NLR receptor networks: filling the gap between evolutionary and mechanistic studies

30 min readSep 13, 2021

Modified from the transcript of a keynote lecture given by Prof. Sophien Kamoun (The Sainsbury Laboratory, UK) at the international PhD course ‘Pathobiomes and plant immunity’ (June 7–16, 2021, online) hosted by the Department of Forest Mycology and Plant Pathology at the Swedish University of Agricultural Sciences. The talk is available on YouTube at

Cite as Kamoun, S. (2021). NLR receptor networks: filling the gap between evolutionary and mechanistic studies. Zenodo

In the absence of an evolutionary perspective, a mechanistic understanding of any aspect of biology is unlikely to explain the messiness of biological systems. For example, let’s say you study wing function and your model system is penguins. You would publish very good papers showing that wings function in swimming and many other interesting tidbits about the rather atypical penguins. However, you would be very confused by many features of wings, like their morphology, some of the mechanisms by which wings develop and how the muscles are structured. Therefore, by studying that one single organism and not bringing in an evolutionary perspective or a comparative approach, you will have a very narrow view of what wings do.

“Biologists must constantly keep in mind that what they see was not designed, but rather evolved. It might be thought that evolution would play a large part in guiding biological research, but this is far from the case.” — Francis Crick, What Mad Pursuit (1990).

This view has been around for a long time, but many of the people who do mechanistic research haven’t really integrated that thinking until recently. In the last five to ten years my lab has decided to focus on filling the gap between mechanistic research and evolutionary biology in the study of molecular plant–microbe interactions (MPMI), a budding field nicknamed ‘evoMPMI’. The post-genomics era presents a lot of opportunities to combine these two, formerly separate, approaches and generate new perspectives on how we do research in the biological sciences. Below, I describe my lab’s work in this field, starting with a brief introduction to the plant immune system, then describing our work on the proteins that recognize pathogens and initiate key immune responses.

The plant immune system

One dogma in plant pathology is that most plants are resistant to most plant pathogens. Disease is an exception and plants are so good at fighting off most pathogens because they have a very effective, complex immune system.

All kinds of pathogens and pests secrete effectors that modulate host processes

Let’s start from the pathogen side. We know that all kinds of pathogens — fungi or oomycetes, bacteria, nematodes, aphids and so on — secrete effectors that modulate host processes in different compartments of the interface between the pathogen and the plant.

Now, on the plant side, plants have immune receptors, some on the cell surface and some intracellular, that perceive the molecules coming from the pathogens — the effectors. This immune surveillance system is very effective at detecting perturbations caused by pathogens. More often than not, these are indirect perturbations, and the plant has evolved to say, “Yes, I have a pathogen now; I’m facing a pathogen that’s attacking me.” In that case, the effectors become liabilities, if you like, for the pathogen: they trip the wire and activate immunity.

Effectors can “trip the wire” and activate plant immune receptors

These immune receptors basically are sitting there in a plant cell in an inactive, resting state. When a pathogen effector is detected, then they switch on into their active form. This activates a bunch of different defence responses and results, more often than not, in a hypersensitive cell death response, which produces programmed cell death around the infection.

We can experimentally visualize this response using this wonderful model plant, Nicotiana benthamiana or ‘benthi’. We do assays where we infiltrate the leaf with agrobacteria to coexpress an effector and its cognate receptor. When there is a matching interaction, we get the cell death response, and we have a positive readout for immunity.

The immune receptors are the first layer of the surveillance system, and they’re extremely heavily studied. There are multiple classes, but the ones we’ll talk about today are the NLRs, or intracellular sensors of invading pathogens. NLRs are multi-domain proteins. They have a central domain, the NB-ARC domain, which acts as a switch: NLR proteins get switched on through conformational changes resulting from an ADP to ATP switch in this nucleotide-binding domain. Plant NLR proteins have additional domains, including a variable N-terminal domain. This domain defines the different types of NLR proteins: for example, CC-NLR proteins have an N-terminal coiled-coil domain and TIR-type NLRs have a Toll/Interleukin-1 Receptor homology domain.

Activated NLRs oligomerize into resistosomes

The first NLRs were identified back in the early 90s. As scientists identified more and more NLR proteins, which all looked fairly similar on the sequence level and had similar domains, the assumption was that they had similar functions. Indeed, this is a very heavily studied protein family, with over 450 experimentally validated NLRs. Now, not to give any spoilers here, but in a theme that will recur in this article, evolutionary comparisons (and a ton of other research) showed that there was a lot more to NLR function.

