyrosine kinase inhibitors irrespective DNA mutations as biomarkers of therapy response
of EGFR mutational status. Not all mutations in the same gene have the same bio­
Over the past decade, several clinical and laboratory observations have led to the
logical properties and the same clinical implications.
recommendation that clinical tumour genomic profiling is necessary for all patients
Somatic mutations are subclassified into those that are
with locally advanced and metastatic lung cancer to guide treatment selection. First,
the subsequent identification and clinical validation of additional targetable oncogenic oncogenic (drivers) versus those that are biologically
alterations in ALK195, ROS1 (ref.196), RET197, BRAF198, ERBB2 (refs199,200) and MET201,202 in a inert (passengers)32,33. Among the drivers, a subset are
largely mutually exclusive pattern in patients with lung cancer suggested that these predictive biomarkers of drug response or so-​called
and other not yet druggable oncogenic alterations, such as KRAS mutations, delineated clinically actionable mutations. The success of precision
distinct subsets of patients with lung cancer whose prognosis, clinical characteristics oncology strategies requires clinicians to distinguish
and response to therapy were dictated in part by the presence or absence of these clinically actionable molecular alterations from benign
recurrent genomic alterations196,203. second, clinical trials in patients with lung and variants and understand how differences in the biolog­
colorectal cancer suggested that eGFr-​targeted therapies were not only ineffective ical properties of individual mutant alleles influence
but could be harmful to patients with eGFr wild-​type tumours owing to their toxicities
treatment response and patient outcomes.
or drug-​induced acceleration of tumour growth85,188,204,205. the latter possibility, while
initially deemed unlikely by some, was supported by concurrent studies indicating that Currently, most mutations that are clinically actiona­
raF inhibitors could paradoxically activate MaPK pathway signalling in BraF wild-​type ble are the targets of small-​molecule kinase inhibitors or
cells and accelerate the growth of occult skin and other BraF wild-​type cancers206–209. antibodies that bind to cell surface receptors. Actionable
as the tumour type with the largest number of genes for which targeted therapies have mutations may also increase sensitivity to drugs that
been us Food and Drug administration (FDa) approved, the need to identify multiple function through a synthetic lethal mechanism such as
lung cancer-​specific and tumour-​agnostic biomarkers of drug response in large numbers poly(ADP-​ribose) polymerase (PARP) inhibitors, which
of patients with lung cancer has prompted the development and clinical adoption of are most effective in tumours with loss-​of-​function
larger next-​generation sequencing-​based tumour and cell-​free DNa sequencing panels. mutations in genes that mediate homologous
recombination-​based DNA repair such as BRCA1 and
In turn, the modest incremental cost of adding addi­ BRCA2 (refs34,35). Gene mutations in DNA repair path­
tional cancer genes to NGS-​based diagnostic panels has ways have also been shown to be predictive of response
made the development of drugs targeting increasingly to cytotoxic chemotherapies36,37 and immunotherapy29,38.
smaller and molecularly defined subsets of patients
with cancer logistically and financially feasible. The Challenges of variant interpretation. A major hurdle to
efficient clinical development of inhibitors effective in the broader adoption of precision oncology is the diffi­
Cell-​free DNA cancers driven by rare genomic mutations required the culty in distinguishing functional from benign variants,
(cfDNA). In the context of this concurrent development of novel clinical trial designs even in well-​studied cancer genes. In the case of onco­
article, circulating tumour such as basket trials, a study design in which eligibility genes, functional mutations typically enhance the enzy­
DNA, that is, DNA fragments
is based on mutational status instead of organ site26–28. matic activity or induce oncoprotein signalling through
shed by the tumour into
the blood. With the subsequent validation in basket studies of dimerization or as a result of altered affinity for and acti­
tumour-​agnostic biomarkers of drug response, such vation of downstream effectors. As only a small subset
Basket trials as microsatellite instability29 and NTRK gene fusions30, of the potential mutations in an oncogene are capable of
A clinical trial design that many oncologists now believe that NGS-​based tumour inducing activation via these mechanisms, population-​
prospectively accrues patients
with a specific molecular
genomic profiling should be offered to all patients with based studies have been used to identify likely gain-​of-​
alteration irrespective of cancer who are not candidates for curative-​intent local function mutations based on their recurrence, so-​called
tumour type. or systemic therapy. mutational hotspots39–42. Although mutational recurrence

www.nature.com/nrg

0123456789();:
 Reviews

within a population is typically a result of positive inherently mutable sites, for example, APOBEC3A-​
selection for a functional phenotype, non-​functional mediated deamination of cytosine within DNA short
recurrent variants (‘passenger hotspots’) can arise at hairpin loops42,43.
a

His1047Arg Ser249Cys
Leu858Arg Glu545Lys
Arg248Cys Tyr373Cys
Glu542Lys

EGFR PI3Kα FGFR3 ETV6 TRKC

b c
Therapy
EGFR
PIK3CA
ETV6–NTRK3
MSI-H and TMB-H

PML–RARA BRCA1/2
BCR–ABL1 CHEK2
+ IDH1 NF1
RET
TSC1/2
Histopathology NGS
Diagnosis Heritable risk
ETV6–RUNX1 BAP1
NPM1 CDH1 STK11
Diagnosis EWS–FLI1 APC
TP53

versus

CDH1

120 130 120 130

Bladder cancer plasmacytoid variant Pat GATAAATCTAGTCTTATTTCC Pat GATAAATCTAGTCTTATTTCC
Mat Deleted Mat GATAAATCTGGTCTTATTTCC

Breast ca
(42 yo)

+ +
Breast ca Ovarian ca
PML RARα NPM1 (50 yo) (39 yo)

Pancreatic ca
Acute promyelocytic leukaemia NPM1-mutant acute myeloid leukaemia (49 yo) Breast ca (41 yo)

Fig. 1 | clinical applications of tumour sequencing. Tumour profiling can germline mutations, such as inactivating mutations in the tumour
contribute to patient care by identifying mutations or structural variants suppressor genes BRCA1 and BRCA2, are both predictive of drug response
that are predictive biomarkers of drug response (part a), assist with cancer and associated with increased heritable risk. Germline mutations of CDH1
subtype diagnosis (part b), or confer increased heritable cancer risk (part c). are associated with increased risk of developing hereditary diffuse gastric
As highlighted in part a, predictive biomarkers of drug response can be cancer, whereas somatic CDH1 mutations are pathognomonic of the
either tumour subtype specific or tumour agnostic. Some somatic plasmacytoid variant subtype of bladder cancer. EGFR, epidermal growth
alterations, such as the BCR–ABL1 translocation, are both diagnostic for a factor receptor; MSI-​H, microsatellite instability-​high; NGS, next-​generation
cancer subtype and predictive biomarkers of drug response. A subset of sequencing; TMB-​H, tumour mutational burden-​high; yo, years old.

Nature Reviews | Genetics

0123456789();:
Reviews

Mutational signatures
For tumour suppressor genes, protein-​truncating Of particular concern is that patients may be treated
Patterns of base pair variants (frameshift, nonsense or splice site muta­ with the wrong therapy based on a failure by clinicians
substitutions or structural tions) are often presumed to be oncogenic (classified to recognize that certain mutations in an actionable can­
abnormalities that are often as ‘likely oncogenic’) based on their ability to promote cer gene are either inert or intrinsically resistant to the
characteristic of exogenous
nonsense-​mediated mRNA decay44. Oncogenic and drug chosen. For example, while many EGFR exon 20
or endogenous mutational
processes (such as tobacco likely oncogenic mutations are therefore typically dis­ mutations are oncogenic, they are typically insensitive
smoking or DNA repair tributed throughout the coding sequence of tumour to the FDA-​approved EGFR inhibitors erlotinib and
pathway mutations). suppressor genes, making computational inferences osimertinib54. The challenge of clinical variant interpreta­
such as recurrence less useful for distinguishing between tion has prompted the development of physician support
Somatic mutations
Non-​heritable mutations that
functional and benign variants. tools such as OncoKB55, CIViC (Clinical Interpretation
may arise in any cell except Neomorphic mutations confer novel phenotypes of Variants in Cancer) 56, the Jackson Laboratory
germ cells (sperm or ova). that do not simply potentiate or impair wild-​type pro­ Clinical Knowledgebase (JAX-​CKB)57, the Precision
Although somatic mutations tein function. Therefore, failure to observe a canoni­ Medicine Knowledge Base (PMKB) 58, the Cancer
have classically been defined
cal phenotype in laboratory-​based biological studies Genome Interpreter Cancer Biomarkers database
as those found specifically
in tumour but not in healthy does not always signify that a mutation is a benign (CGI) 59 and Personalized Cancer Therapy (PCT)
cells, accumulation of somatic variant. For example, while a primary function of the resource60, which seek to catalogue the known biological
mutations is common in PIK3R1-​encoded p85α subunit of PI3 kinase is to reg­ properties and clinical implications of individual mutant
non-​transformed, histologically
ulate the p110α catalytic product of the PIK3CA locus, alleles. As our understanding of the clinical implications
normal-​appearing cells as
patients age.
the mutant protein arising from a common recurrent of individual mutations is constantly evolving, and as
PIK3R1-​truncating mutation (p.Arg348∗) is unable to new investigational drugs enter clinical testing and are
Drivers bind to p110α. Instead, it augments cell survival by act­ either approved by regulatory agencies or fail to show
Mutations that enhance ing as a scaffold for signalling intermediaries within the efficacy in genetically defined populations, these vari­
tumour cell fitness by providing
a growth or survival advantage.
JNK pathway, resulting in increased JNK and MAPK ant interpretation tools need to be continuously updated
pathway activity45,46. to remain comprehensive and accurate through expert
Passengers Gain-​of-​function mutations in the same oncogene curation, crowdsourcing or a combination of the two.
Biologically inert mutations can also have distinct biological properties that result
with no impact on tumour
in differences in drug sensitivity. Mutant allele-​specific The current landscape of clinical actionability. A long-​
cell fitness.
drug sensitivities may be due to differences in drug standing criticism of the precision oncology field is that
Clinically actionable binding affinity or, in the case of BRAF, differences its proponents often overstate the clinical actionability
mutations in the ability of the mutant to promote BRAF dimer of individual genes or genomic variants. Mutations that
A subset of driver mutations formation47. Further complicating the task of variant are clinically validated and FDA-​recognized as predictive
that are predictive biomarkers
classification, mutations can have different pheno­ biomarkers of drug response are often grouped together
of drug response or resistance.
types depending on tumour lineage or co-​mutational as clinically actionable with mutations identified as the
Synthetic lethal mechanism context 48. For example, laboratory studies indicate putative basis for outlier exceptional responses or var­
An interaction between two that the RAF inhibitor vemurafenib potently inhib­ iants that correlate with enhanced sensitivity in cell-​
genes whereby loss of function
its the BRAF-​V 600E mutant protein in cell-​f ree based screens61,62. To better communicate the strength
of both (due to mutation,
epigenetic silencing or drug
assays49. However, the likelihood that a patient with a of evidence supporting the clinical actionability of indi­
inhibition) results in tumour cell BRAF-​V600E mutant tumour will respond to vemu­ vidual mutant alleles, many variant knowledge bases
death, whereas loss of function rafenib is strongly influenced by tumour type. Whereas stratify genomic alterations based on the level of clinical
or inhibition of either individual the majority of patients with melanoma and histiocyto­ and/or biological data supporting their use as predictive
gene does not.
sis respond to vemurafenib, RAF inhibitor monotherapy biomarkers of drug response or resistance. Optimally,
Mutational hotspots has limited clinical activity in patients with colorectal these databases should also incorporate information as
Mutations identified in a cancer5,50 owing to intrinsic RAF inhibitor resistance to how lineage and co-​mutational context influence the
population of patients with mediated by EGFR-​activated RAF dimer formation. likelihood of clinical response.
cancer more frequently than
This biological insight was the molecular basis for As an example, in the OncoKB knowledge base,
expected by chance.
combination trials of the anti-​EGFR antibody cetux­ level 1 and 2 mutations are those variants established
Germline variants imab and the RAF inhibitor encorafenib, which was through clinical experience (preferably prospective
Heritable mutations that were FDA approved for patients with BRAF-​V600E mutant randomized clinical trials) to be clinically validated pre­
present in germ cells and colorectal cancer in April 2020 (ref.51). In sum, despite dictive biomarkers of drug response to FDA-​approved
consequently found in all cells
large-​scale efforts over the past several decades to define therapies in a specific cancer subtype (Table 1). Lower
of the descendants.
the biological properties of cancer mutations as a guide levels are assigned to mutations with compelling but not
to clinical interpretation, many if not most somatic yet definitive clinical evidence (level 3) or only labora­
mutations identified by clinical tumour genomic pro­ tory evidence (level 4) that the mutation is predictive
filing are variants of unknown biological and clinical of drug sensitivity55. As different mutations in the same
significance. This remains true even for extensively gene can have different phenotypes, individual variants
studied cancer genes such as BRAF, for which targeted in cancer-​associated genes can be assigned to different
inhibitors are approved for clinical use by the FDA. OncoKB levels. Furthermore, as tumour lineage often
The recognition that different mutations within the influences the likelihood of drug response, the same
same gene often have different biological properties has mutation can be assigned to different OncoKB levels in
made it challenging for point-​of-​care clinicians to inter­ different cancer types.
pret the large number of somatic and germline variants In contrast to the germline setting, where inter­
that emerge from clinical tumour genomic profiling52,53. pretation of germline variant pathogenicity has been

