Dana Carroll: Background on Genome Editing

Dana Carroll: Background on Genome Editing

(bright instrumental music) – It’s great to have you all here for the third annual IGI CRISPR Workshop. My job this morning is
to talk to you about some of the background of genome editing, some of the earlier platforms, and some of the aspects of
editing that are pretty general. And, the advent of genome
editing demonstrates, validates a theory of Sydney Brenner’s, which is expressed as,
“Progress in science “depends on new techniques,
new discoveries, “and new ideas, probably in that order.” And this technology, as I’m sure you know, you’re here because of it, has really opened the door to
a lot of avenues of research that weren’t accessible or
as accessible previously. Well, we aren’t the first genome editors. Genome editing has been
around for a very long time. Darwin recognized variation amongst closely-related organisms that reflected adaptations
to their environments, and these finches feed
on different nutrients and different sources of food, and their beaks and head sizes are adapted to their
different sources of food. And this is all based on natural selection amongst natural genome variants. So, that’s a form of
genome editing for sure. Sometimes, humans intervene
and accelerate the rate of editing, and the
development of modern maize from its ancient precursor
teosinte is an example of that. Humans selected for teosinte
plants that had larger ears and larger kernels from
this plant that had rather small ears and not
such succulent kernels to make a very important food
source for people and animals. All of genetics, for many years, was based on natural variants, natural mutations that
arose, some way or another. But in the first half of the 20th century, H.J. Muller and Charlotte Auerbach showed that you could accelerate
the rate of mutogenesis using radiation or chemicals. And both of these groups,
Muller actually solo, but Auerbach in a group, were working with Drosophila and showed that you could generate
new Drosophila mutants, visible mutants of various kinds, by these treatments of the organism. So, those mutations were still random. Even though you could
make ’em arise faster, they were still random mutations. And then it was in the 1980s,
late 70s, into the 1980s, that people developed
gene targeting technology, particularly in yeast and in mice. And this is my Utah
colleague, Mario Capecchi, who developed this procedure
for mouse embryonic stem cells where you make a piece
of DNA in the laboratory that represents the
modification of the target gene that you have in mind, and
you hope what’ll happen is that this piece of DNA, when it’s introduced into the cells, will undergo homologous recombination with the endogenous gene it corresponds to and introduce whatever sequence
change you had intended. The problem was that
many integration events, the vast majority of integration events, went in through the ends of this donor DNA and didn’t go into the right gene at all. So what Capecchi and his colleagues did was to impose a double selection that put a positive selection marker, a Neomycin resistance gene,
between the homologies that were going to be used for integration and a negative selective
marker, a Thymidine kinase gene from Herpes simplex virus, on the outside. So if the donor DNA goes
into the right place, it loses the TK gene but retains Neo. If it goes into the wrong place, it retains both. And integration of the HSVtk gene confers sensitivity to Nucleoside analogs. And so double selection
for a Neomycin analog and this Nucleoside analog allows people to recover the integration
events they want. And for decades, this was
used in mice very effectively. There must be thousands,
maybe tends of thousands, of knock-in and knock-out mice that have been developed
using this sort of technology. The reason this wasn’t
adopted for other organisms and the sort of underlying problem with it was that the raw efficiency of the homologous recombination
events was extremely low. Often, just one in every
million cells or less, and cells from many other organisms weren’t even that effective. What we knew at the time was that, if there’s a double-strand break in the target, that that
will stimulate the process of homologous recombination
and kick up the frequency several orders of magnitude. And these numbers aren’t
meant to be exact. They’re just representative. So if you could, if you could
make a break in the target at the same time you
introduced the donor DNA, then you wouldn’t need all
this highfalutin selection to recover the products
that you were interested in. So the problem became, how do you make targeted double-strand breaks in DNA, and how do you develop
reagents that can make a break at any target you choose,
allowing you to change your mind from day to day about which
target you’re going after? And without belaboring
it, I’m sure you know that there are three platforms that fulfill these criteria these days: the zinc finger nucleases, and TALENs, and the CRISPR-Cas system. And all of these platforms
are capable of making very efficient and very
