Miao-Ping Chien, Principal Investigator, Department of Molecular Genetics, Erasmus University Medical Center, joins us to discuss microscopy-based functional and spatial single-cell sequencing, therapy-resistant cancer cells and using AI for multi-omics data integration.
Please note the transcript has been edited for brevity and clarity.
FLG: Hello and welcome to the latest ‘A Spotlight On’ interview. Today I’m joined by Miao-Ping Chien, and we’re going to be talking about Miao’s work on microscopy-based functional and spatial single-cell sequencing. Miao, if you could please introduce yourself and tell everyone a little about what you do.
Miao-Ping Chien: Thank you, Miyako. My name is Miao-Ping Chien, and I am currently an Assistant Professor at Erasmus University Medical Centre in the Department of Medical Genetics. I’m also a Principal Investigator at the Oncology Institute. So, a really short introduction about what we are doing. Our group actually develops single-cell technologies for cancer research. More specifically, we actually combine multiple multidisciplinary techniques including advanced imaging, computational analysis and single-cell techniques to study rare and aggressive cancer cells.
FLG: Thank you. Why were you drawn to studying therapy-resistant cancer cells?
Miao-Ping Chien: That’s a good question. We’re interested in those rare and aggressive cancer cells that are responsible for tumour metastasis or therapy resistance because we know at this stage, cancers are not really curable. Part of the reason we cannot completely cure cancers is that after the treatment, there are always small subpopulations of cancer cells that survive; we give these surviving cell populations a collective term, aggressive cancer cells. These cells survive after the treatment and over a period of time, they would relapse and regrow, sometimes become even more aggressive than before. So, that’s the reason we’re interested in studying those small populations of resistant cells.
FLG: Could you tell us about the challenges in identifying and analysing these aggressive cancer cells?
Miao-Ping Chien: So, people in this field are aware of the existence of these types of cells, but in the past few decades, the majority of techniques we used to study these cells mainly involved looking at specific markers. If we know the markers of cells, we can use techniques like cell sorting, for example, where you use an antibody with a probe to target certain cells and then those cells can be separated using the FACS machine, for instance, and this works quite well to a certain extent. People also use these techniques to isolate resistant cells, but the thing is, it’s not thorough enough because not every cell will express those markers. So purely using those markers to identify or study those cells is simply not enough.
Because of this, the approach we are using in our group is basically to look at these cells not based on their markers, but on their behaviour. So, we really profile, study and analyse individual cells’ behaviour, and from this data we can then identify the subpopulations that are potentially more aggressive, after which we further isolate and profile these cells. The challenge here is really to capture the dynamic information of cell behaviour. And collecting this type of information is very challenging using currently available technologies. So that, to me, is one of the main challenges, but it has been tackled by the technology developed in our group.
FLG: Like you said, your lab developed a variety of multidisciplinary approaches to investigate these rare, aggressive cancer cells. Could you give us a quick overview of the different tools that you have developed?
Miao-Ping Chien: Yes, of course. As mentioned, our group is multidisciplinary, and we operate within four different research pillars. One is advanced imaging, we actually built our own custom-built microscope which allows us to screen really large quantities of cells. We’re talking about tens, hundreds of thousands of cells in one single field of view, which is very impressive. Importantly, despite being able to see large amounts of cells, we don’t lose any spatial resolution. This allows us to look at individual cell information. So we can see many cells, but don’t lose the single-cell or sub-ceullar information. This kind of combination is really hard to have using commercial microscopes, and that’s part of the reason we have our own setup. In this setup we also implemented a device so that no matter what kind of cells you identify, you can also pinpoint and photo-label those cells, and then the cells can be isolated accordingly. That is part of the reason we have this capability to image and screen a bunch of cells, and from there we’re able to identify and isolate certain subpopulations of cells. That’s the first tool we use in our group, advanced imaging.
The second part of our process is, once we create a lot of big imaging data, we also need to have a really robust and advanced image analysis algorithm because, as I also mentioned earlier, we are interested in the behaviours of individual cells. We need to have an algorithm that can help us analyse individual cells or the imaging data we acquired on the fly, and this is also very challenging. For that, we also scripted our own image analysis to be able to instantly profile the data and instantly export individual cells’ features, and we can then use that information for further cell isolation.
So that’s the second part, and the third part will be single-cell technology development. We use quite a lot of single-cell RNA sequencing, and we started to use single-cell genomic sequencing and proteomic profiling as well. Those techniques have been developed quite well in the field, and we kind of adapt it to fit with our technique, our pipelines. So that’s the third tool we use, and the last one is bioinformatic analysis. We create quite a lot of data, and this data is also quite different than the standard single-cell sequencing you will normally get. Because of this, we also need to adapt a little bit of the algorithms and use them to identify potential targets or hidden driving mechanisms that cannot be identified using current analysis methods. Those are basically the four different types of tools we use.