www.nature.com/nrg

0123456789();:
 Reviews

Table 1 | OncoKB standard-​care biomarkers

Genetic biomarker indication targeted therapy
Level of evidence 1
BCR–ABL1 fusion ALL, CML Imatinib, dasatinib, bosutinib, nilotinib
ABL1 Thr315Ile Ponatinib
ALK fusions NSCLC Crizotinib, ceritinib, alectinib
ALK fusions and oncogenic mutations NSCLC Brigatinib, lorlatinib
ATM, BARD1, BRIP1, CDK12, CHEK1/CHEK2, Prostate cancer Olaparib
FANCL, PALB2, RAD51B/RAD51C/RAD51D,
RAD54L loss-​of-​function mutations
BRCA1/BRCA2 loss-​of-​function mutations Ovarian cancer, peritoneal Olaparib, rucaparib, nirapariba
serous carcinoma,
prostate cancer
BRAF Val600Glu Anaplastic thyroid cancer, Dabrafenib + trametinib
NSCLC
Colorectal cancer Encorafenib + cetuximab
BRAF Val600Glu/Lys Melanoma Dabrafenib + trametinib,
encorafenib + binimetinib,
vemurafenib + cobimetinib, vemurafenib + c
obimetinib + atezolizumab
BRAF Val600 Erdheim–Chester disease Vemurafenib
EGFR exon 19 deletions, EGFR Leu858Arg NSCLC Erlotinib, gefitinib, afatinib, dacomitinib,
osimertinib
EGFR Gly719, Leu861Gln, Ser768Ile Afatinib
EGFR Thr790Met Osimertinib
ERBB2 amplification Breast cancer Trastuzumab ± chemotherapy
± pertuzumab/capecitabine + tucatinib,
ado-​trastuzumab emtansine, lapatinib
+ letrozole/capecitabine, neratinib
± capecitabine, fam-​trastuzumab
deruxtecan-​nxki
Oesophagogastric cancer Trastuzumab, fam-​trastuzumab deruxtecan-​
nxki
EZH2 Ala692Val, Tyr646Cys/Phe/ Follicular lymphoma Tazemetostat
Hys/Asn/Ser
FGFR2 fusions Cholangiocarcinoma Pemigatinib
FGFR2 fusions and FGFR3 fusions and Bladder cancer Erdafitinib
hotspots (FGFR3 Gly370Cys, Arg248Cys,
Ser249Cys, Tyr373Cys)
FLT3 internal tandem duplication and FLT3 Acute myeloid leukaemia Gilteritinib
oncogenic mutants, including Asp835 and
Ile836
IDH1/IDH2 oncogenic mutations Bladder cancer Ivosidenib, enasidenib
KIT exons 11, 9, 17, KIT Val654Ala and KIT Gastrointestinal stromal Imatinib, sunitinib, regorafenib, ripretinib
Thr670Ile oncogenic mutations tumours
KRAS wild-​type Colorectal cancer Cetuximab, panitumumab, regorafenib
MET exon 14 splice/exon-​skipping NSCLC Capmatinib
mutations
NF1 loss-​of-​function mutations Neurofibroma Selumetinib
MSI-​H/TMB-​H All solid tumours Pembrolizumab
MSI-​H Colorectal cancer Ipilimumab + nivolumab, nivolumab
NTRK1/NTRK2/NTRK3 fusions All solid tumours Larotrectinib, entrectinib
COL1A1–PDGFB fusion Dermatofibrosarcoma Imatinib
protuberans
FIP1L1–PDGFRA fusion CEL-​NOS Imatinib
PDGFRA/PDGFRB fusions Myelodysplastic/ Imatinib
myeloproliferative
neoplasms

Nature Reviews | Genetics

0123456789();:
Reviews

Table 1 (cont.) | OncoKB standard-​care biomarkers

Genetic biomarker indication targeted therapy
Level of evidence 1 (cont.)
PDGFRα Asp842Val, Asp842Tyr, Gastrointestinal stromal Avapritinib
Asp842_Hys845del, Asp842_Hys845insVal tumour
PIK3CA oncogenic mutations ER+/HER2− breast cancer Alpelisib + fulvestrant
RET fusions NSCLC Selpercatinib, pralsetinib
Thyroid cancer
RET oncogenic mutations Medullary thyroid cancer Selpercatinib, pralsetinib
ROS1 fusions NSCLC Crizotinib, entrectinib
SMARCB1 loss-​of-​function mutations Epithelioid sarcoma Tazemetostat
TSC1/TSC2 loss-​of-​function mutations Subependymal giant cell Everolimus
astrocytomas
Level of evidence 2
BCR–ABL1 fusion ALL Bosutinib, nilotinib
ABL1 Glu255Lys, Glu255Val, Phe317Cys, ALL, CML Bosutinib, dasatinib, nilotinib
Phe317Ile, Phe317Leu, Phe317Val,
Phe359Cys, Phe359Ile, Phe359Val,
Thr315Ala, Tyr253His
ALK fusions Inflammatory Crizotinib, ceritinib
myofibroblastic tumour
BRAF Val600Glu Ganglioglioma, hairy Cobimetinib + vemurafenib,
cell leukaemia, pilocytic trametinib + dabrafenib, vemurafenib
astrocytoma, pleomorphic
xanthoastrocytoma
BRCA1/BRCA2 loss-​of-​function mutations Breast cancer Olaparib, talazoparib
CDK4 amplification Dedifferentiated Palbociclib, abemaciclib
liposarcoma,
well-​differentiated
liposarcoma
EGFR Ala763_Tyr764insPheGlnGluAlab NSCLC Erlotinib
ERBB2 amplification Colorectal cancer Pertuzumab + trastuzumab,
lapatinib + trastuzumab
Uterine serous carcinoma/ Trastuzumab + carboplatin–taxol regimen
uterine papillary serous
carcinoma
ERBB2 oncogenic mutations NSCLC Ado-​trastuzumab emtansine
KIT exon 17 oncogenic mutations Gastrointestinal stromal Sorafenib
tumour
KIT oncogenic mutations Melanoma Imatinib, sunitinib
MET exon 14 splice mutations, amplification NSCLC Crizotinib
MET amplification Renal cell carcinoma Cabozantinib
PDGFRA oncogenic mutations, PDGFRα Gastrointestinal stromal Imatinib, dasatinib
Asp842Val tumour
RET fusions NSCLC Cabozantinib
ALL, B cell acute lymphoblastic leukaemia/lymphoma; CEL-​NOS, chronic eosinophilic leukaemia, not otherwise specified;
CML, chronic myelogenous leukaemia; EGFR, epidermal growth factor receptor; MSI-​H, microsatellite instability-​high; NSCLC,
non-​small-​cell lung cancer; TMB-​H, tumour mutational burden-​high. aFDA approval of this drug is mutation-​agnostic. bWhile most
EGFR exon 20 insertions are resistant to FDA-​approved EGFR inhibitors, Ala763_Tyr764insPheGlnGluAla is sensitive to erlotinib.

standardized through a joint effort of the American created working groups to develop guidelines on how
College of Medical Genetics and Genomics (ACMG) best to classify the clinical and biological relevance of
and the Association for Molecular Pathology (AMP)63, to genomic variants identified by clinical tumour genomic
date there is no broad consensus among knowledge bases profiling assays65–67. Additionally, cross-​institutional
and regulators on how best to define clinical actionability consortia such as the Variant Interpretation Cancer
of somatic variants. To address this issue, the AMP, the Consortium (VICC)67 and the somatic variant working
American Society for Clinical Oncology (ASCO)64 and group of ClinGen68 have been formed to standardize best
the European Society of Medical Oncology (ESMO) have practices and define vocabulary that in the future could

www.nature.com/nrg

0123456789();:
 Reviews

Resistance mutations
guide FDA recognition of variant databases as physician in gliomas, pancreatic cancers and mesotheliomas
Mutations that increase support tools69. (Fig. 2c). One could also argue that the recent increase in
tumour cell fitness under To estimate the current fraction of patients with standard-​care actionability overstates to some degree the
the selective pressure of cancer for whom tumour genomic profiling would be benefits of tumour profiling by NGS, as many TMB-​H
a systemic therapy.
predicted to directly influence therapy selection, we ana­ tumours arise in patients with tumour types for which
Penetrance lysed a cohort of 52,069 solid tumour samples sequenced immune checkpoint blockade is clinically indicated irre­
A measure of the proportion as of November 2020 at Memorial Sloan Kettering spective of MSI or TMB status (Fig. 2c; Box 2). However,
of individuals with a mutant (MSK) Cancer Center using the MSK-​IMPACT clini­ tumour profiling with NGS can also benefit patients
allele in a defined population
cal NGS platform20. In this prospective dataset, 92% of by identifying mutations predictive of drug resistance
who manifest the associated
phenotype. For germline
tumours harboured at least one oncogenic mutation, (so-​called resistance mutations) such as KRAS and BRAF
variants associated with and 34% harboured at least one mutation classified as mutations, which are associated with intrinsic resistance
increased cancer predis­ a predictive biomarker of response to an FDA-​approved to standard-​care anti-​EGFR antibodies in patients with
position, the penetrance is or investigational drug based on compelling clinical data colorectal cancer85,86. Clinical tumour genomic profiling
the proportion of patients
(OncoKB, levels 1–3A)55 (Fig. 2a,b). can therefore benefit patients by allowing them to avoid
who develop the associated
cancer type. Although there has been significant focus on the potentially toxic therapies unlikely to result in clinical
use of broad-​panel NGS to identify investigational bio­ benefit based on their specific tumour molecular profile.
markers of drug response, the increasingly widespread In the near term, the most substantial barrier to
adoption of NGS-​based clinical tumour profiling has the broader success of precision oncology paradigms
been driven primarily by the need to identify somatic is not the limited number of genes included in current
and germline mutations that are predictive biomarkers NGS-​based panels but the large number of ‘undruggable’
of response to FDA-​approved therapies. Since 2017, oncogenic mutations identified by tumour genomic pro­
there has been an approximately threefold increase filing. The prospective MSK-​IMPACT experience indi­
from 9%20 to 30% in the fraction of tumours for which a cates that undruggable mitogenic drivers are common
disease-​matched, standard-​care predictive biomarker of in patients who lack clinically actionable alterations.
response to an FDA-​approved therapy would have been Examples of oncogenic mutations that are not currently
identified by tumour NGS, defined as OncoKB levels 1 actionable include those in NF1, PTEN, RB1, STK11 and
and 2 (Fig. 2b). This increase in standard-​care clinical KEAP1 (refs20,66). For some undruggable drivers, down­
actionability is partially the result of recent FDA approvals stream effectors are potentially targetable (for example,
of first-​in-​class selective inhibitors of IDH2 (acute mye­ the use of MEK inhibitors in KRAS-​mutant tumours).
loid leukaemia (AML), 2017 (refs70,71)), IDH1 (AML, 2018 Drugs targeting downstream effectors do not, however,
(ref. 72 ) ), TRKA/TRKB/TRKC (tumour-​a gnostic, typically demonstrate the levels of clinical efficacy of
2018 (ref.6)), PI3Kα (breast cancer, 2019 (ref.73)), FGFR3 drugs that directly inhibit the mutated oncoprotein87,88.
(bladder cancer, 2019 (ref.74)), FGFR2 (cholangiocarci­ Recent promising results with covalent inhibitors of
noma, 2020 (ref.75)) and RET (lung and thyroid cancer, KRAS p.Gly12Cys89 have generated hope that novel
2020 (refs76,77)). The landscape of clinical actionability approaches to drug development including targeted pro­
has also been expanded through FDA approvals of addi­ tein degraders and synthetic lethal approaches will result
tional indications for PARP inhibitors for breast cancer in a continued expansion in the landscape of clinical
(in 2018 (ref.78)), pancreatic cancer (in 2019 (ref.79)) and actionability over the next several years.
prostate cancer (in 2020 (refs35,80)) and RAF and MEK
inhibitors for lung cancer (in 2017 (ref.81)), Erdheim– Integration of somatic and germline profiling
Chester disease (in 2017 (ref.82)), anaplastic thyroid Heritable cancer risk and pharmacogenomics. It has
cancer (in 2018 (ref.83)) and colorectal cancer (in com­ long been recognized that some families and eth­
bination with the anti-​EGFR antibody cetuximab, in nic groups are at greater risk of developing particular
2018 (ref.50)). cancers. For example, individuals of Ashkenazi Jewish
Additionally, three classes of molecular altera­ descent are at increased risk of developing early-​onset
tions (NTRK1/NTRK2/NTRK3 fusions, microsatellite breast and ovarian cancer90,91. By applying segrega­
instability-​high/deficient mismatch repair (MSI-​H/ tion analysis of incidence patterns to a cohort of 1,579
dMMR) and tumour mutational burden-​high (TMB-​H, patients with breast cancer, Newman et al.92 were the
defined as ≥10 mutations per megabase of DNA)) are first to show that an autosomal dominant and highly
now recognized by the FDA as tumour-​agnostic bio­ penetrant susceptibility allele was the basis for familial
markers of drug response. As at least one of these clustering of early-​onset breast cancer cases. Positional
tumour-​a gnostic predictive biomarkers has been cloning subsequently localized the first candidate breast
reported in essentially all cancer subtypes, tumour cancer susceptibility gene, designated BRCA1, to chro­
molecular profiling is now viewed by most oncologists as mosome 17q21 (refs93–95). Clinical testing for patho­
a clinical necessity in all patients with metastatic cancer genic germline variants in BRCA1 and additional high
who require systemic therapy6,84. and moderate penetrance cancer predisposition genes,
Although the recent FDA approvals of additional including BRCA2, CHEK2, PALB2, ATM, VHL, BAP1
genotype-​directed therapies represent clear progress, the and MSH2, among others, is now a component of the
likelihood of identifying a clinically actionable genomic standard management of an expanding number of can­
alteration remains highly tumour-​t ype dependent, cer types96. These tests use a DNA sample derived from
ranging from >80% in gastrointestinal stromal tumours healthy-​appearing (‘normal’) tissue, typically blood, nail
(GIST), and 64% in non-​small-​cell lung cancer, to <10% clippings or a buccal swab. As with tumour genomic

Nature Reviews | Genetics

0123456789();:
Reviews

a Mapping between the OncoKB levels of evidence and the AMP/ASCO/CAP consensus recommendation
AMP/ASCO/CAP
OncoKB levels of evidence variant categorization

FDA-recognized biomarker predictive Tier I: variants of strong b

Standard care 1 of response to an FDA-approved drug clinical significance 7.7%
in this indication
18.6%
Level A evidence
Standard-care biomarker recommended FDA-approved therapy 23.2%
by the NCCN or other professional included in professional 9.5%
2 guidelines
guidelines predictive of response to an
FDA-approved drug in this indication 1.7%
17.3% 4.1%
Level B evidence 17.9%
Compelling clinical evidence supports Well-powered studies
Investigational 3A the biomarker as being predictive of with consensus from
response to a drug in this indication experts in the field

