Session 10: Viral Evolution

Transcript of Part 3: Host Evolution

00:00:00.28	Hello.
00:00:02.01	My name is Harmit Malik,
00:00:03.11	and I'm an evolutionary geneticist
00:00:05.03	studying the evolution of viruses and host genomes
00:00:07.28	at the Fred Hutchinson Cancer Research Center.
00:00:10.15	Today, I'm actually going to tell you
00:00:11.28	about molecular arms races between primate
00:00:14.09	and viral genomes
00:00:15.25	and how we aim to understand
00:00:18.01	the evolutionary rules that take place
00:00:20.02	between these viruses and hosts
00:00:21.24	and what that will tell us about,
00:00:23.13	not just the evolution of ourselves and viruses,
00:00:26.11	but also to design therapeutics interventions
00:00:28.23	to allow us to designed better strategies
00:00:31.06	that are going to be effective against viruses.
00:00:33.22	The work in the field of molecular arms races
00:00:36.25	is really inspired by
00:00:39.10	the character the Red Queen that was introduced to us
00:00:41.25	by Lewis Carroll in his book "Through the Looking Glass",
00:00:44.20	and the Red Queen tells Alice
00:00:46.29	in this sort of nice book
00:00:50.07	that it takes all the running you can do
00:00:52.11	to keep in the same place.
00:00:54.01	Very much the same idea
00:00:55.27	was adopted by the evolutionary biologist
00:00:58.01	Leigh Van Valen,
00:00:59.23	as the Red Queen hypothesis,
00:01:01.28	and he argued that in a system
00:01:04.00	where two entities are constantly competing
00:01:05.22	with each other in this sort of battle
00:01:07.22	for evolutionary supremacy,
00:01:09.15	the only way for this battle to be resolved
00:01:12.08	is just for one party to temporarily win
00:01:14.29	before the other party catches up,
00:01:16.26	and this requires both of these parties
00:01:19.01	to be really running as fast as they can
00:01:21.04	with this really rapid evolutionary signature,
00:01:23.09	formalized as the Red Queen Hypothesis
00:01:25.15	that's been used to invoke
00:01:27.21	all kinds of very important principles
00:01:29.19	in evolutionary biology,
00:01:31.13	including the existence of sex
00:01:33.16	and why we actually evolved
00:01:35.21	to be sexual creatures in the first place.
00:01:38.22	So, if you consider a host-virus interaction,
00:01:41.06	this is an interaction
00:01:43.00	that screams out genetic conflict.
00:01:44.23	This is what we refer to
00:01:46.13	as the usual suspects.
00:01:48.00	It doesn't take a lot of imagination
00:01:49.13	to understand that what is in the best interest of the virus
00:01:52.10	will not always be in the best interest of the host.
00:01:55.03	So, in this cartoon example,
00:01:56.25	you can see that we've got two states described here,
00:01:59.24	you've got the host that is binding the virus
00:02:02.08	on one side,
00:02:03.25	and the virus that has evolved a mutation
00:02:06.03	to evolve away from that recognition
00:02:07.28	by the host immune system.
00:02:09.17	What you'll actually appreciate
00:02:11.20	is that these state transitions,
00:02:13.11	between one state and the other,
00:02:14.28	are really profound but very simple
00:02:16.26	from a mechanistic standpoint.
00:02:18.25	What it might take is just a single amino acid mutation
00:02:21.10	for the virus to gain one step ahead
00:02:23.19	in this battle for evolutionary supremacy.
00:02:26.09	So, the important take-home message
00:02:27.25	from this kind of slide is,
00:02:29.14	one party is always losing
00:02:31.22	this high-stakes evolutionary battle.
00:02:33.13	On the left-hand side
00:02:34.28	you can see that the host is winning,
00:02:36.18	because it is recognizing a viral protein.
00:02:38.14	On the right-hand side
00:02:39.23	you can see that the host is losing,
00:02:41.13	because the virus has acquired the right mutation
00:02:43.22	that allows it to evade detection by the immune system,
00:02:46.15	which basically means that there's never going to be
00:02:49.10	a perfect equilibrium between these two states.
00:02:51.10	Over the course of evolution,
00:02:53.01	and even the course of a single infection in a person,
00:02:56.02	the immune system and the virus
00:02:58.26	are basically locked in this arms race
00:03:01.03	of very rapid evolution
00:03:03.09	and, because one party is always losing,
00:03:05.03	there's always going to be an evolutionary advantage
00:03:07.06	to be gained by innovation.
00:03:09.06	Now, we're going to actually talk about
00:03:11.04	two types of innovation today.
00:03:12.18	In the first part of my talk,
00:03:14.04	which is focused exclusively on how hosts evolve
00:03:16.28	in the face of viral challenges,
00:03:18.28	we're going to specify innovation
00:03:21.10	in protein coding genes,
00:03:23.03	and so, if you consider
00:03:25.05	what a protein coding gene arbitrarily looks like,
00:03:28.11	it's this sort of sequence that I've indicated here,
00:03:31.02	where we've got three triplets, three codons,
00:03:34.00	that specify three amino acids
00:03:35.25	that will be incorporated into the protein
00:03:37.17	that is produced from this gene.
00:03:39.15	Now, you can see on this side,
00:03:41.15	you have a mutation
00:03:43.17	that does not alter the amino acid being encoded.
00:03:45.23	We refer to these as silent or synonymous changes,
00:03:48.27	because from a very sort of rough approximation
00:03:51.17	natural selection is really acting on
00:03:54.01	the protein coding sequences,
00:03:55.19	and here, because the protein coding sequence
00:03:57.10	has not altered, we refer to these as
00:03:59.22	silent or synonymous changes.
