Scientist Stories: Ulf Landegren, Founder of Olink and Pioneer in Proteomics

Donald Rubin told us that more data isn't necessarily better. What we are going to hear now is about omics. There is more data coming inevitably, and to put a perspective on the other omics, we've heard a lot of genomics. We've asked three speakers. The first is Ulf Lundgren from Uppsala. He graduated in 1984, spent 5 years with Leroy Hood at Caltech, came back to Uppsala, has been there since, and is working very broadly on biomarker molecules, small molecules, and also towards the interface with industry. Ulf, I'm very glad you could participate.

So thank you, Rolf, for the kind words and for the invitation to come here, and it's been very interesting. I wasn't able to participate for the full day yesterday, but I joined towards the end, and today I also have learned a lot. But as Rolf says, my focus will be on how to use— what kinds of molecular data that can help us understand disease, and I think the purpose of this meeting is really a very important one. For me, the overarching hope is that we will be able to identify disease at an earlier stage and be able to do something about it.

So that's a very clear goal.

What can we measure that would best allow us to intercede in a disease process before it's gone too far? The pharma industry has some predilection for finding ways of prolonging life, making malignancy a chronic disease is one way they often put it. And that, of course, is an excellent business prospect. But for the patient and for society, of course, it's much better if we can completely avoid it. So anyway, I will be talking about techniques, mainly protein techniques for measuring samples. And everybody's aware that there is a fantastic development. I think some of you represent a very important quality of collecting samples and information about the patients, what happens to them. We also know a lot more about what goes on in our bodies and in our cells, what the signaling cascades are, which proteins might be overexpressed, which ones might be exclusively expressed in tissues that we're interested in. All these things are very important clues to how we can define next-generation biomarkers. But I will argue that the development is even faster in the area of research tools or molecular tools. And I'll show you a version of a slide that I think all of you have seen before. So this is the cost of sequencing over the 15 years since the beginning of this millennium, where we have seen a cost reduction. It's a somewhat unfair comparison, I should say, because of course sequencing becomes much easier once you've done it. For the genome.

But still, we've seen a cost reduction of 5 orders of magnitude from 2001 to 2015 in sequencing.

And if you look at the cost of genotyping, the technology behind the GWAS analysis, the development has been pretty much the same. So where you previously had to pay $1 for one genotype, one SNP, you now get 100,000 for the same amount of money. And that, of course, should influence what we do, both in the sense that we should, of course, take advantage of recent developments, but I also would argue that we need to pay very close attention to where the development is going and what will become possible in the future. I haven't plotted a third curve, but maybe I should do that for protein— costs of protein analysis, but there you would find a very shallow line. Cost has not gone down very much, but that— I believe that proteins will be extremely valuable markers. I think that's almost a trivial statement, that proteins are going to be important for biology and for medicine, but we haven't seen the fantastic development. But I'll illustrate some recent developments there. But more broadly, there's a lot of things we're meshing, and things are growing along many dimensions.

So I like to show There are more and more things that we are taking interest in. DNA and RNA, of course, microRNA, proteins, protein complexes, exosomes, little membrane-coated particles, cell-free DNA, cell-free RNA circulating tumor cells.

We come up with more and more entities that might represent valuable markers.

But for each of them, we're not happy to measure just a handful of RNA, but we want to measure all RNA.

We would want to measure all proteins and so forth.

So it grows in that dimension. And of course, we want to do that for more and more individuals.

I'll get back to that.. And we will measure samples taken from different locales, cerebral spinal fluid as well as plasma and the tear fluid and so forth. But even more importantly, we want to measure each individual many times. With genetics, that's not a factor. You take one measurement and then you know the genotype of the individual. With proteins, you may want to follow them very closely. This has not gone undetected by industry. So some of you will be familiar with—

Some of the companies here, for instance, 23andMe, of course, are well known. Decode.me, I think, are no longer as active.