Standard-care or investigational Level 1 (incl. MSI-H)

3B biomarker predictive of response to Level 1 (only TMB-H)
an FDA-approved or investigational Tier II: variants of potential
drug in another indication clinical significance Level 2
Level 3A
Compelling biological evidence Level C evidence Level 3B
Hypothetical 4 supports the biomarker as being FDA-approved therapies Level 4
predictive of response to a drug for different tumour
types or investigational Drivers without actionability
therapies Only VUS
Multiple small published
Standard care Standard-care biomarker predictive of studies with some
resistance R1 resistance to an FDA-approved drug consensus
in this indication

Level D evidence
Compelling clinical evidence supports Preclinical trials or
Investigational R2 the biomarker as being predictive of
resistance a few case reports
resistance to a drug without consensus

c Gastrointestinal stromal tumour, n=582

Melanoma, n=1,757
Bladder cancer, n=2,245
Non-small-cell lung cancer, n=8,198
Breast cancer, n=6,706
Skin cancer, non-melanoma, n=447
Small cell lung cancer, n=447
Small bowel cancer, n=154
Endometrial cancer, n=2,168
Oesophagogastric cancer, n=1,643
Colorectal cancer, n=4,761
Thyroid cancer, n=803
Prostate cancer, n=3,220
Head and neck cancer, n=628
Soft tissue sarcoma, n=2,057
Cancer of unknown primary, n=1,578
Anal cancer, n=115
Ampullary cancer, n=123
Cervical cancer, n=323
Ovarian cancer, n=1,909
Hepatobiliary cancer, n=1,444
Salivary gland cancer, n=385
Nerve sheath tumour, n=122
Glioma, n=2,198
Appendiceal cancer, n=307
Gastrointestinal neuroendocrine tumour, n=209
Sellar tumour, n=103
Renal cell carcinoma, n=1,061
Uterine sarcoma, n=327
Pancreatic cancer, n=2,879
Mesothelioma, n=403
Peripheral nervous system, n=365
Bone cancer, n=560
Germ cell tumour, n=746
CNS cancer, n=157
Retinoblastoma, n=196
0 25 50 75 100
Samples per tumour type (%)

www.nature.com/nrg

0123456789();:
 Reviews

◀ Fig. 2 | current landscape of clinical actionability. a | OncoKB levels of evidence. identified an association between germline variants in the
Individual mutations or structural alterations are annotated based on the level of genes encoding the cytochrome P450 enzymes CYP2D6
evidence that the alteration is a predictive biomarker of drug response in a specific (ref. 105) and CYP2C8 (ref. 106) with altered tamoxifen
cancer subtype (left panel). Mapping of OncoKB levels of evidence to the Association of metabolism and increased risk of paclitaxel-​associated
Molecular Pathology (AMP)–American Society of Clinical Oncology (ASCO)–College
neuropathy, respectively. Similarly, germline polymor­
of American Pathologists (CAP) evidence-​based variant categorization64 (right panel).
b | Clinical actionability of solid tumour samples prospectively analysed using the MSK-​
phisms in the TYMS gene, which encodes thymidylate
IMPACT clinical next-​generation sequencing tumour profiling assay (n = 52,069). Fraction synthetase, are associated with increased likelihood of
of samples across all solid tumour cancer types that harbour a mutation considered response and toxicity to capecitabine and 5-​fluorouracil,
clinically actionable according to the OncoKB levels of evidence (pie chart). Tumours antimetabolites that competitively bind to and irreversi­
with level 1 alterations based only on the tumour-​agnostic tumour mutational burden-​ bly inhibit thymidylate synthetase107. Although a lack of
high (TMB-​H) approval of pembrolizumab are shown in lighter green (dark green, all expert consensus regarding the clinical utility of these
other level 1 tumours; blue, level 2; dark purple, level 3A; light purple, level 3B; black, pharmacogenomic biomarkers has limited their clinical
level 4; peach, non-​actionable oncogenic drivers only; and light grey, tumours with adoption, germline variants that influence drug metab­
variants of unknown significance (VUS) only). c | Clinical actionability of tumour samples olism or absorption could be easily incorporated into
as a function of common solid tumour types. Similar analysis to part b of solid tumours
future tumour–germline paired NGS panels.
in the prospective MSK-​IMPACT cohort but shown as a function of cancer subtype.
All MSK-​IMPACT sequencing results are made available through the AACR GENIE
Consortium20,66. CNS, central nervous system; MSI-​H, microsatellite instability-​high. Clonal haematopoiesis. Clonal haematopoiesis, which
is associated with increasing age and prior treatment
with radiation or cytotoxic chemotherapies108, is a con­
profiling assays, single-​gene germline tests have grad­ founding factor for the accurate classification of soma­
ually been replaced by multigene NGS-​based panels tic mutations. Clonal haematopoiesis-​derived somatic
designed to identify pathogenic germline variants in mutations can be present in tumour-​only sequencing
a dozen or more heritable cancer-​associated genes. data at variant allele frequencies above the thresholds
Paired analysis of tumour and germline samples typically used for somatic mutation calling. Therefore, in
has long been standard for research-​focused whole-​ the absence of patient-​matched germline DNA sequenc­
exome sequencing (WES) and whole-​genome sequencing ing data (which is typically derived from blood), clonal
(WGS) studies97. Filtering of germline variants using a haematopoiesis-​derived somatic mutations can be mis­
patient-​matched normal DNA sample allows for more classified as tumour somatic mutations109. As a subset of
accurate classification of mutations as somatic versus clonal haematopoiesis mutations are in clinically action­
inherited germline variation. Tumour-​only NGS-​based able cancer genes, tumour-​only profiling may result in
sequencing panels — now widely used as companion patients receiving inappropriate treatment owing to mis­
diagnostic tests for therapy selection — rely on exist­ classification of clonal haematopoiesis-​derived somatic
ing germline databases and computational algorithms variants as tumour-​derived. In this context, incorrect
to distinguish somatic from germline variants97. This treatment selection resulting from misinterpretation
can lead to misclassification of germline variants as of clonal haematopoiesis-​derived actionable mutations
somatic mutations, in particular for racial and ethnic could include the inappropriate use of a drug (for exam­
minority patients, who are currently under-​represented ple, a clonal haematopoiesis-​derived ATM mutation in a
in germline variant databases98. patient with prostate cancer prompting the use of PARP
An equally important benefit of the incorporation inhibitors) or the inappropriate withholding of therapy
of paired germline DNA analysis into tumour genomic (for example, a clonal haematopoiesis-​derived KRAS
profiling platforms is the opportunity to reanalyse the mutation in a patient with metastatic colorectal cancer
paired normal sample to identify pathogenic germline contraindicating the otherwise standard-​care use of the
variants associated with increased heritable cancer risk. anti-​EGFR antibodies cetuximab and panitumumab).
Paired tumour–germline genomic profiling has been As mutational calling at lower allele frequencies is asso­
shown to identify at least one pathogenic or likely patho­ ciated with a greater risk of misclassification of clonal
genic germline variant in a cancer predisposition gene in haematopoiesis-​derived mutations as tumour-​derived,
8–18% of patients with cancer96,99,100. Of note, 25–50% of sequencing of paired cell-​free (plasma) and white blood
patients with cancer with pathogenic germline variants cell (buffy coat) samples will be of particular importance
in population-​based tumour genomic profiling cohorts for cfDNA analyses110.
Whole-​exome sequencing would not have been referred for germline genetic anal­
(WES). Sequencing of all
protein-​coding regions (or
ysis based on historical clinical guidelines96. On this Logistical challenges of integrated somatic–germline
exons) in the genome. basis, there is increasing support among oncologists sequencing. Although the historical focus of germline
for the testing of all patients with cancer for pathogenic genetic testing was to identify variants associated with
Whole-​genome sequencing germline variants to appropriately counsel patients as increased heritable cancer risk, the success of PARP
(WGS). Sequencing of the
to their personal risk of a subsequent cancer and fam­ inhibitors in patients with mutations in genes associ­
entire genome including
non-​coding sequences. ily members as to their risk of harbouring the same ated with homologous recombination repair has led
heritable cancer predisposition101,102. to a convergence in the clinical need for tumour and
Clonal haematopoiesis Chemotherapy response and toxicity has also been germline genetic testing. PARP inhibitors are most effec­
The acquisition of somatic linked to genetic differences in genes that have a role in tive in homologous recombination-​deficient tumours,
genomic alterations in
haematopoietic stem and/or
the absorption or metabolism of cytotoxic and hormo­ in particular, in patients with loss-​of-​function muta­
progenitor cells, resulting in nal therapies103,104. As an example, genome-​wide associ­ tions in BRCA1 and BRCA2 (ref.34). As both somatic
clonal expansion. ation studies (GWAS) of patients with breast cancer have and germline BRCA1 and BRCA2 mutations can confer

Nature Reviews | Genetics

0123456789();:
Reviews

Box 2 | tumour genomic profiling and immunotherapy sensitivity

Microsatellite instability-​high/deficient mismatch repair (Msi-​H/dMMr) alterations in individual oncogenes have also been associated with
and tumour mutational burden-​high (tMB-​H, defined as ≥10 mutations greater or lesser likelihood of immune checkpoint inhibitor response.
per megabase of DNa) are FDa-​recognized predictive biomarkers of As an example, WNT pathway activation is associated with T cell
response to the anti-​PD1 antibody pembrolizumab. Dramatic, durable exclusion and innate resistance to immune checkpoint blockade in
responses have been observed in chemotherapy-​refractory metastatic preclinical models216,217, and clinical data suggest that wNt/β-​catenin
Msi-​H/dMMr tumours in cancer types in which immune checkpoint pathway mutations are a predictive biomarker of immune checkpoint
inhibitors are not broadly active (for example, colorectal and prostate inhibitor resistance in hepatocellular carcinoma218. in the phase iii
cancers)29. a failure to test all patients with metastatic disease for Msi-​H/ clinical trial testing the anti-​PDL1 inhibitor atezolizumab in lung cancer,
dMMr can therefore result in some patients not receiving effective, patients with ALK fusion-​positive or EGFR-​mutant tumours had a low
potentially curative, standard therapy. as next-​generation sequencing likelihood of responding to immune checkpoint blockade, but this may
(NGs)-​based tumour sequencing assays can robustly detect microsatellite simply reflect the low mutational burden of these molecularly defined
instability, the need to efficiently screen all patients with metastatic solid lung cancer subtypes219,220. Host factors such as human leukocyte
tumours for the Msi-​H/dMMr phenotype to assess eligibility for immune antigen (HLa) genotype and variations in the tumour microbiome may
checkpoint inhibitor therapy has been a rationale for the broader also influence sensitivity to immunotherapies and could potentially
adoption of NGs-​based tumour genomic profiling in patients with be assayed by NGs-​based panels221,222. it has also been proposed
treatment-​refractory solid tumours. that tumours with a viral aetiology are more likely to respond to
there is greater disagreement among oncologists as to the immunotherapy223–225, and probes designed to capture viral DNa
appropriateness of the tumour-​agnostic approval of pembrolizumab for are being incorporated into newer NGs-​based tumour genomic
tMB-​H tumours29,84,210 as the label expansion was based largely on a single-​ profiling assays.
arm study of 102 patients, in which a 29% objective response rate was Finally, whole-​exome and whole-​genome sequencing combined with
observed in the tMB-​H cohort211. Many of the responses were, however, machine learning approaches that can predict which mutated peptides
durable, extending over 2 years, with many ongoing212,213. Despite its bind with high-​affinity to autologous HLa molecules have made possible
limitations as a predictive biomarker, the tMB-​H label expansion provides the development of personalized cancer vaccines226,227. in such a scenario,
a new and potentially transformative therapeutic option for patients with tumour NGs functions not as a predictive biomarker of drug response
cancer types in which immune checkpoint inhibitors have yet to establish but rather as the initial step in the development of personalized
a role because response rates are low in biomarker-​unselected patients immunotherapies that target the neoantigens present exclusively in a
(for example, prostate and pancreatic cancers214,215). importantly, the tMB-​H patient’s tumour but not healthy cells. as a more off-​the-​shelf alternative,
approval will also facilitate access to immune checkpoint inhibitor therapy T cell-​based therapies are in development that target shared neoantigens
for patients with rare cancers, for whom enrolment in adequately powered that arise from recurrent mutations in commonly mutated oncogenes228.
clinical trials is difficult. Future studies will be needed to determine whether in sum, while previously viewed as a means to guide the selection of
the underlying mechanism of hypermutation influences the likelihood of targeted therapies, tumour genomic profiling has quickly established
immunotherapy response and to refine the optimal tMB-​H cut-​off, which an additional role in optimizing the use of immunotherapies in patients
in retrospective studies varies as a function of cancer type210. with cancer.

