Single-Cell Analysis: the Deepest Differences

TECHNOLOGY FEATURE THE DEEPEST DIFFERENCES To understand biological heterogeneity, researchers are learning how to profile the molecular contents of individual cells. S. GSCHMEISSNER/SPL

Studies are uncovering the molecular biology of individual tumour cells, such as this breast-cancer cell, seen in a coloured scanning electron micrograph.

BY CHARLOTTE SCHUBERT a single cell was by labelling molecules using Eberwine and others are using sequencing, fluorescence in situ hybridization. “People and techniques such as microfluidics and flow ames Eberwine, a neuroscientist with a were used to using microscopy to look at cytometry, to profile single cells — cataloguing penchant for invention, helped to pioneer RNA; they wanted to see it,” says Eberwine, RNA molecules, sequencing DNA and even Ja technique that is now routine. In the early who works at the University of Pennsylvania profiling metabolites and peptides. 1990s, he sucked the contents out of a single in Philadelphia. Studies indicate how strongly cells can show cell with a pipette, and examined the expres- Things have changed since then. Gene- their individuality. Brain cells may express as sion of a handful of genes using molecular expression analyses leaped forward in the mid- few as 65% of the same genes as their neigh- techniques that amplify RNA. His data veri- 1990s with the invention of the microarray. And bours, according to an unpublished analysis by fied a long-held assumption — that electrical the rise of high-throughput RNA sequencing, Eberwine. In the immune system, cells placed activity in a neuron simultaneously changes or RNA-seq, which spits out the sequences of in the same category on the basis of surface the abundance of multiple RNAs inside it1. thousands of cellular RNAs at once, has enabled markers can express different sets of genes, and But other researchers were sceptical. At the researchers to reveal the collection of active have different responses to vaccines2. And as time, just about the only way to detect RNA in genes in a cell in a single readout. tumour cells evolve, their genomes quickly

become twisted in unusual ways. intact RNA out of a single cell. Eberwine solved Single-cell techniques let researchers track the problem by capturing the material in the and catalogue this heterogeneity. They may same pipette used to measure electrical activity. be the only way to get at some fundamental Now, researchers can use a variety of techniques

questions, such as what makes individual to pick out single cells, from enzymatic diges- S. KIM/M. A. SURANI cells different biochemically and function- tion, which releases cells from tissues, to laser- ally. How much is each cell influenced by its capture microdissection. But it is still tricky, says microenvironment, and what is the role of Eberwine. “A major technical issue is how do stochasticity — random ‘noise’ in the behav- you do that initial capture.” iour of cellular molecules? Eberwine has used his pipette-capture These questions are getting more attention system to study individual warm-sensitive (see page 139). In 2009, Eberwine co-organ- neurons3, which regulate core body tem- ized a meeting on single-cell analysis at Cold perature and underlie fever. Together with Spring Harbor Laboratory in New York, along Tamas Bartfai, a neuroscientist at the Scripps with Sunney Xie, a single-cell biochemist at Research Institute in La Jolla, California, and Harvard University in Cambridge, Massachu- his colleagues, Eberwine examined the cells’ setts. The meeting drew 47 attendees. This July, Studies can probe single cells, such as those in the transcriptomes. The researchers identified 120 people went to the second such meeting. mouse embryo that give rise to gametes (green). transcripts for G-protein-coupled receptors — And the US National Institutes of Health (NIH) potential drug targets — that went undetected has launched an initiative to support single-cell picolitre (1 × 10−12 litres) of cytoplasm. And in screens of pooled cells. techniques (see ‘The NIH gets singular’). some key regulatory molecules are scarce — As techniques improve, they are letting But single-cell analysis is still an emerging just a few, hard-to-detect RNAs can exert a big researchers explore the heterogeneity within field. Many researchers say that protocols from effect on a cell. a cell. By cutting branches, or dendrites, off academic labs are often superior to commercial Many established