The first step in bringing together evolutionary and mechanistic perspectives on NLRs is to get a handle on the diversity of NLRs. When we were stuck at home in lockdown in March 2020, we decided, “Okay, let’s go through the literature and catalogue all the validated NLRs.” So Jiorgos Kourelis did just this, producing a collection we call RefPlantNLR. This reference dataset of over 400 NLRs illustrates basically the diversity of NLRs that have been experimentally validated in 30 genera, more than 10 flowering plants, mostly from Arabidopsis, Solanaceae, and monocots, particularly cereal crops. This survey further illustrates the fact that NLRs detect effectors from pretty much every type of organism that can infect plants, including parasitic plants.

NLR function in pathogen perception and cell death

For over 20 years, since the first NLRs were cloned, it was really a struggle to understand how NLRs function because NLRs don’t necessarily function through your classical sort of MAP kinase pathways. It took really completely new technology — the technology of biophysics, cryo-EM, the structure of activated oligomers — to really figure out how NLRs function. Results from cryo-EM showed that NLRs function through the formation of oligomeric resistosomes, which activate cell death in different ways, depending on the different classes of NLR.

This is wonderful work out of Beijing, China, by Jian-Min Zhou and Jijie Chai’s labs, an absolute landmark discovery in our field. They studied the class of NLRs that carry a coiled-coil domain at the N terminus, termed CC-NLRs, and discovered that the ZAR1 CC-NLR of Arabidopsis, when activated, forms a pentamer with a wheel-type structure, which they called a ‘resistosome’. More recent studies (here and here) identified two new resistosomes formed by NLRs with an N-terminal TIR domain.

Death switch — ZAR1 N-terminal α1 helix undergoes a fold switch that releases a funnel-shaped structure

How do these resistosomes cause cell death? Well, the mechanisms are very different depending on the type of NLR. In the case of the CC-type NLR ZAR1, the N-terminal alpha helix is normally buried in the structure of the inactive complex. However, when ZAR1 is activated, the alpha helix flips so that it protrudes from the oligomer. The oligomer then forms a pore-like structure that is presumably inserted in the plasma membrane where it acts as a channel. In particular, calcium has now been demonstrated to go through this channel, but there could be additional molecules going through this pore. We don’t fully understand how cell death takes place or how the immune response is activated, but this could be through perturbation of the membrane and could also be through this pore-like or channel-like activity.

RPP1 TIR domain is an NADase enzyme that is essential for activation of cell death

The TIR-type NLRs function very differently. These have a TIR domain, which is essentially an enzyme domain, at the N terminus, and function through hydrolysis of NAD. The formation of various compounds from NAD activates the immune response and the cell death response, through an unknown mechanism.

These findings revolutionized our thinking about NLRs. Essentially, 20 years ago, people cloned NLRs. They all looked similar at the sequence level. They had similar domains. The thinking was, “Okay, NLRs: one family, one function, etc.” (Remember the penguins, right?) However, these two different classes of NLRs function through completely unrelated mechanisms, if you like. They might be connected, but that’s another part of the story, which we’ll get to — right after a brief diversion into the history of NLR research.

NLRs and gene-for-gene disease resistance

Let’s go back in time now. What we just covered was basically the state of the art in our understanding of the molecular mechanisms of how NLRs function. But let’s roll back the clock and go back a century ago to Rowland Biffen, a British geneticist who was working at the Plant Breeding Institute, the precursor of the John Innes Centre. He was apparently a very funny character. You can see him with his joke about breeding some very special giant wheat. He was one of the scientists who rediscovered Mendel’s laws of genetics, and he did that by studying rust diseases. He basically discovered R genes. He was the first (or one of the first) to show that there are single genes that were segregating and conferring resistance to a pathogen, in his case, to the yellow rust pathogen.

Pioneers of plant immunity: from Biffen’s resistance genes to Flor’s gene-for-gene model

That was in the early 20th century. Fifty years later, Harold Flor brought in the pathogen perspective with the very influential and very powerful model of the gene-for-gene system. In this model, pathogens have AVRs (what we call effectors today) and if a pathogen carries a particular version of this gene — the AVR effector, let’s call it — and that matches a resistance gene on the plant side, you get this gene-for-gene match or this gene-for-gene interaction: and then you get the resistance. You need both components, both genes, to be present on both sides for resistance and activation of the immune response.

Harold Flor’s model is absolutely fantastic and has been extremely influential, in particular in plant breeding, for example. It has also guided the research because it led to the discovery of R genes and AVR genes through various genetic approaches — both classical and molecular. The reality is, of course, that all these traits have a much more complex genetic architecture that underpins them. Obviously, more than one gene is involved.