PARP inhibitor sensitivity, neither tumour-​only nor at higher risk of cancer121. The finding of a pathogenic
germline-​only profiling can identify all patients who germline variant can also result in distress and anxiety
are candidates for PARP inhibitor therapy. Additional for patients and family members122. These concerns have
examples of germline mutations that are predictive led some experts to propose that counselling by specially
biomarkers of drug response include ALK mutations trained nurses or physicians be required before germline
in children with neuroblastoma111 and RET mutations in genetic testing to ensure that patients fully understand
patients with medullary thyroid cancer112,113. Germline the risks associated with an incidental finding of a path­
mutations in DNA repair pathways that result in ele­ ogenic germline variant. The need to test an expanding
vated TMB have also been associated with increased fraction of patients with cancer for potentially action­
likelihood of response to immunotherapy. For example, able germline genetic alterations as a guide to therapy
most tumours that arise in patients with Lynch syn­ selection has upended the traditional pre-​test counsel­
drome are mismatch-​repair deficient, which is a tumour-​ ling model, as there are currently insufficient genetic
agnostic biomarker of response to the anti-​PD1 antibody counsellors to advise all patients with cancer123,124 (as
pembrolizumab114. Similarly, germline mutations in the opposed to the subset of patients referred to genetic
MBD4 gene, which encodes a DNA glycosylase, are counsellors on the basis of historical clinical guidelines).
associated with hypermutation and immunotherapy A requirement for referral for pre-​test counselling could
response in uveal melanoma115,116. also markedly delay genetic testing and the reporting of
The need for broad-​based germline genetic testing results to treating physicians, which could prove harmful
in patients with cancer as a prelude to treatment selec­ in patients with rapidly progressive metastatic disease
tion has raised several ethical and logistic hurdles. First, who are candidates for PARP, ALK or RET inhibitors125.
several states regulate germline genetic testing over con­ The burden of obtaining consent and advising patients
cerns about the risk of discrimination (for example, with about the risks of germline testing is therefore expected
respect to insurance or employment) targeting patients to fall increasingly on point-​of-​care medical, surgical
with a heritable cancer susceptibility117–120. Privacy con­ and radiation oncologists, many of whom have lim­
cerns linked to accessibility to germline genetic data are ited training in cancer genomics, are less experienced
particularly acute, as results can have an impact not only in interpreting the clinical implications of variants of
on the index cancer patient but also on family members unknown significance and less likely to propose cascade
who may be germline mutation carriers and therefore testing of family members126,127.

www.nature.com/nrg

0123456789();:
 Reviews

Paired tumour–germline sequencing can add to assay by capture-​based NGS sequencing panels through the
costs owing to the need to obtain, extract and analyse inclusion of probes targeting intronic regions, the large
DNA from a normal sample128. In our opinion, these size of some introns and the variable location of fusion
labour and reagent costs are offset by the greater effi­ breakpoints make the tiling of all relevant intronic
ciency of variant interpretation in the setting of paired regions prohibitively expensive. RNA sequencing
tumour and germline genetic sequencing. However, (RNA-​seq) is far better suited for fusion detection and
for many providers, the logistical hurdles and potential can also provide information on gene expression that
delays involved in coordinating collection and shipment cannot be derived from DNA-​based tumour profiling.
of both tumour and healthy tissue samples have been Unfortunately, the generation of high-​quality whole-​
a barrier to the broader adoption of paired tumour– transcriptome RNA-​seq data from the limited and often
germline sequencing129. Despite these logistical hurdles low-​quality archival tumour samples available for most
and the potential risks associated with an incidental patients with cancer has proved challenging. There
finding of a pathogenic germline variant, we expect are also computational challenges with RNA-​seq data
that the need to identify potentially actionable germline associated with batch and sample quality-​associated
variants in an increasing number of cancer types will artefacts136. As an alternative, RNA-​based gene panels
drive the broader adoption of paired tumour–germline have been developed that use anchored multiplex PCR
genomic testing in patients with cancer. technology that can robustly detect gene fusions using
small amounts of FFPE-​derived RNA137. The small num­
Alternatives to panel-​based tumour NGS ber of genes covered by current RNA-​based fusion panels
Whole-​exome and whole-​genome sequencing. A pri­ will likely restrict their use in the near future to the val­
mary limitation of targeted NGS panels is that they fail idation of putative gene fusions detected by DNA-​based
to detect mutations in genes not included in the assay tumour profiling or the analysis of patient samples in
design. In response to declining sequencing costs, panel which the pre-​test suspicion of a gene fusion is high (for
sizes have steadily grown from several dozen to hun­ example, sarcomas and lung cancers in which DNA test­
dreds of cancer-​associated genes20. Although clinical ing failed to detect an oncogenic driver). As an alterna­
WES and WGS assays are offered by select academic tive, exome-​capture RNA-​seq has shown promise for the
laboratories and commercial providers130–133, the wider detection of fusions in archival FFPE tumour samples138.
adoption of clinical WES and WGS has to date been lim­ Finally, diagnostic assays capable of characterizing global
ited for several reasons. One hurdle is that the tumour DNA methylation patterns are likely to prove useful in
material available for many patients is of insufficient refining tumour subtype classification and for iden­
quantity, quality or purity for these broader sequencing tifying the likely primary site in patients with cancers
platforms. In designing clinical NGS platforms, the limi­ of unknown primary139. The latter may be particularly
tations imposed by cost and sequencing capacity require relevant for cfDNA-​based screening platforms as dis­
the balancing of sequencing breadth and depth. Given cussed in the next section. Novel diagnostic platforms
this trade-​off, the higher depth of coverage of targeted that can identify epigenetic modifications or changes
NGS assays provides an advantage over WES and WGS in protein expression or activation that are predictive
assays for maximizing detection of alterations in genes of treatment response are also in development.
that are clinically validated biomarkers of drug response,
in particular, in samples with poor DNA quality or Cell-​free DNA. In contrast to broader tumour sequenc­
significant stromal contamination. More widespread ing platforms such as WES, narrower but ultrasensitive
adoption of WGS will require further reductions in gene panels capable of detecting small quantities of
sequencing costs and technological improvements to circulating tumour-​derived DNA in plasma have been
enable the use of lower-​quality, archival formalin-​fixed, quick to show evidence of clinical utility. Tumour biopsy
paraffin-​embedded (FFPE) tumour tissue. Of note, samples are often of insufficient quantity and quality for
although additional rare and private (that is, genetic tumour genomic profiling, and analysis of a single pri­
alterations specific to one individual) oncogenic gene mary or metastatic tumour site may fail to capture spatial
fusions, non-​coding mutations and structural altera­ heterogeneity or treatment-​associated clonal evolution.
tions that dysregulate gene expression will certainly be Profiling of plasma cfDNA can overcome many of these
discovered in the coming years, the likelihood of identi­ limitations by enabling minimally invasive analysis of
fying a clinically actionable mutation or structural vari­ tumour-​derived DNA that can be serially repeated as
ant by WES or WGS that would not have been detected patients progress from localized to metastatic disease
by current large-​panel NGS assays is low. Therefore, we or develop resistance to systemic therapies. Tumour-​
and others predict that the adoption of clinical WGS derived cfDNA can also be present in cerebrospinal
will be driven more by the need to robustly characterize fluid140, pleural fluid141, ascitic fluid142 or in urine143.
mutational signatures predictive of drug response (for Beyond its use to identify actionable mutations, cfDNA
example, structural signatures of homologous recombi­ analysis has shown promise as a platform to screen for
Capture-​based DNA nation deficiency) rather than by the need to identify occult cancers in high-​risk populations, for the detec­
sequencing incremental discrete oncogenic alterations134,135. tion of minimal residual disease in patients treated with
A method for selectively curative intent, and as a means to quantify treatment
sequencing targeted regions
of the genome using baits that
RNA sequencing. Capture-​based DNA sequencing pan­ response in patients with metastatic disease144–149.
hybridize with specific regions els are not optimal for the detection of oncogenic The challenge of cfDNA analysis is that the quan­
of DNA. gene fusions. Although gene fusions can be detected tity of circulating tumour DNA is typically low, and it

Nature Reviews | Genetics

0123456789();:
Reviews

Clonal mutations
can be difficult to distinguish between tumour-​derived cancers, given the challenges of obtaining high-​quality
Mutations present in all somatic mutations, somatic mutations arising from genomic material from bone biopsies.
of a patient’s cancer cells. clonal haematopoiesis and artefacts induced by DNA In the minimal residual disease setting, where the
oxidation, PCR amplification errors or through the pro­ fraction of tumour-​derived cfDNA is expected to be very
cess of DNA sequencing110. DNA barcoding-​based error low, personalized cfDNA assays designed to detect mul­
suppression methods have recently been developed to tiple clonal mutations identified by tumour sequencing
filter out mutations resulting from PCR or sequencing (some of which may be non-​functional variants) appear
artefacts150 (Fig. 3). These methods rely on a 20–40-​fold to be more sensitive than NGS-​based panels, which are
greater depth of sequencing coverage than needed for better suited to the detection of a limited number of
tumour-​based NGS panels, and therefore cfDNA assays actionable alterations in patients with greater tumour
typically analyse significantly fewer genes. The limited burden or metastatic disease154–156. Finally, there are
sequencing breadth of current cfDNA panels and the low millions of CpG sites across the human genome, and
fraction of tumour-​derived DNA in plasma also make DNA methylation patterns differ as a function of cancer
some genomic features such as deletions and mutational subtype157. Therefore, for screening applications, cfDNA
signatures more challenging to detect in blood. Negative assays designed to detect aberrant patterns of DNA
cfDNA results in individual patients must be interpreted methylation may be more sensitive than NGS-​based
with caution as some tumours do not shed sufficient cfDNA panels, while also providing information as to
DNA to allow for mutation detection using even the the likely primary site of disease to guide subsequent
most sensitive of assays. Tumour and cfDNA sequenc­ imaging or diagnostic endoscopies110.
ing are therefore likely to be complementary strategies
for identifying potentially actionable genomic alterations Allelic configuration and mutational signatures
in patients in need of systemic therapy. Mutational clonality. New somatic alterations arise
Despite their limitations, NGS-​b ased multigene stochastically as a result of tumour cell-​intrinsic defi­
cfDNA assays are rapidly being adopted in lung cancer, as ciencies in DNA replication, repair or chromosomal seg­
oncologists need to screen for multiple targetable genomic regation, or as a result of ongoing exposure to mutagens
alterations in a short time frame, and cfDNA results can such as cigarette smoke or drugs, including cytotoxic
on average be delivered more quickly151. In particular, for chemotherapies, which induce DNA damage. Cancer
patients who had a diagnostic tumour biopsy or surgery genomes are also constantly evolving, with new muta­
at another institution, results from cfDNA testing can be tions and structural variants arising as tumours invade,
available weeks or even months faster than tumour test­ metastasize or adapt to therapy-​induced selective pres­
ing given the logistical delays associated with requesting sures. Ongoing mutation and clonal selection during the
and processing tumour tissue. cfDNA sequencing is also course of a patient’s disease therefore result in complex
likely to be particularly useful for identifying targetable patterns of intra-​lesional and lesion-​to-​lesion genomic
genomic alterations in tumours with a predominant pat­ heterogeneity that may diminish the effectiveness of pre­
tern of osseous spread, such as breast152 and prostate153 cision oncology approaches. Current tumour sequencing

cfDNA molecule
UMI 1 UMI 2
UMIs

True mutation Strand-specific

damage

Duplex family

Error-free consensus

Fig. 3 | Detection of low-frequency mutations in tumour cell-free DnA. As the fraction of cell-​free DNA (cfDNA)
derived from tumour cells is often very low, tumour-​specific somatic mutations are typically present in blood at very low
allele frequencies. For mutations present at low allele frequencies (<1%), ultra-​deep sequencing and error correction are
needed to distinguish PCR and sequencing errors from tumour-​derived somatic variants in blood. Error suppression can
be achieved through the use of unique molecular indices (UMIs) with dual index barcodes, which allow for the filtering of
sequencing artefacts. Red stars represent true somatic mutations, whereas green stars are stand-​specific damage and
white stars sequencing errors. Image courtesy of M. F. Berger, Memorial Sloan Kettering Cancer Center, New York, USA.

