DBiT-seq for High-Spatial-Resolution Multi-Omics Profiling

October 02, 2019

Information

Yale Cancer Center Grand Rounds | September 17, 2019

Rong Fan, PhD Associate Professor of Biomedical Engineering, Yale School of Medicine

ID4518

To CiteDCA Citation Guide

00:00<v Rong>Everyone, thank you,</v>
00:01from the Cancer Center leadership for giving me
00:03this opportunity to share my latest work.
00:07I have been working my entire research career,
00:12for almost 15 years, on cancer.
00:15But the presentation I'm giving today,
00:19it's not much about cancer
00:21and not much about the single cell analysis
00:23I have been working on for almost 10 years.
00:26This is something we haven't published,
00:27it just came out in my lab.
00:29I'm happy to hear feedback from you guys.
00:32So, I think that largely the anomaly
00:37in the omics area recently is,
00:40people can do single cell omics
00:43and multi-omics to understand tumor heterogenetics,
00:49but you really don't have the spatial information anymore.
00:53So the spatial omics kind of came out, or emerged,
00:57to address this challenge.
00:59Over the past couple years, I think largely,
01:01you'll see many different technologies,
01:04but largely, they are all based on just FISH.
01:07The more specific and more precise FISH,
01:10being a single molecule level FISH.
01:13So the shortcomings here, using FISH is,
01:18it's difficult, even my lab work and technology,
01:22I just cannot do it.
01:23This requires very advanced imaging technology,
01:27single-molecule fluorescence.
01:29You need to image over some time
01:32for a very sort of high volume
01:34and genome-scale data you want to collect from one sample,
01:38you probably need to image over days, repeatedly,
01:42to get this sort of large number of genes
01:46analyzed on the same sample.
01:49And also, that's not a sort of unbiased genome-scale,
01:53you really need to know the sequence you want to analyze.
01:58And also, so far, I think no one else talks about
02:02spatial omics and another terminology
02:05people use in this field is this spatial transcriptomics.
02:10It's not so obvious,
02:12how you can extend to other omics measurements using FISH.
02:18So I think the latest breakthrough
02:21came out actually this year,
02:24the two papers published, I think one
02:26just came out last week in Nature Methods,
02:29Another paper a couple of months ago in Science,
02:34to really use the power of Next Generation Sequencing
02:39for spatial omics mapping,
02:41or spatial transcriptome mapping.
02:43So an approach they took actually is quite similar.
02:48So they create sort of a barcoded surface
02:51using the packed beads.
02:54So whoever working in this space
02:56probably know no matter text genomics on
03:01the DropSeq technology, you need a DNA barcoder beads.
03:06So each bead has this thing, the DNA barcode,
03:09to really tell you which messenger is from which cell,
03:13or whether or not they are from the same cell.
03:16They're basically packing the beads
03:17on a monolayer on a glass slide.
03:20And they need to decode the beads,
03:21they need to know which bead has what sequence.
03:26So this decoding process was done
03:28by either SOLiD sequencing, or again,
03:30very much like FISH, you do repeated cell hybridization
03:34and imaging to decode the beads.
03:36That is a very tedious process as well.
03:39But afterwards, you get
03:42sort of a freshly micro-sectioned tissue sample
03:46and you place it on top and you lyse the tissue section
03:49and hopefully, the messenger is released from the cells
03:54in the proximity of the specific bead.
03:58It should be captured only by that bead,
04:00but I don't think the lateral sort of diffusion
04:03can be really avoided.
04:06But at least they saw a pretty good preferential capture
04:09of the messengers from the adjacent cells.
04:14I think this technology published or released in Science,
04:17demonstrate you can do 10 micron resolution
04:21spatial mapping of mRNA transcriptome by sequencing.
04:26And this paper came out last week
04:29demonstrating you can actually use even smaller beads,
04:31like two micron beads, to further sort of reduce
04:36the pixel size and increase the resolution.
04:38But two microns really (mumbles),
04:41the data analysis becomes even more complicated.
04:44And it turns out there have to be multiple beads
04:50to get a quality image.
