Jeffrey Moffitt is an Assistant Professor at Boston Children’s Hospital and Harvard Medical School. His lab uses state-of-the-art spatially resolved single-cell approaches to characterize genome-wide properties of individual cells within tissues. Here, he discusses the importance of understanding the spatial and functional organization of cells within a tissue, using single-cell transcriptomics to build cell atlases, and tells us more about the MERFISH techniques of identifying drug candidates, ensuring diversity in the field and the benefits of a genetics first approach in drug discovery.
Please note the transcript has been edited for brevity and clarity.
FLG: Hello everyone! And welcome to the latest ‘A Spotlight On’ interview. Today, I’m joined by Jeffrey Moffit, who is going to tell us all about spatial biology. So, without further ado, Jeff, could you introduce yourself and tell everyone a little about what you do?
Jeff Moffitt: Certainly. First of all, thank you Lauren, for taking the time to chat with me today. And thank you everybody who’s joining (I guess virtually) for this conversation. My name is Jeff Moffitt, and I’m an assistant professor at Boston Children’s Hospital and Harvard Medical School. My lab started in 2018, so we are a young lab still! I got my PhD in physics from the University of California, Berkeley, where I did biophysics work with Carlos Bustamante, and then I was trained as a postdoc with Xiaowei Zhuang in the Chemistry Department at Harvard. It was in Xiaowei’s lab that I helped co-develop a technology that we call multiplexed, error-robust fluorescence in situ hybridization – or MERFISH for short. This is an image-based approach to spatial transcriptomics. In a nutshell, this is a technology that allows us to image and identify hundreds, to thousands, to 10,000 different RNA molecules with intact, fixed samples – that can be cell culture or slices of a wide variety of different tissues. So, this is a technology that my laboratory is leveraging heavily in a variety of things that we are now starting to tackle.
FLG: That’s great. Thank you so much for that introduction. Can you tell us, what is spatial biology? And what is its role within omics?
Jeff Moffitt: Yeah, that’s an excellent question. In many ways, spatial biology as a term captures many things. But perhaps one way to think about this is the merger of the spatial component of biology that has typically been reserved for technologies like microscopy – techniques that allow us to image where different cells are found within a tissue, and the organization of those tissues or molecules within those tissues, or samples. And [spatial biology] is the merger of that suite of technologies with genomics – things that allow us to probe the complexity of the gene products produced by samples, so technologies like RNA sequencing, or in the past decade or so now, single-cell RNA sequencing. I think everyone who’s tuned into this is probably quite familiar with the tremendous amount of biological insight that those types of omics-scale measures have been able to provide into biology. What’s exciting about this new suite of technologies that are captured in this term ‘spatial biology’, is that they can maintain the strengths, or the discovery potential, that we’ve seen from single-cell and genomic-scale measures, but couple that with the tremendous insight that has been provided historically via microscopy and similar technologies. These provide you with the intricate organization of molecules within cells and cells within tissues, but historically have been limited in the number of molecules that they can probe in a single sample. The merger of these technologies now allows us to make microscopy-style measurements, but with genomic-scale information.
FLG: So why is it so important, not just to understand cell type, but also the spatial and functional organization of cells within a tissue?
Jeff Moffitt: That’s a phenomenal question. And I think this is one of the things that drives excitement in this field. I mean, there’s many answers to that. But if you permit me, I’ll give you an analogy. When I was young, I used to do a lot of auto mechanics with my father. So, imagine that you want to understand how a car engine works – this is a complex machine comprising many different parts. Having a catalogue of all the parts that fit within that engine – pistons or plugs, etc. – and understanding their individual function can provide tremendous insight into how an engine functions, but it’s really only part of the picture. If you were given that catalogue of parts, but no understanding of how they assemble and fit together, I think you can get a sense of the challenge you would have in understanding how the engine actually works. In many ways, that analogy applies to biology, where technologies like single-cell RNA sequencing and single-nucleus sequencing have done a phenomenal job of giving us parts catalogues. We know what the cell types and states are within the samples. But until we understand how they are assembled physically and spatially together, we really can struggle to understand how the behavior and function of individual cells cooperatively gives rise to the function of the tissue as a whole. It’s that ability to both discover these parts and put them together that I think demonstrates the discovery potential and promise of these spatial biology tools.
FLG: That’s a great analogy. I’ve never heard that before! Could you tell us about some of your work to build cellular atlases of complex tissues? And a little about how you use single cell transcriptomics to do this?