www.nature.com/nrg

0123456789();:
 Reviews

reports are mutation-​centric in that mutations and onco­ structural variant signatures that may influence treat­
genic structural variants such as amplification of oncoge­ ment response (Fig. 4b,c). Many tumour suppressor genes
nes, deletions encompassing tumour suppressor genes have a phenotype only in the setting of biallelic inactiva­
and in-​frame oncogenic translocations are reported as tion of both gene copies (the Knudsen’s two-​hit hypothesis).
independent and unrelated events (Fig. 4a). Future NGS Therefore, the ability to assess for loss of heterozygosity
reports will seek to better place mutations in context resulting from deletion of the non-​mutated wild-​type
by reporting their clonality and potential functional allele may provide clinicians with insight as to whether
interaction with co-​occurrent driver mutations (Fig. 4b). targeting a particular tumour suppressor gene mutation
Some driver mutations arise early in tumour devel­ is likely to be effective167. Biallelic inactivation of tumour
Subclonal mutations
opment and are clonal158. Others arise later during suppressor genes can also result from dysregulation of
Mutations present in only
a fraction of a patient’s metastatic progression or as mediators of resistance to epigenetic modifiers such as promoter methylation168,
cancer cells. systemic therapies and are subclonal. For kinase inhib­ and therefore the clinical utility of analysing tumours
itors, antitumour activity is presumed to be greatest for only DNA-​mediated loss of heterozygosity events
Cancer cell fraction in patients in whom the targeted mutation is clonal, without the concurrent ability to detect epigenetically
The estimated fraction of
cancer cells that harbour
as targeting subclonal mutations would be expected to driven gene silencing remains an area of substantial
a specific mutation. result in rapid selection for cancer cells lacking the controversy.
drug-​sensitizing mutation. More broadly, cancers that Although less recognized, the allelic configuration of
Variant allele frequency have a high degree of tumour heterogeneity may be gain-​of-​function mutations can also influence oncogene
The fraction of mutant versus
less sensitive to a variety of systemic therapies includ­ dependence and drug sensitivity. For example, 12–15%
total sequencing reads at the
mutation locus. ing cytotoxic chemotherapies and immunotherapies, of PIK3CA-​mutated breast cancers harbour not one
as the existence of a larger number of genomically dis­ but two mutations in the same PIK3CA allele169. These
Allelic configuration tinct subclones before treatment initiation increases compound PIK3CA mutations activate PI3 kinase sig­
The number of mutant the odds that a therapy-​resistant clone already exists nalling to a greater degree than the individual PIK3CA
and wild-​type alleles, which
because of copy number gain
pre-​treatment159,160. mutations and are predictive of greater sensitivity to
or loss can be less than or Computational algorithms have been developed PI3Kα inhibitors169. Allelic imbalance resulting in gain
greater than two. that leverage the high depth of sequencing coverage of additional copies of mutated oncogenes or loss of the
generated by NGS-​based tumour profiling to convert wild-​type allele can also enhance clonal fitness or modu­
Knudsen’s two-​hit
variant allele frequencies into cancer cell fractions161,162, late sensitivity to targeted therapies167. For example, loss
hypothesis
The hypothesis, proposed by a numerical estimation of the fraction of cancer cells of the wild-​type BRAF allele in BRAF-​V600E-​mutant
Alfred Knudsen in 1971, that harbouring a particular mutation. For single-​nucleotide tumours may confer greater sensitivity to vemurafenib,
for tumour suppressor genes variants (SNVs), cancer cell fraction can be inferred from a selective RAF inhibitor167. A more precise estimation
that are recessive in nature, NGS data using estimates of variant allele frequency, of integer copy number from tumour NGS data could
in order for a phenotype to
manifest, both alleles must
tumour purity and local copy number. The accuracy of also help clinicians interpret the significance of amplifi­
be inactivated (biallelic cancer cell fraction estimates can, however, be affected cations involving targetable oncogenes, as higher levels
inactivation). This may be by tissue quality, sequencing depth and the breadth of of gene amplification may be associated with greater
achieved through multiple the sequencing panel163. The clinical utility of cancer cell oncogene dependence and drug sensitivity161.
mechanisms including
fraction for guiding treatment decisions remains poorly
deletions (either chromosomal
or subchromosomal), defined, and estimates of mutational clonality are not Co-​mutational context. Most tumour sequencing reports
epigenetic silencing routinely included in clinical tumour profiling reports. annotate point mutations, copy number alterations and
or mutation. A major concern over the use of clonality inferences is structural variants such as translocations independently
that the results may not provide an accurate reflection of each other. Co-​o ccurring mutations that confer
Loss of heterozygosity
A common form of allelic
of the patient’s current disease burden. For example, a drug resistance or enhance drug sensitivity are rarely
imbalance in which a drug-​sensitizing mutation that is subclonal in the sur­ highlighted as such. For example, mutations in PTEN
heterozygous somatic gically resected primary tumour site may be clonal in and NF1 have been shown to influence RAF inhibitor
alteration becomes all metastatic sites7. Mutational clonality can also vary response in BRAF-​mutated melanoma170–173. Conversely,
homozygous following
between metastatic sites164. Therefore, the use of clonal tumours with two or more alterations that activate the
loss of the wild-​type allele.
inferences derived from a single tumour biopsy sample same pathways, for example, concurrent TSC1 and NF2
Clonal fitness may give an incorrect assessment of the clonality of an alterations, both of which activate mTORC1, exhibit
The relative growth, survival or actionable mutation in the patient’s overall disease bur­ greater drug sensitivity than tumours with either
metastatic potential advantage den. Subclonal drug-​sensitizing mutations have also alteration alone61. Enhancements to clinical reporting
of a cancer cell clone over
other cancer cells within the
been shown to arise in parallel with sensitizing muta­ detailing epistatic relationships between two or more
tumour or non-​cancer cells. tions in other genes through the process of convergent co-​occurring mutations and highlighting whether two
The term fitness here derives evolution165. Therefore, although retrospective studies mutations in the same gene are present in cis or in trans
its origins from the concept suggest that pre-​existing tumour heterogeneity and may in the future help guide treatment selection or
of natural selection in
mutational clonality often influence drug response7,166, refine prognostication.
evolutionary biology.
how best to incorporate this knowledge into clinical
Integer copy number practice guidelines requires further study. Mutational signatures. Extending analyses beyond con­
The copy number of a gene sideration of individual genetic alterations to patterns
or localized DNA segment Allelic configuration. The methods used to infer the can­ of single, doublet and clustered base pair substitutions,
represented as an integer
value. For missense mutations,
cer cell fraction of somatic mutations from NGS-​based small insertions or deletions and larger structural
the number of mutated and tumour profiling data can provide information as to alterations is likely to provide additional insight into
wild-​type alleles in the cell. the allelic configuration of individual mutations or global the biological and clinical significance of mutations

Nature Reviews | Genetics

0123456789();:
 Reviews

a Allele level
Reference sequence

BRAF KIAA1549
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
A
Gain Loss A

Point mutation
Copy number alterations Indel Translocation breakpoint

b Allelic configuration
Heterozygous Whole-genome Other forms of
oncogenic variant Focal amplifications Genomic losses Copy-neutral LOH imbalance
duplication
Example

or
or

or
Wild type, mutant allele
1,1 1 , 19 0,1 0,2 2,2 0,3

c Structural variants
Simple structural variants Complex structural variants

Reference Tandem Foldback

Deletion Inversion Translocation Insertion Chromothrypsis
chromosome duplication inversion
12 21 13 142 1212 2112 4321

d Mutation signatures
Signature SBS4 20 Signature SBS6 20 Signature SBS7a
% Mutation type probability

10
15 15

10 10
5

5 5

0 0 0
C>A C>G C>T T>A T>C T>G C>A C>G C>T T>A T>C T>G C>A C>G C>T T>A T>C T>G

Fig. 4 | expanding the clinical utility of tumour genomic profiling. genomic variants and single base substitution signatures may provide
a | Current clinical tumour genomic reports are mutation centric with insight into the likely cause of a patient’s cancer or may be more predictive
mutations, small insertions/deletions (indels), gene amplifications, deletions of response to poly(ADP-​r ibose) polymerase (PARP) inhibitors or
and translocations reported as discrete events. b | Future next-​generation immunotherapy than single gene mutational status alone. For example,
sequencing (NGS) reports will seek to report additional genomic features single base substitution (SBS) signature SBS4 is associated with tobacco
such as allelic context (the number of mutant and wild-​type copies of an smoking and likely arises owing to DNA damage by mutagens such as
actionable cancer gene estimated to be present in the cancer cell), which benzo[a]pyrene present in tobacco smoke176, whereas SBS7a is common in
may influence drug response or patient prognosis. c | Broader sequencing skin cancers arising in sun-​exposed areas and is believed to be the result of
platforms such as whole-​genome sequencing can detect the presence of ultraviolet light-​mediated mutagenesis. Mutational signature SBS6 is
complex structural variants such as tandem duplications, foldback associated with defective DNA mismatch repair and is commonly found in
inversions or chromothripsis (thousands of clustered chromosomal tumours with microsatellite instability, which is predictive of response to the
rearrangements in a single genomic region). d | Global patterns of complex immune checkpoint inhibitor pembrolizumab.

www.nature.com/nrg

0123456789();:
 Reviews

in potentially actionable cancer genes174. By combin­ of novel therapies directed at new drug targets. Viewed
ing the six possible SNV classes together with their in this way, NGS-​based tumour profiling has the poten­
trinucleotide contexts, all SNVs can be classified into tial to reduce costs and improve patient outcomes by
1 of 96 possible combinations. Analysis of patterns replacing and improving upon more limited PCR, flu­
of these SNV substitutions has led to the identifica­ orescence in situ hybridization, immunohistochemistry
tion of 81 distinct SNV-​based mutational signatures, and small-​panel NGS-​based analytics.
many of which are linked to specific defects in DNA Improvements to tumour NGS assays, such as the
repair, or drug or toxin exposure175,176. The presence addition of newly identified cancer-​associated genes,
or absence of a mutational signature can aid in the are likely to be incremental and facilitated by declining
clinical interpretation of potentially actionable muta­ sequencing costs. These newer assay versions should
tions. For example, not all tumours with mutations in not need to demonstrate clinical utility. Instead, regu­
the Lynch syndrome-​associated genes MLH1, MSH2, lators should focus on ensuring that the assays used for
MSH6 and PMS2 have evidence of dMMR, but those patient care are accurate, reproducible and performed by
that do often have a distinctive pattern of single base qualified personnel. Such a nimble regulatory structure
substitutions (Fig. 4d). Integration of mutational signa­ would ensure quality while maximizing the potential
ture analysis with an analysis of microsatellite regions for innovation in this rapidly evolving field. A flexi­
using MSIsensor177 or other bioinformatic tools178,179 ble and adaptive approach to the regulation of clinical
can help define the biological and clinical significance tumour profiling assays would also allow individual
of mutations in MLH1, MSH2, MSH6 and PMS2 by hospital-​based and commercial laboratories to quickly
determining whether there is phenotypic evidence of adopt novel methods of DNA extraction, library prepa­
mismatch repair deficiency180. Such phenotypic char­ ration, sequencing or bioinformatic analysis, presuming
acterization of microsatellite instability is likely a more such changes do not negatively impact the sensitiv­
robust predictor of immunotherapy sensitivity than ity or specificity of detecting established biomarkers.
simply the presence or absence of a mutation in a Lynch In sum, regulatory agencies should seek to encourage
syndrome-​associated gene180. the development of more cost-​effective assays that are
The ability to robustly characterize mutational tissue-​efficient and capable of detecting all gene alter­
signatures may prove to be the most clinically signif­ ations and mutational signatures required to guide the
icant incremental benefit of WGS over targeted panel care of patients with cancer, recognizing that no single
sequencing assays. Structural alterations that have been assay developed to date is optimal for all applications in
associated with a poor prognosis or predictive of drug all cancer types.
response include focal tandem duplications resulting
from CDK12 loss-​of-​function mutations181, foldback Conclusion and perspectives
inversions and interstitial deletions that may be charac­ The continued decline in sequencing costs and the iden­
teristic of differential aberrant DNA repair processes in tification of new genomic biomarkers predictive of drug
ovarian cancer182, and reciprocal loss of heterozygosity, response has driven the rapid adoption of multigene
which is commonly observed in germ cell tumours and NGS-​based tumour genomic profiling panels as a com­
may be associated with cisplatin resistance183 (Fig. 4c). ponent of routine cancer care. In addition to the iden­
WGS-​scale assays can also robustly characterize sin­ tification of predictive biomarkers of therapy response,
gle and doublet base substitutions and small insertion tumour genomic profiling can identify somatic and
and deletion signatures that may be more predictive of germline mutations that refine or confirm a patient’s
immunotherapy or PARP inhibitor response than the cancer subtype diagnosis and provide clinicians with
mutational status of individual cancer genes181,182,184. insight into heritable cancer risk and the likelihood of
cancer recurrence and death. The future adoption
Balancing regulation and innovation of broader sequencing panels or WGS methods should
Although the quality assurance goals of governmental enable more accurate assessments of mutational clonal­
regulation of tumour genomic profiling assays is lauda­ ity, allelic context and the identification of mutational
ble, the paradigm of companion diagnostic tests as cur­ and structural signatures that are predictive of drug
rently implemented does not best facilitate the iterative response. The emergence of NGS-​b ased platforms
improvements needed to promote rapid innovation185,186. capable of detecting and analysing tumour-​derived
Of particular concern is the possibility that drug reim­ DNA in plasma will also allow the development of
bursement and drug access may in the future require diagnostic assays to screen high-​risk patients for occult
the use of a specific companion diagnostic test. While cancers, assess for minimal residual disease following
reimbursement for NGS-​based tumour genomic pro­ curative-​intent therapy, monitor and better quantify
filing in patients with cancer has been largely justified treatment response and assess for clonal evolution
to date by the need to screen for predictive biomarkers under the selection pressure of therapy as a prelude to
of drug response, these profiling assays are multipur­ the development of rational combination strategies that
pose diagnostic platforms that can also help ensure the prevent or delay drug resistance.
correct tumour subtype classification, inform prognosis Data-​sharing initiatives that allow for integration of
and determine whether a cancer arose as a result of a tumour genomic profiling data with detailed clinical
heritable cancer predisposition. Multigene NGS panels annotation and treatment response data, such as AACR
also enable the concurrent reporting of established and GENIE66, will be crucial in assessing the clinical utility of
investigational biomarkers, facilitating the development tumour genomic profiling, in particular, in rare cancer

Nature Reviews | Genetics

0123456789();:
Reviews

types. As not all mutations in the same gene have the mutational signatures. A significant expansion in the
same biological effect or clinical significance, improve­ fraction of patients with cancer who benefit from a pre­
ments to variant knowledge bases and clinical reporting cision medicine approach will also require the develop­
will be needed to help communicate to point-​of-​care ment of new therapies effective against tumours driven
clinicians the therapeutic, diagnostic and prognostic by currently undruggable oncogenic drivers.
relevance of individual mutations, epistatic interactions
between co-​o ccurrent mutations and genome-​wide Published online xx xx xxxx

