©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 30 NUMBER 7 JULY 2012 639

Much of modern biomedical research starts from a premise that there is a deeply encrypted code by which cells act together, in pathways and across networks, to maintain cellular tissue and organ health. Adjacent cells might appear the same, but even recently divided daughter cells show variation from their neighbors owing to changes in the microenvironment, the nature of the cells that are adjacent, and exposure to subtle shifts in gradients of growth factors, oxygen and other environmental components. When every cell we look at is potentially unique, how can we discern the principles of the code underlying how cells work alone or together? And what do we do when we come up against a formidable problem that underpins a devastating class of disease—that of cancer heterogeneity—where many of the rules so neatly obeyed in normal cells are rewritten at every cancer cell division?

Traditional biochemical approaches underpinned our initial comprehension of cellular functions. In the early days, assays were often based on lysis or dissolution of complex cell mixtures to enable purification and discernment of their component parts. Using such biochemical methods, and drawing on an understanding of single-cell biology, researchers have shown that in normal cells, ordered signaling events drive cellular responses to internal and external cues leading to phenotypic outcomes, such as differentiation, proliferation, apoptosis and secretion of effector molecules.

A growing body of research now indicates that the biology of single cells is rarely deterministic1 and that even recently divided cells in culture are highly variable1,2. It has become increasingly clear to researchers over the past 15 or so years that, by averaging information across many cells, differences among cells, which may be important in explaining mechanisms, can be lost. Consequently, the caveats in the

conclusions one draws about how any cell type functions when one uses bulk biochemical or molecular approaches quickly pile up when aver-aging cellular information: the system and context and the co-varying features are muddled. Making an assumption that two cells are identical—or that a cell population can be declared homogeneous— is a gross simplification of how biology operates. We now appreciate with far greater clarity how stem cell hierarchies, transcription start sites, cell signaling pathways (and more) all function against a back-drop that assumes that carefully orchestrated single-cell stochastics, in concert with mass action, is what determines outcome.

One illustration of why this matters can be found in the study of mechanisms underpinning the development of cancers. Tumors have long been understood to play by different rules compared with normal cells. They are not merely collections of malignant cells but rather are complex, atavistic entities3 in which a pathologically symbiotic interplay occurs between cancer cells and adjacent immune, inflam-matory, vascular and stromal cells. In solid tumors even the epithelial compartment itself is variable. Our view of tumor heterogeneity has sharpened to a vision where tumors are viewed as a hierarchy of cells, with tumor-initiating cells presumed to be at the top of a local hierar-chy, and cells with lesser, or at least different, potential for unlimited growth descending from that apex. This has led to the key notion that all-too-common treatment failures in several cancers are caused by intrinsically chemoresistant, tumor-initiating cell populations that can reestablish a complete tumor cell hierarchy after treatment4–8.

Cancer starts as one cell, ultimately shields itself in a forest of genetic and epigenetic diversity3 and appears capable, after therapy, of resurrecting itself from just one surviving cell with tumor-initiating cell activity. The extent, then, to which genetically diverse tumor cell subpopulations differ with respect to phenotypes, including renewal capacity and drug resistance, is far from fully characterized. Why are we so far from a more complete understanding? Because we simply have not had the tools to characterize the crucial differences, and just as importantly, sort out which differences are causal and which

From single cells to deep phenotypes in cancerSean C Bendall & Garry P Nolan

In recent years, major advances in single-cell measurement systems have included the introduction of high-throughput versions of traditional flow cytometry that are now capable of measuring intracellular network activity, the emergence of isotope labels that can enable the tracking of a greater variety of cell markers and the development of super-resolution microscopy techniques that allow measurement of RNA expression in single living cells. These technologies will facilitate our capacity to catalog and bring order to the inherent diversity present in cancer cell populations. Alongside these developments, new computational approaches that mine deep data sets are facilitating the visualization of the shape of the data and enabling the extraction of meaningful outputs. These applications have the potential to reveal new insights into cancer biology at the intersections of stem cell function, tumor-initiating cells and multilineage tumor development. In the clinic, they may also prove important not only in the development of new diagnostic modalities but also in understanding how the emergence of tumor cell clones harboring different sets of mutations predispose patients to relapse or disease progression.

Baxter Laboratory for Stem Cell Biology, Department of Microbiology & Immunology, Stanford University School of Medicine, Stanford, California, USA. Correspondence should be addressed to G.P.N. (gnolan@stanford.edu).

Published online 10 July 2012; doi:10.1038/nbt.2283

r e v i e wnp

g©

201

2 N

atur

e A

mer

ica,

Inc.

All

right

s re

serv

ed.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

640  VOLUME 30 NUMBER 7 JULY 2012 nature biotechnology

r e v i e w

mutations are passengers of the diversification process at the heart of cancer pathology. Finally, it is critical to understand the extent to which the kinds of heterogeneity drive treatment decisions. For instance, depending upon where in a gastric cancer specimen one measures HER2, a different clinical outcome might be predicated for responsiveness to trastuzumab (Herceptin)9. Thus, even an incipient cancer cannot be assumed to be driven by a common underlying set of genes. Analysis of one part of a tumor might determine it is treatable by a different drug compared with analysis of a different part of a tumor.

At the heart of these inquiries is the extent to which observable phenotypes in cancer and genomic heterogeneity are linked at the single-cell level. That is, do various mutations work in concert to converge on a limited number of cancer cell behaviors and/or pheno-types? How far away from individual mutations should we step to understand how they work in concert to drive common cancer phenotypes and outcomes3? It is well understood that gross genetic abnormalities result in distinct cancer cell behavior. For instance, mutations resulting in constitutive regulatory kinase expression can lead to uncontrolled cellular growth but also provide specific thera-peutic targets10. Other mutations can stretch the phenotypic bounda-ries (CD34−) of the cancer-initiating cell in acute myeloid leukemia (AML)11. But the fact remains that there are not an infinite number of wholly different biological or clinical behaviors in any given cancer—meaning that despite the overwhelming number of mutations we observe in cancers12,13, genetic constraints on the system define the kinds of mutations ‘permitted’, which in turn limit the phenotypes of a cancer and still allow that cancer to remain viable14.

In this review, we discuss prospects for new technologies aimed at studying heterogeneity in single cells in tumors. In particular, we sum-marize the uses of ligand- or pharmaceutical-based interventions that act by known signaling mechanisms on particular cell subsets—whether stromal, nurse cells, tumor-initiating cells or other processes—to under-stand the role of such cell subsets in tumor maintenance and behavior. Although major advances have also been made in traditional micro-scopic techniques for picturing and enumerating cancer diversity, these are not covered here.