04:54So interesting, when we visited their data,
04:57we found although they can see sort of anatomic or
05:01histological structure of different cells in a tissue,
05:05but it is almost impossible to visualize individual genes
05:08because the number of genes they can detect per pixel
05:10is extremely sparse, about like 100, 200 genes per spot.
05:16If you tried to image on individual genes
05:19across on pixel's entire tissue,
05:22the data totally is sort of not that meaningful at all.
05:26So what we can do is fundamentally different,
05:30I'm not about to say too much in the technical details,
05:34but this is totally different.
05:36We don't use beads and we just need
05:38a bunch of reagents with this device.
05:41And although we have been working
05:44on microfluids for years,
05:45but I don't like complicate microfluids like you guys.
05:50So this device, basically, you just place PDMS
05:53on top of your tissue and your clamp it, that's it.
05:56That's everything you need to do
05:58to deal with the microfluids.
06:00Afterwards, you just pipette your reagent to the host.
06:04So in the data, the validation data we have shown
06:07is we use sort of pan-messenger RNA FISH
06:11to visualize the individual tissue pixels
06:14we eventually are able to sequence
06:18with the spatial resolution.
06:21So we found we can get a very nice 10 micron pixel,
06:24as shown here if you zoom in.
06:26And then also interestingly,
06:29we saw sort of in the tissues
06:31after we process with our barcoding strategy,
06:34our barcoding approach, show some topological features.
06:38Even under optical microscope
06:41you can see where your individual pixels
06:44are located on the tissue.
06:45And worth noting, so this is sort of exactly the same tissue
06:50we're gonna take for sequencing,
06:53rather than the previous methods
06:56that always have to compare to an adjacent tissue.
07:00They are not able to get any good image
07:02from the same tissue at all.
07:04Also, the tissue sample we analyzed,
07:08they are just a formaldehyde-fixed tissue sample
07:12on a glass slide.
07:13So if you have a freezer of those samples banked
07:16in your freezer, we can look at those samples as well.
07:21We don't have to use sort of frozen tissue block
07:24and a fresh section to put on our slide.
07:31So we did some quantitative analysis
07:33of how many cells we can get per pixel,
07:37using this DAPI staining.
07:39And also, we were also concerned whether or not
07:44each pixel is distinct molecular barcode,
07:47we can put on or some sort of diffusion between the pixel
07:52that might cause cross contamination.
07:54We quantified a diffusion distance,
07:56we found it using the fluorophores basically.
07:59So we found the diffusion distance
08:00is actually just one micro meter,
08:04which suggests we can potentially
08:07further reduce the pixel size and increase the resolution
08:12to about like two micron using our technology.
08:17So the feature size matched
08:21the sort of the microfluid design very well.
08:24And the number of cells we can get
08:27in the 10 micron pixel size device is about 1.7 cells,
08:32we're really getting close
08:34to single cell level spatial omics.
08:38As I kinda alluded a little bit earlier,
08:42so the qualitative data, very important.
08:44So we compared our data to the Slide-seq data
08:48published earlier this year.
08:50So for the number of genes they can detect per pixel,
08:52about the size, 10 micron
08:54and then the number of genes we detected
08:57by using our technology.
08:58So really all that (mumbles) increase,
09:01in terms of how many genes,
09:03how many transcripts we can detect.
09:05About two years, three years ago,
09:07similar technology, sort of barcoded surface,
09:11basically capture of messenger RNAs
09:13for spatial transcriptome mapping
09:15was published in Science 2016.
09:19But that was very low spatial resolution,
09:21about 150 micron, but in that data,
09:24when you look at how many genes they can detect,
09:26that's about the same as what we can do.
09:29But the resolution is much, much lower.
09:32Or if you calculated sort of an area per pixel,
09:35it's 100 times larger than what we have.
09:39So I was very excited about this sort of data quality,
09:43which really enabled on the following slides,
09:46we can really visualize individual genes
09:48rather than using extremely sophisticated informatics
09:52to identify genes just to visualize
09:57the different cells types.
09:59We can actually interrogate every single genes
10:02across the entire tissue map.
10:06So when we first start with this,
10:11I'm extremely excited about
10:13tumor micro environment feature.
10:14But we decide to pick something
10:16that's well characterized,
10:18people know what cell types are there.