Arivale is a company founded by Lee Hood, which is a so-called wellness company. So you pay a yearly fee and they allow you to send in samples and they analyze them and they tell you if you're doing well or if you should change your diet or start exercising more. Verily here. It's Google's life science company. And there's a number of them. Theranos, some of you know that they have had a somewhat rocky ride. But anyway, a lot of the development going forward will not happen, I think, in academia alone. Or at least, if you work in academia, you also have to somehow have a relation to industry. And for us, that's been a very important element. So by way of disclaimer, but also to show how we work, this is a slide that illustrates companies spun out from our lab.

Parallele was a genotyping company acquired by Affymetrix, which is now part of Thermo Fisher. Olink Bioscience, I will talk about that company. And from Olink Bioscience, we spun out the company we now call Olink Proteomics, which is offering the protein technology that I will spend some time on. But this is a prenatal testing company. This is a sepsis company. And this is a drug company.

So we've been very active trying to take our technology forward and make sure that somebody uses it.

And this is a list of companies that have been licensing our technology.

And I think going forward, that will be a very important element. It's very sad if you develop a technology and nobody ends up using it. And technologies are a little too hard for labs to set them up themselves. And in terms of measuring protein and protein in plasma, this is a slide that illustrates the challenges you're facing.

So here you have the ranges of concentrations of protein in cells, and this is the range of concentrations of protein in plasma. Albumin, you have several tens of milligrams per mL. Interleukin-6, you're down to picograms per mL, which is 10 orders of magnitude below.

What that should make you realize is that there is a fantastic problem of specificity. How can you find these needles in a haystack? Any assay, in order to be successful has to bring to bear a fantastic amount of specificity. And I think there is currently a bit of a Wild West among assays, so not all are of the same high quality. And I think this will play out and there will be clearer what assays work. In terms of the assay that we have developed and that now is commercialized by Olink Proteomics, the company discloses a lot of the information.

So this is taken from the website. So maybe you should see this as an advertisement, but I will try to be as objective as I can. So this is a slide of illustration of the concentration ranges going from 100 picogram per mL down up to 10 to the 3 picograms, 1 nanogram per mL for this particular marker, and then others are measured at higher ranges. So this is a dynamic range. For every marker, you can see the performance and so forth. And the assay consists in taking 1 microliter of plasma or cell lysate or cerebral spinal fluid or whatever and mixing that with 4 microliters of the antibody cocktail, letting it incubate overnight or 1 hour at room temperature, and then doing 1 more addition, and then adding them to a microfluidic plate like this. So here you can add 96 different patient samples having been assayed with these panels, and here you would add the 96 different PCR primer pairs that are specific for each of these proteins. And by microfluidics, each sample will encounter each of the 96 different PCR primers. And in the end, you will have determined 96 proteins, 92 for controls, for 96 samples. And so far, in Uppsala alone, we have measured 200,000 samples.

One reason we think this is a good technology is that it does depend on the dual recognition.

You have to have two antibodies agreeing that the protein is present. But in other dual recognition assays, it has proven difficult to multiplex because you end up with lots of possible cross-reactions that tend to increase the background and decrease your sensitivity. In our case, we can ignore anything but the right assays. So when we compare a single assay with a multiplex assay, the performance is very similar, and therefore we can multiplex.

And currently we're limited by the readout we've chosen. I borrowed the slide from a colleague to show data on the concentration ranges.

So by mass spectrometry, you can measure lots of protein, but they have to be typically in a relatively high concentration range. And with the proximity extension assay, you are on average maybe 4 orders of magnitude below.

And I think really that's where we will be looking for proteins, protein markers, things that are expressed generally at relatively low levels because the more abundant proteins have already been quite well examined. I put in the slide also by a colleague of mine.

I'm not part of this study, but Ulf Würlenstein has been using one of the panels here. It's a cancer panel. For a group of 1,000 individuals from northern Sweden for whom he had a lot of other information. He had total GWAS data. He had known, of course, the age, sex, smoking habits, news habit, and blood pressure, waistline, and so forth.

So it's possible now to say what are the covariates that may influence the protein levels. And on the left, you see that the top SNP, the highest-ranking SNP, accounted for a lot of the variation.

And the typical top SNP would be located where the gene is encoded in the genome. Over here you have growth hormone, which has a big contribution from the sex, of course. The sexes differ in levels of growth hormone.