Jeff Moffitt: Yeah, absolutely. I can give you a little bit of an overview of the type of things that we’re doing in my lab and the work that I did while a postdoc in Xiaowei Zhuang’s laboratory. For example, one very illustrative example of the potential of cellular atlasing with MERFISH and spatial transcriptomics in general, comes out of the work that I did as a postdoc, where, in collaboration with Catherine Dulac’s laboratory, we took a region of the mouse brain known as the preoptic area, or preoptic region, in the hypothalamus, and we used MERFISH – this image-based approach that allows us to image and identify single RNA molecules within samples – to profile about 150 different RNAs simultaneously within individual cells. By selecting those RNAs so that they contain classically defined markers of different cell types, as well as panels of genes that are functionally relevant to the way in which neurons in this region of brain communicate, we can basically define all of the cell types that one would anticipate to find in this region of the brain, as well as define a really truly remarkable diversity of neuronal types in this small region of the brain. And because we were imaging everything within intact slices, we basically (for free) got an atlas of this tissue – we knew exactly where each of these cell types were found, what they were next to, and, importantly for the hypothalamus, there had been a tremendous amount of work previously to define different anatomical regions within the hypothalamus and functions for the neurons within those. And so we could now define molecularly, and cellularly, the diversity of neurons that comprise these different regions.
Also, because RNA expression itself can report on cellular state, we could actually start to functionally annotate the different types of cells found in this region. For example, we could stimulate these mice with a variety of cues that trigger instinctive behaviors that are known to be controlled in part by circuits in that region of the brain. Then we could look for RNAs that are expressed in neurons that have recently fired – what are known as immediate early genes – and by looking at which neurons were expressing those genes (proxies for neuronal activity), we can determine which neurons were involved in the circuits that control the instinctive responses to behavior. This allowed us to define the parts (the cell types) and their organization within the tissue, and also to begin to define their functional role, and the different instinctive behaviors that are controlled by regions of this part of the brain.
Now, we are doing the same type of thing in a wide variety of tissues in my laboratory. We’re collaborating very broadly, in tissues ranging from the mouse brain, the other portions of the mouse, the nervous system… we have collaborations in human tissues, lymph nodes, and lymphatic tissues, we have a variety of collaborations and other human tissues as well. My lab itself is focused mainly on looking at aspects of the interface between microbial communities and the host. And one of our major focuses is on the gut microbiome, and how microbial communities and spatial structure within the host can shape interactions across this interface, both in homeostasis and health, but also when problems arise and the diseases that result from this. Most of that work is still unpublished, but we are very excited to use this type of technology and extend it in ways that allow us to do similar things to what I described in the mouse brain, in both the mouse and human gut.
FLG: It’s a really exciting area. You mentioned MERFISH, this technique that you co-developed, would you mind telling us a little more about how that actually works?
Jeff Moffitt: Yeah, absolutely. So MERFISH, at its core, is based on a very powerful technique that had been developed many, many years earlier. It’s a technology known as single-molecule fluorescence in situ hybridization, or single-molecule FISH. This is a technology in which you take and create fluorescently labeled DNA oligonucleotides that are complementary in sequence to regions of the RNA that you want to target. If you create a large number of these, each targeting a different region of that RNA, you can then take a sample, you can fix it, poke holes in the membrane (or permeabilize it) and then hybridize on these probes. And then base pairing of those probes to the complementary region on the RNA will actually bind the fluorophore to that RNA. By having many probes, you can concentrate them at each and every molecular copy of that RNA. So now, if you image that sample with an epifluorescence or confocal microscope, you’ll see a bright fluorescent spot. And that spot is the signal generated by a single RNA molecule. What this means is you can now actually do spatial transcriptomics – you can count how many spots you see within a cell that tells you how many copies of that RNA, and you can see where that cell is within the context of the whole, because you’re imaging the sample. This very powerful technique has been used to great effect to answer a wide variety of questions, but historically, it’s been limited in its multiplexing. This is because, if you want to target multiple RNAs, the standard approach has been to use multiple colors – each RNA gets a different color. But there’s only so many colors you can see in a microscope. The key insight that we developed in MERFISH was that we were going to replace the colors that we discriminate individual RNAs with, with barcodes. Now, every different RNA gets its own barcode. And then we have a process by which we can read out those barcodes optically. So individual molecules still generate a fluorescent spot that tells us where they are, but this optical barcode is what tells us the identity of that RNA.