FLG: A lot of multi-omics data integration there, a lot of focus on single-cell data and spatial resolution as well as combining that with image analysis. I’m assuming a lot of machine learning and AI is involved, too. Could you tell us a little bit about the microscopy-based part of your own work and elaborate on the functional and spatial single-cell sequencing technique?
Miao-Ping Chien: First, we developed what we call microscopy based functional single-cell sequencing. Through this, we can profile or sequence individual cells based on their functional features observed under a microscope, and that’s why we call it functional single-cell sequencing. We’re able to isolate cells based on any functional feature visualised under a microscope. Say we screened 10,000 cells under a single field of view, from there we’d identify maybe 100 cells that display, aggressive migration. We’d then want to isolate those 100 cells, and can use the photolabelling technique I mentioned earlier to label and isolate the cells of interest for downstream single-cell sequencing. The way we do that is: you screen a bunch of cells and then perform real-time image analysis. You then identify the cells you want followed by photolabeling and cell isolation. And I’ve used the example of aggressive migration here but you can do this with any feature you can imagine. We’ve isolated cells before based on their abnormal DNA damage response to radiation, where we looked at DNA damage responses of individual cells after radiation. From there, you can also see heterogeneous populations. We can also look at abnormal cell division, because under the microscope you can see how cells perform their mitosis. From observing those regular and irregular mitosis features, you can already see which cells are abnormal and which cells are normal, and so we can then isolate the cells we want to study.
You can also see which immune cells interact with cancer cells, which do not, which can kill cancer cells, which cannot, et cetera. You can observe these interactions under a microscope, and can then selectively separate those cells. I mentioned the techniques to profile cells based on the function of features, and another function concerns the fact that cells are located in different spatial locations. That’s how we expand our technology to spatial omics, because under this setup, you can also see where those cells are located and which area they’re in, as well as what cells they interact with. So we can also immediately identify this information, and this can also be used. That’s why our technique can also be applied to spatial profiling.
FLG: What’s really novel about your approach is you’re looking at behavioural and functional hallmarks to isolate cells, and that obviously has an advantage over epitope-dependent forms of isolation, like you mentioned. What about microfluidics? Because that’s a big conversation at the moment. Do you take advantage of new advances in microfluidics in terms of isolating cells?
Miao-Ping Chien: Yeah, so that’s a good question. We currently don’t need microfluids – microfluids are perfect when you don’t necessarily need to isolate specific cells with certain markers or phenotypes. When you do this, you basically extract or dissociate your samples into single-cell formats and then you feed the dissociated cells through this microfluid. They will then be encapsulated in droplets, and then the RNA will be converted to cDNA, and after this, you can create your library for sequencing. In that setting, microfluids work perfectly and it also enables sequencing tens of thousands of cells simultaneously. From the expression profiles of those tens of thousands of individual cells, you can identify different cell types. This allows you to see different combinations of cells in different treatment conditions. So that’s how people use the microfluid system.
But for us, we don’t really need that because we are more interested in studying subpopulations of cells that we know are aggressive, and we want to know what their profiles are. Using microfluidics, you cannot have that selection power because you randomly sequence a bunch of cells, but you don’t know in the end if the cells you profile display aggressive features; you can’t make that link. So what we do is we first screen under the microscope, and we know which cells are the aggressive ones. We selectively isolate those aggressive cells and we sort them; we often use plate-based sequencing because our cells are quite rare and 384 wells are enough for us to study them. In that case, we basically narrow down to the ones we want. After this, we do in depth sequencing.
FLG: Thank you for elaborating on that. That’s really interesting. You also use machine learning and AI for your analysis, so could you expand on the machine learning and AI approach you use, as well as the advances in machine learning that excite you most and whether or not you think it’s going to be applied more and more for data integration or multi-omics?
Miao-Ping Chien: That’s a very good question. AI has birthed some really useful techniques, but as you also mentioned, it has its own challenges as well. We do apply AI in our research, and earlier, I talked about four different pillars in our research group. We currently implement AI not only for image analysis, but also for bioinformatic analysis. In terms of image analysis, when we curate large image data, we need to instantly process them to recognise, detect and track cells, This needs to be done quite accurately to avoid contamination. So, that’s part of the reason why we implement AI, to improve our detection precision. So we do impart cell segmentation techniques using AI. The second aspect of AI in our research is about bioinformatics. Many people also use AI or machine learning to dig out useful information from sequencing data and try to see if they can identify some hidden markers or target genes.