00:04:01.14	In contrast, you can see here we have,
00:04:03.21	again, a single amino acid mutation,
00:04:05.28	which has altered one of the amino acids
00:04:08.13	that's being encoded,
00:04:10.05	so-called non-synonymous or replacement changes.
00:04:12.21	Now, both of these
00:04:14.25	are sort of equal likelihood mutations. Y
00:04:16.21	ou can actually have a synonymous mutation
00:04:18.16	or a non-synonymous mutation,
00:04:20.08	but you can appreciate that, based on the genetic code,
00:04:22.18	you're much more likely
00:04:24.22	to see an amino acid-altering mutation,
00:04:26.18	just by random chance alone.
00:04:29.18	So, consider the sort of situation
00:04:32.19	where you actually had a gene,
00:04:34.23	we refer to these as pseudogenes,
00:04:36.29	that at some point in their evolutionary history
00:04:39.02	encoded for a particular protein.
00:04:41.19	Now, if you consider this gene
00:04:44.09	now in its current degenerate form,
00:04:46.22	let's say built from the chimpanzee genome
00:04:48.16	versus the human genome,
00:04:50.09	and we were to just roughly calculate
00:04:52.07	the number of synonymous changes
00:04:54.15	versus replacement changes,
00:04:56.17	we have to correct for the fact
00:04:58.10	that there are many more possible replacement changes,
00:05:01.01	so when you normalize for that correction
00:05:03.02	you will find that, because this gene no longer codes
00:05:05.18	for a protein,
00:05:07.09	the rate of synonymous changes
00:05:08.24	and the rate of replacement changes
00:05:10.15	are roughly equal,
00:05:11.25	and that's because selection has stopped worrying
00:05:14.13	about this part of the genome
00:05:16.02	in terms of its protein-coding capacity.
00:05:17.24	It has tolerated both mutations,
00:05:19.22	and they roughly go to fixation
00:05:21.20	in a fairly random fashion.
00:05:23.15	Now, for most genes in the genome,
00:05:25.08	you do care about the final product being produced,
00:05:27.16	which is the amino acid sequence
00:05:29.19	of the resulting protein.
00:05:31.07	So, here I have this hypothetical example
00:05:33.14	where you have a protein-coding gene
00:05:36.14	that is basically representing these triplets of codons,
00:05:39.12	and what you'll see is there's a lot more blue changes,
00:05:42.11	or non-amino acid altering or silent changes,
00:05:46.23	and very rarely do you see something
00:05:48.22	which looks like a replacement
00:05:51.13	or a non-synonymous change.
00:05:53.09	The net result is that,
00:05:55.00	regardless of all of this change at the nucleotide level,
00:05:57.16	the amino acid sequence remains STEVE,
00:06:00.01	because STEVE is really what is being selected for
00:06:03.00	by evolution.
00:06:04.11	Very rarely do you see a deviation
00:06:06.13	from this optimal amino acid sequence.
00:06:08.16	For instance, we can see SiEVE coming in,
00:06:10.29	in terms of this sort of grammar.
00:06:13.00	The net result is not that
00:06:16.20	we should infer that mutation has now stopped
00:06:19.06	hitting the replacement sites.
00:06:20.27	What we infer from this is,
00:06:22.28	because mutation has introduced changes
00:06:24.22	in both replacement and synonymous positions,
00:06:27.19	the fact that we don't see replacement changes
00:06:30.01	over the course of evolution
00:06:31.26	is an indication that natural selection
00:06:34.03	acted upon these changes,
00:06:35.21	deemed them deleterious,
00:06:37.18	and removed them from the population
00:06:39.10	before they had a chance to really spread
00:06:41.28	in the population,
00:06:43.21	which means mutations is not really causing
00:06:45.19	this bias between the blue and the red changes.
00:06:48.06	It's actually natural selection,
00:06:50.02	and, more specifically, purifying selection,
00:06:52.06	that is acting to purify the population
00:06:54.16	from these presumed deleterious mutations.
00:06:56.26	The net result is,
00:06:58.21	if you were to now compare the rate of
00:07:01.02	synonymous and replacement changes,
00:07:02.29	we will find that the rate of replacement changes
00:07:04.25	is actually much lower than synonymous changes,
00:07:07.26	regardless of the fact that both of these changes
00:07:10.00	were introduced in roughly the same proportion.
00:07:13.11	My lab is actually interested in the other
00:07:15.00	class of genes that emerges
00:07:16.20	from these kinds of analyses.
00:07:18.05	Here again, now, we have a triplet code of sequences
00:07:21.03	that encodes for my name in amino acid code,
00:07:24.18	and what we will see when we compare
00:07:26.26	across this sequence
00:07:28.27	is that there are a lot more red changes
00:07:30.27	than blue changes,
00:07:32.09	in fact a lot more red changes than what you'd expect,
00:07:34.20	even by chance alone.
00:07:36.19	It's in fact easier to align these sequences
00:07:38.20	at the nucleotide level
00:07:40.12	than it is to align them at the amino acid,
00:07:43.03	where my name can change to a popular car model
00:07:45.11	very quickly,
00:07:46.29	because every mutation
00:07:48.29	has a high likelihood of altering the amino acid
00:07:51.03	being encoded,
00:07:52.21	and this is exactly the signature we see
00:07:54.14	when you have an interface
00:07:56.19	that is precisely at the interface
00:07:58.08	between a host and a virus conflict,
00:07:59.29	and that's because every single one of
00:08:02.05	these amino acid mutations
00:08:04.19	is potentially beneficial
00:08:06.01	and has been acted upon by natural selection
00:08:09.07	to increase their rate of fixation in the population,
00:08:12.15	hence the term diversifying selection.