techniques are only now neurons, Eberwine kits. “With any nascent field, there are lots of being applied to single cells. Fluorescent tag- “It’s the and his colleagues different approaches,” says Eberwine. “People ging and microscopy can be used to analyse individual cell have discovered that are trying lots of things to see if they can make molecules that have already been charac- that makes a RNA in dendrites the techniques more sensitive, more represent- terized. To profile previously unexamined can retain nucleotide ative of the state of a cell, easier and cheaper.” molecules, there is transcriptome analysis — decision.” sequences that target cataloguing the set of RNAs expressed in a cell the RNA to that loca- PROFILING HETEROGENEITY — as well as high-throughput methods based tion4. They could not have found such infor- The classic biochemical approach is limited, on microfluidics or flow cytometry. But getting mation by analysing the RNA of an entire say single-cell researchers. Grinding up and such techniques to work on single cells is not neuron. analysing the contents of large pools of cells easy, says Schroeder, whose research involves Developing technology will produce even — a procedure undertaken in thousands of long-term imaging of individual bone-marrow more fresh data. Most biologists will need to labs every day — averages out the results, says cells. Single-cell applications are, he says, “at work closely with computational biologists to Timm Schroeder, director of the Institute of least one level more demanding and complex evaluate the huge data sets that will result from Stem Cell Research at the German Center for than the conventional approaches”. cataloguing thousands of molecules in numer- Environmental Health in Munich. But, says And the unexplored biological terrain is ous single-cell experiments. Schroeder, “it’s the individual cell that makes vast. “We don’t even know what we are getting a decision” such as whether to fire an electrical into in terms of heterogeneity,” says Sherman EXTRA-LOUD AMPLIFICATION impulse, migrate or differentiate into a new Weissman, a geneticist at Yale University in Perhaps the best-known single-cell profiling cell type. New Haven, Connecticut. technique is transcriptomics. Azim Surani, a Looking at single cells to uncover the impetus Whatever the study, the first step is generally developmental biologist at the University of for such decisions means doing fussy experi- getting hold of the cells. When Eberwine first Cambridge, UK, uses this method to examine ments on a very small object: a cell might span struggled with studying gene expression in a cells of the early embryo, which are hard to about 10 micrometres and contain less than 1 neuron 20 years ago, it was difficult even to get study in large batches because they are so rare. He is tracing how, such cells turn into pluripo- tent embryonic stem cells in culture. Surani has adapted a single-cell protocol for The NIH gets singular the polymerase chain reaction (PCR) to work with RNA-seq. To do this, he has collaborated The challenges of single-cell analysis have Maryland. The programme will fund with technical experts such as Kaiqin Lao, a caught the attention of the US National new techniques in areas ranging from molecular cell biologist at Applied Biosystems Institutes of Health (NIH). The agency has microscopy to biochemistry, and foster in Foster City, California (a subsidiary of Life launched a programme to fund advances their commercialization. The NIH also sees Technologies in Carlsbad). In the cells of the in single-cell research, with a budget of a big need for tools to examine cells in their early mouse embryo, the team detected the around US$90 million over five years natural environment. expression of some 12,300 genes — 75% more from the NIH Common Fund, which Many of the techniques need an extra than were detected by microarray techniques5. backs science that crosses disciplines. push. “It’s still really difficult for individual Lao says he can now get his PCR technique to Grant applications are due early next labs to move into that area; the group of work with 1 picogram of RNA — one-tenth the year, and the NIH expects to make the researchers who work on this is still highly amount of RNA in a typical cell. first awards by September 2012, says specialized,” says Beckel-Mitchener. “If you The published protocol amplifies molecules Andrea Beckel-Mitchener, a programme want to reach the next level you really have only if they are no more than 3 kilobases long, officer at the NIH campus in Bethesda, to push the envelope.” C.S. so it misses about 40% of transcripts, says Lao. He and his colleagues are using different

