As of today, 85,074 point mutations that change coding sense in 18,724 genes have been collected for screening, and almost all of those have been subjected to the screening pipeline in full. These mutations came from a pool of 44,906 G3 mice, examined over a period of two years. These G3 mice belonged to 1,587 pedigrees. In all, the mutations were queried a total of 6.6 million times by the computer. That means, in every mouse, every mutation (of which there might be 60 or so), was checked in 150 different screens, testing to see whether the null hypothesis of non-association in recessive, additive, or dominant models of inheritance could be rejected. This is a fairly computationally intensive process, and it requires a cluster computer for adequate speed.
We take it somewhat arbitrarily that a good test of the effect of a mutation is to look at it in the homozygous state three times or more. Obviously more testing permits the detection of more subtle phenotypic effects. But a robust phenotype can generally be detected with three tests. By that criterion we have examined 10.2% of all genes in a putative null state, and 37.7% of all genes in either a putative null or probably damaged state. All this means that we have examined 24% of all genes in the mouse genome, having damaged them sufficiently to detect a phenotype if one could be made through damage, and having checked the mutant alleles three times or more in the homozygous state.
It is possible for us to project how this will go over time. At the present time in the progress of our mutagenesis effort we are right here [22:15]. We have created coding changes in something like 74% of all genes. This is the red curve for probably null or probably damaging; this is the curve for putative null alleles, and this is the curve for mutation to phenovariance—in all cases, with three times examination. We can project how far we will have to go to reach, let’s say, 33% mutation to phenovariance. That should happen sometime around the end of this year, or by the time we have reached a total of approximately 68,000 G3 mice. We can also estimate saturation by the number of mutations that have been screened, or by time (assuming a certain rate of mouse production). All of these curves are predicated on our present rate of screening, which is about 600 mice per week. That means testing about 1,000 mutations per week for phenotypic effects in 150 screens.
In the course of screening, we find things that are known, of course. For the adaptive immune screens alone I made a list this morning. I see that we have found by phenotype a total of 74 genes that were previously known to be needed for adaptive immune function or response, because mutations in these genes created phenotypes in our mice. Of course we are not interested in things that are already known. We are interested in things that are unknown. I estimate that approximately the same number of mutations that have effects that are unknown have also been found and verified. So there is a great deal still to learn about what is needed for the adaptive immune response.
How do we keep track of our mutations and their effects? We developed software that lets us parse the mutations and query them according to gene name or screen name or particular groups of mice. We can also parse according to the predicted effects of mutations, whether they are nonsense or critical splicing errors, or other. We can restrict our search to pedigrees with greater than a certain number of mice. We can restrict our search according to the number of homozygotes for any mutation. We set the P value of what we believe is significant and wish to pursue.
I will give you an example from the innate immune screens, where we have recently had some success. Many of you have heard of the NLRP3 inflammasome. NLRP3 is involved in most forms of inflammation—notably in gout, and also in certain diseases such as cold-induced auto-inflammatory disease, or neonatal onset multisystem inflammatory disease (NOMID). It is the protein that organizes the processing of interleukin-1β, an inflammatory cytokine that is secreted and helps to generalize the inflammatory response. We wanted to find all kinds of mutations that affect NLRP3 inflammasome function, either increasing interferon-β production or abrogating it in response to a defined stimulus. Here we focus on that particular screen [25:46]. I won’t go through what trimmed and untrimmed are. We restrict the search to >15 as the total number of mice in the pedigree. We insist on seeing three or more homozygotes for a mutation, and we set the P value at 0.005.
When we click [26:08], we find a list of genes. In fact, we retrieve 61 mutations in 60 genes from 46 pedigrees. Some of these will be familiar and some will not. Here is one that is familiar. We mutated NLRP3 itself. In fact, we hit the gene nine times in our collection of mutations, making nine separate mutations. Sometimes one has multiple pedigrees with the very same mutation. Usually that is the result of inheritance from a common G0 progenitor. You see the co-ordinates of the mutation, you see the computer’s declarations about it, that it is a missense allele, probably damaging. You can move over to the right and you see that there are in this pedigree five homozygous wild-types, five heterozygotes, five homozygous mutant mice. By the additive model of inheritance (in other words, semi-dominant model of inheritance), the mutation passes our criterion for the P value.