00:08:15.11	In contrast to purifying selection,
00:08:17.08	natural selection is increasing
00:08:19.10	the amino acid diversity
00:08:21.07	of these protein-coding genes.
00:08:22.27	As a result, what we have, again,
00:08:24.22	is an apparent rate of replacement changes,
00:08:26.24	kA or dN,
00:08:28.13	which is increased
00:08:30.12	over the apparent rate of synonymous changes.
00:08:32.17	Once again, this is not a bias
00:08:34.25	that is introduced by mutation.
00:08:36.02	This is simply a different selective sieve
00:08:38.08	that is acted upon by natural selection.
00:08:41.15	This term diversifying selection
00:08:43.22	is also referred to as positive selection
00:08:45.19	or adaptive evolution.
00:08:47.03	I'll use these terms interchangeably,
00:08:48.29	and they're only different in the context of the tempo
00:08:51.02	with which these changes happen.
00:08:53.14	Now, if you were to take these characteristics
00:08:55.25	of replacement rates and synonymous rates
00:08:57.29	and calculate them for all genes
00:08:59.29	that we can compare between three sets of species,
00:09:02.25	our own species genome,
00:09:04.26	the rhesus macaque,
00:09:06.18	or the chimpanzee genome,
00:09:08.16	what we have is this very nice histogram
00:09:11.02	which really reflects the selective constraints
00:09:13.20	that have acted on all the protein-coding genes
00:09:16.10	within our genome.
00:09:18.01	What you'll see is there's a large number of genes
00:09:20.24	in the left-hand side of this histogram,
00:09:23.01	which means for the bulk of the genes in the human genome,
00:09:25.13	purifying selection,
00:09:27.04	or a dearth of replacement changes,
00:09:28.28	is really what is going on.
00:09:30.19	We are very interested in this sort of
00:09:32.26	small blip of genes right here
00:09:35.04	where you actually have a very small set of genes,
00:09:37.16	which even at the whole-gene level
00:09:39.09	have undergone much faster replacement changes,
00:09:41.10	almost breaking the speed limit of evolution, if you will,
00:09:44.16	to increase because of this diversity.
00:09:46.16	And when you take a really close look at
00:09:48.25	this category of genes,
00:09:50.13	immunity genes are really overrepresented,
00:09:52.06	as you might expect,
00:09:53.19	because these genes have been acted upon
00:09:55.14	repeatedly by natural selection.
00:09:57.21	So, we're going to consider
00:09:59.08	a very specialized case of an arms race
00:10:01.08	in today's seminar,
00:10:03.07	and this arms race ensues when a viral protein
00:10:05.14	begins to antagonize an antiviral protein,
00:10:08.21	and in this example the viral protein antagonism
00:10:11.08	is going to force the antiviral protein
00:10:13.11	to evolve to a state which this viral protein
00:10:16.27	can no longer defeat,
00:10:18.14	which will now force this viral protein
00:10:20.01	to evolve rapidly in order to restore its antagonism.
00:10:22.22	And this, in a microcosm,
00:10:24.27	is one step of this arms race,
00:10:26.29	where both the host protein
00:10:28.26	as well as the viral proteins have evolved
00:10:30.27	in these subsequent arms race interactions.
00:10:33.24	Now, what we're going to consider today
00:10:35.29	is a specialized example of this antagonism,
00:10:38.00	when the viral that is being used to antagonize
00:10:42.02	the host antiviral protein
00:10:43.28	is itself a host protein.
00:10:46.05	So, we are now basically considering
00:10:48.02	how would the host be able to distinguish
00:10:50.11	between an antagonism
00:10:52.07	that is caused by a viral mimic
00:10:54.12	versus its interaction with its own host proteins,
00:10:56.25	and that's the problem we'd like to address today,
00:10:59.04	which is, how do host genomes
00:11:01.12	confront and overcome, if they can,
00:11:04.11	the challenge of pathogen mimicry?
00:11:06.21	In today's seminar,
00:11:08.07	we're going to focus on a very specific example
00:11:10.07	of viral antagonism
00:11:11.18	that is mediated by mimicry,
00:11:13.13	and this example involves the host antiviral protein,
00:11:16.15	protein kinase R (PKR).
00:11:18.08	So, protein kinase R
00:11:20.01	is actually expressed when the organism senses
00:11:22.17	it's under some sort of viral attack
00:11:24.27	by virtue of an interferon detection pathway,
00:11:27.09	but it's actually produced as an inactive monomer,
00:11:29.16	which means it can no longer activate itself
00:11:31.24	as a kinase,
00:11:33.14	which is in the process of putting phosphate moieties
00:11:36.05	onto other proteins.
00:11:37.25	However, if this particular cell
00:11:39.27	happens to be infected by a virus,
00:11:41.23	that is detected by the fact that
00:11:45.01	there will now be double-stranded RNA in the cytoplasm,
00:11:47.05	which should not be case unless the cell
00:11:49.19	was under viral attack,
00:11:51.09	and what PKR will do is
00:11:53.08	it will use the signature of double-stranded RNA
00:11:55.06	to dimerize and activate itself as a kinase
00:11:57.22	whose primary substrate
00:11:59.21	is this protein eIF2α,
00:12:01.24	which stands for elongation initiation factor 2α,
00:12:05.04	which is a very important control step
00:12:07.22	to initiate protein production through the ribosome.
