Single-cell biology is an increasing area of interest for researchers as more and more studies show that cell populations — even those from the same organ or tissue — are not as homogeneous as previously thought.
Studying gene expression at the single-cell level is of particular interest to some researchers because it can provide much better insight into molecular mechanisms than gene expression studies of pooled cells. Rather than assessing the average gene expression over a set of heterogeneous cells, the ability to quantify transcription activity within one cell provides a clear picture of the biological pathways at work.
Anders Ståhlberg, a researcher at the Lundberg Laboratory for Cancer Research at the University of Gothenberg in Sweden, has been exploring the use of reverse transcription quantitative real-time PCR for single-cell gene expression analysis for several years. An overview of methods he and his colleagues have developed for this application is currently in press with the journal Methods.
PCR Insider recently caught up with Ståhlberg to discuss the challenges of real-time PCR-based single cell analysis, as well as some key applications for this approach. The following is an edited version of the interview.
What led you to start looking into PCR-based single-cell expression profiling?
We started from the beginning with normal gene expression profiling like many people are doing with microarrays to check the status of a clinical sample or a biological sample. However, when it comes to diagnostics, for example, people run into a lot of problems [with standard gene expression profiling] because when you take the samples you have rather heterogeneous materials, so they are contaminated by different amounts of unwanted cell types, or different numbers of cells, et cetera. So that, of course, will bias what sort of output you get.
Therefore, if you can actually select the cells you want to look at and can find the cells that you're interested in, you will gain much more information from your study.
So we did a lot of traditional gene expression measurements, and in many cases we found that it was not enough.
Your paper notes that the single-cell approach is still not widely adopted. Why is this the case?
For a long time it was more or less impossible to do it because of technical limitations. You can do different types of in situ hybridization, for example, but that requires quite advanced equipment and specialized experience, and you usually can only measure a few characteristics at a time. For example, if you do single-cell analysis on the protein level, you usually use some type of fluorescent protein and then you're restricted to just red, green, and yellow, and that's quite tricky.
I think the problem is also that people don't [understand] what sort of data you actually will get from [a single-cell experiment]. Because doing measurements on a cell population will give very different results than a single-cell measurement. With a single-cell measurement you should get very variable data because … the mRNA for a specific gene goes up and down all the time in a cell. And people usually think about gene expression as a steady-state value for a specific cell or a specific cell type, but that is not true. We've shown that with very sensitive methods for specific genes.
So by just knowing this fact that it should be variable, you can use that fact and adapt it to using real-time PCR and measure many genes. The benefit with our method is that it's quite easy to change the genes you're interested in looking at. And also, you can do pre-amplification. You can look at all transcripts. We usually do around 10 genes, and if we do pre-amplification we maybe do 50 genes or something like that.
Microarrays … can be good if you do some kind of screen and then you can go on with real-time PCR. If you use microarrays it becomes quite expensive if you want to do many cells. If you know that you have this heterogeneity within the same cell type, then you maybe want to look at at least 30, 40, or 50 cells, and then for one sample you're running 50 microarrays, which becomes quite expansive. And then, if you have many samples you want to analyze, there are very few who can do this kind of experiment.
But if you do it with our approach, it's quite feasible for many labs because [they] have the equipment. The thing is that you need to know what you're doing and really have optimized all the different steps — everything from how you take care of the cell, how you collect the cells, how you lyse the cells, how you reverse transcribe the cells — all these different steps you really need to have full control over. So, by putting it all together it's actually not that tricky to do these kinds of measurements today.
What would you consider to be the key differences between a standard real-time PCR gene expression protocol and a single-cell protocol?
The main difference [with single-cell analysis] is that you can't do any extraction from the beginning because you have so few transcripts. Depending on what kind of cells you have, maybe you have a few hundred thousand transcripts, and this may be five, ten thousand different genes that are expressed at a specific time point in a specific cell type. That means that there are not that many transcripts that you can detect, so you cannot really extract the sample and purify. If you do normal purification, as is standard for all other normal measurements, you maybe lose 90 percent of the material.
[ pagebreak ]
So we have looked into this quite closely and we actually lyse the cell. This means that you have to disrupt the cell walls and then make the mRNA accessible. And then it's important to know that you don't have any inhibition from this step and that you really get hold of all the transcripts you are interested in.