If we want to look at all of the mutations, we can do so by clicking on that value [27:21]. Here is the one of interest. If you mouse over it, you see that this is NLRP3. This is a scale showing the likelihood of association between phenotype and genotype, given random assortment. In other words, there is a likelihood of nearly one in 10–6 that the association would occur by chance.
If you did not know anything about this gene, you could click on it again, and the computer has already calculated a lot of information for you with links to the MGI database and other sources of detailed information. It has drawn a gene model and also a protein model. This is the protein model in SMART format [27:59], and it shows that this is a missense mutation of isoleucine to asparagine at position 293. The model is interactive.
You can also look at the gene model by clicking on the link, and you see that the mutation is in exon 5, and not predicted to effect splicing.
As to the phenotypic data, if you left-click on the point [28:26] you see this is the performance of homozygotes, which are poor producers of IL-1β, heterozygotes which are intermediate, and homozygotes for the reference allele. These are all from a common pedigree and can be taken as littermates. It was the allele-additive relationship that made the computer flag the mutation. The wild-type is shown for different purposes, not for comparison.
There are also mutations that would seem new to most of us. Many of you would not guess that a protein called NEK7 is important for the inflammasome response. We have four separate alleles of NEK7. This is a nonsense allele, and we see that the pedigree is quite large. There are a total of 42 G3 mice, and by an additive model the Manhattan plot shows a linkage peak. NEK7 is a member of a kinase family called the NEMA kinases, or the “never in mitosis” kinases. The mutation in question is a premature stop codon. The NEMA kinases are associated with mitosis. They are known to be required for assembly of the spindle apparatus and involved in abscission of cells during late mitosis. Yet nothing was known about their involvement in inflammasome function.
The homozygotes show poor IL-1β responses. Heterozygotes perform better; homozygous wild-type mice perform best of all.
That might not be enough to convince you. Perhaps you are wondering about the other pedigrees we have and what they show. What of the other alleles of NEK7? Fortunately, the computer knows this. It realizes when there are multiple alleles and groups all of the data into superpedigrees. Eventually all of the genes will be included in superpedigrees. At present 61% of all genes have two or more alleles and are in superpedigrees. With multiple alleles one gains confidence about the strength of associations.
Some superpedigrees are spectacular. This is a particular unknown gene that I won’t talk about today. A mutation within it affects the number of CD8 cells. We are looking here at phenotypic performance pooled across about 16 pedigrees. In this case, always the same mutation was involved. This was a case where there was multiple transmission to many G1 mice. Notice there are hundreds of mutations now. Only one of them is above the Bonferroni correction line. That’s the gene in question [31:11].
If you click on it, you can observe the phenotypic performance of this allele in all the different pedigrees, with color coding of each. You see the CD8 count is shifted quite dramatically in homozygotes as compared to heterozygotes or mice with the reference allele. You would not really need to target this gene in view of these data. You could be quite confident that this was the causative mutation. But to guard against the possibility that an unseen mutation might be responsible, we target the gene nonetheless.
The situation isn’t quite so good for NEK7, but NEK7 is above the Bonferroni correction line. Each of three different mutations that reached homozygosity shows a diminished inflammasome response compared to the heterozygote or reference allele populations. So we can feel fairly confident in proceeding and knocking out the gene.
We observed the NEK7 protein in cells from homozygotes for the ENU allele, which we named Cuties, is practically gone. The mRNA is eliminated by nonsense-mediated decay, and what little truncated protein is translated is apparently unstable. When we did knock NEK7 out by CRISPR targeting, we found again that the protein was gone. Furthermore, the NLRP3 inflammasome does not work properly; one can normally activate it with Nigericin, with ATP, or with Alum, all of which are inflammatory stimuli specific for NLRP3. On the other hand, other inflammasomes, such as Nlrc4 which responds to flagellin, or Aim2, which responds to Poly(dA:dT), operate normally in the absence of NEK7—there is no difference in phenotypic performance. We knocked down the gene in human mononuclear cells, and we found that there, too, the inflammasome was suppressed in cells from two different donors.
Notice that IL-18 as well as IL-1β is suppressed in Cuties mice or knockout mice. In this case we are looking at another end-point of the NLRP3 inflammasome. We find that cytotoxicity is diminished. This is called pyroptosis, and it is diminished using all of these stimuli. On the other hand, TNF production and IL-6 production are unimpaired in Cuties mice.