You can think about these barcodes in many different ways, but the simplest way I think, is to consider them as a binary barcode. One RNA is given 110. Another one is given 101 and then another 011. And the power of binary barcodes, of course, is that every time you add one more bit, you could effectively double the number of RNAs that you could discriminate. This means the number of barcodes and the number RNAs that you can distinguish, it just explodes as the number of bits increases. The insight that we had with MERFISH is that instead of doing one round of staining, and imaging with single-molecule FISH, we could do many rounds. In its simplest form, you could think about one round a single-molecule FISH representing a single bit. If an RNA is fluorescent, in that round, it gets a one. If it’s not, it gets a zero. Your first image tells you the value of the binary entry in the first bit, then you can remove that signal, re-stain the sample, and now you can again use fluorescence on/off signals to tell you the value of the second bit. The fluorescence on/off pattern developed across a series of single-molecule FISH images is the physical embodiment of this binary barcode. It’s the optical barcode.
There are many details of how we actually implement this that are critical to how this scales. The first important aspect of it is that adding one more round of single-molecule FISH adds a bit, doubles the number of barcodes. But the problem with this type of combinatorial labeling approach is it is very sensitive to error. If you don’t have a molecule that’s bright enough to be called a molecule when it should, then you can think about that as misreading a one as a zero, and those type of errors, while infrequent, compound very quickly. We recognized right in the implementation of MERFISH that we had to have a way to handle those errors. By conceptualizing these barcodes as binary barcodes, we can basically borrow from decades of understanding of how you handle noise and binary communication. And so, what we use are slightly modified forms of error robust and corrected encoding schemes. These are schemes first developed in the context of information theory and used extensively in computer science that can allow you to identify when a bit has been corrupted, and in certain circumstances actually uncorrupt and correct that bit. And so we leverage those type of barcodes to actually encode our RNA identity. And that gives us very high performance with this technology.
FLG: That’s really interesting. It’s amazing how you adapted that to MERFISH. As more of a broader overview, what are some of the advantages of being able to directly image genome-wide properties of individual cells within tissues?
Jeff Moffitt: I think there are a variety of advantages, and perhaps that also depends on contrasting these with other methods for spatial biology. As I’m sure you are very aware, now is a very exciting time for the field because there’s a growing array of approaches to gain genomic-scale information while registering that information in space. There’s a suite of technologies that allow you to do that in a spatially resolved capture approach – so you can essentially capture molecules on a surface that have been spatially barcoded in a certain way, and those barcodes tell you where those molecules came from. Those are very powerful techniques, they’ve been providing great biological insight, but they’ve been historically limited in their resolution.
The advantage of what we’re doing, in which we generate fluorescent signals directly from individual RNAs, is that we are able to leverage the highest optical resolution possible. We can resolve the location of the RNAs to the 100-nanometer scale or better – the subcellular scale. This means that not only can you unambiguously assign RNAs within individual cells and get true single-cell expression profiles, you can actually understand the structure of those RNAs within cells. There’s a whole host of post-transcriptional regulatory mechanisms that are driven by the internal location of RNAs within a cell. And there are diseases that are cued by mislocalization, not just misexpression. So this ability to resolve RNAs at that length scale is very powerful because it opens up a wide variety of biological questions that aren’t possible with spatial biology techniques that have a more modest resolution.
The other advantage, of course, is that you can tie many images together so that you have this 100-nanometer (or better) scale resolution. But you can have that across cm2 areas, or tissue areas. This means you can span a really remarkable dynamic range in length from nanometer to centimeter. And that’s critical, because the biology of tissues often covers that length scale, where you have critical interactions that happen between individual cells and even individual molecular interactions between those cells. But of course, those occur in the context of cellular neighborhoods and in tissue architectures. Having an ability to span that length scale in a single measurement, or that range of length scales, is important for understanding the biology of many tissue-level questions.
Another advantage that’s perhaps less well appreciated is that because you’re generating signals from the molecules within the sample themselves, you can have a very high capture efficiency. What that means is, if you had, say, 100 copies of the molecule in that sample, the capture efficiency or the detection efficiency would be a measure of how many of those you actually detect. If you detected 90 out of those 100, you’d have a 90% detection or capture efficiency. For methods that remove the molecules from the sample in order to characterize them, and their spatial location post-hoc through sequencing, like these spatial capture methods, historically have had much more modest capture efficiencies. But with MERFISH because you’re imaging these molecules in generating the signals within the sample without ever removing the RNAs, without ever having to convert the RNAs to another type of molecule like DNA, these capture efficiencies can be very high – almost 100%. That’s important because there are whole swathes of RNAs that are very important for key aspects of biology. And yet, they’re expressed at a couple of copies per cell, when they’re actually expressed in those cells. Those can include transcription factors, receptors or other categories. And those have been historically challenging categories of genes to analyze with technologies that have more modest capture efficiencies. This ability to actually get very, very high capture efficiencies opens up aspects of biology with MERFISH that is challenging to do with other techniques.