We do that too, but one of the differences for our research is that we train our algorithms differently. AI is very good at identifying genes that are shared or distinct across different conditions or across different samples, but in our technique, we have the annotation information. We know which ones are and aren’t aggressive. So by training the algorithm and providing these annotations, we can more straightforwardly identify the genes that are distinct, and uniquely expressed in aggressive cells and not expressed in non-aggressive cancer cells. This is the way we apply AI, because we have the annotation of each cell. When you don’t have this annotation, you can still use AI to try to extract information, but that will be quite challenging and messy because any variation between samples will complicate the training and lead to unreliable outcomes. We do see some promising results using AI, but the reason it can work for us is because we have this very clear annotation at the beginning. So that’s another way we apply AI in bioinformatic analysis.
The third one we have recently implemented is to, as I mentioned earlier, look at individual cells’ behaviours. Quite often we will rely on recognisable features like migration, morphology and location, and now we are also training AI to observe images and to identify the cells that, for example, haven’t displayed aggressive features yet and so have not been identified as such, but they are destined to become aggressive cells. We’re training this programme to be able to identify the cells at earlier time before they become aggressive cells, and the benefit of this is we can get more comprehensive information about how this cell will eventually drive to aggressive features. This kind of thing is the third application of using AI in our group.
FLG: So obviously, with this AI and your analysis, you’re looking at cells that develop their features over time, and you’re looking at different behavioural changes that occur over time. You mentioned that you look at everything in real-time. Do you think more studies will start to use this real-time imaging instead of snapshot imaging? What advantages does real-time have over snapshot, and what challenges are there in other studies that don’t do real-time data analysis?
Miao-Ping Chien: That’s a great question. One of the things I mentioned earlier was that in the past, people studied those aggressive or therapy resistant cells based on markers, and by doing that you’re gathering snapshot, static information. Although this information is already very powerful, what we do is to offer additional information because some cells just don’t have these universal markers, and some have no markers at all. So we started to look at the behaviour changes, and that requires real-time imaging. What people can get out of this is additional information, in addition to what we have studied based on the static information, is to have more comprehensive understanding about the driving mechanisms of those aggressive cells, and that’s the main advantage I think it provides.
The challenging part is the real-time imaging, so after acquiring this large image data, one of the challenging elements is to process the data in a real-time fashion. We really need this because it would allow us to immediately identify cells after data acquisition. That process is very important, otherwise the target cells that you’re interested in might already migrate away from the original location, making it difficult to separate and isolate them. We have developed this algorithm and hopefully will release it soon publicly, because we just submitted it. In the paper, it details how you can implement the algorithm in your experiments and setups. Regarding real-time image analysis, the second part will be to do with the fact that from the data, we need to be able to extract cellular features defined at the beginning of the analysis. That information will be further used for downstream cell isolation. I started to see more and more people using this kind of information for cell selection in their own studies. In terms of the real-time image analysis, these are the challenges I see and hopefully, in the near future, our going-to-be-published paper will help the community.
FLG: Thank you so much. Following on from this, you use real-time to analyse cell behaviours and there have been advances in temporal analysis for multi omics, like live cell recording of dynamic transcriptional events that are occurring over time. Not so much with proteomics, but a little bit of genomics and a little to do with average blood flow. So do you take measurements at intervals, or do you integrate these temporal analysis technologies into your research? Could you just discuss that a little?
Miao-Ping Chien: I actually have one collaborator in a different institute in the Netherlands. She’s also looking at dynamic transcriptional events, and we are actually communicating for a collaborative project in this aspect. As I mentioned earlier, anything you can observe under a microscope, we will have a capability to isolate those cells. In this particular example you mentioned, we are monitoring the cells that have different dynamics of transcription. We can analyse this and distinguish which population of cells has more stochastic transcription properties compared to other cells, for example. So we can also extract that sort of information from the image analysis and ideally, we’d then be able to isolate those cells accordingly.
FLG: Obviously, there have also been advances in spatial omics or increasing resolution, and there’s been a lot of work done in that area. Do you think that we’ve reached a plateau when it comes to spatial omics, or do you think we could still push that frontier further, and there are going to be further advances? What do you think about that?
Miao-Ping Chien: It definitely hasn’t plateaued yet. The field of spatial omics is really changing very rapidly, in that every six months to every year, you’ll see big jumps happen. Personally, I’m actually very impressed by the progress in this field. I think the ultimate goal people want to have for spatial omics is to have what you can do with current single-cell omics. What are the benefits of single-cell omics? We would like to have in-depth sequencing profiles of tens of thousands of cells, and to have tens of thousands of genes per cell, and to have single-cell resolution. This is what the current state-of-the art single-cell omics techniques can reach, but not for spatial omics yet.