Fluorescence-based flow cytometryPossibly the first of the high-throughput devices that discerned differ-ences between cells, by simultaneously measuring multiple events per cell, was the flow cytometer. Although many flow cytometer designs have been created over the years, one of the most influential of these is the fluorescence-activated cell sorter (FACS) invented at Stanford University in the laboratory of Leonard and Leonore Herzenberg15,16. FACS and its competitors and relatives—even in the early 1980s—was a synergistic blend of physics, biology, informatics, computation and efforts in data visualization. Arguably, FACS and its associated com-puter hardware were the first of the ’omics tools. Very large data sets (even by today’s standards) comprising millions of cells and multi-ple events per cell were the norm. Flow cytometry was, in essence, the technical forebear of ‘systems immunology’ in that it provided the most detailed data sets of as many of the immune compartments one desired to study at any one time. What was missing until recently were the computational tools to mine the meaning in the data.

From its inception, the aim of flow cytometry was to measure multiple events per cell and correlate them with biological mechanism, or disease processes, in animal models and humans. The device made an a priori assumption of nonhomogeneity of a target cell population. The family of instruments grouped under the term ‘flow cytometer’ now extends to microfluidics that miniaturize the components, lasers

and detectors into lab-on-a-chip devices17,18. As such, the instru-ments set a high standard for very quantitative and high-throughput multiparameter measurements of single-cell attributes. Because of its strengths, flow cytometry has been a mainstay of immuno-logy and the study of hematological malignancies for more than 40 years (Table 1).

Leukemias and lymphomas were the earliest cancers studied using flow cytometry analysis. Panels of cell-surface markers standard for the characterization of cell subsets in human peripheral blood mono-nuclear cells or bone marrow revealed unusual expression patterns or cell features in the hematopoietic malignancy compared with normal, standard samples. Just as a pathologist can look at a field of cells and grade a tumor’s stage, in part, by the disorder of the cells relative to each other and other cellular attributes, so an analysis of FACS data can identify disrupted expression patterns of proteins on cells in cancer. For instance, levels of protein expression are distorted compared to normal, healthy controls, but flow cytometry demonstrated that, in addition, proteins never normally seen together can be co-expressed on the same cell—suggesting disarray at the level of normal gene expression programs driving differentiation and signaling. Whereas heterogeneity in solid tumors has traditionally been determined by microscopy, some studies have used flow cytometry, to visualize hetero-geneity, such as with prostate cancer (and cross-documented by immu-nohistochemistry) with as many as 15 individual markers19.

Flow cytometry has been applied to detecting tumor stem cell sub-sets in liquid and solid tumors9,20,21 by cell sorting. Bonnet and Dick22 demonstrated that a prospectively isolated minor population (lineage negative, CD38−, CD34+) of human AML cells exclusively initiated the disease in an immune-deficient animal model. This paradigm of leukemia-initiating cells, more broadly thought of as cancer stem cells or tumor-initiating cells, has since been observed in several other solid-tissue human malignancies23,24. Being able to target neoplastic cell populations based on unique surface-marker expression patterns has provided a new avenue in cancer research to study directly the cells thought to be at the root of the disease. FACS analysis and sort-ing provides the utility to both isolate and assay such cells and their capabilities. The importance of cancer stem cells, the surface markers used to define them and their rarity in various tumors remains a topic of great debate25.

Another avenue taken by FACS analysis in dissecting tumor het-erogeneity has focused more on signaling behavior and its relation-ship to disease outcome. This approach pays less attention to cellular surface-marker phenotype and underlying sequences of gene muta-tions and more to the signaling (phosphorylation) networks in the cells that are often regulators of common chemotherapeutic targets. The first example of this was demonstrated in human AML where phosphorylation in response to extracellular factors not only revealed unique regulatory networks subdividing patients but was also corre-lated to both patient outcome and mutational status26. More recently, similar studies employing single-cell phosphorylation analysis by flow cytometry have classified patients suspected of having juvenile myelomonocytic leukemia27 and predicted outcome in human B-cell lymphoma28 and drug sensitivity in chronic lymphocytic leukemia29.

The ability to discern heterogeneity in signaling became possible in the early 2000s by profiling the activities of kinases and networks at the single-cell level26,30,31 through so-called phospho-flow. This led to new applications of the flow cytometer as an instrument with proteomics applications capable of probing intracellular circuitry32 on fixed cells. This technique is primarily used to analyze fixed (dead) cells, and therefore gave up one of the key utilities of FACS—that of live cell sorting. It is possible, however, to sort these fixed cells for genomic

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 30 NUMBER 7 JULY 2012 641

r e v i e w

and transcriptomic profiling (Y. Goltsev & G.P.N., unpublished data), but the live-cell analysis remains out of reach for high-parameter, single-cell signaling applications with current technologies.

A generic template for the use of phospho-flow (and other perturbation-response systems) started with the notion that the basal state of a cellular system is a homeostatic, but flexible, resting state26. A cell at homeostasis reflects the sum of its genetic and environmental history. The cell might be nominally at rest, but it is poised to adapt to fresh challenges. The biochemical differences that reflect what a cell might do are often lost in stochastic noise of homeostatic feedback networks. Thus, a solution is to grossly perturb the system of cells with input stimuli, thereby taking advantage of the built-in amplification processes of signaling systems. In this manner, cells are forced to make strong choices to reveal major conduits by which information flows within and between cells. It turns out that this information can be catalogued (Fig. 1), used to create diagnostic predictors at the single-cell level26–29,33 and mathematically mined to elucidate cell subsets and responses that correlate to orthologous or surrogate phenotypes. Other correlates include clinical outcomes or drug response34, and reconstruction of cellular networks32,35,36. Finally, highly correlated multiplex measurements can be used to discern whether ‘heteroge-neous’ cancers have a discoverable recapitulating structure (B. Neel, J. Stewart, W. Fantl & G.P.N., unpublished data), thereby uncovering previously unknown intercellular relationships linking the functions of nonadjacent cells of complex tissues (M. Clausson, D. Koller & G.P.N., unpublished data).

Probing intracellular and intercellular relationships in this context is achieved using what would be recognized as a classical medical perturbation test, but at a cellular scale. The approach is related, for example, to a stress test for heart functions or to a reflex test on your knee for nervous system function. Although the mechanism is not always directly inferred, perturbation and its readouts narrow the pro-cesses in a cellular system to those on which further inquiries should focus. In cancer, pathways stray from standard or ‘normal’, but there remain rules by which cellular systems operate such that we can com-pare a cancer cell’s biology to that of normal cells and determine the extent of the differences between them. Single-cell perturbation and cytometric interrogation reveals cell-by-cell response patterns to per-turbations (by challenging the regulatory wiring of each cell) and then determining the extent to which the cancer cell population is divergent from normal. One key finding is that signaling differences can correlate significantly with outcomes and therefore the development of diagnostic tests for cancer could be predicated on such evidence26,37–39.