00:12:10.22	However, when PKR will phosphorylate eIF2α,
00:12:13.25	this essentially blocks protein production.
00:12:16.08	So, the cell's response to detecting itself
00:12:19.17	under viral attack is,
00:12:21.07	"I'm going to stop all protein production
00:12:23.05	so that I do not become a virus production factory."
00:12:26.04	This can be a very effective
00:12:27.29	and a very potent block to viral production,
00:12:30.17	and so what viruses have had to come up with
00:12:33.12	is several clever means by which
00:12:35.25	they can actually inhibit the PKR reaction.
00:12:37.21	Some viruses, for instance, inhibit the dimerization of PKR.
00:12:40.27	Some viruses will actually hide away
00:12:43.08	all the double-stranded RNA they produce,
00:12:45.02	whereas some viruses actually
00:12:47.08	will encode a phosphatase that specifically
00:12:49.23	takes out the phosphate residue
00:12:51.28	that is put on by PRK,
00:12:53.15	and perhaps the cleverest model
00:12:55.20	comes from the hepatitis C viruses
00:12:57.25	that actually allow PKR to block protein production,
00:13:00.15	to essentially block all manner of host protein production,
00:13:03.15	but will now nonetheless
00:13:05.20	carry on their own protein production
00:13:07.15	in an eIF2α-independent fashion,
00:13:09.20	really highlighting the clever inventions
00:13:12.00	that are really forced upon
00:13:13.27	by virtue of these Darwinian arms races.
00:13:15.25	In todays' seminar, we're actually going to focus on
00:13:18.03	only one of these antagonists,
00:13:20.02	which is encoded by the poxvirus class proteins,
00:13:23.10	which include smallpox and vaccinia virus,
00:13:26.11	and this is a protein called K3L,
00:13:28.14	which acts as a competitive and non-competitive inhibitor,
00:13:31.08	essentially breaking the interaction
00:13:33.12	between PKR and eIF2α,
00:13:36.12	which basically allows the virus to restore protein production
00:13:40.02	and go on with its life cycle.
00:13:42.05	So, we actually started this by looking at what this arms race,
00:13:45.12	with the potential for multiple antagonists from viruses,
00:13:48.15	has done to PKR evolution.
00:13:50.19	And so, to do this,
00:13:52.15	we actually sequenced the PKR gene
00:13:54.10	from a panel of primates,
00:13:56.03	which includes homonoids,
00:13:57.22	including humans, great apes, as well as gibbons,
00:13:59.28	old world monkeys,
00:14:01.20	which includes things like rhesus macaques,
00:14:03.14	and new world monkeys,
00:14:05.05	which are primates that populate Central and South America.
00:14:07.24	And when we do the sequence,
00:14:09.28	we can actually reconstruct the evolutionally history
00:14:12.25	of essentially every step and every codon
00:14:15.08	across the PKR phylogeny,
00:14:17.04	and so what we see in these numbers here
00:14:19.21	are those dN/dS or kA/kS signatures
00:14:22.24	that I talked about.
00:14:24.19	When when we have very low numbers
00:14:26.22	like this number 0.2, here,
00:14:28.19	that's an indication of not very much happening
00:14:30.19	at the protein evolution level.
00:14:32.11	In contrast, we have some amazing examples
00:14:34.16	like this lineage in old world monkeys,
00:14:36.18	where we actually have 22 replacement changes
00:14:39.17	without a single synonymous change happening.
00:14:42.08	That's a really profound signal
00:14:44.06	of multiple staccato replacement changes
00:14:46.17	occurring in the course of evolution,
00:14:48.14	in a very, very short time frame,
00:14:50.09	really highlighting the very intense
00:14:52.16	and very episodic evolutionary pressures
00:14:55.11	that have acted on PKR
00:14:57.17	over the course of the last 35 million years
00:14:59.22	of primate evolution.
00:15:01.13	If you were now to sort of turn this around
00:15:03.16	and squish it codon by codon,
00:15:05.20	we essentially get a landscape
00:15:07.24	of how PKR has been influenced
00:15:09.28	by positive selection.
00:15:11.11	All of these tick marks that I've shown
00:15:13.07	above the PKR protein
00:15:15.09	are individual codons that have recurrently evolved
00:15:17.15	under positive selection,
00:15:19.06	and you can see that, in the case of PKR,
00:15:21.25	these are really spread throughout the entire protein motif of PKR,
00:15:25.01	including in the N-terminal domain,
00:15:27.19	in this linker domain or the spacer region,
00:15:29.20	as well as in the kinase domain,
00:15:32.00	which actually carries out the very important step
00:15:34.09	of eIF2α phosphorylation.
00:15:37.26	And the reason we think that there's been such
00:15:40.07	dramatic and such widespread positive selection
00:15:42.15	is because multiple viruses
00:15:44.10	actually antagonize completely different domains of PKR
00:15:47.12	in order to mediate their antagonism of PKR.
00:15:50.12	So, what we're gonna focus on today
00:15:52.19	is just one of these antagonists,
00:15:54.05	which is, again, these poxviral antagonist K3L,
00:15:56.27	that actually specifically antagonize
00:15:59.12	the kinase domain of PKR.
00:16:01.25	So, the reason I've been spending so much time
00:16:03.28	discussing K3L with you
00:16:05.27	is because K3L is a special antagonist.
00:16:08.28	It actually is an evolutionary-derived mimic
00:16:12.24	which used to be eIF2α,
00:16:14.29	which means that at some point in poxviral evolution,
00:16:18.08	poxvirus actually stole eIF2α from a mammalian host,
00:16:22.13	and have whittled it away to become this perfect mimic,
00:16:25.29	in order to break PKR's interaction with the eIF2α substrate.