1 DECEMBER 2011 | VOL 480 | NATURE | 135 © 2011 Macmillan Publishers Limited. All rights reserved TECHNOLOGY SINGLE-CELL ANALYSIS enzymes to increase that; Lao can now amplify TargetAmp from Epicentre Biotechnologies genome and RNA content of a single cancer 10-kilobase transcripts, corresponding to of Madison, Wisconsin (owned by Illumina cell using the company’s technology. about 99% of transcription, he says. of San Diego, California), and MessageAmp Nicholas Navin, a geneticist at the MD Another technique to amplify a cell’s RNA from Ambion of Austin, Texas, which is Anderson Cancer Center in Houston, Texas, is antisense RNA (aRNA), an in vitro tran- owned by Life Technologies. Both can work is one of only a handful of researchers who scription technique from Eberwine and his for single cells, says Eberwine. Companies have sequenced the genomes of single cells colleagues6, in which a cell’s RNA is copied such as NuGEN in San Carlos, California, from eukaryotic organisms. This year, in col- into a stable DNA library, with each DNA and Sigma-Aldrich in St Louis, Missouri (in laboration with Michael Wigler, a geneticist molecule containing partnership with Rubicon Genomics of Ann at Cold Spring Harbor, and his colleagues, a short sequence rec- “Every neuron Arbor, Michigan), have products designed for Navin sequenced the DNA of 100 indi- ognized by an RNA is probably small amounts of RNA, and some say that their vidual cells from different parts of each of polymerase. The different from systems can work for single cells. It is unclear two human breast tumours, tracing how polymerase uses the every other when Life Technologies might release a prod- the cancer evolves as it spreads7. It took sev- DNA library to make neuron.” uct based on Lao’s method, but both Eberwine eral years and cost about $2,000 per cell; the multiple copies of the and Lao report that their single-cell techniques cost has since fallen to about $200 per cell, RNA. are being used successfully in other labs. he says. In the end, Navin was able to reli- Each approach has its advantages, and its ably cover about 6% of the genome of a cell problems. Bias can be introduced to PCR when THE GENOME GAP — enough to assess some larger copy-num- certain sequences dominate during amplifica- Many researchers want to analyse not just the ber aberrations, but not to look at the accu- tion, so approaches based on this technique are transcriptome of a cell, but the underlying mulation of point mutations during tumour less quantitative than aRNA. But aRNA is less genome. This would be particularly relevant evolution. efficient than PCR, and can take days, notes for cancer cells, with their warped DNA, and The limitation, say Navin and other Weissman. Life Technologies is offering US$1 million to researchers, is the technique used to amplify Commercially available aRNA kits include the first researchers to sequence the entire the DNA: whole-genome amplification, which Beyond amplification Unlike DNA and RNA, some cellular and analyse their molecular components. University of Illinois at Urbana–Champaign, molecules cannot be amplified. Very few labs The researchers have adapted the system for instance, have used matrix-assisted have tried to profile proteins, metabolites for the detection and analysis of circulating laser desorption/ionization (MALDI) or peptides in single cells. “The analytical tumour cells and are launching a company mass spectrometry to profile the major challenges are extreme,” says Renato Zenobi, based on the technology: MiCareo, in Taipei, metabolites and peptide neurotransmitters a chemist at the Swiss Federal Institute of Taiwan. released in a large neuron of the sea slug Technology in Zurich. Researchers working with Garry Nolan, Pleurobranchea californica in response to To assess populations of proteins, some a geneticist at Stanford University in hunger16. In unpublished work, Zenobi and researchers have used fluorescent tagging. California, and Scott Tanner, a biomedical his colleagues have generated a map of the Sunney Xie, a single-cell biochemist engineer at the University of Toronto yeast metabolome using MALDI mass- at Harvard University in Cambridge, in Canada, are revamping a tried and spectrometry data from single cells. He says Massachusetts, for