A singular drawback of flow cytometry, or any technique that dis-rupts tissue structure and function in the process of studying living cells, is that cell-cell contact is lost and cells therefore lose (among other things) access to nutrients and become hypo-oxygenated in the process of preparation, as well as subject to mechanical shear forces that might initiate or change intracellular signaling. In whole blood–based assays, this tends not to be a problem, but certain pathways (stress response), might be sensitive to disruption when the cell is taken out of context. Although this is often not a problem in deriving

Table 1 Different approaches for analysis of tumors at the single-cell levelTechnique Sensitivity Speed Resolution Pros Cons Reference

Flow cytometry ~500–107 molecules/cell

~25,000 cells/sec

Single cell Up to 15 parameters/cell Many fluorescent probes for cellular biology Sorting of live cells

Cellular autofluorescence interference Emission spectra interference as multiplexing increases

78

Mass cytometry ~1,500–107 molecules/cell

~1,000 cells/sec

Single cell Currently 42 parameters per cell; ~100 possible No autofluorescence or spectral overlap

For every new parameter, new chemistries are required to attach isotopes to reagents Currently, 70% of the cells ejected from the nebulizer do not maintain full integrity. After ionization, such subcellular fragments do not reach the detector as ionic clouds representing the constituents of a com-plete cell. A new cell nebulizer design is expected to reduce cell loss to only 30%

78–80

Single-cell sequencing

93% of complete genome

10 d/cell Single cell Most of genome can be sequenced Identification of mutations

Sequences prone to possible mutation during early PCR steps

53,54

Single-cell PCR of targeted transcripts

96 transcripts/well 96 cells/ plate/4 h

Single cell Possible 1,000 cells/day Relative quantification, Absolute quantification with controls

Only 96 transcripts per cell reported 63

Transcriptome 1 cellular exome or 107 transcripts

2–3 d/cell Single cell Quantitative sequence counting of all transcripts Splice-o-forms quantified Point mutations identified

Limited by exome selection method 53,54

MALDI-imaginga ~10−15 mols/µm2 ~1 sec/pixel ~50 µm pixel size

Theoretically hundreds of different molecular species (protein and small molecule) can be analyzed Tissue structure intact

Reporter masses <250 Da difficult to observe due to ‘matrix effects’ Fragmentation and molecular abducts complicate interpretation

65

SIMS/MIMSa ~1012–1016 atoms/cm2 for trace elements

~1 ms/pixel ~0.05 µm pixel size

Currently able to determine biologically labeled isotope ratios, such as N, C, O Subcellular resolution Tissue structure largely intact

Fragmentation and molecular abducts complicate interpretation Most instrument configurations limited to <10 analytes per scan

67,75,76

Laser ablation ICP mass spectrometrya

~102 ppb for lanthanides

~1 sec/pixel 4 µm pixel size

Tissue structure intact Relatively simple to interpret

Limited to the analysis of only elemental constituents and reporters (see mass cytometry)

81,82

SMIS, secondary ion mass spectrometry; MIMS, multi-isotope imaging mass spectrometry.aResolution, speed and sensitivity are interdependent. Sensitivity can be increased by increasing pixel size (lower resolution) and increasing scan dwell time (lower speed).

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

642  VOLUME 30 NUMBER 7 JULY 2012 nature biotechnology

r e v i e w

mechanism (witness many decades of suc-cessful research despite the disruptions)—and we have long found that even cell lines harbor a decent memory of their origin—it remains a potent caveat in any single-cell study.

FACS-based single-cell approaches have contributed immensely in the deciphering of what kinds of cells a tumor might contain and has provided some knowledge of their roles in tumor composition (Table 1). The question remains as to how we will merge mutational analysis, phenotypic diversity and single-cell signaling heterogeneity (or a combination of all three with other methods) to drive development of more effective therapies for various cancers as well as develop robust predictors of how and when such therapies should be applied.

Cytometry time of flight (CyTOF)The strength of traditional flow cytometry—fluorescence—has also been its Achilles heel. Fluorophores, the workhorses of flow cytom-etry, in concert with lasers tuned to their excitation spectra, have been unmatched interrogators of single-cell activities. However, multiple fluorophores don’t always play well with each other in the sandbox of a single cell passing by four lasers in millisecond time frames. The exci-tation spectra of multiple fluorophores measured simultaneously can ‘bleed’ into each other’s detection channels—obscuring measurements. Unfortunately, the careful design of reagent panels, optimized lasers and filters for excitation, along with compensation for spectral overlap is no match for the need of scientists wishing to quantify more than a canonical maximum of 10–15 cell-associated events. After all, when you are watching the guy down the hall measure tens of thousands of mRNAs per experiment, it’s hard not to be a little jealous. The introduction of mass spectrometry now promises to change that.

The mass spectrometer has been one of the signature tools of ‘proteomics’ over the past decade40. Quantitative, robust, versa-tile and sensitive at the level of the number of ions counted, it has many applications in biochemical analyses. That is until one wants to measure low-abundance protein or RNA constituents of single cells. Simply put, the mass spectrometer most familiar to us does not (today) have the sensitivity to measure rare proteins or their isoforms at the number of copies present in single cells. Scott Tanner at the University of Toronto saw an opportunity to adapt a form of mass spectrometry termed inductively coupled plasma (ICP) mass spectrometry to the measurement of events on and within cells41. The idea was to tag antibodies with rare isotopes of elements not nor-mally found in cells, stain cells with those tagged antibodies and then pass those cells—a single cell at a time—rapidly through a 7,500 K argon plasma41,42. There, the ions of what were once the individual cells pass into a time-of-flight (TOF) mass spectrometer tuned to

the elemental weight range of the isotopes used to tag the antibodies bound to the cells. Every molecule within each individual cell is completely atomized and ionized—and then those ions falling within a specific mass range are quantified by Tanner’s newly developed instrument CyTOF41 (Fig. 2). The summed levels of all isotopes from each cell are digitized and associated with that cell to create a spreadsheet of cell-by-cell information that is completely analogous to a flow cytometry data file (only much bigger).

Using this ‘mass cytometry’ approach (Fig. 2), Tanner and colleagues phenotyped and profiled complex primary human bone marrow cell populations by cell surface, phosphoprotein and nucleic acid constituents with 35 simultaneous measurements on millions of cells per sample42. The depth of connectivity within the data sets collected allowed discovery of some of the principles by which the system is organized. By collecting enough cell surface events in a progressive system, such as normal hematopoietic development in human bone marrow42,43, they reconstructed the most likely lineage relationships between cell types. This led to a computational reca-pitulation of a familiar representation by means of a human hemato-poietic progression tree42–44. By simultaneously measuring multiple intracellular responses to perturbations, they created a global immune posture or profile that reflected a cell-by-cell, subset-by-subset ‘phenotype’ of cell function.