And over here you see that age has a big impact on many of the protein markers. And Ulf has continued doing this work and applied it also backwards, sort of, to use information about protein patterns to try to determine what was the age of the individual who left the sample.

In terms of biobanks, so this is meant to be a cartoon of a biobank. You have a number of individuals. Some of them end up red, meaning that they develop the disease. Some of them remain blue, meaning that they remain healthy. And you have one sample from each of them. Some samples may have been taken after the view of the disease. Some have been taken before. And using this kind of information from lots of people, you can generate these kinds of map of GWAS data. So these are GWAS hits. I think many of you have seen this picture, the GWAS hits for some common diseases across the human chromosome.

With proteins, I think it will look different. We will look at biobanks that are much denser, much more individuals.

And we will have to have lots of samples. And particularly interesting are the samples I've marked with Xs, so samples taken before onset of disease when you might still have been able to do something about it. But of course, it's difficult to go back in time and collect those samples.

So the only way to make sure that you have samples when you would have wanted to make the diagnosis is to collect samples frequently. And that will incur a cost.

So we've been looking into a technique for reducing the cost of collecting biobanks, namely just taking a blood spot and then out of that cutting out a little disk of 1.2 millimeters and measuring the 96 proteins or 92 proteins and 4 controls. And here we compare the data to data generated when we take 1 microliter of fresh whole blood. And as you can see, there is a pretty good correlation. It's not as good correlation as if you measured fresh blood against fresh blood. So then you would have more or less a straight line. But still, all proteins are still— seem to be— or almost all that are detectable in fresh blood are detectable also in the dried bloods, more or less.

This means that we have the potential to build extremely extensive and inexpensive biobanks. You can even have people prick themselves in the finger and send it in. You don't need a nurse. You don't need cold storage. But in terms of cold storage, we were interested to see how well they would store. In this regard, I'm very grateful to Christian Wiem in the audience who helped us and put us in contact with Mats Maelbye in Denmark. We had access to biobanks both from Norway— from Denmark and from Sweden. But before I get into that, I'll just show you that when we looked at samples—

well, in these biobanks, we looked at samples taken 40 years ago, 30, 20, 10, and half a year or zero years ago, fresh samples.

And these were the newborn spots, the Guthrie spots or the PKU spots.

So we went into those biobanks and got samples out, 5 individuals from each. And we could show that for some of the proteins, levels were very stable through the years, whereas for others, there was a clear decline over the years.

And we could investigate that a little bit more clearly in comparing the Danish biobank with the Swedish.

The Danes started a little later than we did. They started in the '80s, so we started in the '70s. But here we compare '85, '95, and 2000, and so forth.. And as you can see, there is a trend here that the Swedish samples go bad, so the levels tend to decrease. The red— I mean, the redder, the higher levels. And you can see that there is a trend, downward trend, meaning that you can measure more and more protein at higher levels the younger the samples are, whereas the Danish samples are relatively preserved. And this is shown in a different way here. So here we see how many of the proteins That gave us still almost 100%, between 90 and 100% of the signal still. So we compare the numbers, and we find there that, for instance, half a year in a Swedish biobank is actually worse than 10 years in a Danish biobank.

And the reason for that is that the Danish biobanks have been stored at -20, and ours are stored at +4. So the lesson, if you want to use this for long-term storage, is that probably you still need to keep them very cold. But still, I think it's encouraging to note that we will be able to construct these very large biobanks. Going on to another area, two of my PhD students realized that this very high sensitivity and low requirement— whoop, what's that? Oh. Low requirement for sample meant that we could use very small samples.

And so Spiros Darmanis, who is now at Stanford, and Carolyn Gallant, who still with me, they demonstrated that you can take a single cell and divide it into aliquots and measure protein composition in one half and measure the RNA composition in the other half.

And using that, we can now determine the single cell levels. I want to illustrate this as one approach of biological analyses that I think are going to be increasingly important. We will not be able to just grind up tissue and measure, but we want to have cellular resolution.

And it is indeed possible to measure protein composition in individual cells.