So the main difference is actually in the beginning: how you take care of the cell, how you lyse the cell. Then in the middle when you run the reverse transcription and the qPCR, that is quite similar to [standard PCR studies], although the efficiency of the reverse transcription is very important. When you convert the mRNA to cDNA, if this efficiency is very low — say 1 percent — and you have 100 transcripts, that means that you have only 1 cDNA copy instead of 100. Then of course it will be very hard to do a reliable quantification, so it's important to have high efficiency in all the steps.
And also, when you measure a specific gene, more or less all genes should give a similar value. The transcript level in a cell will always go up and down, and at some time point most cells will have a certain number of transcripts that is below your detection level because the detection level will vary between genes with maybe five copies for one gene and 20 for another. But the transcript may vary in expression level between zero and 200 transcripts. That means that you should not get results in all of the single cells that you're analyzing. You maybe should have values in 80 percent of the cells, but a zero value doesn't mean that it failed; it just means that there are maybe 10 copies or lower.
So you need to know these kinds of things. Many [make] the mistake here [in thinking] that there is some kind of error occurring, but that is usually not the problem.
Another issue is that people in traditional gene expression profiling want to do some kind of normalization, which means that you want to normalize all the samples you're comparing, and that traditional normalization usually means that you normalize it to some genes that are expressed at a constant level. But of course all the genes are going up and down, so that's not valid on a single-cell basis. So it's hard for people to go from normal gene expression profiling to single-cell measurement without knowing these kinds of things. They can run into all kinds of different problems.
Is there an alternative to normalization for single-cell analysis?
The good thing is that you know that you're measuring just one cell, so the main problem when you do these kinds of experiments is how to know that you're getting all the mRNA out of the cell or not. That is very hard to determine. I have not done these experiments, but one way would be to inject a known amount of an alien mRNA molecule, and then you do this in a lot of cells and see if you get out the same amount when you lyse them … but that is something that no one has done. That is a little bit tricky to do.
The normal way is actually that you know that you have one cell. With traditional [gene expression profiling], you maybe take 10,000 cells and then you extract [the DNA], but then you lose a lot of material so you can't really compare it to 10,000 cells any more, so therefore you go on with these reference genes [and normalization]. So therefore the value you actually get out is copy numbers for one specific cell.
So either you can deliver the data as the absolute number of cDNA molecules, and that's quite easy, or you can determine the RT efficiency. That requires some more experiments, but it can be done. Quite a lot has been done in this area of research, so it's quite nice to know how many copies we're talking about, and not just relative values. Relative values [are] what you get when you do normal gene expression profiling — something is up or down. But here we can actually say that we have 100 copies per cell. And our estimate is always a lower estimate, so we know that if we've determined that we have 100 cDNA molecules or 100 mRNA molecules, that means that we have this number or more, because if we lose something we don't know, but if we have 100 we know that we have at least 100.
What do you consider the key applications for single-cell gene expression analysis?
Well, it's my research field, but … identifying cancer stem cells [is one promising area]. Many tumors are known to arise from cancer stem cells, but there are very few, so if you can look in the individual cells and identify these cancer stem cells, you can see what makes them different from other cells, and then you can also compare that to circulating tumor cells in the blood.
Another thing is in stem cell differentiation. For example, if you want to do insulin-producing beta cells for transplantation to treat type 1 diabetes, then of course in many labs you do these in vitro, so you start with undifferentiated human embryonic stem cells for these induced pluripotent stem cells and then you differentiate them. But of course not all cells will be differentiated, so it can be quite important to look at the in vitro cells: Do these cells that we want to transplant look the same as other cells? We don't want to just harvest all of them because you know just by looking at them that some are differentiated and some are not. So it can be a characterization tool to investigate the differentiation between different kinds of stem cells.
So it's a very powerful tool for characterization. It could be used in pathology … to do more refined diagnosis.
[ pagebreak ]
These are the two applications I'm working on, but you can also think about more general use. For example, a few years ago we did a single-cell study on pancreatic cells — the cells that produce insulin, the beta cells. So we took cells and we treated them with glucose, and what we saw then was that it's not really that you activate all cells, it's more that you're turning on more and more cells. So if you have a low glucose level a few cells are activated. And then if you get more glucose you activate more and more cells. So it's not a uniform increase, it's that you're turning on more and more cells. And that's quite general for all different kinds of organs — that they behave differently to stimuli.