A sweet spot for the application of CyTOF at present is in broad profiling (Fig. 2) of immune cells, inflammatory cell states and cancers. For instance, one first takes primary cell materials (blood, bone marrow or tumor tissue) and dissociates it into single cells. As noted above for phospho-flow and traditional flow cytometry, you start by interrogating the cells with a variety of environmental stimuli, such as cytokines or growth factors. The key is diversity of relevant stimuli to perturb and provoke the cells—and their intracellular signaling apparatus—into action. The workflow then proceeds to staining of the cells with antibody panels that probe the biology under question. These can include pluripotency panels (e.g., Nanog, c-myc, SOX2), cell cycle45 (such as with IdU, cyclin B1, cyclin A, p-histone H3, Rb, p21 and p27), apoptosis, DNA damage, kinases and cell-surface CD markers for most major immune subsets from mouse and human,

Phosphorylated S

TA

T5

Unstimulated

G-CSF stimulated

Patient A Patient B Patient C Patient Z

Started high(high basal)

Responsive tostimulation(potentiated)

G-CSF

p-STAT5 p-STAT3

G-CSF

p-STAT5 p-STAT3

G-CSF

p-STAT5 p-STAT3

Phosphorylated STAT3

Mixed clinical outcomes Poor clinical outcomes Better clinical outcomes

Figure 1 Signaling-responsiveness–dependent heterogeneity can correlate with signaling outcomes26. Tumors observed to be relatively homogeneous by surface marker expression, as well as intracellular signaling states (unstimulated, top row), can instead be observed to comprise multiple cell subsets (granulocyte colony stimulating factor (G-CSF) activated, bottom row of FACS plots), which, upon retrospective analysis may correlate with a clinical outcome26,37–39. (Reproduced from Fig. 2, ref. 77, with permission.)

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 30 NUMBER 7 JULY 2012 643

r e v i e w

transcription factors and epigenetic modifi-cations. In our laboratory we can interrogate 45 parameters per cell, and with novel statisti-cal approaches and applications of informa-tion theory, we should be able to double that number (K. Sachs & G.P.N., unpublished data). It is even possible to detect mRNA at the single-cell level using branch-chain amplification techniques simultaneously with measurement of surface proteins and phospho-epitopes, for example (Y. Goltsev & G.P.N., unpublished data). Luckily, CyTOF benefits from three decades of fluorescence-based innovation that paves the way for adapting probes to using isotopes as tags. To date, most of the major human and murine surface markers, numer-ous intracellular phospho-epitopes, as well as many apoptosis, cell cycle and DNA damage markers have been adapted for CyTOF42,44–46 (G.P.N., unpublished data). Of course, for at least the near future, cer-tain cellular features will remain the sole provenance of fluorescence, but in such cases fluorescence cytometer–mass cytometer hybrids are already envisioned to step into that breach.

At the end of a mass cytometry workflow, one has an alarmingly large data set that requires interpretation. Luckily, the problem is solvable by going back to basics. Traditional flow cytometry has for 30 years relied on modestly scaled extensions of histograms or ‘dot plots’ where single cells are plotted on graphs that depict the level of expression of one or two parameters. As the number of parameters increases, the number of two-dimensional plots required increases considerably, leading to a visualization problem—our primate brains can’t readily ‘see’ five or more dimensions.

Analytic approaches that tackle this complexity43 are related to other dimensionality reduction problems in science and are an area of considerable history and constant innovation. The utility of under-standing this high dimensionality rests in the often-unappreciated power of correlated measurements. To know that a marker is co-expressed in distinct or discoverable patterns across a set of cell types (especially minor subpopulations) or that a principle determines such expression patterns allows one to propose that cells exist as subsets, or express regulatory modules of signaling proteins (phospho-protein activation events, for instance).

Now that we can achieve this greater, more wholly connected picture of a tumor’s proteins and signaling, the challenge remains

to computationally compare this new image of cancer across both patients and tumor types to determine whether this new level of single-cell resolution yields an improved understanding of the disease process or can better inform therapeutic interventions. So, if single-cell data from 10 to 100 dimensions is exciting, it’s about to get better, or worse, depending on your perspective.

Single-cell DNA and RNA sequencingWhere cytometry technology has tended to move at a slow pace (until recently), sequencing technology has leapt forward, with each innova-tion more creative than the last. Initially, the advent of semiquantitative DNA microarrays for transcriptome profiling opened up the possibil-ity of hypothesis-generating experiments in cancer. Sophisticated sta-tistics allowed us to mine the data for expression patterns that gave rise to gene expression patterns for prediction of clinical outcome and classification47,48. At the same time, despite lacking single-cell resolu-tion, gene copy number analysis informed by array-based techniques has also raised the question of clonal heterogeneity and evolution during disease progression. For example, in two studies using genome-wide DNA copy number analysis of pre–B cell acute lymphoblastic leuke-mia49,50, a minor clone (cell population), seemingly ignored during disease diagnosis, became dominant as the cancer progressed or the patient relapsed. Studies like these foreshadowed the potential of next-generation sequencing approaches in providing a higher resolution view of cancer evolution as well as the recurrent mutations that drive it.

At least three general single-cell sequencing techniques have been reported: first, whole-genome sequencing (WGS); second, whole mRNA transcriptome sequencing; and third, targeted sequencing of DNA regions (that is, exome sequencing) or mRNA transcripts. Epigenetic modifications and nontranslated RNAs could be included here, but the former we will treat as a subapplication of WGS, whereas the latter has been extensively discussed with RNA sequencing else-where51–54 (Table 1).

Using earlier incarnations of next-generation sequencing plat-forms, the first WGS analysis of karyotypically normal human AML

Antibodieslabeled with

elemental isotopes

Cell 1 3,8,9,7...8Cell 2 1,8,6,5...4Cell 3 9,9,4,5...7 . . . .

ElementA B C D...J

Mass

Cell 1

Cell 2

Cell 3

ICP

Nebulizer

Quadrupole

Time-of-flight

Integrate per cell

Heavy (>100 Da)Reporter atomic ions

Light (<100 Da)Overly abundant ions

.FCS file

NN

NO

OO

O

OO

OO

O

OO

Gd

H H

Analysis104

104

103

103

102

102

10

10

–10

–10

0

0

Ele

men

t A

Element B

Figure 2 Most antibodies, either against surface proteins delineating immune cells or intracellular targets, can be adapted for use with the CyTOF instrument. First, a polymeric chelator is loaded with cations of a stable heavy metal isotope. The loaded polymer is then conjugated to the antibody of interest using chemistry similar to that of fluorophores or biotin. Free ions and unconjugated polymers are removed by size-based purification. Panels of appropriate antibodies are assembled for staining much like one would for a traditional fluorescence-based panel. The advantage is that not only is there no compensation to worry about from interfering channels, but there is no autofluorescence in mass cytometry that might obscure weak signals41. Cells are passed through an ICP TOF mass spectrometer and the individual ions are counted and then parsed into a flow cytometry file format. This file is then capable of being analyzed by traditional flow cytometry software or more advanced software as required42,43. (Parts of this figure are based on Fig. 1, ref. 42.)