Beyond the single gene: the genome as a system

This issue of genetic complexity or metagenes and complex genetic architectures is difficult to tackle with classical genetics. You have issues like redundancy, epistasis, also sometimes phenotyping quantitative traits. These issues are very common in biology and they complicate genetic analysis.

What geneticists do is they simplify the system and ignore essentially all these other genes to focus on the one gene that is variable in the population they study. This has been really fantastic and led to many great discoveries. But perhaps it’s time to move beyond that and start thinking about interactions not simply in gene-for-gene systems but in terms of complex systems. Genomics offers us this opportunity.

Beyond the gene-for-gene interaction model

When the first R genes and the first AVRs were discovered, people assumed immediately that it’s a gene-for-gene interaction in the biochemical sense. The simplest model would be that the matching genes would encode a ligand and a receptor. You have a ligand that binds the receptor, activates the receptor, and then that results in a response: the immune response known as the hypersensitive response (HR) cell death.

This biochemical version of the gene-for-gene model has stuck around for quite a while. But now we know that reality is often more complicated. The diagram above shows just how complex the immune system can be. If you think of the effectors in red and the plant components in blue, then the reality is probably more like this extended network of effector vs. host proteins than the oversimplified gene-for-gene model.

Beyond the gene-for-gene model: receptor networks underpin plant immunity

The reality is that plants have very complex immune networks as the above diagram, taken from this review article, illustrates.

We have indirect detection of pathogen effectors through host proteins known as guards and decoys and even more complex versions of these models.

We have also cases where the receptors themselves — this is a very important point I would make and I will get back to — have diversified to start forming receptor networks. And these receptors start specializing either in either sensing the pathogen or in executing the immune response. So essentially we have sub-functionalization of the receptor into these very specialized activities.

And then of course we have the downstream components, which are also highly variable. They can participate in different responses, depending on the type of NLR that is activated, and also depending on the cell that is being attacked or based on whether the cell is actually the one that is infected, the neighbouring cell, etc. We’re just starting to scratch the surface of these concepts.

Now, of course, every signalling pathway is a network. But what’s really exciting here in NLR immune receptor biology is that most of these receptors form networks among themselves; they form receptor networks. The receptors themselves are interacting with each other to underpin immunity in this very complicated fashion.

Sensor NLRs and helper NLRs

Many NLRs require other NLRs to activate immunity

To illustrate the concept of receptor networks more specifically, we first need to split NLRs into so-called sensor NLRs and helper or executor NLRs. As you might expect, the sensor NLRs recognize the presence of the pathogen (directly or indirectly) and the executor NLRs induce the downstream responses.

We have cases where there is a one-to-one relationship, where there is one sensor interacting with one executor, one helper. In other cases, we have many sensors interacting with many helpers. That’s for example the NRC network that we discovered that I will tell you a little bit more about.

To give you the punch line, the evolutionary model we developed is that plants started with a multifunctional receptor. So the ancestral state of this receptor is to do both jobs: sense the pathogen, and execute cell death. Then throughout evolutionary time, these multifunctional ancestral receptors have sub-functionalized into specialized sensors or into helpers.

In some cases, the sensors have acquired completely new domains that function as baits or decoys for detecting the effectors. This is a really interesting bit, because these sensors have specialized to such degree that they have evolved away from the classical architecture of an NLR and acquired new domains, new extraneous domains, that are specialized in sensing the pathogen. Those type of sensors rely on helpers to execute the cell death.

Asymmetrical evolution of NLR proteins: from multifunctional singletons to pairs to networks

Okay, so to wrap up the evolutionary model, we have transitions from multifunctional ancestral singleton, or functionally singleton receptors, into pairs and networks of specialized NLR proteins. Now let me introduce you to the network we work on.

The NRC network

Studying the NRC network is a project that took my lab by storm. This is now a project that’s five or six years old. I’ll tell you how. It’s getting too complicated to actually illustrate on a slide, but basically, all the sensors shown here are mostly encoded by classical R genes. These are the genes that vary in the population and can be introgressed to bring in resistance. For example, you can introduce Rpi-blb2 into potato and get very specific resistance to Phytophthora infestans carrying Avrblb2. Similarly, we have Rx against Potato virus X and Bs2 against Xanthomonas. We have a number of Phytophthora R proteins: Rpi-blb2, also R1. The network here shows classical R genes that were cloned over the years by various people through classical genetics. They function against all types of pathogens and pests.