00:16:30.00	Now, what is really remarkable about this interaction
00:16:32.12	is that it not just happened once
00:16:34.25	in mammals,
00:16:36.08	but it's happened on three separate occasions
00:16:38.27	with three completely independent lineages
00:16:40.24	of double-stranded DNA viruses,
00:16:42.20	each of them acquiring a K3L-like mimic
00:16:44.20	from their own version of eIF2α.
00:16:47.16	So, this really highlights the very, very successful
00:16:50.11	strategy of mimicry that is encoded by pathogens,
00:16:54.09	and really, from an evolutionary standpoint,
00:16:56.16	the strategy of mimicry
00:16:58.19	and overcoming mimicry
00:17:00.05	is a debate that's really been going on
00:17:02.10	for a very, very long time,
00:17:04.02	going back all the way to Henry Walter Bates,
00:17:06.04	who really first detected evidence for mimicry
00:17:09.13	in these butterflies in the Amazon,
00:17:11.24	where we have model butterflies
00:17:14.10	that are basically poisonous,
00:17:16.13	and so they're avoided by predators
00:17:18.22	who can use their coloration patterns
00:17:20.21	as an indication to...
00:17:22.17	as a warning signal to stay away from them,
00:17:24.16	and mimic butterflies
00:17:26.10	that actually don't encode a poison at all,
00:17:28.05	but take advantage of this coloration pattern,
00:17:30.11	and mimic the coloration pattern,
00:17:32.08	to take all the advantages of avoidance from predators,
00:17:35.09	without actually having to encode
00:17:37.07	any of the toxins that are required.
00:17:39.19	Now, this is actually quite a really great strategy
00:17:41.27	for the mimic.
00:17:43.06	It's not so good for the model,
00:17:45.03	because as the mimics start increasing in frequency
00:17:47.03	and the predators start eating more and more butterflies
00:17:48.29	that look like this, but are quite tasty,
00:17:51.12	they will lose their avoidance of the predators,
00:17:53.16	which means that the success of the mimic
00:17:56.08	is directly, inversely correlated
00:17:58.22	with the success of the model.
00:18:01.01	And very much the same thing might be going on
00:18:03.00	at a molecular level, we feel,
00:18:04.27	where eIF2α is acting as a model protein,
00:18:07.17	which is being mimicked by this poxviral mimic K3L
00:18:10.27	in order to defeat
00:18:13.22	the PKR-eIF2α immunity response.
00:18:16.29	So, if you were to sort of rephrase
00:18:18.27	the challenge of mimicry,
00:18:20.22	it is that the PKR kinase domain
00:18:22.29	needs to bind and maintain its interaction with eIF2α,
00:18:26.22	while avoiding its interaction with the mimic,
00:18:29.12	which really is evolutionarily being selected
00:18:31.19	to look like eIF2α
00:18:33.11	from the viral perspective,
00:18:34.28	and you can see in this crystal structure
00:18:37.03	that the structures of the PKR interaction domain
00:18:39.27	between K3L and eIF2α
00:18:42.01	are almost completely super-alignable,
00:18:43.23	so how is it that PKR is able to acquire
00:18:46.23	the ability to discriminate between these two?
00:18:49.12	As I've already told you,
00:18:51.08	one of the strategies that PKR is using is very rapid evolution,
00:18:54.26	it's got that at its disposal,
00:18:56.17	and this is just a sliding window plot of dN/dS
00:18:59.12	over the entire protein of PKR,
00:19:01.15	and what you see here is that
00:19:03.14	there is not even a single domain
00:19:05.20	where the dN/dS signature drops below one,
00:19:07.26	which means pretty much every domain of PKR
00:19:10.06	is evolving under positive selection
00:19:12.06	in this comparison between human and rhesus PKR.
00:19:15.08	It's really remarkable how profound the signal is,
00:19:18.14	because when we compare PKR
00:19:20.06	to its closest relative kinase, PERK,
00:19:22.14	which is not involved in antiviral immunity,
00:19:25.04	you can see that the signature is completely profound
00:19:27.23	of purifying selection,
00:19:29.17	and not of positive selection.
00:19:31.15	And this actually gets even more interesting
00:19:33.06	when you look at eIF2α,
00:19:34.22	which is the substrate for PKR,
00:19:36.25	because eIF2α is so important for translation
00:19:41.02	that it has not tolerated any amino acid changes
00:19:43.07	over the course of evolution.
00:19:44.28	You might be actually wondering where the red line went,
00:19:47.12	and actually the red line is exactly on zero,
00:19:50.05	because no amino acid changes have occurred
00:19:52.12	over the course of primate evolution.
00:19:54.05	So, in a way, you can view this
00:19:56.19	as a very high-stakes game of rock-paper-scissors,
00:19:59.28	except eIF2α is always playing rock,
00:20:03.14	and so it would seem that mimic
00:20:05.25	would have a very, very simple game,
00:20:08.04	which is to mimic an unchanging protein
00:20:10.05	and stay there.
00:20:12.06	We wondered whether that was actually the case,
00:20:13.29	because, first of all, we've actually survived poxviruses,
00:20:16.25	and secondly, this suggested that
00:20:19.23	PKR might have some adaptive routes
00:20:21.18	in order to escape mimicry.
00:20:23.06	Furthermore, if it was the case that K3L
00:20:25.05	was simply evolving to an optimal mimic status,
00:20:28.04	we might actually presume
00:20:30.09	that K3L should now be under purifying selection,
00:20:32.14	having optimized for this role in mimicry.