All cells are not identical. Some cells are activated easier by some signals and others by other signals, and that can be quite important for disease research and also for the general understanding of what's triggering different responses in the body. Because otherwise you just have this rough estimate — you take the whole liver and do a transcription profile, but then you don't know if it's a certain cell in a specific region that [is responding] or if it's all the cells that [are responding]. So that is also a quite interesting field to apply this.
[Another application is identifying] new subtypes of cells. Cell types are usually defined by morphology — you have a yes or no upon staining. But using single-cell analysis, you can actually get the quantitative value of different [genes]. We have found that there are different cell types depending on the level of genes that are expressed. Meaning, for example, that if we're looking at 10 genes, all are on in both cell types, but we can define two different cell types just from the level of the genes being expressed. That can't be done by any other method than single-cell measurement. Because usually you stain them and then see if a marker is on or off. So I think a lot of things can be done here.
There is quite a lot of research going on in single-cell biology. It's actually a quite rapidly growing field of research. It used to be that at conferences there were one or two talks about single-cell biology, but now there are entire conferences about single-cell biology and different approaches.
Many people know that it's very important to do this, but there have been many technical limitations, and usually they don't have any equipment to do it. So now, [with real-time PCR], you can use more standard equipment, but you just need to know how to do it and how to optimize all the steps.
That also makes it easier to get it accepted for diagnostic purposes, because if you needed very specialized equipment that no one has, it would be quite hard to implement it in more routine work.
Is single-cell PCR currently being used in diagnostic applications, or is that something that's more longer term?
I'm working with human tumor cells at the moment, but it's mainly at the research level. What we're doing is starting to identify cancer stem cells and see where they come from, what they look like, and we're also interested in looking at circulating tumor cells in the blood and seeing how similar these cells are to the tumor. Can we find a pattern [regarding] what cells are going into the blood?
These approaches to do the circulating tumor cells are very interesting for the clinic because lately some commercial kits have been approved so they can be used in the clinic as a prognostic for treatment. If you can see how many circulating tumor cells you have in the blood, that can also predict the outcome of the patient and what kind of treatment they can have. We're taking that a little bit further so we actually want to look at these cells that are identified and see what they are, how they behave. We do it both in vitro, we transplant cells into mice, and then we will also have access to clinical samples in the hospital, so that is the last step.
Do you envision this being used as a standard diagnostic tool, or would the fact that it requires a bit more expertise than a standard PCR workflow limit its adoption?
I think it has the potential to be [used as a diagnostic tool], because normal gene expression profiling is routine in many diagnostic laboratories today, so this doesn’t really require much more equipment. It's more know-how and doing it in the right way. So therefore, I think it can have a future there. Exactly how it will look is one thing, but as a research tool it's already there. For routine [diagnostic] work it may also have a use.
When we learn more about how to classify tumors, for example, if this turns out to be very important for how we should treat patients, then of course it will not be that hard to do the measurements if we know exactly what we should look for. And if we know what to look for, then doing the experiments will not be that hard.
Another good thing is that you need very little sample, so you don't need to do surgery. You can do a fine needle aspirate [or blood sample]. I can imagine that you might take a blood sample and if the profile can predict how you should treat patients it will be very useful and not traumatic for patients. So I think it can be very useful for more refined diagnosis in the near future, once we learn more about this heterogeneity.
What's next for you and your work in this area?
We have started to identify these cancer stem cells, so we're actually screening for these cells and looking more into them to see how different they are from other cells. We have a different kind of model system so that we can use it both in vitro in mice and then we also want to take this into human as the final thing.
We are interested in the biology behind the tumors we are studying. We want to understand the molecular mechanism, so we do a lot of basic research. Today we have set up all the experimental techniques that we want to have and they're working fine, so we're at the stage where we're doing a lot of experiments.
We're also working on some parallel projects where we're identifying different cell types in different organs, but that's more basic characterization in different areas and determining the degree of differentiation between different stem cells and things like that.
So there are actually two parts there: one understanding cancer stem cells in tumors and the other is basic characterization of different cells, especially stem cells.