NRC network — a CC-NLR network that mediates immunity to diverse plant pathogens

What we discovered — what we stumbled upon, really — is that all of these individual R proteins in this NRC network require NRC helpers, and these NRCs are also NLRs. For simplicity here I’m showing you NRC2, 3, and 4. Again, the system is actually more complex than that, but for simplicity, I’m showing you only these major NRCs. As you can see, there is redundancy in the system, Rpi-blb2, for example, only requires NRC4 and doesn’t work with the other NRCs, but if you take Rx then you have NRC2, NCR3, and NRC4 equally redundantly working with Rx. This redundancy explains why it’s been tricky to identify these components, because if you have three genes basically redundantly functioning in a response, it’s going to be really hard to tackle that using classical genetics or even using metagenetics.

New way of doing business — effectoromics pipelines led us to discover AVR effectors back in the mid-2000s

The reason we started working on this is really purely accidental. This is something I really love because we’ve been following the science here, hopping from one finding to another logical step, and not knowing where things will take us. Years ago, we started working on identifying the AVR effector from the potato blight pathogen Phytophthora infestans that matches the Rpi-blb genes of the wild potato species Solanum bulbocastanum. They were of commercial interest at the time because BASF Plant Science wanted to bring to market a transgenic potato with these two R genes Rpi-blb1 and Rpi-blb2. At that time, we just had access to the genome sequence of the pathogen, and we cloned these two AVR genes using functional screens with what we call an effectoromics pipeline, mining effector sequences from the genome and co-expressing them with the R gene in N. benthamiana. With this approach, we identified AVRblb2, a member of a gene family in Phytophthora infestans that activates the R protein Rpi-blb2.

It was this R protein–effector pair that gave us our lucky break in accessing the NRC network. Chih-hang Wu, a PhD student at the time, joined the lab and was interested in studying the AVRblb2–Rpi-blb2 association. That’s when he identified NRC4 somewhat accidentally while he was doing genetic analyses on components associated with Rpi-blb2. There was also quite a bit of luck associated with this project, because NRC4 is not redundant. Rpi-blb2 only works with NRC4. If we were working on a redundant interaction, say on Rx, for example, or Prf, we probably wouldn’t have easily identified the downstream components.

Afterwards, Chih-hang, through some really powerful comparative genomics and evolutionary analysis, identified that there is indeed a network here beyond the Rpi-blb2 to NRC4 connection. The reason he could identify and predict that there is a network is he noticed that there is a phylogenetic structure among the interacting components. Indeed, looking at the phylogeny of NLRs from a whole bunch of different plant species, mostly asterids, shows that the NRCs form a tight well-supported clade. These are the helpers, and then what’s really remarkable is all the NRC-dependent NLRs, all those R protein sensors, fall into this one big expanded clade, which is a sister clade to the NRC clade. This network is massively expanded in Solanaceae and some other asterids — in some of the species, fifty percent of the NLRs belong to this superclade of NRCs and their R sensors. Moreover, the very well-supported branches suggest a multifunctional ancestor for the R genes, and the clade of the helpers. The NRC superclade and this network probably evolved from a singleton that duplicated into sensor–helper pair and then went on to expand in asterid species and Solanaceae in particular into this massive network.

The NRC network emerged in asterid plants from an NLR pair ~100 MYA

That was a truly exciting discovery. We could estimate that the expansion at least of the network occurred about 100 million years ago, before most asterid species split from each other. And then again it’s a massive expansion. And as you expect, the R genes — the sensor NLRs that evolved to detect bacteria, viruses, effectors, aphids, nematodes, what have you — these have diversified over these hundred million years into all these activities, whereas the helpers, the executors, remained somewhat constrained and had limited expansion. We’ll get back to this topic later.

Now, one criticism we had in the early days when we presented this network was, “Well, okay, you have a network, but big deal. I mean, every signalling pathway is a network and networks are very common especially from the perspective of developmental biology.” And in fact it was a developmental biologist who made the case that networks function in signal conversions to compute an optimal response. Basically, the cell would be taking information coming from the environment, coming from multiple signals, and computing them into the appropriate response. So basically, networks function in signal conversions, adaptation to the environment, amplification of the signal, etc. It is not necessarily that you have redundancy in the strictest sense but rather that these different components are part of a network that computes the best response from multiple signals.

How redundant is the NRC network? Signal convergence/amplification vs. insulated pathways

It was a fair point but this is not the case in the NRC network, where we have truly redundant and insulated pathways. These NRC pathways, even though they are working with the same sensors, are actually insulated from each other, not partially redundant or anything like that. With Rx, for example, we could show that NRC2, 3, and 4 are equally able to confer full resistance to the virus independently of each other. Recently we also developed biochemical evidence that supports the view that these are independent insulated pathways. For example, we can show that NRC2, even though it’s redundant for some R proteins with NRC3 and NRC4, does not physically associate with NRC3 and NRC4, but only forms a complex with itself. We conclude that these are truly biochemically separated pathways that are equally redundant.