And note in this case which cells have seen— these are glioma cells. And some of them were treated with an experimental therapeutic agent, BMP4. And some were not treated. And it was possible to distinguish them clearly.

In a collaboration using this method with the Broad Institute and Aviv Regev and colleagues there, We developed the technology further so we can take automatically the single cell, lyse it in a microfluidic device made by Fluidigm, and then in the same aliquot measure both RNA and protein.

And then we could ask the question, given that we have a certain number of genes we're interested in, do we derive more information by measuring their protein products or their RNA products?

And as you can see, the protein measurements distinguished better in this case where the cells had seen 4 bolsters compared to the RNA assay, which was a little bit more murky.

So I illustrate that as another new avenue to biomarker analysis at the cellular resolution. Another way to have cellular resolution, of course, is to do microscopy. So I'll also say something about this.

We have developed a version of the proximity ligation assay where we can use pairs of antibodies with DNA. That template the formation of a DNA circle which is locally amplified. And in this paper, Ulla Söderberg has, together with his students, has established actually 9 different assays involving the HER2 protein and its partner protein, in this case HER2, the epidermal growth factor receptor, HER2 interacting with itself or HER2 interacting with HER3, and observed in the breast cancer the distribution.

And you can see that the protein interactions are more prominent in the tumor tissue compared to the normal tissue. So you can now have an extended range of biomarkers, in this case protein interactions and also protein modifications.

We can also demonstrate that just as in solution phase assay, dual recognition improves assays. Here we used a polyclonal antibody to look for a protein which is highly expressed in pancreas. Digestive enzyme, lipolytic enzyme, and it's hardly present at all in the uterus and not at all in the spleen. But the antibody could detect signal by immunohistochemistry in all three tissues. So these are artificial signals. When we just divided that polyclonal antibody into two aliquots and required that two components would bind together, then we got a nice staining of the pancreas and no staining of the uterus and the spleen.

So you can improve specificity also. And other people— this technology is now in use in thousands of labs around the world.

And this is a slide from a colleague in Florida who's developed an assay for the epidermal growth factor receptor and the protein called GRB2, demonstrating that higher levels of this pair of protein in the lung cancer predicts better survival.

Whereas if you just measure the level of epidermal growth factor receptor, then there's no predictive value there. I should say we have improved on the method recently.

This is the old assay where we get signals. In this case, we look at beta-catenin and E-cadherin in epidermal tissue. There are signals and less signals here in the subcutis. But with the newer design, which we call Unfold, we have much brighter signals, so the technology is improving.

Another more recent development of this is to replace the antibodies that we normally use as affinity probe with low molecular weight compounds. So taking a drug, in this case a kinase inhibitor like dasatinib or gefitinib, and just putting a DNA strand on it and using that as our affinity probe to see, can we now see where does the drug bind in tissues?

And I just put in one slide from the work of my PhD student Abdullah Al-Amin, showing that indeed you can demonstrate in the breast cancer tissue now that the dasatinib probe is binding very well. The gefitinib in this case does not bind as much. There is some red signals for those of you who can see that.

But this is a way of evaluating the possibility of a drug to have its effect in treating the patient. Two more applications of this imaging version of the test where we require pairs of antibodies to bind together to give rise to local amplification signal.

The first example here is one where Lisa Löhr, who just recently presented her thesis working with her supervisor Masood Kamali, she developed an assay against the well-known fusion protein BCR-ABL, seen in chronic myelogenous leukemia. So one antibody binds the BCR component, one binds ABL. And she can demonstrate that by using this kind of assay in permeabilized cells and then using flow cytometry, she's able to detect the malignant cells. In this case, in the newly diagnosed case of CML in the chronic phase, you have 85.9% of the blood cells having this malignant-specific fusion protein, whereas a minority of the white blood cells in the blood are lacking this. In comparison with a healthy individual, you have no malignant cells having this fusion protein. And similarly, in another case, this is a control and this is a patient, in this case, having 16.4%.

So this is a convenient way of monitoring malignancy. And you can now use flow cytometry also for these more complex molecular events. Another study that Lisa did together with Masoud Kamali Moghadam is to set up assays where we could target exosomes. So little membrane-coated particles, typically you measure them in bulk, but then you can't tell the different types apart.