instance, has harnessed true high-throughput technique — flow that he can detect metabolites in the low a microfluidic device to quantify the set of cytometry, which relies on the detection of attomole range (around 1 million molecules), proteins and RNA in the Escherichia coli fluorescent tags, and generally measures and can identify more than 200 correlations bacterium, using a fluorescent-protein six to ten parameters simultaneously. The — for instance, whether increases in one reporter library12. team has developed ‘mass cytometry’. The metabolite correlate with decreases in Sherman Weissman, a geneticist at Yale technique can simultaneously measure another. University in New Haven, Connecticut, has 34 parameters in a single cell15, including Other researchers use different mass hopes for ribosome profiling, a molecular protein phosphorylation and cell-surface spectrometry methods. Cynthia McMurray, technique to monitor protein production. antigens, and has the potential for some a neuroscientist at Lawrence Berkeley The procedure involves obtaining a 100. The instrument that they use evaluates National Laboratory in Berkeley, California, sequence of RNAs (about 28 nucleotides up to 1,000 cells per second, and relies on is constructing single-cell metabolic maps long) that have been sequestered by antibody-bound tags consisting of rare-earth from thin sections of brain, comparing protein-making ribosomes13,14. The method elements detected by mass spectrometry. healthy, aged and diseased mice using works with pools of cells, but it may have It is sold by DVS Sciences in Sunnyvale, nanostructure-initiator mass spectrometry. potential for single cells, says Weissman. California, and costs US$600,000. She ultimately aims to create a three- Daniel Chiu, a chemist at the University Mass spectrometry is routinely used for dimensional metabolic map of the brain. of Washington in Seattle, has a device that profiling proteins and metabolites, but does Zenobi is negotiating with Sigma Aldrich can add reagents to label a cell or organelle not have the sensitivity of optical techniques in St Louis, Missouri, about commercializing and send its contents — one molecule at such as fluorescence. Very few proteins a micrometre-scale device17 that prepares a time — to an instrument that can count are present at more than 100,000 copies multiple single cells on an array for fluorescent molecules11. His team has per cell, so most are below the limit of analysis by MALDI mass spectrometry. The hooked this up to another microfluidic detection for commercially available mass device may be available as early as next system: a high-throughput device to isolate spectrometers. But researchers are making year. “Once people realize these types of single cells from a population. With this progress, mainly with more abundant analyses can be done, they will jump on two-step system, the group can isolate cells metabolites and peptides. Researchers at the them,” he says. C.S.

relies on an enzyme that copies some genomic regions but skips others. By tweaking this step, Navin says, he is now exceeding 50% coverage of the genome of a human cell, although his D. MCCARTHY/SPL work has not yet been published. Navin is not the only one tackling this problem. At the Cold Spring Harbor Meeting this year, Xie said that he and his colleagues had been able to sequence 85% of the genome of a mammalian cell. The paper describing it has not yet been published — but researchers who have seen the data are impressed. “Sunney nailed it,” says Lao. This is welcome news to researchers such as Fred Gage, a neuroscientist at the Salk Institute for Biological Sciences in San Diego, California, who wants to sequence individual neurons. He has found that long interspersed elements (LINEs) — DNA sequences that can move around in the genome — form new insertions when neurons are born from neuronal Researchers are going beyond single cells to profile parts of cells, such as neuronal extensions. stem cells8. Every neuron probably contains unique LINE insertions, with most cells having Several companies are working on these deploys the vortex trap to fuse droplets and between 80 and 300. “Every neuron is probably goals, offering miniature devices that inte- change the concentration of reagents11. different from every other neuron,” says Gage. grate multiple steps for the high-throughput Chiu’s lab has also developed microfluidic analysis of single cells. Fluidigm of South devices for quantifying