FLG: What is the value of enabling simultaneous imaging of the transcriptome and the proteome?
Jeff Moffitt: That’s an excellent question. To provide some background, MERFISH is an RNA imaging technology, but we have been able to couple this with proteomic imaging. We do that by leveraging phenomenal advances in oligo-tagged antibodies. So you can tag an antibody with a DNA molecule, and when you stain your sample, the act of reading out the location of the antibody is now just hybridizing on a fluorescently labeled oligo complementary to the tag that’s on the antibody. We’ve turned the immunofluorescence problem into a hybridization problem. Now, this exact same hybridization chemistry that’s used to read out these optical barcodes associated with RNAs can be immediately leveraged to read out proteins. My lab, and the work I’ve done before, has been more focused on a modest scale of multiplexing and proteomics. But these oligo-labeled antibodies with other special biology techniques have been really pushed towards the scale of 50 Maybe 100 different proteins, and the combination with MERFISH, I think would be quite straightforward.
The advantage of layering proteomics is perhaps multi-fold. The first is that it’s just a practical aspect. When you have a technology like single-cell sequencing, where you dissociate out the cells, capture single cells, extract the RNAs, you get the encapsulation of RNAs within a cell, basically, for free – you get it from the physical boundaries of the cell itself. But with image-based methods, you have a slice of tissue, you see the location of every RNA, but you don’t necessarily know a priori how to partition those RNAs into individual cells. That’s what’s called the cell segmentation problem – looking at an image and defining where the boundaries of each cell are located. My lab has been working extensively on this problem, we have a collaboration with Peter Kharchenko’s lab that has been quite successful in developing algorithms to handle this. But one of the ways that proteomics is powerful is that you can include immunofluorescence stains against markers of cell boundaries, cell surface markers, and so you can use immunofluorescence to help define the boundaries of these cells to partition RNAs into. And that’s a very powerful way in which these technologies can be combined. That’s kind of like a practical aspect, that’s giving us the ability to overcome a technical challenge, not to layer on new biological insights. And there, I think there’s a lot of promise for proteomics. There have been beautiful studies where you can include antibodies to different histone modifications. And so now, you can report on aspects of the epigenetic state of that cell while also reading out the transcriptional state as well. One could imagine doing the same for the phosphorylation state of signaling proteins, or even the localization state of transcription factors. By layering in proteomics, I think what we will see is the ability to add new biological dimensions that aren’t as immediately read out by transcriptomic measures. And that I think is the true promise of this combination, though I think much of that is still to be realized.
FLG: That’s fascinating, and it’ll be great to see how it develops in the future. You mentioned that a lot of your research is around understanding host-microbiome interactions. From your perspective, why is this such an exciting area? And perhaps could you touch on some of the work you’re doing in this area?
Jeff Moffitt: What I’ll say is that this is a new direction for me and my laboratory. It’s a topic that I’ve found fascinating for quite some time, and I’m excited to be able to dive into this area. There are many reasons why I think it’s interesting, but most of those are personal for me. I mean, it’s a fascinating topic. In some sense, I think we’ve seen over the past decade of beautiful sequencing work by a wide variety of groups that microbial communities play a very important role in the homeostasis of the organism within which they reside. And they can play fundamental roles in almost all phenotypes, whether it’s nutrient absorption and digestion and availability, whether it’s education of the immune system, whether it’s neurological state and the development of neurological disorders. And I think that, you know, we’re really seeing this field move from correlative studies into really mechanistic studies – understanding the bacterial communities or individual bacteria and the small molecules they make, and the ways they interact with the host that drive this type of phenotypic outcome in the host. And for us, what we are hoping to be able to do is to provide a window into aspects of this interface that have been perhaps historically under-studied because of challenges in the technology. And that’s just the spatial organization of this. And so that can range from understanding what are the actual ecosystems at the micron scale and the gut? And how does that shape microbial communities? And in turn, how does that shape the dynamics within these communities? To questions of what happens when this interaction between host and microbe takes a turn for the worse and you have infection or pathogenesis? And these are places where you can have very local interactions that remodel both the microbiome and the host. For us, being able to come into this field with a tool that allows us to look in space at the organization of this interface, in both homeostasis and disease, and to reveal the gene expression changes that happen in that context, I think is very exciting because it offers a relatively new window into this great system that many labs have already made phenomenal progress in understanding.