So here, by using this local amplification and targeting several proteins on the exosomes, it was possible to distinguish, in this case, exosomes from the prostate, so-called prostasomes, that we have previously shown may represent interesting biomarkers for prostate cancer. And in this case, distinguishing them from exosomes from a cell line, U937.

Okay, I have a few more minutes, I think. So I'll end by showing something a little bit more complex and novel. But I mentioned that one of the entities that are attracting great interest now as possible biomarkers is cell-free DNA. In pregnant women, some of the cell-free DNA will derive from the fetus, so you can diagnose trisomias by a simple blood draw. That was the purpose of the company VanaDIS that some of my students recently commercialized and sold to PerkinElmer. You can also look in tumor patients to see, can you find the tumor-specific mutations, the somatic mutations on the tumors as a way of measuring if the tumor is still in the body, just like we have been measuring the BCR-ABL transcripts for a long time. Now you can try at least to do it for mutations. And in order to do so, my PhD student Lei Shen has developed a technique.

Maybe I won't go into great detail, but he briefly amplifies in cell-free DNA in plasma sequences that he knows are mutated in the tumor. He then circularizes the molecules, Once they are circularized by ligation, you can replicate them, and the rolling circle replication will give you a product that now has 1,000 copies either of the wild-type sequence marked with a green C here, or in this case, a mutant sequence. So you'll have 1,000 copies of this T here. Next, you can take these rolling circle products and now interrogate them with another probe that forms a circle if it encounters the right sequence. This is what we call a padlock probe., and the green ones will recognize the wild-type sequence, the red ones the mutant sequence. So now you have 1,000 of these circles encircling the rolling circle product. If you now replicate these, then it turns out that you can make another 1,000 copies of each of these circles. So you've got 1,000 copies here for every molecule, and now you get 1,000 copies of all the copies, so you have 1 million copies. And this builds a bundle of DNA, intensely fluorescent because you can now have up to 1 million fluorophores per particle.

And it turns out that these molecules can be directly fed into flow cytometer, normally used to measure cells, but here instead we can use it to count events as single molecules.

By electron microscopy, this is what our one-generation rolling circle product looks like when it curls up. This is a second generation, it's what we call a super-RCA or super-rolling circle amplification product, and this has dimensions of around 10 microns or so, so dimensions of a cell. By now targeting mutation known to be present in a patient cell, we can get flow cytometric data like this. At the bottom you see normal individual, all cells are down here. There are no— sorry, not cells, but no molecules have the mutant-specific signature which would be located here. In a patient, by contrast, the bulk of the molecules are located here where the wild type should be. But the mutants are up in this corner, and you get an exact answer back. So the wild type counted molecules were 3,585,604, and the mutant events in this case was 6,128,000.

So you can measure this as a measure of the tumor burden in the patient. Is the tumor still around in the body? Are the numbers going down or up upon treatment?

This is a somewhat larger compilation of data where we looked for 8, or Lei Shen looked for 8 different mutations in a group of patients, always finding the right mutation in the patients and no mutation in the wild type. But the level of mutation converted to numbers per mL vary widely. Some were present in 1 copy per mL of blood and some were present in several hundred copies.

That brings to an end my presentation and I've tried to illustrate that there is a number of modalities that we will be able to measure. Which of them will give us the best insight, I think we have yet to determine. How well do they perform? I think that also has to be played out over the coming years, how well the assays work. But at least I'm hopeful that the kinds of methods that we are developing are able to deliver valuable response— results. So I just want to thank our collaborators.

Some work I showed had been done by Olink in Uppsala. Mats Melby and Christian Wehm have been very helpful to secure the PKU samples from Denmark. Alex Gjenshaft, Alex Schalek, and Avi Vregev were our collaborators for the single-cell work at the Broad Institute. Frederic Pontien and Matthias Ullian with the Human Proteome Atlas were our collaborators on the in situ work on antibodies. Rickard Runzenquist at my department, Uppsala, has been very helpful with the CML diagnostics via BCR-ABL fusion proteins. Susanne Garbrison is an expert on exosomes, a collaborator with Massoud and myself on this. And Lars Johansson and Thomas Lundbeck are drug developers at the Chemical Biology Central Sweden who have been extremely helpful with the work I illustrated of detecting the binding in situ of tyrosine kinase inhibitors.