specific double-strand breaks in genomic targets in
essentially any organism. Let me just go quickly through where these platforms came from. The ZFNs, which were the
first of those three, came from studies on the
restriction enzyme FokI, and Chandrasegaran and his colleagues figured out that the
recognition and cleavage domains of FokI were physically separable and they said, well, if we can get this non-specific nuclease domain from FokI away from its
natural recognition domain, we can put other recognition domains on. And amongst the ones they chose
were a set of zinc fingers. Zinc fingers naturally occur in eukaryotic sequence-specific DNA binding
transcription factors. Each module of these transcription factors is a finger that recognizes principally three base pairs of DNA,
and successive fingers recognize successive
triplets in the DNA target. So you could make the FokI cleavage domain cut at other places in DNA by assembling different combinations
of zinc fingers on them. And so we got involved in this
right after that first paper was published and showed
that this nucleus domain has to dimerize in order to cut DNA, and because the dimer
interface is weak, it takes two sets of zinc fingers to
bring these cleavage domains close enough together so
that they will cut DNA. Each finger recognizes three
base pairs principally. You can change the specificity by changing the identity of these fingers, and we demonstrated that
we could get cleavage and recombination with
a chromatinized target, and also, I think importantly, since this cleavage
domain came from bacteria, we showed that we could
actually get cleavage mutagenesis at the cut site and targeted gene replacement in whole Drosophila fruit flies. So, those are the zinc finger nucleases. The TALENs arose from studies
of bacterial plant pathogens of the genus Xanthomonas, and these bacteria produce
proteins that they secrete into the plant host cell
that go into the host cell, bind upstream of host gene, and regulate its transcription in a way that promotes the bacterial infection. And when enough of these combinations of transcription factor, which is called a transcription
activator-like effector, that’s where TALE comes from, and their targets had been studied, it emerged that there’s
a code of recognition. One module, one 34-amino
acid module in the protein, recognizes one base pair of DNA depending on what amino acids
are in these two positions in each repeat. And so there’s this
robust recognition code between these TALE
modules and DNA sequences, and it didn’t take a genius
to link some of these TALE domains to the FokI cleavage domain and generate nucleases where the, you now had one module per base pair, and they’re much easier to assemble, and the frequency of success was a lot higher than with the ZFNs. Jennifer’s gonna talk about
CRISPR more in the next talk, but just to remind you
that the CRISPR system comes from a natural
bacterial immune system where small sequence representations of viral genomes and other
passible invading DNAs are in an array in the genome. They get copied into RNA. The RNA is processed and
bound to a specific protein, which then, guided by
this viral sequence here, can cleave and inactivate DNA from a subsequent viral infection. And, of course, we have
adopted this latter part of the CRISPR function to generate these CRISPR/Cas nucleases. In most of the common CRISPR systems that are being used for
editing, the protein is Cas9. The guide sequence and
some Cas9 binding modules have been combined into
a single guide RNA, and recognition occurs by base pairing between the RNA and the DNA target, but only if the sequence
is next to a protospacer adjacent motif, a PAM, which, in the case of the Streptococcus pyogenes, Cas9 is just two g’s in the top strand separated from the RNA
recognition sequence by one base pair. So, knowing how this works
in bacteria allowed us to make an intentional DNA
cleavage system out of it. So why, why did CRISPR
revolutionize genome editing? Well, it’s for these reasons. First of all, it’s really simple to design an RNA for a new target, a guide RNA. You could assign this to a
junior high school student, ’cause all you need to know are the Watson-Crick base pairing rules. There’s one constant
protein, the Cas9 protein, that’s used every time
you use this system. You don’t have to redesign
the protein at all. Once you’ve learned how
to deliver that protein, you can do it over and over again. It’s very easy to produce
the RNA and the Cas9. You’re going to produce RNA in the laboratory this afternoon. You can do it at low
cost with basic skills. You’ll also discover that. You can multiplex the
system very conveniently, because you can make
lots of different RNAs to go with a single protein, and there are many variants of this system now available to give you
additional capabilities. And just before I leave this, I just wanna emphasize the fact that all of these platforms
arose from unexpected sources, by which I mean nobody set
out to develop a genome editing tool when the initial