FLG: It’s an incredibly interesting area. What are some of the goals of your research in this space?
Jeff Moffit: We’re really just getting started in this space. Spatial transcriptomics, in particular MERFISH, needs some extension in order to be able to ask questions across this interface. Much of what my lab is working on now is extending this technology in those ways. This is actually very exciting for us. Throughout my career, I’ve had an opportunity to work in technology development, where we are identifying biological questions that are just beyond our technical capabilities, and then working to build the technologies that open the door for that. This is a very rewarding way to do science, in my opinion, and that’s what we’re trying to do here. By working at this interface, we are pushing ourselves to extend MERFISH and spatial transcriptomics, more broadly, to allow us to ask these types of biological questions that have basically been very difficult to ask previously. That’s very rewarding to me. And I hope that is also very rewarding for our students, because it gives them an opportunity to learn all the skills associated with building and extending a cutting-edge technology, while also using it to do some very interesting biology.
FLG: In a broader sense, then, what are you excited about for the future of spatial biology? Or where do you see the field heading?
Jeff Moffit: I think that there are multiple aspects that are very exciting about this new type of measurement modality that we see with spatial biology, and particularly the image-based transfer domain techniques of MERFISH. To take a very broad perspective, what I think is the most exciting to me is to see this technology that I played a role in developing disseminate out there and be used broadly – that’s very rewarding for someone who’s worked to build new technologies, to see others make discoveries with that technology. I think we’re really starting to see this happen, not just for MERFISH, but for the field of spatial biology as a whole, we’re seeing these technologies move out of the labs that develop them into the hands of others. There will be many different aspects of biology in which this technology has an impact, but perhaps the most obvious one, in my opinion, is that it has the potential to really change our understanding or enrich our understanding of tissue-level function and dysfunction. What I mean by that is – again, returning to this analogy of the car engine – it’s not just that we want to understand how this engine works because it’s a beautiful piece of machinery (though, that is, of course, one aspect of it). If you have an understanding of how it works, when something goes wrong, and you’re faced with a, you know, an odd clicking or some ugly smoke that comes out the tailpipe, you are extraordinarily well posed to create rationally designed diagnostics, and interventions and treatments, because you understand all the parts of the system and how they fit and function together. And that, I think, has been one of the challenges of modern medicine – really being able to stratify disease, and to create rational interventions. It’s really amazing how much has been done. If you just pose the point that we don’t really still know the full set of cell types and states and their organization in basically any human tissue. And again, there’s a tremendous amount of work that’s been done, but these spatial transcriptomic techniques offer us the ability to come in and really provide a more comprehensive view, in many cases, probably just really solidifying the beautiful work that’s been done over decades in defining cell types and states in different tissues, but adding the potential to discover new or rare cell-type states or interactions, and to layer on that really a genomic view – a comprehensive view of the molecules expressed within these cells. I think what that type of cellular atlasing will do is it will give us a healthy reference that allows us to make much more rational diagnostic and prognostic interventions when we have the diversity of diseases that arise from different tissues. I think these technologies have this ability to help us really change the way we approach human disease. That’s a long way off. There’s a lot of work that has to be done to get there. But I find that to be a very exciting potential for these technologies.
FLG: You’ll actually be speaking at the Tri-Omics Summit on September 27 – 29th in Boston, MA. Could you tell us a little about why you’re excited about or why you’re looking forward to this event?
Jeff Moffit: I always love being able to get out and talk with a diverse set of scientists. I think this event is likely to bring in people from many different aspects of biology and technology development. There’s a really great lineup of speakers in the spatial biology space. I’m looking forward to not just being able to share our work and a deeper understanding of this technology with that community, but also to hear more of what they’re doing and how they’re continuing to push this field forward as well. It’s going to be a fun event, and an educational event.
FLG: That’s great. Thank you so much. That’s actually all we’ve got time for today. I’ve certainly learned a lot about spatial transcriptomics and MERFISH in particular. So I’d just like to say thank you again for taking the time to answer my questions. It’s been really great chatting with you. Thank you.
Watch more interviews in our series here.