Tobias Schwerblum finally has been very helpful with the samples for the rare mutant DNA in plasma.

And I'll end by just thanking our supporters, commercial supporters. These are the group leaders in the lab, Ulla Söderberg, who has been a pioneer of in situ work, and Massoud, who has been doing much of the work that I described or coordinating it.

And these are our students. And I'll just end by— showing you a picture of the group and take any questions.

Thank you. Well, thank you. I see that you have SciLife. I was just wondering, is SciLife a platform that is now financed by infrastructure funds in Sweden, or are there any other options for combining what you're doing here, which is, I think, an excellent kind of work, with research projects and inventory research projects to promote this on a large scale?

Yeah. So are you asking how SciLifeLab is operating? The bulk of the work and the bulk of the money which came directly from the government overriding the research council, the bulk of that money goes to the platform. Platform, technology platform, and most of that goes to sequencing, which is a very big effort. At least for a while, Sweden has been the number 2 or number 3 in sequencing firepower in Europe, which I think may be overdoing it. But anyway, there is also money for young starting scientists, so we are recruiting talented people, and we've been very lucky to recruit extremely talented people who are given quite attractive conditions to conduct their work. So there is some funds for research, but there is no real— you can't apply with a research project in mind and hope to get funding for that. If you take advantage of the facilities in Sweden, then as a Swede, you only pay for the running costs. You don't pay for personnel. You don't pay for instruments. So there is a benefit to everybody to use these facilities. But they still have to come up with that part of the money for the disposables and so forth. We are constantly advised to also, at SciLifeLab— I say we because I've been centrally involved in SciLifeLab. I'm no longer very centrally involved. I'm just one of the— involved in two other facilities. But we've been advised to identify flagship projects. The Human Protein Atlas project was seen as a flagship project. But currently, we are not really conducting a single projects. So that is up to the individuals. I should also say that for the— each of the facility— most of the facilities are also funded by the infrastructure funding by the Swedish Research Council. Like the sequencing has a lot of its funding also from the VR, and the Wallenberg Foundation in Sweden has been extremely helpful with some of the facilities also.

I hope that was an answer to your question. Yes, thank you.

Final question from— Jens Petterberg from Oslo. You mentioned that Theranos was— is struggling, and— but some of the concepts that you are presenting are really quite familiar or similar to some of the concepts that Theranos were arguing for and growing on. I mean, the use of small volume samples and also the ability to detect with very high sensitivity small fractions of molecules. I mean, to what extent do you think your— the struggles of Theranos that goes on specificity and sensitivity is an issue that you also will meet up— meet with in your, yeah, struggle to develop assays?

Yeah, so there's intense commercial interest in this space, and Theranos was an example that they were worth $9 billion. Dollars last summer before a Wall Street Journal paper or article about them. I have been monitoring them over the years, and since I've been very much involved in developing technology for DNA and RNA and protein measurement, I was very curious to see their reports. But there was never any documentation what they did, which I found very strange. So in a way, I was kind of relieved to see that they didn't have any particularly revolutionary technology. Of course, it would have been very good if they had that functioning technology, but it was a very unusual way of presenting their technique. They never described what they could do, and in fact, it seems that they were not able to do what they claimed that they could do. In our case, we have constantly published all our results at the website for the protein measurement. You can see very careful documentation, so I think it's very different. There may be overpromises of what you can gain from these kinds of analysis. How important will it be? Will protein markers be less important, or will there be other technologies that are even less expensive or even higher throughput? I'm sure there will be competing techniques. But hopefully we can play that out in the open and demonstrate what we're doing and what we're not doing. So Theranos, I think, was a rather unpleasant example, and I'm not entirely unhappy to see what happened to them. I agree. Okay, thank you. Okay, thank you very much.