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

644  VOLUME 30 NUMBER 7 JULY 2012 nature biotechnology

r e v i e w

cells from a single patient revealed mutations associated with the dis-ease55. As next-generation sequencing approaches rapidly advanced to become cheaper, faster and better, however, so did the depth of genomic characterization of human leukemia. The following year, the addition of multiple patient samples revealed recurrent muta-tions associated with human AML56. Two retrospective, longitudinal studies in the same patient, one from AML diagnosis to subsequent relapse57, the other on the myelodysplastic syndrome to secondary AML transition58 revealed—as previous gene copy number experi-ments had suggested49,50—a clonal heterogeneity in the primary dis-ease that changed as the patient progressed through different stages of disease. By following the frequencies of recurrent somatic and germline mutations using high-coverage WGS, these studies dem-onstrated how an often minor genotypic clone in the initial disease state acquires at least one new mutation, becoming the dominant or only clone in the subsequent relapsed or transitioned, second-ary AML disease. This suggests that the therapy itself could be the selective pressure for further mutations in clonal evolution to a more aggressive, resistant state. Thus far, the majority of studies have been reported as case studies, with a limited number of patients, focusing mostly on hematopoietic malignancies. As in the aforementioned case with tumor-initiating cells, until there is more evidence to sup-port this paradigm (through sampling more patients and tumor types longitudinally), the question remains as to whether it can be informa-tive to the disease mechanism or outcome.

To avoid situations where therapy unwittingly potentiates recur-rence, we must understand the functional consequence of clones har-boring different combinations of mutations predisposing the patient to relapse or progression; this is one of the next great problems of cancer research. Do multiple combinations of mutations present in different clones converge or diverge in terms of cellular behavior (that is, in terms of transcriptional program, cellular phenotype, response to regulatory environment or therapeutics)? To answer this, we will require a novel combination of mutational analysis with more targeted single-cell approaches. One attempt to do this was with fluorescence in situ hybridization (FISH) where known ‘driver’ mutations in acute lymphoblastic leukemia were followed at the single-cell level for the purpose of modeling the clonal architecture of this disease59. To go beyond tracking obvious mutations in a given tumor, advanced cyto-metric approaches could be married with WGS studies for a more comprehensive analysis of other single-cell features. Doing this, one could finally determine whether phenotypic, functional and muta-tional diversity are linked at the single-cell level in cancer.

Although WGS and FISH can probe ‘clonality’ (an implication of single-cell resolution) in cancer, their results still rely on bulk material and only infer clonal information statistically from the sequencing information. Thus, these studies have a sensitivity limit of about 10% of the population for detecting sequence-based clones from the bulk cell mixture data—potentially hiding an unappre-ciated level of genomic complexity represented in the individual cells of most tumors. Two recent studies demonstrated single-cell, whole exome sequencing (WGS targeted at DNA coding regions) in cancer60,61. One tracked mutations in myelodysplastic syndrome60 using whole genome exome sequencing on 90 cells to individu-ally identify novel mutation patterns. The second study focused on clear cell renal carcinoma and revealed that there was actually no predominant clonal subpopulation at the single-cell level, sug-gesting the underlying genotype of this cancer (and others) is more complicated than previously thought60,61. Both of these studies were hampered only by the fact that the cells had not been previously selected by FACS or phenotyped to relate genotypes back to physical

cellular features—certainly an obvious next step that might coalesce certain mutation patterns with cellular phenotypes.

Mapping transcriptional profiles by sequencing at the single-cell level can be more challenging than WGS. With the exception of a qualitative profiling of transcripts in a single human blastomere62, currently there is a paucity of quantitative, single-cell, RNA-Seq data published, largely because of the inability to quantify such low levels of RNA through existing sequencing approaches. Clarke, Quake and colleagues63 undertook an accounting of normal and cancerous colon cells using semiquantitative, microfluidics-based, single-cell amplification of 53 unique genes. Using FACS to isolate cells with defined phenotypes and comparing tumor-initiating cells to those with a similar phenotype yet inability to engraft as cancer, this study revealed a close correlation of cellular phenotype to transcriptional profile. Not only could cancerous cells be mapped back to equivalent ones in healthy tissue, but transcriptional profiles also provided a two-gene classifier that could outperform existing pathological prac-tices in predicting disease outcome. Interestingly, both this study and another investigation of a mouse leukemia model64 demonstrated that (genetically) clonal tumor-initiating cells regenerated the original cellular hierarchy of the primary tumor. In the case of the mouse leukemia study, cytometric analysis of the phenotypically distinct tumor-initiating cells revealed divergent regulatory signaling that when targeted with the appropriate kinase inhibitor uniquely per-turbed the activity of that tumor-initiating cell population alone. Altogether, these observations highlight that mutations alone are not driving tumor heterogeneity; therefore, understanding other cellular characteristics (e.g., gene expression) at the single-cell and/or clonal level will be required.

A near-term goal for single-cell gene expression in cancer is the ability to undertake a fully quantitative RNA-Seq experiment on hundreds of individual cells. Besides gene expression itself, quanti-fication of mRNA splice forms would be highly complementary to genetic abnormalities identified by WGS. Although efforts towards this are underway in a variety of settings, sequencing technology remains sufficiently expensive and time consuming, thus limiting the absolute numbers of cells measured to only a few hundred per sample with single samples taking a week or more. Flow cytometry, and the powerful statistics that can be brought to bear, processes millions of cells at a time. But a future is coming where RNA-Seq and associated methods will begin to rival traditional cytometric approaches. Until then, applications in the immediate future will most likely have to leverage FACS to isolate a limited number of the most compelling tumor cell types for sequencing analysis.

New mass spectrometry–based, single-cell applicationsMass spectrometry in its varied forms has not ceded the task of single-cell measurements to nucleic acids and flow cytometry quite yet. Tandem and high-resolution biological mass spectrometry, as noted earlier, has had an immense presence in cellular analysis. Top-down and bottom-up analysis have interrogated and provided information on the protein composition and in some respects the proteomic relational dependence that reflects cell translation. Even so, in almost all instances, these insights are global, reporting on the ensemble average of cell lysates. Maximal information from limited samples, whether chromatographically separated or discrete spatial images of biologic state, benefit from ‘simultaneous’ analysis. Among the technologies that offer this capability, TOF mass spectrometry continues to hold an important position. Perhaps the impact of mass spectrometric technologies will be first felt in the imaging of tissues, as complementary or even competitive with microscopy. Imaging by

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 30 NUMBER 7 JULY 2012 645

r e v i e w

matrix-assisted laser desorption ionization (MALDI)-TOF has the advantage of being able to measure hundreds of cellular constitu-ents at a time65 in a raster format, wherein individual subregions of a tumor tissue section are successively analyzed and then stitched together into a larger format image (Table 1). Although MALDI-TOF is not yet readily applicable to the resolution of multiple single cells, it is certainly technically feasible with today’s techniques to place cells in an organized raster to achieve single-cell measures of highly expressed metabolites. Thus, although sensitivity remains an issue for MALDI in this application, the field is making progress and we await further developments in the near future. Laser ablation ICP mass spectrometry offers the opportunity to interrogate less-abundant proteins that have been immunologically stained with metal- conjugated antibodies41,66; here again, the spatial resolution is generally insufficient to allow analysis at the cellular level, although technology advances should be anticipated in this direction67.