There are two signals that activate the inflammasome. Signal 1 drives the expression of inflammasome components like NLRP3 itself and also Pro-IL-1β—the target for cleavage by the inflammasome. Signal 2 is generated by reagents like Nigericin or Alum. In a somewhat mysterious way they activate the inflammasome. Signal 1 is unimpaired. Notice that with no treatment little or no NLRP3 or Pro-IL-1β is expressed. With LPS, the first signal, there is a strong response in Cuties mice. On the other hand, signal 2 is very much impaired. Homozygotes show almost no response to LPS+Nigericin or LPS+ATP in terms of secreted IL-1β.
Signal 1 is also unimpaired if you look at hallmarks like mitochondrial oxidative radicals or calcium flux into the cell. These responses are essentially the same in Cuties and wild-type mice.
We next looked at the assembly of NLRP3 and its association with ASC, a downstream protein that recruits caspase 1, which then cleaves Pro-IL-1β to release IL-1β. This can be followed [35:06] by sucrose density gradient ultracentrifugation. We can see a slight shift of NLRP3 when we do this, toward a heavier weight, if we activate with Nigericin. One can also observe ASC recruitment of the complex, and in Cuties mice that is very much impaired compared to wild-type mice. We can also look at the oligomerization of ASC using a cross-linking agent like disuccinimidyl suberate [DSS at 35:36]. In this case you find that with Cuties mice that is also very much impaired.
One can see physical association between NEK7 and NLRP3 as well. NEK7 in normal wild-type cells is distributed like this [35:55] in a sucrose gradient, and, on the other hand, NLRP3 is located toward the high-density end of the gradient. When you activate, NEK7 largely overlies the NLRP3 band, indicating that NEK7 may associate with NLRP3 and co-sediment with it.
Looking at how it might interact more directly by expressing tagged versions of NLRP3 and NEK7, we find that the whole protein does associate and co-precipitates. Furthermore, if we break up NLRP3 into its pyrin domain, nucleotide-binding domain, and leucine-rich repeat portions, NEK7 binds to the leucine-rich repeats of NLRP3.
Mutations affecting the leucine-rich repeats of human NLRP3 result in rare cases of NOMID, because they make a constitutively active protein. Two of these are shown here: G775A and G775R. These mutations cause tighter association with NEK7. On the other hand, we have in our collection a mutation in NLRP3 that diminishes the inflammasome response, and here it diminishes the binding of NEK7 to the leucine-rich repeats. This mutation is also in the leucine-rich repeat moiety of NLRP3.
NEK7 is a kinase, but kinase activity is apparently not required for association between NEK7 and NLRP3. Nor is it required for NEK7 to rescue the Cuties phenotype. If we take knockout cells and transfect them either with wild-type or with a kinase-dead mutant expression construct, in both cases rescue of the IL-1β non-production phenotype is observed.
Finally, NEK7 is also required for NLRP3 inflammasome activation in vivo. Cuties mice don’t mount a normal inflammatory response to the intraperitoneal (IP) injection of urate crystals in terms of recruitment of peritoneal exudate cells or neutronphils or macrophage separately. We have also studied experimental allergic encephalomyelitis (EAE) which is an IL-1β-dependent phenomenon, and the NEK7 mutation suppresses that as well. We are intrigued by the fact that NEK7 is a kinase that is known to be involved in mitosis. We blocked cells at different parts of the mitotic cycle and found that during mitosis it was not possible to induce an inflammasome response. This is a measurement of the activation of caspase 1 in mitotic cells compared to interphase cells. We hypothesize that this failure to activate the inflammasome during mitosis reflects the recruitment of NEK7, an essential inflammasome component, to fulfill duties required elsewhere in the cell during mitosis. The evolutionary explanation for this recruitment might be that it prevents damage to DNA that would occur during inflammasome activation, at a time when chromosomes are condensed and when repair can’t take place as it ordinarily would.
We now view the inflammasome quite differently than before. We would say that in response to signal 2 there might be activation of the NLRP3 molecule itself, and there are a number of proposed mechanisms for that. But it could just as well be that activating signals impinge upon NEK7. Whatever the activation mechanism, it is clear that NEK7 binds to the leucine-rich repeats of NLRP3 and that is necessary for activation of caspase 1 and the generation of IL-1β, IL-18, IL-33, and all that follows downstream in the inflammatory cascade. Nobody has ever managed to crystallize the NLRP3 inflammasome; that may be because they have been missing this important component that affects the solubility of the molecule. We are attempting to do structural studies now on the complex between NEK7 and NLRP3.