research was being done on eukaryotic transcription factors, this bacterial restriction enzyme, these plant transcription factors that contribute to the pathogenicity, and people began to try
to understand what these odd sequences and bacterial
genomes were all about. So it just came from
investigating how the world works that we got these powerful tools, and if we’re going to
get other powerful tools, either for editing purposes
or for other purposes, they’re likely to come from nature, so it’s important to continue figuring out how the world works. Okay. So, we’ve got ZFNs, TALENs, and CRISPRs. What do they do? The only thing they do is
make a double-strand break in the target DNA. Even the marvelous CRISPR
system doesn’t actually do anything after making the break. It has to get out of the way
so the break can be processed, but everything that happens
after the break is made depends on cellular DNA repair activities. And one of the types of repair that we depend on is
homology-dependent repair or homologous recombination, and this is like the Capecchi system that I was talking
about, where a donor DNA that’s provided by the
experimenter can serve as a template for repair
of a double-strand break that’s been induced by
one of these nucleases, leading to sequence replacement. The other process that cells use to repair a double-strand break is called
non-homologous end joining, and there are varieties of
non-homologous end joining. But end joining is sort of
a panic response by cells. They’d rather jam ends back
together with the possible hazard of making a mistake at the junction than leave a fragment of
chromosome that would be lost. So although end joining can be precise, it occasionally makes mistakes, and those are right at the target, so they’re targeted mutations. Somebody whistling at me,
or was that just a phone? So let’s talk for a
minute about what happens in non-homologous end joining. If you, like you’re gonna do this week, you send in nuclease, it’s directed to a target
by its recognition domain and makes a break, and
then cells repair it. And when they repair it by end joining, the mistakes that are made are localized, small
insertions and deletions, and so this is just from an older paper from Keith Joung’s lab. They’re insertions that are pretty small, even as small as a single
base pair sometimes, but they can go up to a few hundred bases, and there are small insertions that occur. Occasionally larger insertions,
but not very frequently. In many situations, you
get a somewhat higher level of insertions than is represented here. One of the characteristics of
the deletions in particular is that some of them are mediated by what we call microhomologies. And the way you recognize a microhomology is that, I’ll use this as an example here, the sequence that’s at the
junction could be assigned either to the left side or the
right side of the junction, because in the target, that TCC was here, and also over here. And the way we think that
happens, almost certainly happens, is that if you make a break
between repeated sequences, even very short repeated sequences, after a little bit of resection
on one strand of the DNA, this… I guess the way it goes, this TGA, reading five prime to
three prime, right to left, can hybridize, can base pair with that TCA from the other strand on
the other side of the break. So, if you chew back one strand, now you have base pairing
possibilities between sequences that were on opposite sides of the break. And if this sets the
register for a new junction, you have a microhomology
mediated deletion event. And Jin-soo Kim’s lab went to some lengths to predict these. If you’re gonna make a
break at a particular site, you can look around that site and see whether there are these microhomologies. Here’s a five-base microhomology
that exists on both sides, and if it’s used to generate a deletion, then that deletion would
be a 20-base pair deletion. And the reason for thinking
about this ahead of time is if there is going to be sort of a dominant
microhomology-mediated deletion, you can assess, just by looking at it, whether it’s going to
be an in-frame deletion or an out-of-frame deletion. Now, 20 base pairs is
not a multiple of three, so if it’s in coding sequence, it will be an out-of-frame deletion, and they just went through and they looked at a bunch of different targets to see how often that might occur. This is just another, another example where you’ve got a gene you wanna mutate, and you could get in with the, with Cas9, either at this site, using that PAM, or at this site, using that PAM. Now, if you go in here, there’s
a seven-base microhomology that would make an 11-base pair deletion. It would be out of frame, but if you went in over here, there’s a five, sorry,
six-base pair microhomology that would generate an in-frame deletion, and in an experiment, it’s actually very, very common. So there’d by some danger, if you’re trying to inactivate a protein, that making a four-amino acid deletion would not inactivate the protein,
so you might wanna choose that target for mutagenesis. It’s just thinking about mechanisms and allowing them to govern
how you choose targets. This is something that