00:20:35.06	Instead, what we actually find
00:20:36.21	when we compare K3L
00:20:39.00	from a panel of poxviruses,
00:20:40.20	is that, very much like PKR
00:20:42.27	shown here on the host side,
00:20:44.18	which is very rapidly evolving,
00:20:46.08	in contrast to eIF2α which is not,
00:20:48.20	K3L happens to be one of the most [quickly]
00:20:52.09	evolving proteins the poxviral genome.
00:20:54.08	So, this is truly an arms race between
00:20:56.19	K3L and PKR.
00:20:58.03	What makes this arms race really interesting
00:21:00.01	is that they're both really evolving
00:21:02.03	to get the attention of eIF2α,
00:21:03.28	which is not changing at all,
00:21:05.29	and so that's what makes the problem of mimicry
00:21:07.29	really interesting from an evolutionary standpoint.
00:21:11.12	So, we wanted to actually have a system
00:21:13.15	in which we could simply assay
00:21:15.20	for the effects of mutations and evolutionary adaptations
00:21:18.17	in a very facile assay,
00:21:20.11	and we actually took advantage of an assay
00:21:22.12	developed first by Tom Dever and Alan Hinnebusch,
00:21:25.03	who recognized that eIF2α
00:21:27.22	is so slow to evolve
00:21:29.15	that if you actually put human PKR in yeast
00:21:32.06	it will actually bind and phosphorylate yeast eIF2α
00:21:34.25	to cause a growth arrest.
00:21:36.26	Now, in this context,
00:21:38.05	if we now also introduce K3L,
00:21:40.11	we have the situation where K3L
00:21:42.18	can give you a readout of whether it's able
00:21:44.20	to defeat PKR or not,
00:21:46.19	based on whether it can rescue the growth inhibition
00:21:49.08	mediated by the PKR expression.
00:21:51.23	So, Nels Elde,
00:21:53.05	who was a postdoc in the lab,
00:21:54.29	actually took this panel of PKR genes
00:21:57.09	from a panel of different primates...
00:21:59.12	homonoids, old world monkeys,
00:22:01.02	and new world monkeys...
00:22:02.20	and he actually just put it into yeast cells,
00:22:04.24	but he put it in a form which could not be turned on.
00:22:07.14	So, when these yeast grow on glucose,
00:22:09.25	because the PKR gene
00:22:11.19	is put on a galactose promoter,
00:22:13.12	it's silenced,
00:22:15.04	and what you can see is that all of these yeast
00:22:17.01	grow perfectly fine.
00:22:18.16	You can see that, even in this serial dilution across,
00:22:20.17	you basically have no growth inhibition.
00:22:22.27	However, as soon as you turn on PKR
00:22:25.17	by putting all of these yeasts onto galactose plates,
00:22:27.27	you can see no yeast growing here,
00:22:30.27	which means all of these PKR alleles
00:22:32.29	have conserved the property of binding
00:22:35.27	and phosphorylating yeast eIF2α,
00:22:37.26	which is remarkable considering the very large degree
00:22:40.20	of evolutionary divergence that we have seen here.
00:22:43.20	Now, I can tell you that this is all because of eIF2α phosphorylation,
00:22:46.17	because in this yeast,
00:22:48.17	if I engineer a mutation in the phosphorylation site
00:22:51.05	all of the growth inhibition goes away,
00:22:53.06	and that's shown in these two panels here.
00:22:55.04	So now, the really interesting question
00:22:57.00	happens when you introduce the viral antagonist.
00:22:59.28	So, what would you predict
00:23:01.24	would happen here if you now introduced
00:23:04.07	the K3L protein from a poxvirus?
00:23:06.13	In this case, we used the vaccinia virus,
00:23:08.28	and what we find is a completely binary response.
00:23:11.27	In some situations, like in the gibbon PKR case,
00:23:15.04	the introduction of the vaccinia K3L
00:23:17.18	completely reverses the growth inhibition that is going on,
00:23:20.23	whereas in the human case,
00:23:23.01	even the presence of K3L,
00:23:25.00	at roughly equal levels of expression,
00:23:27.06	did not overcome the growth inhibition.
00:23:28.26	So, this is exactly like that cartoon example
00:23:31.08	of those two states between hosts and viruses,
00:23:33.27	and what we have in an evolutionary snapshot
00:23:37.01	of both of those states, w
00:23:38.20	here either the host is winning,
00:23:40.18	in which case the growth inhibition goes on,
00:23:42.24	or the virus in winning,
00:23:44.25	in which case the growth inhibition is completely reversed.
00:23:47.10	Now, these are all assays being done in yeast,
00:23:49.17	but we've actually done exactly the same types of assays
00:23:51.29	in vaccinia cells,
00:23:53.27	where we've actually taken either human cells
00:23:56.01	or gibbon cells
00:23:57.16	or orangutan cells
00:23:59.10	and infected them with either a wild type,
00:24:01.10	fully functional vaccinia,
00:24:03.06	or something in which the K3L
00:24:05.26	specifically had been deleted,
00:24:07.21	and what you'll notice is that,
00:24:09.12	in human cells and orangutan cells,
00:24:11.04	it actually doesn't matter
00:24:13.00	whether you've deleted the K3L gene or not,
00:24:14.27	and that's because these species
00:24:17.02	actually have a PKR that's resistant
00:24:19.00	to the K3L antagonism,
00:24:20.25	whereas in the gibbon case,
00:24:22.15	when you delete K3L, you have this 10-fold drop in fitness,
00:24:25.08	which basically is an indication
00:24:27.15	that K3L from vaccinia
00:24:29.21	is acting as a species-specific antagonist
00:24:32.00	of the PKR response.