One mass spectrometry technology that might be competitive with TOF is magnetic sector68–70 with array detection technology. With an atomic mass spectrometry history of nearly a decade, multi-collector mass spectrometry71 offers the potential to measure up to 9 mass channels simultaneously; indeed, the instrumentation has found important use in isotope ratio analysis. The extension to allow many simultaneous channels was further developed by Hieftje and colleagues72–74. Assuming that technological developments (not yet in sight) will allow analyses on the time scale of single-cell ioniza-tion events, we believe this could be a technology to eventually rival CyTOF’s foreseen applications.

With a history that both precedes and is contemporary with the above technologies, secondary ion mass spectrometry and multi-isotope imaging mass spectrometry (MIMS) have been resurgent67,75. In MIMS a sample is ‘sputtered’ with high-intensity ions, such as Cs+. Ion bombardment generates secondary ions from the surface and immediate substructure of the sample, which are manipulated and gathered by ion optics and measured. An advantage is the ability to track isotopically labeled atoms as they are metabolized inside cells. To date, as many as six isotopes (parameters) have been measured at a time (Table 1).

The sensitivity and resolution of MIMS imaging was illustrated in recent work that provided novel insight into the kinetic and spatial properties of actin turnover within the inner ear67,75. Before this, actin filaments in stereocilia were thought to be replaced every 1–2 days in a ‘treadmilling’ process that proceeded from tip to base. MIMS imaging of stereocilia from frogs fed an N15-enriched diet was used to visual-ize incorporation of new actin filaments. The results of this work revealed the process is much slower than previously thought—10% per day—and occurred only at the tips. Similar strategies have been used to analyze fatty acid transport in adipocytes, metabolic activity in bacteria76, nuclear colocalization of DNA and RNA, and interactions between membrane proteins and lipids.

ConclusionsThe individuality of the single cell, and the information it contains, is likely to be the key to therapeutically targeting every cell in a tumor. Sadly, we are currently treating each patient as if they had an average cancer, the clinical phenotype and subsequent treatment of which has been determined by averaging cellular characteristics across thousands of cells and millions of patients. Some of the single-cell techniques described in this review should help us discern how to treat every cancer cell in every patient according to its unique biology. A range of approaches are being developed that measure more, with less effort, on a per-cell basis helping to sort out the heterogeneous

tumors into understandable subsets of cells. With the exception of surface marker analysis, in the near term, we must make do with tech-niques that kill or destroy cells to study them. Although not optimal, there is still much to be gained by such methods.

Of course, it is hardly productive to spend too much time arguing about which technology is the most informative overall—because in most cases that depends upon the question. If the goal is to understand the ‘system’ of cancer, then we should look for ways to merge technol-ogies measuring different cellular attributes. Would it be unrealistic to develop a technology that reads multiple cellular attributes through a common tag? Fluorescence certainly provided an abstract proof that a single measurement platform (emitted light) could be bent to multiple purposes for measuring a variety of cellular phenotypes. We have yet to figure out how multiple types of a single cell’s constituents (DNA, mRNA, noncoding RNA, metabolites and proteins) can be separated quantitatively and then passed to the micro-platform best enabled to measure them. Certainly there have been attempts from single-cell PCR and microfluidic devices of dizzying variety, but effec-tive simultaneous measures or tandem measures are not in practice to date. Already, though, cells can be individually sorted using FACS by surface markers (and other attributes) at medium throughput and sequenced63. Similarly, laser microdissection and sequencing is fairly common. How soon until we see more all-in-one analyzers that are not separate machines, but effective tandem devices?

The inherent heterogeneity in cancer, coupled with the environmen-tal diversity due to unpredictable tissue structures, means that for now we have to ignore the nonuniformity of cancer when it is considered as a system. But lurking in the notion that cancer is a system is an implica-tion of order (no matter how chaotic cancer seems at first, second or third glances). If such order exists, then our answer both diagnostically and therapeutically lies in multiparameter measurements in which we consider the attributes of a single cell (and its interactions with other single cells and their attributes) as a key determinant of a complex interplay wherein a cancer’s weakness might hide.

The days of the single, one-size-fits-all diagnostic is gone. The future is in complexity—or at least a level of complex simplicity that serves as a surrogate of cancer systems we can use to go after the moving target of cancer pathology. Certainly single-cell sequencing will be at the forefront in this new approach. Flow cytometry and its variants are still useful for diagnostic purposes, but the various mass spectrometry technologies will enter the fray as soon as they are sufficiently reproducible for clinical validation.

One long-term goal is to move toward tagless, or near-tagless, measurement in which viable cells are prospectively isolated based on function, gene expression and genotypes and then assayed directly for tumor-initiating capacity. Techniques that probe the three-dimensional composition (architecture and cell types) of tumors, while minimally disrupting cell biology, are also required to gain more information on cell-type location and its relationship to disease. Currently, the only approaches we have readily available (e.g., green fluorescent proteins or other gene expression–based reporters) require genetic modifica-tion of the target cell. Although the tools we have available and on the horizon should keep us busy for the immediate future, it’s up to the next generation of single-cell converts to devise solutions that make what we use today look primitive by comparison.

ACkNowledGmeNtSG.P.N. is supported by the Rachford and Carlota A. Harris Endowed Professorship and grants from U19 AI057229, P01 CA034233, HHSN272200700038C, 1R01CA130826, CIRM DR1-01477 and RB2-01592, NCI RFA CA 09-011, NHLBI-HV-10-05(2), European Commission HEALTH.2010.1.2-1, and the Bill and Melinda Gates Foundation (GF12141-137101). S.C.B. is supported by

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

646  VOLUME 30 NUMBER 7 JULY 2012 nature biotechnology

r e v i e w

the Damon Runyon Cancer Research Foundation Fellowship (DRG-2017-09). The authors would also like to thank M. Angelo for useful discussions pertaining to the information in Table 1.

ComPetING FINANCIAl INteReStSThe authors declare competing financial interests: details are available in the online version of the paper.

Published online at http://www.nature.com/doifinder/10.1038/nbt.2283. reprints and permissions information is available online at http://www.nature.com/reprints/index.html.

1. Spencer, S.L. & Sorger, P.K. Measuring and modeling apoptosis in single cells. Cell 144, 926–939 (2011).

2. Spencer, S.L. et al. Non-genetic origins of cell-to-cell variability in TRAIL-induced apoptosis. Nature 459, 428–432 (2009).

3. Vincent, M. Cancer: a de-repression of a default survival program common to all cells? A life-history perspective on the nature of cancer. Bioessays 34, 72–82 (2012).