10:20So we used mouse embryo
10:22in the earlier stage of organogenesis, it's about 10 days.
10:27We were able to map out, actually, I wanna talk about
10:30a messenger RNA, actually, we can do also
10:33about 22 types of protein simultaneously mapped out
10:38using the same barcoding strategy,
10:41microfluid barcoding strategy.
10:42Showing here, is sort of pan-messenger RNA,
10:45but done by sequencing.
10:46So you can see actually the intensity
10:48of the total signal of the messenger
10:52does reflect (mumbles) in the tissue on the embryo slides.
10:58And here, this average signal of over 22 proteins
11:02we're able to look at as a panel.
11:06That doesn't really correlate that very well,
11:09but I think that makes sense,
11:10because you're not looking at it globally on all proteins,
11:14but the sub panel, it really depends
11:16on what proteins you put in your panel.
11:19Then we did a cluster analysis.
11:21When we look at single cells, we used tSNE,
11:25but here, it does make sense you have to use tSNE
11:27because you know exactly where the spatial location
11:31of every single pixel is.
11:33But the computational algorithm for clustering
11:37is identical, so, but after clustering,
11:39we just put it back on the tissue histological.
11:42The spatial map, we see sort of
11:48about eight clusters over here.
11:50And they pretty much match the anatomic annotation
11:55we got from the eMouseAtlas.
11:58And more interestingly, I think in the eMouseAtlas
12:00you're now able to kind of resolve
12:03a wide stripe the tissue here,
12:06but we saw a very distinct stripe of sort of cell type.
12:10We're still unclear what those cells are,
12:15but probably associated with the mouse
12:18sort of major aorta around the area.
12:24As I mentioned, we are able to visualize individual genes
12:27or individual proteins at a very high quality
12:31across the entire tissue section.
12:35Showing here a couple of genes and couple of proteins.
12:39And overall, I think the protein signal way higher,
12:43it's not a big surprise, this is because you measure
12:46only like 22 rather than genome scale.
12:49But when you compare, you see consistence,
12:52you see concordance and also discordance
12:54between the gene and proteins
12:56people have seen over and over.
12:58And very interestingly, when we look at EpCAM,
13:03it's a very nice concordance
13:06between the protein and messenger RNA
13:08in the EpCAM expression right here.
13:11And this one, I think, this is a microvascular tissue,
13:16microvascular tissue already developed
13:18in mouse embryo at this stage all over the whole body,
13:22we can see they are expressed everywhere,
13:25but we don't see a distinct structure at this resolution,
13:28because this resolution is about 50 micron, not 10 micron.
13:32I will get down to the high resolution data later.
13:35And then we did a sort of validation
13:37to compare our data to immunofluorescence staining
13:41for several selected genes.
13:43And this vasculature, again, you see extensive everywhere.
13:47You see EpCAM exactly the same pattern
13:50as we saw using sequencing.
13:52So just a couple of those locations
13:55showing the expression of the EpCAM.
14:00And another validation is
14:01we've done the sequencing data
14:03and the paper published earlier this year
14:05by Jason Du, from the University of Washington,
14:07they used single cell sequencing to map out
14:10several mouse embryos over different stages.
14:13And then you can basically do a tissue,
14:17a sort of sample tSNE, or sample UMap,
14:20this is not a single cell UMAP, but a sample UMap.
14:23So we found a four sample sequence
14:25actually mapped very well to this
14:28sort of differential or developmental trajectory.
14:32So in here, from their data, this is sort of the E9.5
14:38and that this is E10.5 and we are right in the middle.
14:42Those are kind of a little bit later stages
14:46of the developmental mouse embryos.
14:52And then we used a little bit higher resolution
14:55to look at the embryonic brain.
14:58This is about the entire brain
15:01and a little bit other tissues in the head and the neck.
15:04And also, this one, we didn't know what that is,
15:09but after data analysis, we found that actually
15:11it's a piece of the heart.
15:13And what we see from the protein
15:16and from the messenger RNA is,
15:18again, the messenger RNA atlas
15:20does reflect in the tissue histology very well.