00:24:34.07	So, we wondered whether
00:24:36.04	we could actually gain better molecular insight
00:24:38.11	into how is it that K3L is able to adopt these multiple states
00:24:41.27	by looking at the co-crystal structure
00:24:44.09	of PKR's kinase domain and the eIF2α substrate,
00:24:47.25	which was first actually established
00:24:50.04	by Arvin Dar and Frank Sicheri's lab,
00:24:52.21	and in this co-crystal structure,
00:24:54.21	one of the most important motifs for this interaction
00:24:58.00	happens to be this α-helix that I've shown here
00:25:01.06	as the g-helix.
00:25:02.23	This is effectively like a bird perch
00:25:04.23	onto which PKR will sit...
00:25:06.21	the bird perch on PKR on PRK
00:25:08.15	onto which eIF2α will sit down.
00:25:10.13	If you take a closer look at the α-helix,
00:25:12.02	shown here,
00:25:13.22	there are three residues in particular
00:25:15.15	that are making direct contacts with the backbone of eIF2α.
00:25:18.29	Now, I'll remind you that eIF2α
00:25:20.18	is not changing at all, in fact,
00:25:22.17	functionally equivalent between human and yeast,
00:25:25.19	so you would predict actually
00:25:28.04	that these three residues would be completely frozen
00:25:30.05	in evolution,
00:25:31.20	by virtue of the fact that they have to interact
00:25:33.25	with something that is completely frozen itself,
00:25:36.05	but instead what we find
00:25:38.05	is that these three residues represent some
00:25:40.12	of the fastest evolving residues
00:25:42.19	in PKR's kinase domain.
00:25:44.10	So, the very...
00:25:46.00	sort of combination lock
00:25:48.02	that is responsible for binding eIF2α
00:25:50.00	is the lock that is very rapidly changing.
00:25:52.05	So, somehow all of these combinations of residues
00:25:55.04	at the αg-helix
00:25:57.13	have preserved the property of binding eIF2α,
00:25:59.25	and yet are basically under very strong evolution.
00:26:02.07	So, we wondered whether this is in fact
00:26:05.08	a signature of the fact that this is an interface
00:26:08.12	that has been constantly challenged by viral mimicry,
00:26:11.14	and so, to test that,
00:26:12.28	we again returned to our yeast assay.
00:26:14.25	We have human PKR
00:26:16.17	that is able to continue growth inhibition
00:26:19.07	even in the presence of K3L,
00:26:21.07	gibbon PKR
00:26:22.27	that is completely reversed by the presence of K3L,
00:26:25.11	and now, in the gibbon backbone,
00:26:27.07	if we add single amino acid changes
00:26:29.27	from human into gibbon,
00:26:31.26	what we find is that we can completely reverse
00:26:34.20	the susceptibility phenotype
00:26:36.11	into the resistant phenotype.
00:26:38.06	So, this really highlights two things.
00:26:40.09	First of all,
00:26:42.10	the interface between PKR and eIF2α
00:26:44.20	is really a hotspot for positive selection,
00:26:47.07	and individual residue changes,
00:26:49.13	these single steps in the arms race
00:26:52.15	between PKR and K3L,
00:26:54.06	result in a complete reversal
00:26:56.10	from susceptibility to resistance.
00:26:58.11	Now, this also actually revealed to us
00:27:00.17	something else that we had missed earlier,
00:27:02.16	which is, even though the orangutan PKR
00:27:05.17	is completely resistant to K3L mimicry,
00:27:07.27	the orangutan g-helix
00:27:10.20	is not resistant to mimicry,
00:27:13.04	which means some other component
00:27:15.27	of the PKR backbone in orangutan
00:27:17.25	is actually necessary for mimicry,
00:27:19.25	immediately suggesting that there was another solution
00:27:22.09	to overcoming mimicry
00:27:24.11	that was evident in orangutan,
00:27:26.05	and we actually mapped that residue again
00:27:28.09	to a single residue in this helix αE,
00:27:30.20	very far away from this helix αG
00:27:33.09	which I've been telling you about today.
00:27:35.09	And so, very much like we saw
00:27:38.16	in the human/gibbon αG case,
00:27:41.09	individual residue changes between gibbon and orangutan
00:27:44.21	have the ability to switch from susceptible to resistant
00:27:48.09	and resistant to susceptible.
00:27:51.15	So, again, really highlighting
00:27:53.12	the very significant power of even individual mutations
00:27:56.04	in individual residues.
00:27:57.27	In the human case,
00:27:59.25	what we also sort of observed was...
00:28:02.09	this particular residue is very interesting,
00:28:04.14	because it's actually toggled
00:28:06.11	between leucine and phenylalanine
00:28:08.12	throughout mammalian evolution,
00:28:10.09	really reflecting the fact that there's probably
00:28:12.05	a high degree of evolutionary constraint
00:28:14.09	acting on this protein,
00:28:15.29	and yet it's toggling so as to keep one step ahead
00:28:18.16	of this mimic interface.
00:28:20.12	The human PKR actually has a very good helix αE residue,
00:28:23.17	as well as a helix αG residue,
00:28:25.28	especially against vaccinia,
00:28:27.25	and we actually have to mutate all three of these residues
00:28:29.27	in order to convert the resistant human PKR
00:28:32.10	into a susceptible version.
00:28:34.20	So, what have we learned from our examples
00:28:37.24	of PKR overcoming the mimicry of K3L?
