Dana Pe’er, Garry Noland and single-cell proteomics at AACR


Standing-room only crowd at the 2014 AACR Symposium "Single Cell Analysis of the Tumor"

Standing-room only crowd at the 2014 AACR Symposium “Single Cell Analysis of the Tumor”

During AACR a number of great sessions were presented at ‘meet the expert’ sessions at 7am in the morning. One benefit of coming out West from the East Coast is not being able to stay up past 10:30pm or so local time, and waking up on my own at 4:30am local time every day. (A friend from the NCI told me he doesn’t like to ‘ping-pong between time zones’, and I whole-heartedly agree!)

At one of the these 7am sessions was one by Dana Pe’er of Columbia University, entitled “Understanding tumor heterogeneity using 40 markers at single cell resolution”. I thought: intriguing title, this should be interesting.

By using a combination of flow sorting and an atomic mass spectrometer this technology is called a ‘CyTOF mass cytometer’ that uses heavy-metal labelled antibodies to decorate single cells. Single cells are bound with the antibody-isotope conjugates, sprayed as single-cell droplets into inductively coupled plasma. This plasma stream (5000K to 7500K) is created by passing argon gas through an induction coil and high radio-frequency electric current.

Due to the speed (both the flow cytometry technology and mass spec on the order of 1,000 cells/sec), this technology enables the simultaneous examination of 45 different antibody-metal tag conjugates, all measured individually in single cells.

A little context here: way back in the late 1950’s flow cytometry was first developed, and the field of cell biology changed significantly with the development of the first flow cytometers into the market in the 1970’s. I remember after finishing graduate school (okay I’m definitely dating myself) and while at the Veterans Administration Medical Center of San Francisco Division of Immunology, the large laboratory I was working in purchased an advanced flow cytometer (this was about 1989). It was an amazing instrument – expensive for any laboratory at that time (over $100,000), it was a luxury few labs could afford, but this laboratory was a prominent one working on NFkB and NK cells at that time. Anyway this huge instrument had its own dedicated room, and dedicated technician to run it.

Fast forward to the early 2000’s, when quantum-dot labeled antibodies, multiple Alexa fluors (courtesy of Molecular Probes), and multiple lasers can measure 6-12 different antibodies simultaneously using a state-of-the-art flow cytometer (this review in 2004 by Pefettor et al Nat Rev Immunol refers to 17 colors.) However trying to push the multiplexing ability further is hampered by spectral overlap (from the emission spectra of the fluors), that can cause variability in the results.

A technology out of the University of Toronto and supported by funding from the Ontario Institute for Cancer Research (some of the background is here in this Lou et al Angew Chem Int Ed Engl paper) turned into a company called DVS Sciences, and they launched the first version of their instrument called the CyTOF® Mass Cytometer in 2009. This instrument was further developed, and in 2011 a very nice Science paper reported a 34 independent parameter measurement system of single cells from the entire hematopoetic spectrum. From that paper, Figure 4 just blows my mind – since it is available as open-access, I’ll take the liberty to show you what I’m talking about.

Figure 2B from Bendall et al Science 2011, "Single-Cell Mass Cytometry of Differential Immune and Drug Responses Across a Human Hematopoietic Continuum"

Figure 2B from Bendall et al Science 2011, “Single-Cell Mass Cytometry of Differential Immune and Drug Responses Across a Human Hematopoietic Continuum”

You can see from figure 2B here the plot only looks at one marker, the CD45RA (which is the ‘expression % maximum’ in the color key on the bottom right), and is the result of narrowing down the 34 parameters and the algorithm chose out 13 cell-surface antigents. Hematopoetic Stem Cells are at the top of the diagram (‘HSC’) and every downstream cell lineage, from NK cells, to naïve CD8+ T-cells to CD4+ T-cells to erythrocytes to monocytes to different forms of platelets. So you have in a flattened 2D plot, the representation of 15 cell-surface markers, and each color represents the level of CD45RA as one example.

I am reminded of the huge wall posters that antibody companies provide with large trees of progression, and all the cluster of differentiation – CD – nomenclature named in leukocyte development. Looking around I found this poster library on Abcam’s website with links to many of these types of PDF charts, and they produced this one for T-cells with the journal Nature as a sponsored poster.)

In this system these lanthanide-labeled antibodies separate very cleanly, and give a much better quantitative measurement compared to flow cytometry’s ‘optical-based’ dependence upon fluorescence. One drawback is that these cells are ablated during the measurement, while with flow cytometry the cells can be gated (sorted) appropriately and then used for other purposes downstream (such as whole transcriptome sequencing, whole genome or exome sequencing, or other analyses). Another drawback that the Science paper mentions is lower sensitivity, although if you look at their comparison data you can easily tell how similar the data is qualitatively.

Dr. Pe’er’s talk began with a question: how does one visualize 30 dimensional space? She showed a somewhat humorous slide with a 20×20 grid of small 3D plots too small to determine the details. A 3D principal components plot (wiki reference to image) can give four dimensions, using color as the fourth dimension. If video is used, those colors can change over time, adding a fifth dimension (the 3D shapes and colors changing over time). But how do you look at the seventh, eighth, ninth dimension?