This is important, because redundancy is rife in biology. But why is redundancy so common in biology? Let’s delve into this in the next section.

Redundancy in plant immunity: robustness

Why are immune receptor networks redundant?

Why do we have redundancy? Why do we have two eyes? Why do we breathe with our nose and mouth? Of course, redundancy brings robustness. For example, if I get a cold and I cannot breathe with my nose, I can still breathe with my mouth. So redundancy allows the system to be more resilient in terms of dealing with perturbations from the environment, etc. In our case, with plant immune systems, what could be the perturbation, what could interfere with immunity? How about immune suppression by pathogens? If the plant has redundant core elements, then it’s more likely to evade suppression by pathogens.

Very early in this project, when we discovered the network, we developed a very powerful hypothesis: given that we have this large network in solanaceous plants, if we were to take solanaceous pathogens — bacteria, nematodes, oomycetes, aphids, etc. — we should find effectors that are going after these core elements. Because if you can suppress the NRCs, you’re basically taking out of action maybe half of the NLRs. So NRCs should be a great target for a pathogen to go after and suppress. Lida Derevnina joined the lab at that time; she wrote a Marie Skłodowska-Curie fellowship on the topic and she got funded, so she was very excited to focus on this line of inquiry. And I’m really glad that this project has now reached fruition. Lida set up the screen with various solanaceous pathogens and pests and obtained collections of effectors from several collaborators. She designed a screen where she looked for suppression of Prf and Rpi-blb2. These are two R proteins, NLR sensors that are dependent on different NRCs. To cut a long story short, she found five candidates that came out of the screen, three from the cyst nematode Globodera rostochiensis and two from the oomycete Phytophthora infestans. Then she went on to test whether these are functioning at the NRC level or even downstream. There was a very easy trick to do that, because we know how to introduce mutants into the NRC is to make them autoactive or autoimmune, which is when they are active in the absence of a matching R gene or matching sensor. Mutations that make them permanently switched on.

Pathogen effectors that suppress helper NLR hubs in the NRC network

What Lida found then is these two effectors AVRcap1b from Phytophthora infestans and SPRYSEC15 from the potato cyst nematode are potent suppressors of autoactive NRC2 and NRC3 that carry these mutations that make them autoimmune or autoactive. On the other hand, they could not suppress NRC4, so there is specificity in this activity. They are suppressing NRC2, and NRC3, but not NRC4.

SPRYSEC15, the nematode effector, turned out to be very interesting, because this effector directly binds NRC2 and NRC3 but does not strongly bind NRC4. For example, the nematode effector associated with NRC2 and NRC3 in co-immunoprecipitation experiments where we are detecting protein — protein interactions but has a hardly detectable interaction with NRC4. Lida went on to show that the binding happens in the NB-ARC domain of NRC2. A blind yeast two-hybrid screen for protein — protein interactions performed by the company Hybrigenics also identified NRCs as SPRYSEC15 interactors. Work primarily by Abbas Maqbool confirmed these results, showing that the NB-ARC domain of an NRC and this SPRYSEC effector form a protein complex in vitro. We recently published this work and the team should be commended for all these orthogonal experiments with three different methods to show that SPRYSEC15 directly bind NRCs. That’s how science should proceed, on a sure footing so we can build on the results we have.

Mauricio (Mau) Contreras, a PhD student in the lab, also obtained crystals of SPRYSEC15 in complex with the NB-ARC domain with the hope of getting the structure. But this turned out to be heartbreaking for Mau and everyone else, because although he generated three rounds of these crystals, they never diffracted well enough to get the structure. I just want to share this with you because this is also how science proceeds and it can be very frustrating. After many years of work on this structure, we still have no data to show for it. We know that these two proteins form a complex in vitro, but structural biology is an all-or-nothing type of research. And in this case, despite all this work, we still don’t have a confirmed structure and we don’t yet fully understand how this effector interacts with the NRC. We’re still at the drawing board. Mau still has a couple of years to go with his PhD 😏.