4. Greaves, M. & Maley, C.C. Clonal evolution in cancer. Nature 481, 306–313 (2012).

5. Visvader, J.E. Cells of origin in cancer. Nature 469, 314–322 (2011).6. Medema, J.P. & Vermeulen, L. Microenvironmental regulation of stem cells in

intestinal homeostasis and cancer. Nature 474, 318–326 (2011).7. Wang, X. et al. A luminal epithelial stem cell that is a cell of origin for prostate

cancer. Nature 461, 495–500 (2009).8. Knoepfler, P. Journal club. A cell biologist looks at the risk and promise of a new

insight into stem cells and cancer. Nature 457, 361 (2009).9. Yang, J. et al. Intratumoral heterogeneity determines discordant results of diagnostic

tests for human epidermal growth factor receptor (HER) 2 in gastric cancer specimens. Cell Biochem. Biophys. 62, 221–228 (2012).

10. Zhang, J., Yang, P.L. & Gray, N.S. Targeting cancer with small molecule kinase inhibitors. Nat. Rev. Cancer 9, 28–39 (2009).

11. Taussig, D.C. et al. Leukemia-initiating cells from some acute myeloid leukemia patients with mutated nucleophosmin reside in the CD34(-) fraction. Blood 115, 1976–1984 (2010).

12. Yap, T.A. et al. Intratumor heterogeneity: seeing the wood for the trees. Sci. Transl. Med. 4, 127ps10 (2012).

13. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012).

14. Longo, D.L. Tumor heterogeneity and personalized medicine. N. Engl. J. Med. 366, 956–957 (2012).

15. Cantor, H. et al. Characterization of subpopulations of T lymphocytes. I. Separation and functional studies of peripheral T-cells binding different amounts of fluorescent anti-Thy 1.2 (theta) antibody using a fluorescence-activated cell sorter (FACS). Cell. Immunol. 15, 180–196 (1975).

16. Parks, D.R. et al. Antigen-specific identification and cloning of hybridomas with a fluorescence-activated cell sorter. Proc. Natl. Acad. Sci. USA 76, 1962–1966 (1979).

17. Werner, M. et al. Microfluidic array cytometer based on refractive optical tweezers for parallel trapping, imaging and sorting of individual cells. Lab Chip 11, 2432–2439 (2011).

18. Wlodkowic, D. & Darzynkiewicz, Z. Rise of the micromachines: microfluidics and the future of cytometry. Methods Cell Biol. 102, 105–125 (2011).

19. Liu, A.Y., Roudier, M.P. & True, L.D. Heterogeneity in primary and metastatic prostate cancer as defined by cell surface CD profile. Am. J. Pathol. 165, 1543–1556 (2004).

20. Bragado, P. et al. Analysis of marker-defined HNSCC subpopulations reveals a dynamic regulation of tumor initiating properties. PLoS ONE 7, e29974 (2012).

21. Choijamts, B. et al. CD133+ cancer stem cell-like cells derived from uterine carcinosarcoma (malignant mixed Mullerian tumor). Stem Cells 29, 1485–1495 (2011).

22. Bonnet, D. & Dick, J.E. Human acute myeloid leukemia is organized as a hierarchy that originates from a primitive hematopoietic cell. Nat. Med. 3, 730–737 (1997).

23. Cho, R.W. & Clarke, M.F. Recent advances in cancer stem cells. Curr. Opin. Genet. Dev. 18, 48–53 (2008).

24. Lobo, N.A. et al. The biology of cancer stem cells. Annu. Rev. Cell Dev. Biol. 23, 675–699 (2007).

25. Nguyen, L.V. et al. Cancer stem cells: an evolving concept. Nat. Rev. Cancer 12, 133–143 (2012).

26. Irish, J.M. et al. Single cell profiling of potentiated phospho-protein networks in cancer cells. Cell 118, 217–228 (2004).

27. Kotecha, N. et al. Single-cell profiling identifies aberrant STAT5 activation in myeloid malignancies with specific clinical and biologic correlates. Cancer Cell 14, 335–343 (2008).

28. Irish, J.M. et al. B-cell signaling networks reveal a negative prognostic human lymphoma cell subset that emerges during tumor progression. Proc. Natl. Acad. Sci. USA 107, 12747–12754 (2010).

29. Palazzo, A.L. et al. Association of reactive oxygen species-mediated signal transduction with in vitro apoptosis sensitivity in chronic lymphocytic leukemia B cells. PLoS ONE 6, e24592 (2011).

30. Perez, O.D. & Nolan, G.P. Resistance is futile: assimilation of cellular machinery by HIV-1. Immunity 15, 687–690 (2001).

31. Krutzik, P.O. & Nolan, G.P. Intracellular phospho-protein staining techniques for flow cytometry: monitoring single cell signaling events. Cytometry A 55, 61–70 (2003).

32. Sachs, K. et al. Causal protein-signaling networks derived from multiparameter single-cell data. Science 308, 523–529 (2005).

33. Rosen, D.B. et al. Distinct patterns of DNA damage response and apoptosis correlate with Jak/Stat and PI3kinase response profiles in human acute myelogenous leukemia. PLoS ONE 5, e12405 (2010).

34. Krutzik, P.O. et al. High-content single-cell drug screening with phosphospecific flow cytometry. Nat. Chem. Biol. 4, 132–142 (2008).

35. Sachs, K. et al. Characterization of patient specific signaling via augmentation of Bayesian networks with disease and patient state nodes. Conf. Proc. IEEE Eng. Med. Biol. Soc. 2009, 6624–6627 (2009).

36. Sachs, K. et al. Learning signaling network structures with sparsely distributed data. J. Comput. Biol. 16, 201–212 (2009).

37. Rosen, D.B. et al. Assessing signaling pathways associated with in vitro resistance to cytotoxic agents in AML. Leuk. Res. 900–904 (2012).

38. Cesano, A. et al. Functional pathway analysis in acute myeloid leukemia using single cell network profiling assay: effect of specimen source (bone marrow or peripheral blood) on assay readouts. Cytometry B Clin. Cytom. 82, 158–172 (2012).

39. Longo, D.M. et al. Single-cell network profiling of peripheral blood mononuclear cells from healthy donors reveals age- and race-associated differences in immune signaling pathway activation. J. Immunol. 188, 1717–1725 (2012).

40. Yates, J.R., Ruse, C.I. & Nakorchevsky, A. Proteomics by mass spectrometry: approaches, advances, and applications. Annu. Rev. Biomed. Eng. 11, 49–79 (2009).

41. Bandura, D.R. et al. Mass cytometry: technique for real time single cell multitarget immunoassay based on inductively coupled plasma time-of-flight mass spectrometry. Anal. Chem. 81, 6813–6822 (2009).

42. Bendall, S.C. et al. Single-cell mass cytometry of differential immune and drug responses across a human hematopoietic continuum. Science 332, 687–696 (2011).

43. Qiu, P. et al. Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE. Nat. Biotechnol. 29, 886–891 (2011).

44. Newell, E.W. et al. Cytometry by time-of-flight shows combinatorial cytokine expression and virus-specific cell niches within a continuum of CD8+ T cell phenotypes. Immunity 36, 142–152 (2012).

45. Behbehani, G. et al. Single cell mass cytometry adapted to measurements of the cell cycle. Cytometry A (in the press).