people have begun to notice in genome editing events, now that there are enough examples. This is from a recent paper looking at mutagenesis of the same target in different cell types and making graphical ways of
illustrating mutations you get. So, this is just going
into a particular target in a particular type of
cells and just saying, well, 12% of all the new mutations are, I think this is a single-base insertion, and then nine percent are
a particular deletion. And then this longer deletion
represents nine percent. Sorry about how small the type is. You’ll get copies of these slides, and you can go look at the papers later. But then you go to a different cell types. These are all established cell lines, human embryonic kidney. These are urethral leukemia cells, and these are colon
cancer cells, I believe. But in all the different
cell types, this particular guide RNA gives the same
pattern of mutations, and you can represent it this way, or you can represent it this way, as sort of a pseudo gel. You get this one base insertion. That happens because Cas9 often makes a single
base offset in the cut. It very often makes a blunt cut, cutting both strands at the same site, but sometimes it makes a cut that leaves a one-base
five prime overhang, and if that’s filled in
before blunt joining, you get a one-base insertion. So that’s probably where
most of those come from. Some of the common deletions are mediated by microhomologies,
but not all of them. And if you then go and you look at a lot of different guide RNAs, each column here is a different guide
and a different target, you find that they fall into classes. And this class gives deletion
of very similar lengths, and this is another way
of representing that. So there are some aspects of the outcome you get from non-homologous end joining that are specific to the guide RNA and not completely random. So, just something to be aware of. One thing you can do with
non-homologous end joining is to use two guide RNAs
to make cuts some distance from each other an make large deletions. This is from Jin-soo Kim’s lab where they started with a guide RNA that makes a cut in the
CCR5 gene in human cells and then used different guide
RNAs to make a second break, 10KB, 30KB, or 100KB away, and then they could isolate the deletions of all of those sequences
between those breaks. And one of the interesting
things about the junctions in these deletions is that
quite a number of them were precise junctions, and
the very bottom line here is an example of that, where… Sorry. It’s this line, the
next to the bottom line. So one of the… Oh, now I’m confusing myself. It is the bottom line, all right. So, that’s the PAM that’s being used. It’s GGT in the other strand. And the break is made
three base pairs away, right before that T. On the other side, there’s
the PAM that’s being used, and the cut that’s being made
is three base pairs away, right after that A, and of the eight junction
sequences that they determined, seven of them were precise
deletions between those sites. And what this illustrates, and there are more data
supporting this now, is that non-homologous end joining can make very precise junctions, can join things up without, without a deletion at the break site. This is a deletion between break sites, but not at the break site. If this is done at a single cut site, you reconstruct the target, and so Cas9 can cut again. But if you make two cuts
and they’re rejoined, now neither of the guide sequences is there and can be used. Okay. So, how do we, how do we look at the insertions and
deletions, the in-dels that are generated by
non-homologous end joining? What you’re gonna do is this experiment. You’re going to take Cas9 protein and a specific guide RNA. I don’t believe you have donor
DNAs, but you could have. Put them into cells. You wait. Something happens behind the fig leaf. You hope it’ll be good or you. After a little while,
you isolate genomic DNA, amplify the target, and
what’s happened here is going to create changes. And so you’ll have some
wild-type sequences and some altered
sequences, and you may have quite a number of
different altered sequences from a collection of cells. So you take this mixture of PCR products, denature them, and reanneal them, and then strands from
different initial PCR products will come back together, and
you’ll get some homoduplexes, plus some heteroduplexes. And if you run these on a gel, you run them on an agarose
gel, they’ll all run in pretty much the same place unless
there are large deletions. But if you cut this mixture
with T7 endonuclease, it only makes a cut
when it sees mismatches. So, you’ll cut the heteroduplexes, leave the homoduplexes intact, and by looking at the
proportion of cleaved products, you get an assessment of how
much editing there has been. And this is an example
of running those products on an agarose gel without T7E1, run as a nice, tight band. Wit the endonuclease treatment,
you get two fragments, and these are very high-frequency
mutagenesis examples because there’s very little
of the homoduplex left, and lots of cleavage. I don’t know if we can
turn the lights down. Possible to do that? Thank you.