Whatever works! Pathogens convergently evolved to target the NRC network

To sum up this work, the key point here is that pathogens have evolved to target the NRC network at multiple levels — as the saying goes, in evolution it’s whatever works, right? We have SPRYSEC15, which we know binds both the inactive and activated forms of the NRC resistosomes, presumably. We also found Phytophthora AVRcap1b indirectly suppresses the NRC response by binding another host protein TOL9a, which is involved in endosome or vesicle trafficking. Again here, we don’t understand the full mechanism yet, but Mau has been studying this and he found out that TOL9a is important for the suppression activity of the AVRcap1b. So maybe this protein is acting downstream of the activated resistosomes. Also we have the three other effectors that we didn’t have time to study in detail yet but that seem to act at the level of Rpi-blb2–NRC4 interaction — not downstream — and they suppress the activity of this NLR pair upstream of NRC4.

The conclusion and the conceptual message here — going back to my point about redundancy — is that perhaps one reason why the NRCs have evolved to become redundant was to evade suppression by pathogen effectors. For example, let’s take Solanum demissum, the native wild potato host of Phytophthora infestans that occurs in Mexico. Phytophthora infestans is a global pathogen in agricultural systems, but it originates from central Mexico — an area known as Toluca Valley — and infects wild Solanum species like Solanum demissum. Solanum demissum is a hexaploid and it has a very large family of NRCs. So our suspicion — and we haven’t had time to actually get into this project — is that co-evolution between effectors targeting these NRCs and NRCs evading the suppression is driving the emergence of multiple NRCs to evade suppression while keeping on interacting with the different sensor NLRs. That would at least be one possible explanation for why we have a network in this NLR system.

Redundancy of NRCs means that when one is suppressed by a pathogen, another one would do the job. The same reason why we have two engines in an airplane (and three engines in those old Tupolev Tu-154 aircrafts). If one engine fails, the plane wouldn’t crash. The system overall would be more resilient to the perturbations caused by external agents, in our case the pathogens.

Redundancy in plant immunity: evolvability

Let’s now talk about the second benefit of redundancy: evolvability. What is evolvability? Evolvability is another really important concept in biology, especially in the context of biological networks. Evolvability is essentially when you have a network, this complex system can be much more efficient and has a lot more room to vary and evolve.

I’m going to give you another metaphor because I love metaphors and I think metaphors are always good to illustrate the point. (Remember the penguins?) So think about a restaurant. Imagine you have a restaurant, and you have only one person running that restaurant. I’ve seen places like that. You know, you get into some small village in Italy, and there will be a restaurant, and this guy will be cooking, hosting you at the table, getting you the check, doing everything, right? So that guy has a really hard time managing a lot of people in that restaurant. That would not be a very efficient system.

But if you have several people involved, and they specialize in different activities… one person is cooking at the kitchen, one person is waiting at the table, one person is the cashier, getting you the bill, etc., that system becomes much more efficient, much more flexible, much more evolvable, if you like. That system can accommodate now a lot more diversity. You can have a bigger restaurant; you can have more people coming to the restaurant; you can serve more types of dishes; it’s easier to deal with specific requests from the diners etc.

Similarly, if you have one receptor doing everything, that receptor is going to be constrained as to what it can actually evolve into. For example, it’s going to be hard to acquire a whole new domain and at the same time maintain the ability to signal and cause cell death through the classical resistosome pathway. However, if you uncouple recognition from signalling into different proteins that are interacting together as a network, that system becomes more evolvable. There is a lot more room for that complex system to diversify.

Indeed, this is what we think has happened: we started with multifunctional ancestors, and these have evolved into specialized receptors (the sensors) and specialized executors (the helpers). Check the network cartoon above and see how much the sensor NLRs diversified into detecting all different types of pathogens and pests. Perhaps that would have been much more difficult to evolve if they also had to maintain the capacity to execute the cell death response rather than rely on NRC helpers to do that?

Evolution of NLRs and their networks

What about the data? Can we challenge this model? Can we look at the diversity of NLRs we have, interpret them from this evolutionary framework, and ask the question: what are the molecular consequences of transitioning from a singleton into paired and networked receptors that are specialized? Can we see biochemical evidence for this transition, and what are the biochemical signatures of this sub-functionalization?

Let’s go back to ZAR1. Remember the ZAR1 resistosome and how the N-terminal alpha helix is normally buried when the receptor is inactive, but it flips out when ZAR1 is activated (let’s call it the death switch, if you like). This conformational change results in a funnel formed by five of these alpha helices, five different ZAR1 proteins coming together. They form a pore or channel that perturbs the membrane and functions as a calcium or anion channel.