1. Sobin, L. H. The international histological classification 24. Odegaard, J. I. et al. Validation of a plasma-​based 47. Yao, Z. et al. BRAF mutants evade ERK-​dependent
of tumours. Bull. World Health Organ. 59, 813–819 comprehensive cancer genotyping assay utilizing feedback by different mechanisms that determine
(1981). orthogonal tissue- and plasma-​based methodologies. their sensitivity to pharmacologic inhibition.
2. Hoadley, K. A. et al. Multiplatform analysis of 12 Clin. Cancer Res. 24, 3539–3549 (2018). Cancer Cell 28, 370–383 (2015).
cancer types reveals molecular classification within 25. Brannon, A. R. et al. Enhanced specificity of high 48. Zabransky, D. J. et al. HER2 missense mutations have
and across tissues of origin. Cell 158, 929–944 sensitivity somatic variant profiling in cell-​free DNA distinct effects on oncogenic signaling and migration.
(2014). via paired normal sequencing: design, validation, and Proc. Natl Acad. Sci. USA 112, E6205–E6214
3. Bouwman, P. & Jonkers, J. The effects of deregulated clinical experience of the MSK-​ACCESS liquid biopsy (2015).
DNA damage signalling on cancer chemotherapy assay. Preprint at bioRxiv https://doi.org/10.21203/ 49. Poulikakos, P. I., Zhang, C., Bollag, G., Shokat, K. M.
response and resistance. Nat. Rev. Cancer 12, rs.3.rs-120695/v1 (2020). & Rosen, N. RAF inhibitors transactivate RAF dimers
587–598 (2012). 26. Tao, J. J., Schram, A. M. & Hyman, D. M. Basket and ERK signalling in cells with wild-​type BRAF.
4. Yarchoan, M., Hopkins, A. & Jaffee, E. M. Tumor studies: redefining clinical trials in the era of genome-​ Nature 464, 427–430 (2010).
mutational burden and response rate to PD-1 driven oncology. Annu. Rev. Med. 69, 319–331 50. Kopetz, S. et al. Encorafenib, binimetinib, and
inhibition. N. Engl. J. Med. 377, 2500–2501 (2017). (2018). cetuximab in BRAF V600E-​mutated colorectal cancer.
5. Hyman, D. M. et al. Vemurafenib in multiple 27. Redig, A. J. & Jänne, P. A. Basket trials and the N. Engl. J. Med. 381, 1632–1643 (2019).
nonmelanoma cancers with BRAF V600 mutations. evolution of clinical trial design in an era of genomic 51. Corcoran, R. B. et al. Combined BRAF, EGFR, and
N. Engl. J. Med. 373, 726–736 (2015). medicine. J. Clin. Oncol. 33, 975–977 (2015). MEK inhibition in patients with BRAFV600E-​mutant
6. Drilon, A. et al. Efficacy of larotrectinib in TRK fusion-​ 28. Woodcock, J. & LaVange, L. M. Master protocols to colorectal cancer. Cancer Discov. 8, 428–443 (2018).
positive cancers in adults and children. N. Engl. J. Med. study multiple therapies, multiple diseases, or both. 52. Gray, S. W., Hicks-​Courant, K., Cronin, A., Rollins, B. J.
378, 731–739 (2018). N. Engl. J. Med. 377, 62–70 (2017). & Weeks, J. C. Physicians’ attitudes about multiplex
7. Hyman, D. M. et al. AKT inhibition in solid tumors 29. Le, D. T. et al. PD-1 blockade in tumors with tumor genomic testing. J. Clin. Oncol. 32, 1317–1323
with AKT1 mutations. J. Clin. Oncol. 35, 2251–2259 mismatch-​repair deficiency. N. Engl. J. Med. 372, (2014).
(2017). 2509–2520 (2015). 53. Schram, A. M. et al. Oncologist use and perception
8. Hungerford, D. A. & Nowell, P. C. A minute 30. Drilon, A. E. et al. A phase II basket study of the oral of large panel next-​generation tumor sequencing.
chromosome in human chronic granulocytic leukemia. TRK inhibitor LOXO-101 in adult subjects with NTRK Ann. Oncol. 28, 2298–2304 (2017).
Science 132, 1497–1499 (1960). fusion-​positive tumors. J. Clin. Oncol. 34, 54. Wu, J.-Y. et al. Lung cancer with epidermal growth
9. de Klein, A. et al. A cellular oncogene is translocated TPS2599–TPS2599 (2016). factor receptor exon 20 mutations is associated with
to the Philadelphia chromosome in chronic myelocytic 31. Berger, M. F. & Mardis, E. R. The emerging clinical poor gefitinib treatment response. Clin. Cancer Res.
leukaemia. Nature 300, 765–767 (1982). relevance of genomics in cancer medicine. Nat. Rev. 14, 4877–4882 (2008).
10. Kantarjian, H. et al. Improved survival in chronic Clin. Oncol. 15, 353–365 (2018). 55. Chakravarty, D. et al. OncoKB: a precision oncology
myeloid leukemia since the introduction of imatinib 32. Vogelstein, B. et al. Cancer genome landscapes. knowledge base. JCO Precis. Oncol. https://doi.org/
therapy: a single-​institution historical experience. Science 339, 1546–1558 (2013). 10.1200/PO.17.00011 (2017).
Blood 119, 1981–1987 (2012). 33. Bailey, M. H. et al. Comprehensive characterization 56. Griffith, M. et al. CIViC is a community knowledgebase
11. Prasad, V. Perspective: the precision-​oncology illusion. of cancer driver genes and mutations. Cell 173, for expert crowdsourcing the clinical interpretation of
Nature 537, S63 (2016). 371–385.e18 (2018). variants in cancer. Nat. Genet. 49, 170–174 (2017).
12. Hyman, D. M., Taylor, B. S. & Baselga, J. Implementing 34. Farmer, H. et al. Targeting the DNA repair defect in 57. Patterson, S. E. et al. The clinical trial landscape in
genome-​driven oncology. Cell 168, 584–599 (2017). BRCA mutant cells as a therapeutic strategy. Nature oncology and connectivity of somatic mutational
13. Pan, Y. et al. ALK, ROS1 and RET fusions in 1139 lung 434, 917–921 (2005). profiles to targeted therapies. Hum. Genomics 10, 4
adenocarcinomas: a comprehensive study of common 35. Mateo, J. et al. DNA-​repair defects and olaparib in (2016).
and fusion pattern-​specific clinicopathologic, histologic metastatic prostate cancer. N. Engl. J. Med. 373, 58. Huang, L. et al. The cancer precision medicine
and cytologic features. Lung Cancer 84, 121–126 1697–1708 (2015). knowledge base for structured clinical-​grade
(2014). 36. Van Allen, E. M. et al. Somatic ERCC2 mutations mutations and interpretations. J. Am. Med. Inform.
14. Mansfield, E. A. FDA perspective on companion correlate with cisplatin sensitivity in muscle-​invasive Assoc. 24, 513–519 (2017).
diagnostics: an evolving paradigm. Clin. Cancer Res. urothelial carcinoma. Cancer Discov. 4, 1140–1153 59. Tamborero, D. et al. Cancer genome interpreter
20, 1453–1457 (2014). (2014). annotates the biological and clinical relevance of
15. Kris, M. G. et al. Using multiplexed assays of 37. Li, Q. et al. ERCC2 helicase domain mutations confer tumor alterations. Genome Med. 10, 25 (2018).
oncogenic drivers in lung cancers to select targeted nucleotide excision repair deficiency and drive 60. Dumbrava, E. I. & Meric-​Bernstam, F. Personalized
drugs. JAMA 311, 1998–2006 (2014). cisplatin sensitivity in muscle-​invasive bladder cancer. cancer therapy-​leveraging a knowledge base for
16. Jordan, E. J. et al. Prospective comprehensive Clin. Cancer Res. 25, 977–988 (2019). clinical decision-​making. Cold Spring Harb. Mol. Case
molecular characterization of lung adenocarcinomas 38. Kelderman, S., Schumacher, T. N. & Kvistborg, P. Stud. 4, a001578 (2018).
for efficient patient matching to approved and Mismatch repair-​deficient cancers are targets for 61. Iyer, G. et al. Genome sequencing identifies a basis for
emerging therapies. Cancer Discov. 7, 596–609 anti-​PD-1 therapy. Cancer Cell 28, 11–13 (2015). everolimus sensitivity. Science 338, 221 (2012).
(2017). 39. Chang, M. T. et al. Identifying recurrent mutations 62. Ross, J. S. et al. Comprehensive genomic profiling
17. Su, Z. et al. A platform for rapid detection of multiple in cancer reveals widespread lineage diversity and of carcinoma of unknown primary site: new routes
oncogenic mutations with relevance to targeted mutational specificity. Nat. Biotechnol. 34, 155–163 to targeted therapies. JAMA Oncol. 1, 40–49
therapy in non-​small-cell lung cancer. J. Mol. Diagn. (2016). (2015).
13, 74–84 (2011). 40. Chang, M. T. et al. Accelerating discovery of functional 63. Richards, S. et al. Standards and guidelines for the
18. MacConaill, L. E. et al. Prospective enterprise-​level mutant alleles in cancer. Cancer Discov. 8, 174–183 interpretation of sequence variants: a joint consensus
molecular genotyping of a cohort of cancer patients. (2018). recommendation of the American College of Medical
J. Mol. Diagn. 16, 660–672 (2014). 41. Hanrahan, A. J. et al. Leveraging systematic functional Genetics and Genomics and the Association for
19. Li, T., Kung, H.-J., Mack, P. C. & Gandara, D. R. analysis to benchmark an in silico framework Molecular Pathology. Genet. Med. 17, 405–424
Genotyping and genomic profiling of non-​small-cell distinguishes driver from passenger MEK mutants (2015).
lung cancer: implications for current and future in cancer. Cancer Res. 80, 4233–301 (2020). 64. Li, M. M. et al. Standards and guidelines for the
therapies. J. Clin. Oncol. 31, 1039–1049 (2013). 42. Hess, J. M. et al. Passenger hotspot mutations in interpretation and reporting of sequence variants
20. Zehir, A. et al. Mutational landscape of metastatic cancer. Cancer Cell 36, 288–301.e14 (2019). in cancer: a joint consensus recommendation of the
cancer revealed from prospective clinical sequencing 43. Buisson, R. et al. Passenger hotspot mutations in Association for Molecular Pathology, American Society
of 10,000 patients. Nat. Med. 23, 703–713 (2017). cancer driven by APOBEC3A and mesoscale genomic of Clinical Oncology, and College of American
21. Sholl, L. M. et al. Institutional implementation of features. Science 364, eaaw2872 (2019). Pathologists. J. Mol. Diagn. 19, 4–23 (2017).
clinical tumor profiling on an unselected cancer 44. Holbrook, J. A., Neu-​Yilik, G., Hentze, M. W. & 65. Mateo, J. et al. A framework to rank genomic
population. JCI Insight 1, e87062 (2016). Kulozik, A. E. Nonsense-​mediated decay approaches alterations as targets for cancer precision medicine:
22. Meric-​Bernstam, F. et al. Feasibility of large-​scale the clinic. Nat. Genet. 36, 801–808 (2004). the ESMO scale for clinical actionability of molecular
genomic testing to facilitate enrollment onto 45. Cheung, L. W. T. et al. High frequency of PIK3R1 and targets (ESCAT). Ann. Oncol. 29, 1895–1902 (2018).
genomically matched clinical trials. J. Clin. Oncol. 33, PIK3R2 mutations in endometrial cancer elucidates 66. AACR Project GENIE Consortium. AACR Project
2753–2762 (2015). a novel mechanism for regulation of PTEN protein GENIE: powering precision medicine through an
23. Stockley, T. L. et al. Molecular profiling of advanced stability. Cancer Discov. 1, 170–185 (2011). international consortium. Cancer Discov. 7, 818–831
solid tumors and patient outcomes with genotype-​ 46. Cheung, L. W. T. et al. Naturally occurring neomorphic (2017).
matched clinical trials: the Princess Margaret PIK3R1 mutations activate the MAPK pathway, 67. Wagner, A. H. et al. A harmonized meta-​knowledgebase
IMPACT/COMPACT trial. Genome Med. 8, 109 dictating therapeutic response to MAPK pathway of clinical interpretations of somatic genomic variants
(2016). inhibitors. Cancer Cell 26, 479–494 (2014). in cancer. Nat. Genet. 52, 448–457 (2020).