That’s great. Actually, we could leave it there, unless people are likely to fall asleep. If I hear snoring, we’ll
turn the lights on. So, there you can see the cut products. Now, if you run those same samples on a high-percentage polyacrylamide gel, the homoduplexes run
where you’d expect them. But there are a bunch of others species that represent the heteroduplexes. Some of them run close
to the homoduplexes. Some of them run pretty far away. And we know that those are heteroduplexes, because on cleavage, they now fall down where we expected them,
farther down the gel. And I’m pretty sure what’s happening is that these heteroduplexes
have some flexibility. Instead of being rigid, short DNA rods, they have a mismatch in the middle, and they get hung up
because of this flexibility. They get hung up on the
tight polyacrylamide matrix. So you don’t even have to do
a T7 endonuclease cleavage in order to see these if they’re there in pretty high numbers. Another way you can look
at the heteroduplexes is to take advantage of the
fact that they’re going to have different melting temperature
than the homoduplexes. This is from a paper from a group at Utah using high-resolution melt analysis. There’s a procedure that Eva Brinkman and colleagues published
three years ago called TIDE, and here, this is a procedure where you rely on computation to figure out what’s going on for you. So you take those PCR fragments. You don’t denature them. You just take the PCR fragments after the experiment, the genomic PCR, and you run a Sanger
sequencing reaction, okay? Now, if nothing had happened, you’d get this nice,
clean sequencing read, and that would tell you
nothing had happened. But if you had achieved some mutagenesis and you now had end join products there, reading from one side, the
sequence would be looking just fine until you got to the point where the cut had occurred,
or close to where that was. And now, you start getting
what you would think of as a lousy read, mixed
sequences, mixed nucleotides, at pretty much every position, but they have devised a
computational protocol that’ll allow you to
deconvolute that sequence read into sensible, more or
less sensible, analysis of the products that had
actually been generated, and they can give you an output that says, well, you’ve got 17% five-base deletion. You’ve got some one-base insertions. You’ve got some one-base deletions, and in the middle, you’ve got some that haven’t
been affected at all. And this is possible because
you know the sequence on both sides of where the cut was, and so you can try to
rationalize the mixed sequences. And of course, if you have
access to next generation sequencing, you can just
take those PCR products and sequence ’em all and then look to see how many reads there were for
every particular sequence, and this is something from a study we did here that was
just published last year. Okay. So, that’s non-homologous end joining. Let’s talk for a few minutes
about homologous repair. So, in order to get homologous repair, you have to have homology
on both sides of the break. If you only have homology
on one side of the break, the cell can’t use that as
a bandaid to fix the break. But if you have homology
on both sides of the break, you have a chance of getting
homologous repair to occur. When people use long double-strand donors, they typically put a few
hundreds base pairs of homology on each side of the break, and sometimes some
non-homologous sequences in the middle if you’re
trying to make an insertion. Depending on the system, you may use a linear or
a circular representation of this donor. We found in Drosophila
embryos that linears didn’t work at all, but
circles worked fine. In other systems, linears
worked better than circles. So, system-dependent. Single-strand oligos work quite well. You need to have homology at
least at one end of the break, and you may want to put sequence
changes into your donor DNA that would prevent recutting. So an example would be,
your donor could include a mutation in the PAM for the
particular target sequence that you were going after,
and when that mutation was incorporated, the
target couldn’t be recut. So just to illustrate this point, that you need homology on at
least one side of the break, this is a series of donors
we used in Drosophila, and we were getting good
incorporation of these donors. Here’s the cut site, and this donor worked fine. It had a deletion over on one side. This donor worked fine. There was a deletion from the
cut site off to this side, but this donor didn’t work at all, ’cause it had no homology
close to the break. And what happens in homologous repair is that the ends of the break
go searching for homology, and they search in a sister chromatid or a homologous chromosome
that are native to the cell, and they search in your
donor DNA for homology. And the length of DNA on
either side of the break that’s involved in that
search is not infinite. So you have to have homology in the donor to at least one side of the break, or things aren’t gonna work very well. This is an example of
using a single-stranded oligo nucleotide as a donor. Chris Richardson, who’s a post-doc in the IGI lab just across the street, reasoned that the two
strands of the target after cleavage are not
equivalent to each other. One of them is bound by
the guide RNA an probably pretty tightly by Cas9 protein
while the other strand, at least in the drawing, is
kind of waving in the breeze. So maybe, if we specifically
design our donor oligo nucleotide, which is in green here, so that it has the correct
polarity to hybridize to this display strand, and if we place the sequence
changes that we’re interested in downstream of that