00:28:40.17	The first really important lesson we learned
00:28:43.06	is that multiple domains of PKR
00:28:45.08	need to be under rapid evolution
00:28:47.09	in order to overcome mimicry.
00:28:48.27	Again, as I pointed out,
00:28:50.18	this is a rock-paper-scissors game,
00:28:52.11	and if only one particular domain
00:28:54.06	was under rapid evolution,
00:28:55.26	K3L would have a much easier task
00:28:57.17	antagonizing and mimicking this interface.
00:28:59.19	The fact that multiple residues
00:29:01.11	in multiple domains
00:29:03.06	are actually rapidly evolving
00:29:04.26	allows these domains to really take turns
00:29:06.28	in antagonizing...
00:29:08.22	overcoming the antagonism of K3L.
00:29:10.14	And, what appears to be the first evolutionary step
00:29:12.26	when PKR encounters this mimicry
00:29:15.10	is actually a negative affinity,
00:29:17.16	where PKR loses affinity,
00:29:19.11	not just to eIF2α,
00:29:21.12	but also to K3L,
00:29:23.11	and then it restores its affinity
00:29:25.08	by interactions in another domain.
00:29:28.08	So, this also implies that there
00:29:30.24	must be extraordinary flexibility for PKR
00:29:32.26	to basically recognize a substrate
00:29:34.23	that really has undergone no changes
00:29:36.24	over the course of evolution.
00:29:38.18	So, just as an example of this flexibility,
00:29:41.03	here again we've the orangutan G-helix
00:29:43.15	in a gibbon backbone,
00:29:45.24	and you can see this is actually susceptible to mimicry,
00:29:49.00	but you can see here, now,
00:29:50.27	because of the growth of this yeast colony,
00:29:52.25	this is telling us that this particular chimeric version of PKR
00:29:56.03	is also not doing a good job of recognizing its substrate,
00:29:59.24	and yet the orangutan backbone
00:30:02.01	has completely restored the binding to eIF2α
00:30:04.23	as well as overcome mimicry,
00:30:06.15	which means something else
00:30:08.15	in the orangutan backbone was sufficient
00:30:10.15	to restore the weakness of this G-helix
00:30:12.28	over the course of these evolutionary arms races.
00:30:15.16	So, this is great,
00:30:17.06	we've learned rules by which PKR
00:30:19.04	might actually overcome mimicry,
00:30:20.24	but this is also sort of a sobering reminder
00:30:22.21	that this overcoming of mimicry
00:30:24.29
00:30:26.24	comes at a cost. So, if you were to look at the αG helix from PKR
00:30:29.05	and three other kinases,
00:30:31.08	whose primary substrate is eIF2α,
00:30:33.11	we'll notice that PKR
00:30:35.18	is the only kinase where we see this dramatic rapid evolution.
00:30:37.25	We don't see if for these three other kinases,
00:30:40.13	which means these kinases
00:30:42.11	have had the evolutionary luxury
00:30:44.20	to optimize to an optimal binding of eIF2α
00:30:48.15	and essentially stay there,
00:30:50.20	preserve their optimal binding
00:30:52.22	by virtue of purifying selection.
00:30:54.15	PKR no longer has that luxury,
00:30:56.18	because as it gets more and more optimal
00:30:58.20	for eIF2α recognition,
00:31:00.23	it gets more and more susceptible
00:31:02.16	for K3L antagonizing it as a mimic.
00:31:05.28	So instead, PKR's evolutionary solution
00:31:08.17	has been to back away from this optimal mimicry
00:31:11.02	in order to gain more of this adaptive landscape
00:31:13.14	that keeps it one step ahead
00:31:15.28	of the virus in terms of these arms races.
00:31:18.02	This a very important sort of consideration
00:31:20.29	because it's not just antiviral genes that face mimicry.
00:31:24.18	This is a slide in which we show that
00:31:28.16	absolutely essential processes in the cell,
00:31:30.24	the cytoskeleton,
00:31:32.14	membrane trafficking,
00:31:34.01	even the cell cycle and apoptosis,
00:31:36.06	all absolutely fundamental housekeeping processes in the cell,
00:31:38.22	are all hijacked by some form of pathogen mimicry.
00:31:42.04	It's worth considering that...
00:31:44.13	what are the evolutionary pressures that have been placed on all of these processes,
00:31:47.09	as they basically tried to survive the mimicry imposed by the pathogen?
00:31:50.20	And, even though they're acquired
00:31:53.22	really great adaptations to overcome this mimicry,
00:31:56.05	some of these alleles might actually be compromised
00:31:59.05	in terms of their housekeeping function
00:32:01.28	- for the function that they were originally intended for.
00:32:03.29	And so, it's not only the fact that the mimic
00:32:06.15	is actually imposing evolutionary adaptation,
00:32:08.28	it might be pushing some of these genes away
00:32:11.20	from their optimal state for cellular function.
00:32:15.00	So, with that I'm going to end this part of the talk.
00:32:17.21	I'd like to really acknowledge Nels Elde,
00:32:20.03	who was a former postdoc in the lab
00:32:22.04	who has his own lab at the University of Utah now,
00:32:24.11	and two very talented technicians,
00:32:25.27	Emily Baker and Michael Eickbush.
00:32:28.03	And this work was done in collaboration
00:32:30.06	with my colleague Adam Geballe,
00:32:32.07	and Stephanie Child in his lab really did all of the viral work
00:32:35.08	that I've discussed.
00:32:36.27	I'd really like to thank our funding sources,
00:32:38.22	and I thank you for your attention.