www.nature.com/nrg

0123456789();:
 Reviews

68. Ritter, D. I. et al. Somatic cancer variant curation and early-​onset breast cancer patients among Ashkenazi 117. Schneid, T. in Discrimination Law Issues for the
harmonization through consensus minimum variant women. Am. J. Hum. Genet. 60, 505–514 (1997). Safety Professional (ed. Schneid, T.) 161–194
level data. Genome Med. 8, 117 (2016). 92. Newman, B., Austin, M. A., Lee, M. & King, M. C. (CRC Press, 2011).
69. Center for Devices & Radiological Health. FDA Inheritance of human breast cancer: evidence for 118. National Human Genome Research Institute. Genetic
recognition of public human genetic variant autosomal dominant transmission in high-​risk families. Information Nondiscrimination Act (GINA) of 2008.
databases. FDA https://www.fda.gov/medical-​devices/ Proc. Natl Acad. Sci. USA 85, 3044–3048 (1988). NIH https://www.genome.gov/24519851/genetic-​
precision-​medicine/fda-​recognition-public-​human- 93. Hall, J. M. et al. Linkage of early-​onset familial information-nondiscrimination-​act-of-2008 (2008).
genetic-​variant-databases (2019). breast cancer to chromosome 17q21. Science 250, 119. Gniady, J. A. Regulating direct-​to-consumer genetic
70. Stein, E. M. et al. Enasidenib in mutant IDH2 relapsed 1684–1689 (1990). testing: protecting the consumer without quashing
or refractory acute myeloid leukemia. Blood 130, 94. King, M.-C. ‘The race’ to clone BRCA1. Science 343, a medical revolution. Fordham Law Rev. 76,
722–731 (2017). 1462–1465 (2014). 2429–2475 (2008).
71. Pollyea, D. A. et al. Enasidenib, an inhibitor of mutant 95. Miki, Y. et al. A strong candidate for the breast and 120. Ferreira-​Gonzalez, A. et al. US system of oversight for
IDH2 proteins, induces durable remissions in older ovarian cancer susceptibility gene BRCA1. Science genetic testing: a report from the Secretary’s Advisory
patients with newly diagnosed acute myeloid 266, 66–71 (1994). Committee on Genetics, Health and Society. Per. Med.
leukemia. Leukemia 33, 2575–2584 (2019). 96. Mandelker, D. et al. Mutation detection in patients 5, 521–528 (2008).
72. DiNardo, C. D. et al. Durable remissions with with advanced cancer by universal sequencing of 121. Lolkema, M. P. et al. Ethical, legal, and counseling
ivosidenib in IDH1-mutated relapsed or refractory cancer-​related genes in tumor and normal DNA vs challenges surrounding the return of genetic results
AML. N. Engl. J. Med. 378, 2386–2398 (2018). guideline-​based germline testing. JAMA 318, in oncology. J. Clin. Oncol. 31, 1842–1848 (2013).
73. André, F. et al. Alpelisib for PIK3CA-​mutated, hormone 825–835 (2017). 122. Li, M. M. et al. Points to consider for reporting of
receptor-​positive advanced breast cancer. N. Engl. 97. Garofalo, A. et al. The impact of tumor profiling germline variation in patients undergoing tumor
J. Med. 380, 1929–1940 (2019). approaches and genomic data strategies for cancer testing: a statement of the American College of
74. Loriot, Y. et al. Erdafitinib in locally advanced or precision medicine. Genome Med. 8, 79 (2016). Medical Genetics and Genomics (ACMG). Genet. Med.
metastatic urothelial carcinoma. N. Engl. J. Med. 381, 98. Jones, S. et al. Personalized genomic analyses for 22, 1142–1148 (2020).
338–348 (2019). cancer mutation discovery and interpretation. 123. Kurian, A. W. et al. Genetic testing and counseling
75. Abou-​Alfa, G. K. et al. Pemigatinib for Sci. Transl. Med. 7, 283ra53 (2015). among patients with newly diagnosed breast cancer.
previously treated, locally advanced or metastatic 99. Zhang, J. et al. Germline mutations in predisposition JAMA 317, 531–534 (2017).
cholangiocarcinoma: a multicentre, open-​label, phase genes in pediatric cancer. N. Engl. J. Med. 373, 124. McNamara, D. Shortage of genetic counselors in face
2 study. Lancet Oncol. 21, 671–684 (2020). 2336–2346 (2015). of growing need. Medscape https://www.medscape.
76. Drilon, A. et al. PL02.08 registrational results of 100. Schrader, K. A. et al. Germline variants in targeted com/viewarticle/877135 (2017).
LIBRETTO-001: a phase 1/2 trial of LOXO-292 in tumor sequencing using matched normal DNA. 125. Eisen, A. et al. Genetic assessment wait time
patients with RET fusion-​positive lung cancers. JAMA Oncol. 2, 104–111 (2016). indicators in the high risk ontario breast screening
J. Thorac. Oncol. 14, S6–S7 (2019). 101. Domchek, S. M. Germline genetic testing for breast program. Mol. Genet. Genomic Med. 6, 213–223
77. Wirth, L. et al. LBA93 - registrational results of cancer: Which patients? What genes? Genet. Med. 22, (2018).
LOXO-292 in patients with RET-​altered thyroid 698–700 (2020). 126. Culver, J. O., Hull, J. L., Dunne, D. F. & Burke, W.
cancers. Ann. Oncol. 30, v933 (2019). 102. Konstantinopoulos, P. A. et al. Germline and somatic Oncologists’ opinions on genetic testing for breast and
78. Litton, J. K. et al. Talazoparib in patients with tumor testing in epithelial ovarian cancer: ASCO ovarian cancer. Genet. Med. 3, 120–125 (2001).
advanced breast cancer and a germline BRCA guideline. J. Clin. Oncol. 38, 1222–1245 (2020). 127. Teng, I. & Spigelman, A. Attitudes and knowledge of
mutation. N. Engl. J. Med. 379, 753–763 (2018). 103. McLeod, H. L. Cancer pharmacogenomics: early medical practitioners to hereditary cancer clinics and
79. Golan, T. et al. Maintenance olaparib for germline promise, but concerted effort needed. Science 339, cancer genetic testing. Fam. Cancer 13, 311–324
BRCA-​mutated metastatic pancreatic cancer. 1563–1566 (2013). (2014).
N. Engl. J. Med. 381, 317–327 (2019). 104. Wang, L., McLeod, H. L. & Weinshilboum, R. M. 128. Teer, J. K. et al. Evaluating somatic tumor mutation
80. Mateo, J. et al. Olaparib in patients with metastatic Genomics and drug response. N. Engl. J. Med. 364, detection without matched normal samples.
castration-​resistant prostate cancer with DNA repair 1144–1153 (2011). Hum. Genomics 11, 22 (2017).
gene aberrations (TOPARP-​B): a multicentre, open-​ 105. Irvin, W. J. Jr. et al. Genotype-​guided tamoxifen dosing 129. Damodaran, S., Berger, M. F. & Roychowdhury, S.
label, randomised, phase 2 trial. Lancet Oncol. 21, increases active metabolite exposure in women with Clinical tumor sequencing: opportunities and
162–174 (2020). reduced CYP2D6 metabolism: a multicenter study. challenges for precision cancer medicine. Am. Soc.
81. Planchard, D. et al. Dabrafenib plus trametinib in J. Clin. Oncol. 29, 3232–3239 (2011). Clin. Oncol. Educ. Book 35, e175–e182 (2015).
patients with previously untreated BRAFV600E-​ 106. Hertz, D. L. et al. CYP2C8* 3 predicts benefit/risk 130. Robinson, D. R. et al. Integrative clinical genomics of
mutant metastatic non-​small-cell lung cancer: an open-​ profile in breast cancer patients receiving neoadjuvant metastatic cancer. Nature 548, 297–303 (2017).
label, phase 2 trial. Lancet Oncol. 18, 1307–1316 paclitaxel. Breast Cancer Res. Treat. 134, 401–410 131. Van Allen, E. M. et al. Whole-​exome sequencing and
(2017). (2012). clinical interpretation of formalin-​fixed, paraffin-​
82. Diamond, E. L. et al. Vemurafenib for BRAF V600- 107. Pullarkat, S. T. et al. Thymidylate synthase gene embedded tumor samples to guide precision cancer
mutant erdheim-​chester disease and langerhans cell polymorphism determines response and toxicity of medicine. Nat. Med. 20, 682–688 (2014).
histiocytosis: analysis of data from the histology-​ 5-FU chemotherapy. Pharmacogenomics J. 1, 65–70 132. Van Allen, E. M. et al. A comparative assessment
independent, phase 2, open-​label VE-​BASKET study. (2001). of clinical whole exome and transcriptome profiling
JAMA Oncol. 4, 384–388 (2018). 108. Coombs, C. C. et al. Therapy-​related clonal across sequencing centers: implications for precision
83. Subbiah, V. et al. Dabrafenib and trametinib hematopoiesis in patients with non-​hematologic cancer medicine. Oncotarget 7, 52888–52899
treatment in patients with locally advanced or cancers is common and associated with adverse (2016).
metastatic BRAF V600-mutant anaplastic thyroid clinical outcomes. Cell Stem Cell 21, 374–382.e4 133. Robbe, P. et al. Clinical whole-​genome sequencing
cancer. J. Clin. Oncol. 36, 7–13 (2018). (2017). from routine formalin-​fixed, paraffin-​embedded
84. Goodman, A. M. et al. Tumor mutational burden as an 109. Ptashkin, R. N. et al. Prevalence of clonal specimens: pilot study for the 100,000 Genomes
independent predictor of response to immunotherapy hematopoiesis mutations in tumor-​only clinical Project. Genet. Med. 20, 1196–1205 (2018).
in diverse cancers. Mol. Cancer Ther. 16, 2598–2608 genomic profiling of solid tumors. JAMA Oncol. 4, 134. Staaf, J. et al. Whole-​genome sequencing of triple-​
(2017). 1589–1593 (2018). negative breast cancers in a population-​based clinical
85. Amado, R. G. et al. Wild-​type KRAS is required for 110. Razavi, P. et al. High-​intensity sequencing reveals the study. Nat. Med. 25, 1526–1533 (2019).
panitumumab efficacy in patients with metastatic sources of plasma circulating cell-​free DNA variants. 135. ICGC/TCGA Pan-​Cancer Analysis of Whole Genomes
colorectal cancer. J. Clin. Oncol. 26, 1626–1634 Nat. Med. 25, 1928–1937 (2019). Consortium. Pan-​cancer analysis of whole genomes.
(2008). 111. Carpenter, E. L. & Mossé, Y. P. Targeting ALK in Nature 578, 82–93 (2020).
86. De Roock, W. et al. Effects of KRAS, BRAF, NRAS, neuroblastoma — preclinical and clinical advancements. 136. Conesa, A. et al. A survey of best practices for
and PIK3CA mutations on the efficacy of cetuximab Nat. Rev. Clin. Oncol. 9, 391–399 (2012). RNA-​seq data analysis. Genome Biol. 17, 13 (2016).
plus chemotherapy in chemotherapy-​refractory 112. Wirth, L. J. et al. 1922P Exploratory patient-​reported 137. Zheng, Z. et al. Anchored multiplex PCR for targeted
metastatic colorectal cancer: a retrospective outcomes among patients with RET-​mutant medullary next-​generation sequencing. Nat. Med. 20,
consortium analysis. Lancet Oncol. 11, 753–762 thyroid cancer in LIBRETTO-001: a phase I/II trial of 1479–1484 (2014).
(2010). selpercatinib (LOXO-292). Ann. Oncol. 31 (Suppl. 4), 138. Cieslik, M. et al. The use of exome capture RNA-​seq
87. Chung, V. et al. Effect of selumetinib and MK-2206 vs S1089 (2020). for highly degraded RNA with application to clinical
oxaliplatin and fluorouracil in patients with metastatic 113. Wells, S. A. Jr. et al. Vandetanib in patients with locally cancer sequencing. Genome Res. 25, 1372–1381
pancreatic cancer after prior therapy: SWOG S1115 advanced or metastatic medullary thyroid cancer: a (2015).
study randomized clinical trial. JAMA Oncol. 3, randomized, double-​blind phase III trial. J. Clin. Oncol. 139. Capper, D. et al. DNA methylation-​based classification
516–522 (2017). 30, 134–141 (2012). of central nervous system tumours. Nature 555,
88. Jänne, P. A. et al. Selumetinib plus docetaxel 114. Yurgelun, M. B. & Hampel, H. Recent advances in 469–474 (2018).
compared with docetaxel alone and progression-​free Lynch syndrome: diagnosis, treatment, and cancer 140. Miller, A. M. et al. Tracking tumour evolution in glioma
survival in patients with KRAS-​mutant advanced prevention. Am. Soc. Clin. Oncol. Educ. Book 38, through liquid biopsies of cerebrospinal fluid. Nature
non-​small cell lung cancer: the SELECT-1 randomized 101–109 (2018). 565, 654–658 (2019).
clinical trial. JAMA 317, 1844–1853 (2017). 115. Rodrigues, M. et al. Outlier response to anti-​PD1 in 141. Goto, K. et al. Epidermal growth factor receptor
89. Canon, J. et al. The clinical KRAS(G12C) inhibitor uveal melanoma reveals germline MBD4 mutations mutation status in circulating free DNA in serum: from
AMG 510 drives anti-​tumour immunity. Nature 575, in hypermutated tumors. Nat. Commun. 9, 1866 IPASS, a phase III study of gefitinib or carboplatin/
217–223 (2019). (2018). paclitaxel in non-​small cell lung cancer. J. Thorac.
90. Tonin, P. et al. BRCA1 mutations in Ashkenazi Jewish 116. Johansson, P. A. et al. Correction to: Prolonged Oncol. 7, 115–121 (2012).
women. Am. J. Hum. Genet. 57, 189 (1995). stable disease in a uveal melanoma patient with 142. Husain, H. et al. Cell-​free DNA from ascites and
91. Abeliovich, D. et al. The founder mutations 185delAG germline MBD4 nonsense mutation treated pleural effusions: molecular insights into genomic
and 5382insC in BRCA1 and 6174delT in BRCA2 with pembrolizumab and ipilimumab. Immunogenetics aberrations and disease biology. Mol. Cancer Ther.
appear in 60% of ovarian cancer and 30% of 71, 511 (2019). 16, 948–955 (2017).