ZAR1 and NRC4 share the N-terminal MADA alpha helix

When the paper of Jizong Wang and colleagues describing the resistosome came out in early 2019, we got very excited, because we had some data in the lab suggesting that the N terminus of an NRC is really important for its activities. About a year and a half before that, Hiroaki (Aki) Adachi, who was a postdoc in the lab at the time, performed a transposon mutagenesis screen — a truncation screen, if you like — of NRC4 to find out if we could get shorter versions of this protein that are autoactivated. What Aki discovered is that the very N terminus of NRC4, its first 29 amino acids, are sufficient to trigger the cell death hypersensitive immune response. To be honest, we really didn’t know how to interpret this result. We knew there was something important there, but we just didn’t know how to make sense of it. So it went in the drawer of intriguing results and Aki moved on to projects with better prospects. This is why when we learned about the death switch of ZAR1 and the funnel-like structure of the ZAR1 resistosome, it was an Aha moment for us. Right there we had a clear explanation of the functional importance of the N terminus, and to top it up a mechanism to explain how it would cause cell death. Our excitement is palpable in this video we produced as part of the public outreach associated with the publication of the Wang et al. papers on the ZAR1 resistosome.

Right then, we made the executive decision to drop everything else and focus on studying the N terminus of NRC4. Aki went into full Ninja Warrior mode and to his credit and the credit of many others who joined forces on the project, they could wrapped up the paper and published a preprint in July 2019, just a few days before we presented the work at the Congress of the International Society for Molecular Plant-Microbe Interactions in Glasgow. Next is a summary of this work.

Of course, the first thing we did was to align our NRCs to ZAR1 and guess what? It turns out that the N terminus of the NRCs is actually quite conserved compared to ZAR1. You can actually see in the alignment below that the N-terminal alpha helix is roughly 50 percent identical between ZAR1 and NRC4. So that was really exciting.

MADA motif defines N-terminus of about one fifth of CC-NLRs

We also had this independent evidence suggesting that in the NRCs, the N terminus is important for cell death activity. And then the N-terminal alpha helix turned out to match a motif, which we call the MADA motif, based on the M-A-D-A signature in the consensus amino acid sequence of these NLR proteins. These MADA-type CC-NLRs make up about one-fifth of the CC-NLRs and they are found in CC-NLRs across all plant species, not just in Arabidopsis, asterids, and the Solanaceae, but also in monocots such as in the well studied CC-type NLRs of rice (Pik2) and barley (MLA), which confer resistance to fungal plant pathogens.

α1 helix of ZAR1 and other MADA-CC-NLRs can functionally replace the N-terminus of NRC4

In the lab, we always aim to follow up such observations with wet lab experiments to challenge the hypotheses that arise from the computational analyses. We did just that and tested the functional conservation of the MADA motif. We demonstrated that this is not just sequence conservation but also functional conservation. We can take the N terminus of multiple MADA-type NLRs, including ZAR1 itself, and swap them into NRC4 and those chimeric NRC4 will still function and cause cell death and even can produce resistance to Phytophthora infestans by functioning with the Rpi-blb2 R protein. In this one experiment, the ZAR1–NRC chimera, where the N terminal alpha helix of NRC4 is replaced by the ZAR1 alpha helix, confers resistance to P. infestans. So this is not only sequence conservation but also functional conservation across a large swath of NLRs from many different dicot and monoct species.

Now, back to our NRC network. What was amazing is that we can only find the MADA-type sequence in the NRC executors. But absolutely none of the sensors, the R proteins themselves, carry this MADA signature. In fact, there are two major classes of NRC-dependent sensor NLRs (NRC-S) that cluster in two major phylogenetic clades as shown below. One clade is defined by Rx, Gpa2 andBs2, which belong to the sensors that have the classic three domain CC-NBARC-LRR architecture, but Prf, Sw5, Rpi-blb2, Mi and so on, all these other R proteins, belong to a clade of NRC-dependent sensors that have an N-terminal insertion before the CC domain, before where the MADA would be. And that N-terminal integration, which can be quite large — several hundred amino acids — can be absolutely important for the detection of the pathogen effector. There’s some really nice work on Sw5 coming from the lab of Xiaoroing Tao at Nanjing Agricultural University indicating the importance of this N-terminal extension. And also with Prf, nice work there also, showing the importance of this N-terminal extension.

So these NLR sensors, because they have that N-terminal extension, cannot function through the resistosome-type model. All these sensors have diversified — they lost the MADA sequence simply because they don’t need it. They’re functioning with the NRC helpers, which are the ones that are executing this cell death, and probably functioning like ZAR1 causing hypersensitivity through pore formation and channel activity at the plasma membrane.