15:23And the protein now, is much higher resolution
15:25of 25 micron, you do see some sort of correlation
15:28between tissue histology and protein expression atlas,
15:33but not as so distinct compared to the messenger RNA.
15:39So we were able to visualize
15:41individual proteins essentially,
15:42here are four of them, I think are very interesting.
15:46Again, EPCAM, this is a very high resolution,
15:49you can see very tight clusters of EpCAM expression
15:54in specific tissue regions right here and here
15:56and there's two or three or four.
15:58And the microvasculature, we can see the microvasculature
16:02by sequencing very well.
16:04And when you go to look on the tissue histology,
16:07or maybe I'm not pathology by training,
16:09I just cannot identify where the microvasculature
16:13are located based on the tissue histology.
16:16And the two other proteins, very interesting as well.
16:19This MAdCAM, we found it is a highly enriched
16:22in part of the forebrain, but not entire forebrain.
16:26And we see in CD63 it's widely implicated
16:29in the early stage mouse development.
16:32It's kinda anti-correlated with MAdCAM in other areas,
16:37so we kinda put them together,
16:38you can see their relative correlation each other.
16:44So, again, this technology where we want to validate
16:47to make sure what we saw using sequencing
16:51does match immunofluorescence staining.
16:54So this is from sequencing, this is from sequencing,
16:57this is about microvasculature, this is EpCAM,
17:01this immuno staining, you'll se almost a perfect match.
17:03I was very surprised, this is really a perfect match
17:09of distinct clusters right here, a little bit right here
17:11from immuno staining and we can pick up.
17:14It's only a few, so one single pixel layer thickness
17:19we can pick up very well.
17:21And so now here, you can see
17:22those microvascular network using immuno staining,
17:27which was also observed in our sequencing map atlas.
17:33So I got an interested in,
17:34this particular protein called MAdCAM and asked my poster
17:38to do some differential gene expression sort of.
17:41But the MAdCAM transcripts, it's difficult to see
17:45the sort of spatially distinct expression,
17:48but in the protein data, you can it see very well.
17:50Then we decided to use our sort of
17:53high quality spatial protein data
17:55to guide the differential gene expression
17:57across the entire transcriptome
17:59for different tissue reagents.
18:01So in this case, we're looking at MAdCAM-positive
18:03and a MAdCAM-negative and mapped out the top ranked genes
18:07for MAdCAM-positive region.
18:08This is still ongoing, since I'm still in the stages
18:12of learning developmental pathology,
18:14but what we can see some interesting features.
18:17But in the negative region, clearly,
18:19so this is the heart, turns out, this is kind of heart,
18:22kind of microtube associated proteins.
18:25And this is interesting thing,
18:27we don't really see this protein showed up extensively
18:30in the brain, but some how look like in this local area.
18:35And I have no idea what that is,
18:37but later we figure out that's actually the eye, here.
18:41And then we decided to do even higher resolution,
18:44which is a 10 micron resolution mapping
18:46of a particular region of the brain.
18:49And again, we had no idea where to map now,
18:53we just randomly placed our device on top
18:56and then mapped out this region.
18:58And the red color actually real data,
19:00this basically just pan-messenger RNA data.
19:03You can see the signal relatively uniformed
19:05and not perfect, but that's totally okay,
19:07just like when we do single cellular sequencing,
19:09we always do normalizations.
19:10Then that gives you, as long as your sequencing quality,
19:13sequencing data quality, number of genes you can read out
19:17(mumbles) genes, you can always do normalization
19:20and compare across different pixels.
19:23And as I told you, actually, we can see in the same tissue
19:27sort of after the barcoding and before the sequencing,
19:31we can even just under optical microscope,
19:34we can see individual pixels over here.
19:36And then when my poster showed me this image,
19:40it's okay, you got a key wide fiber over there
19:42very likely, because we saw this
19:45when we used microfluids before.
19:48And I thought that's unfortunate
19:50but anyhow, let's go ahead
19:51and process the sequencing data.
19:54But turns out that's not a key wide fiber
19:56that's really a very thin layer,
19:58actually it's a single cell layer of melanocytes
20:02lining a round the eye field.