She published a method called viSNE in this 2013 Nature Biotechnology paper that takes these many dimensions and and make appropriate ‘clusters’ along these many dimensions, and then developed a method to represent these clusters in 2 linear dimensions plus color. Taking this analysis to AML, she and collaborators were able to take a cohort of 28 samples, and look at 16 cell surface markers, and 15 intracellular functional markers, under different cell signalling perturbations, across 16 million cells.

Think about that for a second – a total of 31 surface and internal proteins measured simultaneously at single-cell resolution with 16 million replicates, in one experiment. She developed several tools to deal with such complex dataset – one called GRAPHITE that she likened to as a ‘Facebook for cells’ (using k-nn with shared neighbors) (Link: insert here). Another tool called Aperture <link here> took the analysis a step beyond an average and a distribution, and took the distribution differences of subpopulations into account.

From what I understand of this, you can take the clustering from viSNE and get a lot of subpopulations due to the hetergeneous nature of AML; she mentioned that with 224 perturbations of the system, they were able to determine 487 unique subpopulations, each with its own distribution. Looking at the difference between distributions thus requires the GRAPHITE tool.

She then showed data on two genes pAkt and pSTAT3, comparing normal and tumor cells and how their signalling gets disrupted in AML cells compared to healthy ones, and that a unique signalling phenotype can arise. By classifying by this signalling phenotype, this has the potential to derive a predictor of AML survivability.

Dr. Pe’er then went on to look at B-cell development chronology. (At the time of the presentation, their manuscript had been accepted for publication by the journal Cell.) They gated cells via flow sorting into four developmental states, and then analyzed each stage for developmental signatures.

She showed that B-cell development is non-linear, and the difficulty in developing a robust method to model this non-linearity is exhibited in the fact that their supplemental information for the manuscript runs some 30 pages long. (I read recently of one Twitter user complaining of needing to slog through a 200+ page supplement.)

For B-cell development they used 25 surface markers, 18 internal markers, 5 stimulus variables, across 10 samples of 20K to 300K cells apiece. They were able to narrow the B-cell developmental subtype to 6 markers (thus ‘conventional’ flow sorting can be used to separate these subtypes) and then used qPCR to validate the VDJ and other immunoglobulin recombination as a function of the accuracy of their developmental distinguishing power.

One question that came up after her talk was whether how the cells were treated before analysis would affect the results. The samples used were frozen for several years beforehand, so they didn’t have control over how each sample was handled before storage.

Another question that came up was the fact that AML samples have an admixture of normal cells with tumor cells, and whether that posed any problems for this technology. She answered that using only two or three dimensions for analysis, yes that would be a problem; but in 12 dimensions it is easy to separate out the tumor signal from the normal.

The following afternoon, Garry Nolan from Stanford University gave a talk during a standing-room-only symposium (pictured above). He entitled his talk “A single cell systems-structured view of immunity and cancer”.

He gave a few more details about the CyTOF method, in terms of speed (1K cells/sec) as well as their use of 45 dimensional analysis, with the potential to go up to 60-80 in the near future.

He mentioned Dana Pe’er’s work for analysis (referring to yet another tool she developed called ‘Wanderlust’), and shared data from the AML set mentioned above but with some additional data around the perturbations and the surface phenotype couples to the intracellular signalling, comparing normal to AML tumor cells. (It appears that both ‘Wanderlust’ and ‘GRAPHITE’ await the upcoming Cell publication.)

It turns out that the stereotypic signaling responses are decoupled from surface markers in AML, in a non-coordinated fashion. Viewing the variables as colors, it was very clear visually in the slide he presented.

Dr. Nolan went on to their next advance in this technology, something called Multiplexed Ion Beam Imaging (MIBI) that is a quantitative alternative to IHC, and it was recently published in Nature Medicine. With this method you can do high-dimensional immunohistochemistry (i.e. in-situ FFPE tissue sections); the paper used 10 different labels simultaneously with a five-log dynamic range, and he mentioned the ability to look at epigenetics (presumably with methyl-specific antibodies). He showed a movie where 3D reconstructions are feasible, and a resolution from 10nm – 1 um. They expect to have in instrument built within six months.

During the question session, he said that mRNA can also be detected with antibodies down to a 5 copies / cell sensitivity. I’m going to have to do my homework on that one – sounds like an alternative to Advanced Cell Diagnostics, which I’ve written up before.

About fourteen years ago as a Product Manager at QIAGEN, I was very involved with recombinant protein expression, purification and labeled antibody detection. At that time QIAGEN’s agreement with Luminex™ was just getting started, Ciphergen’s SELDI-TOF® was getting a lot of attention, and it appeared that protein arrays were just starting to be commercialized. Now with a tool like this one it seems like the promise of proteomics can be fulfilled, but with the remarkable ability to look at  single-cell resolution.

Only afterwards did I realize that DVS Sciences was acquired by Fluidigm only two months ago for some $207M. Alas, I need to expand my reading list to proteomics (again). So many journals, so little time.


About Dale Yuzuki

A sales and marketing professional in the life sciences research-tools area, Dale currently is employed by Olink as their Americas Field Marketing Director. https://olink.com For additional biographical information, please see my LinkedIn profile here: http://www.linkedin.com/in/daleyuzuki and also find me on Twitter @DaleYuzuki.

Leave a comment

Your email address will not be published. Required fields are marked *