Use it or lose it! CC domain of sensor NLRs have become non-functional over evolutionary time

So the bottom line here is: use it or lose it. Another dictum of evolutionary biology. The sensor NLRs didn’t need the MADA sequence anymore, so the MADA sequence has degenerated over time. That’s the model we have. On the one hand, as these NLR proteins went from singletons into specialized pairs, the sensors didn’t need the MADA sequence so it degenerated and became non-functional as a result of the specialization of the NLRs to the sensor activity. And then some of these NLR sensors acquired totally new N-terminal extensions to help detect the pathogens; these new domains definitely should make them unable to function through typical ZAR1-type resistosomes. On the other hand, the helpers — through all this evolutionary time, dating back to monocots, dicots, and what have you — retained the MADA sequence over long evolutionary time. And the MADA sequence has retained its conserved position at the N terminus, which is critical from the perspective of the resistosome model and the death switch.

Molecular basis of functional specialization during CC-NLR evolution from singletons to networks

So, you can see now, as this evolutionary diversification happened, we also had NLRs transition into new functional categories and take on these specialized roles by sub-functionalization. These roles also come with biochemical signatures, where the protein takes on very different, very distinct evolutionary paths by functionally specializing into either the sensing activity or the helper or executor activity. And that makes this network a much more evolvable system, that probably keeps up better with rapidly evolving pathogens. These guys — the NRC-S — are now freed from previous functional constraints to co-evolve with pathogen effectors, acquiring new domains, acquiring new mutations. May be in some of the sensors the CC domain now is directly baiting the effector. It could have happened, we’ll look for that. And it wouldn’t impact the CC domain as much as it would in a helper NLR. The CC domain of the sensors can now diversify and become dedicated to capturing effectors without being constrained by the necessity to execute cell death. Just like when that Italian restaurant transitioned from a single person to a full team of specialized staff, the staff would have more capacity to develop and add tasks to a specific activity. Beautiful, isn’t it?

Take-home messages and penguins, redux

If you want to understand NLR diversity, if you want to understand how these proteins function, you simply cannot ignore the evolutionary perspective. You’re not going to be able to understand what these proteins do if you focus on one protein, one model system, and ignore the wider evolutionary perspective, ignore the fact that these NLRs form receptor networks, and ignore the fact that many of their features are degenerated or have diversified way beyond their ancestral function.

So if I have one take-home message from the whole talk, it’s this. I want you to remember the penguins. It’s important to do comparative studies to avoid a myopic view — and that’s what we aim to do in evoMPMI. Biology isn’t just about understanding the mechanisms by which organisms function, but also about figuring out how they became what they are. I want you to appreciate the diversity of structures and functions that come with what we call NLR immune receptors. We need to move the field beyond the restrictive uniform view of NLR structure and function. It’s time to question holistic concepts such as effector-triggered immunity and appreciate to its just measure the wide diversity of NLRs and the different ways they contribute to immunity. These proteins are rapidly evolving, and they also have evolved very rapidly both by gaining and losing molecular activities and biochemical functions.

I hope this opens the door to a great discussion.

Thanks to the people who did the work

A lot of people contributed to this project and several of them were mentioned above. Chih-hang Wu is the one who really started this work, and Chih-hang is now at Academia Sinica. He was a PhD student when he started this work and it was truly incredible how we literally stumbled onto the NRC network while collaborating with Ahmed Abd-El-Haliem and Jack Vossen. He was a very influential PhD student because now, more than half of my lab works on the NRC network in some way or another. Lida Derevnina is another major contributor. She is the one who did the suppressor screen, and she’ll be moving to the Crop Science centre in Cambridge in January 2022 to continue working on suppressors of NLRs with an emphasis on nematodes. Aki (Hiroaki Adachi) joined the project later but he’s the one who discovered the MADA motif. He’s also now just started his lab at NARA Institute of Science and Technology (NAIST) in Japan. The three of them will be working on NLRs so I look forward to collaborate with them and follow their progress. Mau (Mauricio Contreras) has done a lot in different studies related to NLR biology and is a current PhD student. Keep an eye on his upcoming papers. Jiorgos Kourelis is the guy behind RefPlantNLR. Abbas Maqbool is our biochemist structural biologist. I didn’t talk about blast diseases today… This is the work of the BLASTOFF Team and that’s for another day. But I should also thank several other people in the team: Adeline Harant, Joe Win and also some new people who joined us in 2021. In particular Hsuan Pai has already had quite a lot of influence and contributed a lot of work to Team NLR. And of course, many thanks to my colleagues at The Sainsbury Lab.

I’m very grateful to Jennifer Mach and Monica Harrington at PlantEditors for turning the transcript of the talk into the readable version posted above.

[September 17, 2021. Modified with minor edits and to fix typos]




Biologist; passionate about science, plant pathogens, genomics, and evolution; open science advocate; loves travel, food, and sports; nomad and hunter-gatherer.