46. Fienberg, H.G., Simonds. E.F. Fantl. W.J. Nolan. G.P. & Bodenmiller, B. A platinum-based covalent viability reagent for single-cell mass cytometry. Cytometry A 81, 467–475 (2012).

47. Hoshida, Y. et al. Integrative transcriptome analysis reveals common molecular subclasses of human hepatocellular carcinoma. Cancer Res. 69, 7385–7392 (2009).

48. Hoshida, Y. et al. Gene expression in fixed tissues and outcome in hepatocellular carcinoma. N. Engl. J. Med. 359, 1995–2004 (2008).

49. Mullighan, C.G. et al. Genomic analysis of the clonal origins of relapsed acute lymphoblastic leukemia. Science 322, 1377–1380 (2008).

50. Notta, F. et al. Evolution of human BCR-ABL1 lymphoblastic leukaemia-initiating cells. Nature 469, 362–367 (2011).

51. Kapranov, P., Ozsolak, F. & Milos, P.M. Profiling of short RNAs using Helicos single-molecule sequencing. Methods Mol. Biol. 822, 219–232 (2012).

52. Nagalakshmi, U. et al. The transcriptional landscape of the yeast genome defined by RNA sequencing. Science 320, 1344–1349 (2008).

53. Ozsolak, F. & Milos, P.M. RNA sequencing: advances, challenges and opportunities. Nat. Rev. Genet. 12, 87–98 (2011).

54. Wang, Z., Gerstein, M. & Snyder, M. RNA-Seq: a revolutionary tool for transcriptomics. Nat. Rev. Genet. 10, 57–63 (2009).

55. Ley, T.J. et al. DNA sequencing of a cytogenetically normal acute myeloid leukaemia genome. Nature 456, 66–72 (2008).

56. Mardis, E.R. et al. Recurring mutations found by sequencing an acute myeloid leukemia genome. N. Engl. J. Med. 361, 1058–1066 (2009).

57. Ding, L. et al. Clonal evolution in relapsed acute myeloid leukaemia revealed by whole-genome sequencing. Nature 481, 506–510 (2012).

58. Walter, M.J. et al. Clonal architecture of secondary acute myeloid leukemia. N. Engl. J. Med. 366, 1090–1098 (2012).

59. Anderson, K. et al. Genetic variegation of clonal architecture and propagating cells in leukaemia. Nature 469, 356–361 (2011).

60. Hou, Y. et al. Single-cell exome sequencing and monoclonal evolution of a JAK2-negative myeloproliferative neoplasm. Cell 148, 873–885 (2012).

61. Xu, X. et al. Single-cell exome sequencing reveals single-nucleotide mutation characteristics of a kidney tumor. Cell 148, 886–895 (2012).

62. Tang, F. et al. mRNA-Seq whole-transcriptome analysis of a single cell. Nat. Methods 6, 377–382 (2009).

63. Dalerba, P. et al. Single-cell dissection of transcriptional heterogeneity in human colon tumors. Nat. Biotechnol. 29, 1120–1127 (2011).

64. Gibbs, K.D. Jr. et al. Decoupling of tumor-initiating activity from stable immunophenotype in HoxA9-Meis1-driven AML. Cell Stem Cell 10, 210–217 (2012).

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.

©20

12 N

atu

re A

mer

ica,

Inc.

All

rig

hts

res

erve

d.

nature biotechnology  VOLUME 30 NUMBER 7 JULY 2012 647

r e v i e w

65. Cornett, D.S. et al. MALDI imaging mass spectrometry: molecular snapshots of biochemical systems. Nat. Methods 4, 828–833 (2007).

66. Ornatsky, O.I. et al. Study of cell antigens and intracellular DNA by identification of element-containing labels and metallointercalators using inductively coupled plasma mass spectrometry. Anal. Chem. 80, 2539–2547 (2008).

67. Steinhauser, M.L. et al. Multi-isotope imaging mass spectrometry quantifies stem cell division and metabolism. Nature 481, 516–519 (2012).

68. Engelhard, C. Inductively coupled plasma mass spectrometry: recent trends and developments. Anal. Bioanal. Chem. 399, 213–219 (2011).

69. Nishiguchi, M. et al. Ion optical evaluation of a miniature double-focusing mass spectrograph. Eur. J. Mass Spectrom. (Chichester, Eng.) 14, 7–15 (2008).

70. Schilling, G.D. et al. Continuous simultaneous detection in mass spectrometry. Anal. Chem. 79, 7662–7668 (2007).

71. De Stefano, J.A. et al. Analysis of Pneumocystis carinii cyst wall. II. Sugar composition. J. Protozool. 37, 436–441 (1990).

72. Barnes, J.H. 4th. et al. Characterization of a focal plane camera fitted to a Mattauch-Herzog geometry mass spectrograph. 2. Use with an inductively coupled plasma. Anal. Chem. 76, 2531–2536 (2004).

73. Barnes, J.H. 4th. et al. Use of a novel array detector for the direct analysis of solid samples by laser ablation inductively coupled plasma sector-field mass spectrometry. J. Am. Soc. Mass Spectrom. 15, 769–776 (2004).

74. Barnes, J.H. 4th. et al. Characterization of a focal plane camera fitted to a Mattauch-Herzog geometry mass spectrograph. 1. Use with a glow-discharge source. Anal. Chem. 74, 5327–5332 (2002).

75. Zhang, D.S. et al. Multi-isotope imaging mass spectrometry reveals slow protein turnover in hair-cell stereocilia. Nature 481, 520–524 (2012).

76. Lechene, C. et al. High-resolution quantitative imaging of mammalian and bacterial cells using stable isotope mass spectrometry. J. Biol. 5, 20 (2006).

77. Irish, J.M., Kotecha, N. & Nolan, G.P. Mapping normal and cancer cell signalling networks: towards single-cell proteomics. Nat. Rev. Cancer 6, 146–155 (2006).

78. Bendall, S.C. et al. A deep profiler’s guide to cytometry. Trends Immunol. published online, doi:10.1016/j.it.2012.02.010 (2 April 2012).

79. Ghosn, E.E. et al. Distinct B-cell lineage commitment distinguishes adult bone marrow hematopoietic stem cells. Proc. Natl. Acad. Sci. USA 109, 5394–5398 (2012).

80. Tung, J.W. et al. Modern flow cytometry: a practical approach. Clin. Lab. Med. 27, 453–468 (2007).

81. Giesen, C. et al. Multiplexed immunohistochemical detection of tumor markers in breast cancer tissue using laser ablation inductively coupled plasma mass spectrometry. Anal. Chem. 83, 8177–8183 (2011).

82. Moreno-Gordaliza, E. et al. Elemental bioimaging in kidney by LA-ICP-MS as a tool to study nephrotoxicity and renal protective strategies in cisplatin therapies. Anal. Chem. 83, 7933–7940 (2011).

npg

© 2

012

Nat

ure

Am

eric

a, In

c. A

ll rig

hts

rese

rved

.