Nature Reviews | Genetics

0123456789();:
Reviews

143. Chang, H. W. et al. Urinary cell-​free DNA as a potential 171. Xing, F. et al. Concurrent loss of the PTEN and RB1 196. Davies, K. D. et al. Identifying and targeting ROS1
tumor marker for bladder cancer. Int. J. Biol. Markers tumor suppressors attenuates RAF dependence in gene fusions in non-​small cell lung cancer. Clin. Cancer
22, 287–294 (2007). melanomas harboring (V600E)BRAF. Oncogene 31, Res. 18, 4570–4579 (2012).
144. Diehl, F. et al. Circulating mutant DNA to assess tumor 446–457 (2012). 197. Takeuchi, K. et al. RET, ROS1 and ALK fusions in lung
dynamics. Nat. Med. 14, 985–990 (2008). 172. Whittaker, S. R. et al. A genome-​scale RNA cancer. Nat. Med. 18, 378–381 (2012).
145. Bettegowda, C. et al. Detection of circulating tumor interference screen implicates NF1 loss in resistance 198. Gautschi, O. et al. Targeted therapy for patients with
DNA in early- and late-​stage human malignancies. to RAF inhibition. Cancer Discov. 3, 350–362 (2013). BRAF-​mutant lung cancer: results from the European
Sci. Transl. Med. 6, 224ra24 (2014). 173. Nissan, M. H. et al. Loss of NF1 in cutaneous EURAF cohort. J. Thorac. Oncol. 10, 1451–1457
146. Lecomte, T. et al. Detection of free-​circulating melanoma is associated with RAS activation and (2015).
tumor-​associated DNA in plasma of colorectal MEK dependence. Cancer Res. 74, 2340–2350 199. Stephens, P. et al. Lung cancer: intragenic ERBB2
cancer patients and its association with prognosis. (2014). kinase mutations in tumours. Nature 431, 525–526
Int. J. Cancer 100, 542–548 (2002). 174. Alexandrov, L. B. & Stratton, M. R. Mutational (2004).
147. Murtaza, M. et al. Non-​invasive analysis of acquired signatures: the patterns of somatic mutations hidden 200. Shimamura, T. et al. Non-​small-cell lung cancer
resistance to cancer therapy by sequencing of plasma in cancer genomes. Curr. Opin. Genet. Dev. 24, 52–60 and Ba/F3 transformed cells harboring the ERBB2
DNA. Nature 497, 108–112 (2013). (2014). G776insV_G/C mutation are sensitive to the
148. Dawson, S.-J. et al. Analysis of circulating tumor DNA 175. Alexandrov, L. B. et al. Signatures of mutational dual-​specific epidermal growth factor receptor
to monitor metastatic breast cancer. N. Engl. J. Med. processes in human cancer. Nature 500, 415–421 and ERBB2 inhibitor HKI-272. Cancer Res. 66,
368, 1199–1209 (2013). (2013). 6487–6491 (2006).
149. Garcia-​Murillas, I. et al. Mutation tracking in 176. Alexandrov, L. B. et al. The repertoire of mutational 201. The Cancer Genome Atlas Research Network.
circulating tumor DNA predicts relapse in early signatures in human cancer. Nature 578, 94–101 Comprehensive molecular profiling of lung
breast cancer. Sci. Transl. Med. 7, 302ra133 (2015). (2020). adenocarcinoma. Nature 511, 543–550 (2014).
150. Kinde, I., Wu, J., Papadopoulos, N., Kinzler, K. W. 177. Niu, B. et al. MSIsensor: microsatellite instability 202. Paik, P. K. et al. Response to MET inhibitors in
& Vogelstein, B. Detection and quantification of detection using paired tumor-​normal sequence data. patients with stage IV lung adenocarcinomas
rare mutations with massively parallel sequencing. Bioinformatics 30, 1015–1016 (2014). harboring MET mutations causing exon 14 skipping.
Proc. Natl Acad. Sci. USA 108, 9530–9535 (2011). 178. Escudié, F. et al. MIAmS: microsatellite instability Cancer Discov. 8, 842–849 (2015).
151. Oxnard, G. R. et al. Association between plasma detection on NGS amplicons data. Bioinformatics 203. Govindan, R. et al. Genomic landscape of non-​small
genotyping and outcomes of treatment with https://doi.org/10.1093/bioinformatics/btz797 cell lung cancer in smokers and never-smokers. Cell
osimertinib (AZD9291) in advanced non-​small-cell (2019). 150, 1121–1134 (2012).
lung cancer. J. Clin. Oncol. 34, 3375–3382 (2016). 179. Huang, M. N. et al. MSIseq: software for assessing 204. Douillard, J.-Y. et al. Randomized, phase III
152. Shaw, J. A. et al. Mutation analysis of cell-​free DNA microsatellite instability from catalogs of somatic trial of panitumumab with infusional fluorouracil,
and single circulating tumor cells in metastatic breast mutations. Sci. Rep. 5, 13321 (2015). leucovorin, and oxaliplatin (FOLFOX4) versus
cancer patients with high circulating tumor cell counts. 180. Abida, W. et al. Analysis of the prevalence of FOLFOX4 alone as first-​line treatment in patients
Clin. Cancer Res. 23, 88–96 (2017). microsatellite instability in prostate cancer and with previously untreated metastatic colorectal cancer:
153. Goodall, J. et al. Circulating cell-​free DNA to guide response to immune checkpoint blockade. the PRIME study. J. Clin. Oncol. 28, 4697–4705
prostate cancer treatment with PARP inhibition. JAMA Oncol. 5, 471–478 (2019). (2010).
Cancer Discov. 7, 1006–1017 (2017). 181. Wu, Y.-M. et al. Inactivation of CDK12 delineates 205. Bokemeyer, C. et al. Efficacy according to biomarker
154. Reinert, T. et al. Analysis of plasma cell-​free DNA by a distinct immunogenic class of advanced prostate status of cetuximab plus FOLFOX-4 as first-​line
ultradeep sequencing in patients with stages I to III cancer. Cell 173, 1770–1782.e14 (2018). treatment for metastatic colorectal cancer: the OPUS
colorectal cancer. JAMA Oncol. 5, 1124–1131 (2019). 182. Wang, Y. K. et al. Genomic consequences of aberrant study. Ann. Oncol. 22, 1535–1546 (2011).
155. Christensen, E. et al. Early detection of metastatic DNA repair mechanisms stratify ovarian cancer 206. Joseph, E. W. et al. The RAF inhibitor PLX4032
relapse and monitoring of therapeutic efficacy by ultra-​ histotypes. Nat. Genet. 49, 856–865 (2017). inhibits ERK signaling and tumor cell proliferation
deep sequencing of plasma cell-​free DNA in patients 183. Taylor-​Weiner, A. et al. Genomic evolution and in a V600E BRAF-​selective manner. Proc. Natl Acad.
with urothelial bladder carcinoma. J. Clin. Oncol. 37, chemoresistance in germ-​cell tumours. Nature 540, Sci. USA 107, 14903–14908 (2010).
1547–1557 (2019). 114–118 (2016). 207. Su, F. et al. RAS mutations in cutaneous squamous-​cell
156. Coombes, R. C. et al. Personalized detection of 184. Angus, L. et al. The genomic landscape of metastatic carcinomas in patients treated with BRAF inhibitors.
circulating tumor DNA antedates breast cancer breast cancer highlights changes in mutation and N. Engl. J. Med. 366, 207–215 (2012).
metastatic recurrence. Clin. Cancer Res. 25, signature frequencies. Nat. Genet. 51, 1450–1458 208. Callahan, M. K. et al. Progression of RAS-​mutant
4255–4263 (2019). (2019). leukemia during RAF inhibitor treatment. N. Engl.
157. Hao, X. et al. DNA methylation markers for diagnosis 185. Salgado, R. et al. Addressing the dichotomy between J. Med. 367, 2316–2321 (2012).
and prognosis of common cancers. Proc. Natl Acad. individual and societal approaches to personalised 209. Sanchez-​Laorden, B. et al. BRAF inhibitors induce
Sci. USA 114, 7414–7419 (2017). medicine in oncology. Eur. J. Cancer 114, 128–136 metastasis in RAS mutant or inhibitor-​resistant
158. Nowell, P. C. The clonal evolution of tumor cell (2019). melanoma cells by reactivating MEK and ERK
populations. Science 194, 23–28 (1976). 186. Salgado, R. et al. How current assay approval policies signaling. Sci. Signal. 7, ra30 (2014).
159. McGranahan, N. et al. Clonal neoantigens elicit are leading to unintended imprecision medicine. 210. Samstein, R. M. et al. Tumor mutational load predicts
T cell immunoreactivity and sensitivity to immune Lancet Oncol. 21, 1399–1401 (2020). survival after immunotherapy across multiple cancer
checkpoint blockade. Science 351, 1463–1469 (2016). 187. Rosell, R. et al. Erlotinib versus standard types. Nat. Genet. 51, 202–206 (2019).
160. Landau, D. A. et al. Evolution and impact of subclonal chemotherapy as first-​line treatment for European 211. Marabelle, A. et al. Association of tumour mutational
mutations in chronic lymphocytic leukemia. Cell 152, patients with advanced EGFR mutation-​positive burden with outcomes in patients with advanced solid
714–726 (2013). non-​small-cell lung cancer (EURTAC): a multicentre, tumours treated with pembrolizumab: prospective
161. Carter, S. L. et al. Absolute quantification of somatic open-​label, randomised phase 3 trial. Lancet Oncol. biomarker analysis of the multicohort, open-​label,
DNA alterations in human cancer. Nat. Biotechnol. 13, 239–246 (2012). phase 2 KEYNOTE-158 study. Lancet Oncol. 21,
30, 413–421 (2012). 188. Mok, T. S. et al. Gefitinib or carboplatin-​paclitaxel in 1353–1365 (2020).
162. Shen, R. & Seshan, V. E. FACETS: allele-​specific copy pulmonary adenocarcinoma. N. Engl. J. Med. 361, 212. Hellmann, M. D. et al. Tumor mutational burden and
number and clonal heterogeneity analysis tool for 947–957 (2009). efficacy of nivolumab monotherapy and in combination
high-​throughput DNA sequencing. Nucleic Acids Res. 189. Fukuoka, M. et al. Multi-​institutional randomized with ipilimumab in small-​cell lung cancer. Cancer Cell
44, e131 (2016). phase II trial of gefitinib for previously treated patients 33, 853–861.e4 (2018).
163. Tarabichi, M. et al. A practical guide to cancer with advanced non-​small-cell lung cancer (The IDEAL 1 213. Marabelle, A. et al. 1192O - Association of tumour
subclonal reconstruction from DNA sequencing. Trial) [corrected]. J. Clin. Oncol. 21, 2237–2246 mutational burden with outcomes in patients
Nat. Methods 18, 144–155 (2021). (2003). with select advanced solid tumours treated with
164. Gerlinger, M. et al. Intratumor heterogeneity 190. Kris, M. G. et al. Efficacy of gefitinib, an inhibitor of the pembrolizumab in KEYNOTE-158. Ann. Oncol. 30,
and branched evolution revealed by multiregion epidermal growth factor receptor tyrosine kinase, in v477–v478 (2019).
sequencing. N. Engl. J. Med. 366, 883–892 (2012). symptomatic patients with non-​small cell lung cancer: 214. Graff, J. N. et al. A phase II single-​arm study of
165. Voss, M. H. et al. Tumor genetic analyses of patients a randomized trial. JAMA 290, 2149–2158 (2003). pembrolizumab with enzalutamide in men with
with metastatic renal cell carcinoma and extended 191. Paez, J. G. et al. EGFR mutations in lung cancer: metastatic castration-​resistant prostate cancer
benefit from mTOR inhibitor therapy. Clin. Cancer Res. correlation with clinical response to gefitinib therapy. progressing on enzalutamide alone. J Immunother.
20, 1955–1964 (2014). Science 304, 1497–1500 (2004). Cancer 8, e000642 (2020).
166. Hyman, D. M. et al. HER kinase inhibition in patients 192. Lynch, T. J. et al. Activating mutations in the epidermal 215. O’Reilly, E. M. et al. Durvalumab with or without
with HER2- and HER3-mutant cancers. Nature 554, growth factor receptor underlying responsiveness of tremelimumab for patients with metastatic pancreatic
189–194 (2018). non-​small-cell lung cancer to gefitinib. N. Engl. J. Med. ductal adenocarcinoma: a phase 2 randomized clinical
167. Bielski, C. M. et al. Widespread selection for oncogenic 350, 2129–2139 (2004). trial. JAMA Oncol. 5, 1431–1438 (2019).
mutant allele imbalance in cancer. Cancer Cell 34, 193. Pao, W. et al. EGF receptor gene mutations are 216. Spranger, S., Bao, R. & Gajewski, T. F. Melanoma-​
852–862.e4 (2018). common in lung cancers from ‘never smokers’ and intrinsic β-​catenin signalling prevents anti-​tumour
168. Jones, P. A. & Baylin, S. B. The fundamental role of are associated with sensitivity of tumors to gefitinib immunity. Nature 523, 231–235 (2015).
epigenetic events in cancer. Nat. Rev. Genet. 3, and erlotinib. Proc. Natl Acad. Sci. USA 101, 217. Xiao, Q. et al. DKK2 imparts tumor immunity
415–428 (2002). 13306–13311 (2004). evasion through β-​catenin-independent suppression
169. Vasan, N. et al. Double PIK3CA mutations in cis 194. Hann, C. L. & Brahmer, J. R. ‘Who should receive of cytotoxic immune-​cell activation. Nat. Med. 24,
increase oncogenicity and sensitivity to PI3Kα epidermal growth factor receptor inhibitors for non-​ 262–270 (2018).
inhibitors. Science 366, 714–723 (2019). small cell lung cancer and when?’. Curr. Treat. Options 218. Harding, J. J. et al. Prospective genotyping of
170. Paraiso, K. H. T. et al. PTEN loss confers BRAF Oncol. 8, 28–37 (2007). hepatocellular carcinoma: clinical implications of
inhibitor resistance to melanoma cells through the 195. Soda, M. et al. Identification of the transforming next-​generation sequencing for matching patients to
suppression of BIM expression. Cancer Res. 71, EML4-ALK fusion gene in non-​small-cell lung cancer. targeted and immune therapies. Clin. Cancer Res. 25,
2750–2760 (2011). Nature 448, 561–566 (2007). 2116–2126 (2019).

www.nature.com/nrg

0123456789();:
 Reviews

219. Peters, S. et al. Phase II trial of atezolizumab as 224. Janjigian, Y. Y. et al. Genetic predictors of response particularly M. Ladanyi, H. Zhang and R. Kundra, for their
first-​line or subsequent therapy for patients with to systemic therapy in esophagogastric cancer. assistance and insightful suggestions.
programmed death-​ligand 1-selected advanced Cancer Discov. 8, 49–58 (2018).
non–small-​cell lung cancer (BIRCH). J. Clin. Oncol. 35, 225. Kim, S. T. et al. Comprehensive molecular Author contributions
2781–2789 (2017). characterization of clinical responses to PD-1 The authors contributed equally to all aspects of the article.
220. Haratani, K. et al. Tumor immune microenvironment inhibition in metastatic gastric cancer. Nat. Med. 24,
Competing interests
and nivolumab efficacy in EGFR mutation-​positive 1449–1458 (2018).
D.B.S. serves on the Scientific Advisory Board for Loxo
non-​small-cell lung cancer based on T790M status 226. Ott, P. A. et al. An immunogenic personal neoantigen
Oncology at Eli Lilly, Pfizer, Scorpion Therapeutics, BridgeBio
after disease progression during EGFR-​TKI treatment. vaccine for patients with melanoma. Nature 547,
and Vividion Therapeutics, owned stock at Loxo Oncology at
Ann. Oncol. 28, 1532–1539 (2017). 217–221 (2017).
Eli Lilly and Scorpion Therapeutics, and received honoraria
221. Chowell, D. et al. Patient HLA class I genotype 227. Keskin, D. B. et al. Neoantigen vaccine generates
from Illumina and Eli Lilly. D.C. declares no competing interests.
influences cancer response to checkpoint intratumoral T cell responses in phase Ib glioblastoma
blockade immunotherapy. Science 359, trial. Nature 565, 234–239 (2019). Peer review information
582–587 (2018). 228. Wang, R. F. & Rosenberg, S. A. Human tumor antigens Nature Reviews Genetics thanks J. Martens and the other,
222. Gopalakrishnan, V. et al. Gut microbiome modulates for cancer vaccine development. Immunol. Rev. 170, anonymous, reviewer(s) for their contribution to the peer
response to anti-​PD-1 immunotherapy in melanoma 85–100 (1999). review of this work.
patients. Science 359, 97–103 (2018).
223. D’Angelo, S. P. et al. Avelumab in patients with Acknowledgements Publisher’s note
previously treated metastatic Merkel cell carcinoma: The authors thank M. F. Berger, H. Al-​A hmadie, C. Ho, Springer Nature remains neutral with regard to jurisdictional
long-​term data and biomarker analyses from the S. Sethi, A. Zehir and S. Nandakumar for their invaluable claims in published maps and institutional affiliations.
single-​arm phase 2 JAVELIN Merkel 200 trial. contributions to the figures. They are also grateful to the
J. Immunother. Cancer 8, e000674 (2020). Molecular Diagnostics Service and the OncoKB team, © Springer Nature Limited 2021

Nature Reviews | Genetics

0123456789();: