Accelerating the path to understanding cancer research samples with 10x Cloud Analysis: An interview with Dr. Timothy Shuen
Clinical research samples, experimental design, and data analysis are top of mind for Timothy Shuen, PhD, Senior Research Fellow in the Division of Medical Oncology at the National Cancer Centre Singapore. It makes sense: he oversees an integrated, operational workflow between the clinic and the lab, supporting physicians with powerful experimental capabilities, including high-resolution single cell and spatial transcriptomics analysis, that can answer different translational questions. As Dr. Shuen said, “How can we understand cancer and design better treatments for cancer patients?”
There’s huge potential for data-driven insights into real clinical challenges through the center’s approach to translational research—and emerging efficiencies for data analysis are also accelerating the path to insights for the team. Dr. Shuen is an avid user of 10x Cloud Analysis, a free, web-based application for running Cell Ranger and Space Ranger pipelines on the 10x Genomics cloud infrastructure. The tool has complemented the center’s bioinformatics team, relieving a busy analysis queue and providing results in just a couple of hours for their single cell and spatial datasets.1
We had the privilege of talking with Dr. Shuen about how he has integrated Cloud Analysis into his research workflow, as well as the impact of the cancer center’s integrated approach to studying clinical research samples. Keep reading to learn how going deep into the basic biology of cancer tissues has helped them reach a deeper understanding of the tumor microenvironment and the cellular interactions driving response and resistance to immunotherapy.
Can you tell us more about your role and your current research projects?
Dr. Shuen: I interact with a lot of patient samples. I need to design the operational workflow: how can we go from patient consent, to operating theaters, to receiving the specimen? And how can we work with the surgeons, pathologists, and other physicians? Then, I make sense of the samples. In the lab, my main role is to design what type of experiments we need to do [on these research samples]—for example, single cell transcriptomics—based on the scientific and translational questions we are asking.
My main focus is virus-related cancers. For example, Epstein-Barr virus (EBV)-related cancers, which are associated with nasopharyngeal cancers, EBV-positive gastric cancer, EBV-positive lung cancer, etc. Another virus type is hepatitis—it’s linked to liver cancer, which can result from hepatitis B, hepatitis C, or even a non-viral cause, such as fatty liver disease. The third virus component we work on is human papillomavirus (HPV). HPV is associated with cervical cancer and the rising clinical burden of HPV-related head and neck cancers.
We are trying to understand the tumor microenvironment and what the roles of these viruses are in the cancers. Also, how do the immune cells, viral components, and tumor cells all interact? So we’re interested in the interactome of these cancers and its impact on treatment outcomes. For example, survival, relapse-free survival, or even resistance to immunotherapies.
We assess anti-PD1-based immunotherapies and even CAR T cells or adoptive T-cell transfer technology here. The question is, in different cancer contexts, what responses are we seeing and what type of tumor microenvironment is associated with treatment resistance? Then, based on the findings, together with the clinicians, we want to develop next-generation precision immunotherapies.
It could be anti-PD1-based combination therapies or checkpoint inhibitors combined with T-cell treatment. My role is more to understand mechanistically or translationally why certain treatments fail or work well, specifically in the context of these virus-related cancers.
How have you been using single cell and/or spatial technology to support your research? What are some of your major findings?
Early on, we used the first generation of 10x Genomics single cell RNA-seq. Since then I’ve been using the newest generations, including GEM-X. Single cell technology is really powerful and helpful to us because we work on virus-associated cancers. When we could only rely on bulk RNA sequencing before, or whole exome and genome sequencing, we always lost the granularity of the data. With single cell RNA-seq, we gain more knowledge of the cancers—what type of immune components are there, what type of tumor-associated endothelial cells or cancer-associated fibroblasts are there, and how do they interact?
We’ve used single cell RNA-seq to profile the cancers from different time points. Given a certain specimen, what type of tumor microenvironment are we seeing? We’ve also profiled PBMCs before, during, and after certain immunotherapy treatments.
I kind of fell in love with this new technology because it’s really helpful and very robust. When I talk about robustness and reproducibility, I mean that I don't have any issues. There is minimal batch-to-batch variation. So on a daily basis, day in and day out, I rely on 10x Genomics products.
In our lab, sometimes the sample comes in unexpectedly. My boss may suddenly call me—we have one very precious sample coming in tonight or that we need to be on standby to receive—then we will get the workflow ready, and just follow the protocol. We have the Chromium iX in the lab, so we can easily generate GEMs anytime when needed.
We have 19 people in the lab and sometimes we take turns to process the samples and conduct library prep, but the robustness of the product is such that we can swap people and the data quality is always there. We’ve optimized the protocol even from how we harvest and homogenize the tissue, to how we handle the single cell data, so now it's really standardized.
So far we have been using single cell RNA-seq, and we went from version one to now GEM-X. We also tried the Visium HD assay recently, and we have more samples being analyzed now. I’m quite excited about the technology and how it can tell us a lot about the trajectory of the treatment response before, during, and after treatment based on PBMCs or tumor samples.
I can give you one example of what I’m most excited about: we used single cell RNA-sequencing to profile patient PBMCs and the T-cell product that we generated in the lab. Our lab is one of the main labs in the cancer center to do adoptive T-cell transfer. We draw blood from the patient, then isolate the T cells to strengthen them to recognize EBV, which is the antigen. After expansion of the T cells, we infuse them back into the patient downstairs in the context of a clinical trial.
We have done some compassionate use cases as well, meaning that the patient failed all types of standard treatment, chemotherapy, or radiation therapy—they failed all types of treatments and lost hope, and so they approached the cancer center here.
From the trial and also this compassionate-use case, we found that T-cell treatment works very well for the EBV-positive nose and lung cancer patients. When we re-analyzed the samples—the T cells and the PBMCs before, during, and after treatment—we used the 5’ gene expression kit with barcoded antibodies, barcoded dextramers, and V(D)J analysis to understand the antigen-specific T-cell clonality. Before we did the T-cell infusion, we could already see the T-cell clonotypes in the patients, and those are EBV-specific. But at baseline before the T-cell treatment, the patient had very little EBV-specific T cells. Upon expansion, we could generate an army force of EBV-targeting T cells in large number.
The exciting thing is that one particular patient survived 9 years after the T-cell treatment, and the EBV-specific TCR clonotypes could still be found in the PBMCs using V(D)J T-cell receptor clonality analysis. And we found that the CDR3 motif was exactly the same, before, during, and after the treatment.
Our historical record is that most of the patients cannot survive longer than 4 years at this advanced stage, unfortunately. But in this situation, one particular patient survived 9 years. We managed to obtain the 9-years post-infusion PBMCs, and, at that particular time point, we could still observe that the clonotype was there. So that gave us evidence that the T-cell treatment works very well. Without this single cell technology, we really can't tease out what T cells are there.
But unfortunately, the main reason why we knew that this patient was 9-years cancer free is that his cancer came back. It was cancer in the neck lymph node region. He failed all types of treatments afterwards, and unfortunately, in the end, he passed away. Based on analysis from single cell and spatial, we realized that the recurrence tissue in his neck actually became quite immunosuppressive.
We profiled the tumor and found the tumor had certain evasion mechanisms. The tumor cells started to lose HLA class I molecules. Without HLA class I, then there is no antigen to cross-present for the T-cell receptor (TCR) to recognize. We believe that the tumor evolved to have low or very minimal HLA class I expression.
Secondly, based on sample profiling, we found that the tumor had other immune checkpoints beyond PD1. Those checkpoints were unrelated to TCR signaling pathways. In the future, if we observe similar patterns from other patients, that may suggest that we should profile their tumors and possibly design a next-generation treatment. It might not be T-cell based and anti-PD1-based treatment; but instead it might be natural-killer or gamma-delta T-cell treatment, or other treatment options.
So these findings really opened a new area for us to brainstorm what could be [developed] next, besides just using PD1 or our standard T-cell treatment.
You said you were seeing an immunosuppressive tumor profile. Going back to your use cases for Visium HD spatial technology, do you see that technology coming into play as you evaluate the immune contexture of these tumors? Generally, how do you see spatial technology adding to and supporting this research?
My projects are always related to solid tumors, not blood cancers or liquid tumors. For solid tumors, we know that, even if we do T-cell treatment, the key question is whether the T cells go to the site. We can have a huge army force, but if the army cannot penetrate the sites, there is no battle. We waste a lot of energy to generate the army.
Here, with the spatial technology, and with the V(D)J clonality analysis tools as well, we can see what kind of T cells are at the site of the battlefield. We can know if maybe the T cells have been exhausted at the site already. Are they interacting with the tumor cells, endothelial cells, or fibroblast cells, which may become a barrier that prevents the T cells from penetrating? [This data could inform how] we can also engineer the T cells to equip them to have better penetration power.
With this much information, we can think about more and leverage the science to give us deeper knowledge of the cancer.
We always say that the first phase is to understand the cancer. Then, the second phase is to understand the response, resistance, or recurrence mechanisms. Then, the third stage will be to develop new treatments for all of these cancers.
I wanted to explore your use of 10x Cloud Analysis for the next part of our interview. How are you using it? What specific challenges did you face before using it and how has Cloud Analysis helped you overcome some of those challenges?
It’s a great relief to me that I have the 10x Cloud Analysis platform because, as I mentioned, we have a lot of single cell data and even spatial data now. Just a few years back, we were always stuck in the situation where we had so much data.
Day in, day out, we generate a lot of libraries—gene expression or V(D)J or Visium HD data. 10x has done great, providing Cell Ranger and Space Ranger, it's quite straightforward. To us, as the end user, it’s like a black box. We just put in the data, then process, then the data will come out—which is great, because I'm not trained as a bioinformatician, I don't know the command line. I know the principle, but I don't technically know how to instruct the computer to process from a raw FASTQ file to a Loupe Browser file.
I rely on my bioinformatician colleague, but I only have one. And we only have one high-performing PC, which is great—we can still process a lot of data. But, besides single cell data, we also have a lot of bulk RNA-seq and whole exome sequencing data, and we also need to predict neoantigens and do cancer vaccine design. Our lab is computationally heavy in this way.
So since it first came up, this 10x Cloud Analysis—I think we have uploaded more than 300 libraries. Now, it's really a no-brainer situation: I just upload all the FASTQ files, based on the project, to the 10x Cloud server. It's really easy. I either drag the file into the browser or I use a very straightforward command line in my terminal that I copy and paste from the 10x website. Then it's just linked to my laptop, and I upload the data or download the data from the 10x Cloud.
The beauty of the 10x Cloud is that it's super fast and a lot of pipelines can run in parallel. For example, if I have multiple projects, after I upload all these files, I can run them in parallel.
For a single cell transcriptomics library, it usually takes the server 1.5 hours to process the data. For Visium HD data, maybe slightly longer, maybe 3 hours per library. But I can just walk away and do other things, then come back. Then, I check my account and usually the Cell Ranger or Space Ranger output file is ready. I download them and can immediately see the data.
So the 10x Cloud really shortens the processing time for us. As a bench scientist, we love to see the data. After we generate a single cell library from the sample, we’re always curious—what happened to the sample. Did we do anything wrong? Will the data be very exciting?
Also, not only can you access the multi-pipeline, in the 10x process there is the aggregate option as well. I can aggregate all the samples together—so like the clinical research samples, before, during, and after treatment, or all the gene expression or all the TCR data together. That helps me make sense of the data more easily.
I'm curious if you could quantify what that time difference is for you now using 10x Cloud Analysis. You have one bioinformatician with limited bandwidth and a lot of other computational needs. What was that timeline for you before Cloud Analysis?
It depends on the queue, but usually from my experience, one week will be minimal. To be fair to my colleagues, they are very hardworking. But we only have one high-performance PC, and we have other concurrent projects.
Having said that, for that week, I wouldn’t know if there were any errors in the data. I wouldn’t know if we need to do any troubleshooting yet. Cloud Analysis is really helpful in that way too, that within 1.5–3 hours we will have the results. If anything happened, if I accidentally used the wrong reference file or, maybe, the data has a problem where we need to change the parameters, within an hour, I can go back and run the analysis again with new parameters. So the entire turnaround time is overall faster. To my surprise, the record from my 10x Cloud account to handle one V(D)J library is only 7 minutes.
So, in brief, the fastest would probably have been 2 to 3 days depending on the number of libraries, potentially up to a week. But now our run time is maybe just a few hours per library, and I can easily access the Cloud server any time, anywhere.
Sometimes I'm a bit impatient when it comes to data analysis. So sometimes, when I travel, I will use my phone to check the progress of the Cell Ranger processing. Then, when I reach the office, for example, I know it's around the time that the data will be ready.
Are there any other benefits that you’ve found in using 10x Cloud Analysis?
It has a user-friendly interface. The analogy I always say is if you know how to use an e-commerce platform like Amazon, then you can use the 10x Cloud because it's really straightforward.
I can also access the latest versions of Cell Ranger and Space Ranger. Previously, we would always have to keep updating our program locally, but the issue with that is it may introduce some unintentional batch variation, if some of our old data was processed with the old version of Cell Ranger. Now with 10x Cloud, I can standardize all the pipelines. And I can downgrade to not necessarily always be in the latest version. I can always go back if I need to.
So it’s really helpful. And the main limitation is just my uploading speed at my house here now!
Would you have any advice for other researchers who may be interested in using 10x Cloud Analysis and are just starting out?
My advice to new users would be:
- Small datasets should be used first, so you can get used to the 10x Cloud interface and the features or parameters you may need for your projects. This will help speed up the process when you have more libraries for analysis under the same project.
- Process a few libraries using 10x Cloud Analysis in parallel and compare it to what you have been doing before. I asked my colleague to run cellranger v3.1 for me using our lab's high-performance PC in parallel and recorded and compared the time taken and the output data. This will give you the security and assurance that there is no influence on data quality. This "head-to-head comparison" in our hands showed that the data output by both methods was exactly the same, but 10x Cloud Analysis processes faster when more libraries are being analyzed concurrently.
- FASTQ files uploaded to one project cannot be transferred to or shared with another project under the same account. It would be advisable to consolidate the needed FASTQ files and put them under the same folder in your own laptop before uploading using the 10x Cloud command line interface.
- To overcome any experience of a slow upload speed, it works well for me to use the command line interface, as 10x technical staff have instructed me to do. Basically, I just copy the command line shown in different projects and follow simple instructions shown on the 10x website to paste the command line to the terminal in my Mac. Then the upload speed is often improved significantly (it sometimes reaches ~300MB/sec upload speed).
Is there anything else that you'd like to say that you haven’t been able to yet?
When it comes to single cell or spatial, 10x Genomics products are always my go-to. It's really hassle free. I don't need to bother about the technical issues or the product quality.
I know that you guys have been improving the Cloud server; at first it was only for single cell, but now it also includes spatial, and I’ve seen that more features will be added. I always try to catch up with the latest [software] updates, but this is a fast moving train. With the 10x Cloud platform and the local sales and technical support team here, it's really easy.
Now, I don't need to bother as much about what type of experiment I need to do. I can focus more on the scientific angle, what questions we need to ask. So I think that helps us to pay much more attention to data analysis instead of data generation.
This interview was edited for length and clarity. We’d like to thank Dr. Shuen for his insights into 10x Cloud Analysis and his passion and drive to improve cancer treatment through translational research!
Keep exploring some of the resources he discussed in this blog:
1Disclaimer: 10x products are intended for research use only, not intended for any clinical diagnostic procedures.