fluorescently tagged ONE DEVICE, LOTS OF INFORMATION San Francisco, California, markets a micro molecules, and for detecting and analysing cells Questions on single cells often lead fluidic system that can simultaneously analyse that are rare in a population, such as tumour researchers into difficult experimental ter- 96 genes in 96 individual cells using quanti- cells circulating in the blood (see ‘Beyond rain. For help in navigating such tricky tative PCR. Fluidigm systems have been amplification’). territory, Gage recommends collaborating deployed to uncover previously unrecog- Ultimately, a combination of techniques will with the best technical experts; he is work- nized subsets of immune cells2, and to exam- be necessary for researchers to attain their goal ing with Roger Lasken, a leader in sequencing ine variability in the response of single cells to of measuring multiple parameters in a single, unculturable microbes at the J. Craig Venter cytokine signalling9. RainDance Technologies living cell. “The more parameters you can Institute in San Diego. of Lexington, Massachusetts, also sells micro- define — the transcriptome, the peptide-ome, But many researchers venturing into scale kits to analyse single cells. how a cell looks, how it responds to drugs — single-cell analysis will be on their own, so Usability and high throughput are a boon for the more information you are going to get out,” techniques will have to become more auto- miniature devices. Integrated steps also help to says Eberwine. mated, integrated and kit-like, says Jonathan conserve precious samples, and small volumes Eberwine is confident that these methods Sweedler, a chemist at the University of Illinois can aid the dynamics of biochemical reactions will emerge, even if it takes years. “I think we at Urbana–Champaign. “Researchers will be — for instance, they can reduce amplification will be successful,” he says, “and if we are not, able to buy a device that has 48 steps incorpo- bias in PCR reactions, notes Stephen Quake, a somebody else will be.” ■ rated into one platform,” he says. Widespread bioengineer at Stanford University in Califor- uptake of single-cell analysis will also require nia, and co-founder of Fluidigm. Charlotte Schubert is a freelance writer high-throughput analyses of dozens to thou- “Working with small volumes gives you based in Seattle, Washington. sands of cells to tease out measurement errors some real technical advantages,” says Quake, 1. Mackler, S. A., Brooks, B. P. & Eberwine, J. H. Neuron from real heterogeneity. whose lab is harnessing microfluidics to 9, 539–548 (1992). develop a technique for single-cell transcrip- 2. Flatz, L. et al. Proc. Natl Acad. Sci. USA 108, tomics, and has created a device to isolate and 5724–5729 (2011). 10 3. Eberwine, J. & Bartfai, T. Pharmacol Ther. 129, sequence single chromosomes . 241–259 (2011). 4. Buckley, P. T. et al. Neuron 69, 877–884 (2011). T. SCHOCHET T. PUTTING IT ALL TOGETHER 5. Tang, F. et al. Nature Meth. 6, 377–382 (2009). But high-throughput techniques will be 6. Van Gelder, R. N. et al. Proc. Natl Acad. Sci. USA 87, 1663–1667 (1990). limited if what they measure is too simple. To 7. Navin, N. et al. Nature 472, 90–94 (2011). grasp how a cell works, “you need to under- 8. Coufal, N. G. et al. Nature 460, 1127–1131 (2009). stand not just chemistry, but spatial and tem- 9. Tay, S. et al. Nature 466, 267–271 (2010). 10. Fan, H. C., Wang, J., Potanina, A. & Quake, S. R. poral information”, says Daniel Chiu, a chemist Nature Biotechnol. 29, 51–57 (2011). at the University of Washington in Seattle. To 11. Chiu, D. T. & Lorenz, R. M. Acc. Chem. Res. 42, integrate these analyses, his lab combines 649–658 (2009). microfluidics, nanomaterials and optics. 12. Taniguchi, Y. Science 329, 533–538 (2010). 13. Ingolia, N. T., Lareau, L. F. &Weissman, J. S. Cell 147, Chiu’s team has developed a technique for 789–802 (2011). single-cell nanosurgery using a ‘vortex trap’, an 14. Ingolia, N. T., Ghaemmaghami, S., Newman, J. R. S. optical method that can manipulate organelles & Weissman, J. S. Science 324, 218–223 (2009). 15. Bendall, S. C. et al. Science 332, 687–696 (2011). or liquid droplets. The group has isolated sin- 16. Hatcher, N. G. et al. J. Neurochem. 104, 1358–1363 James Eberwine’s ingenuity in pioneering single- gle mitochondria from cells and prepared (2008). cell protocols has led to dozens of patents. them for analysis on a ‘droplet nanolab’, which 17. Urban, P. L. Lab Chip 10, 3206–3209 (2010).