20:05At this stage, the eye field actually,
20:07it's a very, very early stage only,
20:09called the eye vesicle an even no optical caps,
20:13it's the optical vesicle.
20:15So we can see, very distinctly, a group of genes
20:19strongly enriched inside the eye
20:22and also lining around the eye, optical vesicle.
20:27And then when we put them together,
20:31a little bit more structures you can see.
20:33For example in Pax6 enriched pretty much
20:36in an entire eye field
20:38but also in this region is optical nerve fiber.
20:43But here this protein, only expressed in the eye,
20:47but also other tissue type but not so much optical fiber.
20:51You can see this very well at a very high resolution,
20:54it's really about a single cell resolution.
20:57So, okay, when you look at it carefully,
20:59you see some yellow spots over here.
21:01That means the Pax6 and the Pmel are actually co-expressed
21:05in those kinda melanoblast cells but this one is not.
21:09The Six6 is not expressed,
21:11only within the eye, optical vesicle.
21:15If you further zoom in, you can see
21:18the sort of gene expression within the vesicle
21:21and also individual pixels, every little square here.
21:24So we can overlay the tissue image
21:26and the transcriptome data.
21:29So we noticed one gene which
21:33is strongly enriched right here,
21:36very strongly differential expression spatially.
21:39We're all curious what this gene does.
21:42We did sort of,
21:46this time they're still global, gene differential analysis.
21:49We saw only top ranked genes and these two showed up.
21:55But we found their functioning on a top ranked pathways,
22:01to some degree, okay, except those ones,
22:04to some degree, are mutually exclusive.
22:06And then later we realized
22:10but that has never been observed before,
22:12I don't have sort of last year's data to support.
22:15But it seems like those cells
22:21sort of characterized by this particular gene,
22:23later on are gonna determine the development of the lens.
22:27And those cells, even at this stage,
22:30you don't see any morphological difference,
22:32they already predetermined to develop
22:35the retina and the photo receptor cells.
22:39And then we were able to basically
22:41just put out those pictures obviously
22:43and compare it to those to perform
22:45a differential gene expression analysis.
22:47And another surprise, now this gene just showed up
22:51extremely differentially expressed.
22:54But we see many other genes that were very interesting.
22:58We still try to look into the details.
23:01So they are kinda enriched on the left side.
23:04Eventually, very likely,
23:07they will contribute to the photo receptor cell development.
23:13Okay, so even though we're able
23:15to visualize individual genes,
23:17we don't have to use the gene cell enrichment
23:19to identify different tissue types,
23:21but we had a challenge in particular
23:24in this kind of eye field region,
23:27due to our lack of knowledge in mouse embryonic development.
23:31But it'll be great if some computational pipeline
23:35can automatically identify different features,
23:37tissue features.
23:38That's what we demonstrate as well.
23:41So using this automatic automated
23:44feature identification pipeline,
23:47we were able to identify actually 20 different features
23:49in this very small region of the brain
23:52around the eye field.
23:54I just will show you some of those,
23:58you can see not just the eye, actually you can see
24:00very already development of the ear
24:03based on the sort of gene expression,
24:07but histologically, you cannot see any difference at all.
24:12But we also look at entire mouse embryo the E10.
24:18We're able to identify about 20 different features.
24:21But we're asking, so if at later stage
24:25many other organs begin to develop,
24:26whether or not this pipeline can identify many more
24:29tissue features or tissue subtypes.
24:33That turns out that that's right.
24:36And using E12, we're now able to cover entire embryo
24:39actually just the lower part of the body,
24:43we identify about 40 different features already.
24:47So this is a very high resolution as well.
24:51Okay, I'm gonna just summarize
24:54back to my sort of, ,the main interest in cancer.
24:59So I believe this enabling platform,
25:02we demonstrate can do protein and the transcripts.
25:05But actually, in my lab, another post I'm working on,
25:08so spatial, high spatial resolution epigenomics.
25:11I believe we can do high res,
25:13high spatial resolution ATAC,
25:15high spatial resolution CHIP-seq.
25:17And the application is extremely broad
25:19and the cancer is put right in the middle
25:21because that's really my main focus.
25:25I will like to thank people in my lab who work on this
25:28and thank you for your attention.