Obviously, spatial omics techniques consist of spatial information, but the combination of these three properties that I just talked about, a large quantity of cells with in depth sequencing profiles and with a single-cell resolution, doesn’t exist in current spatial omics methods yet. That’s basically what this field is going towards. So if you look at an individual element, like having single-cell resolution? Yes, there are some spatial omics techniques providing single-cell resolution, but they can only profile, hundreds of genes, or maybe a maximum of 1000 genes, and that’s it. So it’s still different, but I have no doubt that in the near future people will be able to reach that goal, including my group. In our research, the technique we are developing is basically to combine these three aspects without losing their spatial information.
FLG: Thank you very much. Going back to your labs, when you’re working on your analysis and isolating these aggressive cancer cells, do you still miss some of these rare aggressive cancer cells? If so, how are you trying to isolate those hard to spot aggressive cancer cells?
Miao-Ping Chien: Great question. During this whole isolation process, we definitely lose some cells for sure. As long as we have a minimum of 50 to 100 good quality cells after the whole process including screening, isolation and profiling, it will be sufficient. Even though we lose some cells during this process, as long as we can capture at least 50 or 100 good quality cells in the end, we’ll be able to extract useful information. The cells with phenotypes we cannot yet observe under a microscope will be missed, but that is part of our plan for AI implementation that I talked about earlier. We want to develop methods so that they’re not only based on recognisable features seen under a microscope; we also want to have a programme that can detect the cells before they display these aggressive features. With a tool like this, we can also identify, isolate and sequence those rare and to-be-aggressive cells. So that’s something else we’re also developing and implementing
FLG: Thank you so much. So, we talked about where spatial omics is going and how we’re trying to get that to the same level as single-cell, what do you see is the next frontier for single-cell specifically, what’s the next step to keep up the momentum?
Miao-Ping Chien: Many people have developed great single-cell-omics techniques, so I would say it’s quite accessible now. However, compared to bulk cell sequencing, it’s still not that accessible. So I think the next frontier in terms of single-cell omics is to reach levels of accessibility and commonality similar to those of bulk cell sequencing. Not only for RNA sequencing, but also for single-cell genomic sequencing, epigenomic sequencing and hopefully single-cell proteomics as well. I think the next step, after single-omics alone techniques becoming more accessible, will be to have easy accessibility for multiple ‘omics’ at a single-cell level. Another important part will be analysis. There are a lot of well-developed analysis algorithms out there in the community, but it’s still not that standardized yet. So this is another thing I think people will work on in the coming years. The last part of development, I think, will be to integrate these single cell-omics methods with spatial omics. Spatial omics has its own development, and I don’t think we’re reaching a plateau yet, but single-cell development has plateaued a little. Because of this, I think the next exciting development will be to bridge these two together. So that would be the three aspects I see for the next frontiers of single-cell omics developments.
FLG: Thank you so much, that’s all been really interesting. You’re going to be speaking at the Tri-Omics Summit in London, and I wanted to ask why you’re looking forward to this Tri-Omics Summit?
Miao-Ping Chien: The programme is really exciting, and guest speakers are going to be touching on lots of interesting topics. To be more specific, on the focus day on Tuesday October 18th, there are many talks not only introducing advanced techniques and data analysis, but also introducing how these techniques can be translated to clinical applications, how they can be applied to better understand disease formation and to identify hidden therapeutic targets or biomarkers. Those kinds of talks are basically covered on that focus day. Plus, many people from pharma companies and biotech companies are also going to be there to give a presentation. So I thought that’s quite exciting for me, and the events on the 19th and 20th October look equally exciting. Not only will they cover the most really most advanced single-cell omics and spatial omics techniques, but they’ll also cover quite a lot of different methods of data analysis including AI-assisted analysis methods. There are also a lot of good networking opportunities during the event. So overall, I think your organisation has really arranged an exciting programme, and I’m really looking forward to it.
FLG: Thank you so much! So, what can attendees expect to learn from your talks at the summit?
Miao-Ping Chien: I think, first of all, they will definitely learn how we can apply our microscopy-based functional single-cell sequencing techniques to identify and isolate rare and aggressive cell populations based on their behaviours and characteristics. I will also introduce how we can use that technique to identify otherwise missed targets and molecular mechanisms. The last element will consist of how we can extend this technique to spatial-omics profiling, so I’ll also talk about how we can combine these methods for expansion to the spatial profiling field with the capability of profiling many cells with in-depth sequencing and single-cell resolution. Those are the techniques and applications that I’ll cover in my talk.
FLG: Thank you very much.
Watch more interviews in our series here.