three-prime end from the target, maybe that will enhance the incorporation of sequences from the donor. And down here, you can see
that there are some examples where that was absolutely true. So, this green strand
corresponds to that donor strand, and the sort of purplish
is making a donor strand that’s complimentary to
this strand of the target, and the green one is working
better in several examples than the purple-y one is. So another aspect of donor
design that can be used to enhance your chances of getting the outcome that you want. It’s hard to get long,
single-stranded donor DNAs made for you synthetically, so this group went through a more elaborate process to make single-stranded donor DNAs. And the particular ones they were making, they were using a
micro-RNA target I think, so, this isn’t particularly important. But what they do is either from a plasmid or from a PCR product
that has a promoter for T7 RNA polymerase, they would
make an RNA transcript, reverse-transcribe this with a primer from a constant region,
and then degrade the RNA. And now you’ve got a single strand of DNA that you can use as a donor, and they have examples where
this has worked quite well, particularly in mouse embryos. This is a procedure that people have used. It’s sort of semi-homologous
recombination. It’s a procedure to use
short homologous sequences to get things in at one or two breaks. And this is from Alex Paix
in Geraldine Seydoux’s lab at Johns Hopkins. So, what they found was if they
wanted to make a gene fusion and they made a Cas9 cut at the site, at the three-prime end of a gene, where they wanted to fuse
a fluorescent protein coding sequence, if they
provided just 35 base pairs of homology in the donor DNA, that was enough to give them
efficient incorporation. And if they made these sequences, these homologous sequences too long, the efficiency actually went down. And they have, in their papers, a sensible explanation for
this based on experiments. I’ll refer you to those papers. But they could use this for an insertion, a replacement. This is one where they’re
completely knocking out a gene and replacing it with a green
fluorescent protein gene, and here’s another example where they’re turning GFP
into RFP with a donor, again, with 35 base pairs
of homology on either side. And they’ve told me that
the now have this working, not just in C. elegans where
the original experiments were done, but also in
cultured mammalian cells. And this group from Japan
is doing something similar. They’re relying on even
shorter homologies, and they believe that this is based on the microhomology-mediated events that I was talking about before that occur naturally around
a double-strand break. So, if microhomology-mediated
events occur, why can’t you use that? So what they do is they
make a break in the target, and around the… GOI is gene of interest. They also make cuts at the borders of five to 25 base pairs
of homology to the target. And then you have these
little microhomologies at the end of the donor
that correspond to sequences at the end of the break at
the target, and those are then used for incorporation
of your gene of interest. So, a variety of ways to
go about getting insertions that don’t depend on long homologies. So, you can get all
sorts of events happening with these editing approaches. We talked about large and small deletions, local small deletions and insertions, what we call in-dels, just through nonhomologous end joining. You can get local small
sequence replacements using short oligo nucleotide donors. You get longer sequence replacements using longer double-strand donors. You can get gene
insertions and gene fusions using longer donors with
longer or shorter homologies to the target, large deletions. You can either put the
deletion into the donor and let it carry that in, or
you can, you can make two cuts and get a large deletion
around those two cuts. And people have even
used the two-cut method to make chromosome translocations. Maria Jasin’s lab, as far as I know, was the first one to do this. So you make one cut on one chromosome and another cut on another chromosome, and amongst the end
join events that occurs, that occur, there are going to be ones that link one part of one chromosome to a part of the other chromosome where the two breaks had occurred. And people have used this to
model chromosome translocations that occur commonly in human cancers. All right. So, we’ve got end joining. Now we’ve got homologous repair. One of the frustrations
that people have encountered is that in many cell types,
the end join products tend to outnumber the
homology-dependent products by quite a lot. So what do you do about that? How can you shift the
balance into your favor? What we did many years ago now
was to say, well, what if we, what if we disable one of the components of a non-homologous end joining pathway? We know a lot of the gene
products that are involved in mediating non-homologous end joining. What if we knock out DNA ligase 4? DNA ligase 4, as far as anyone knows, it’s one of several ligases in the cell that’s only involved in
double-strand break repair. So, what if we knocked that out? Would that help? In the Drosophila, the
answer was absolutely. So this was back in the old ZFN days, but we did the same experiment, providing a donor DNA, along with the ZFNs for a particular target
in Drosophila embryos, and then we just looked to
see, amongst the products that we got out, the new mutant products, what proportion of them were
due to homologous repair? What proportion to end joining? And the answer was, in the wild type, about 20% of the new products
were homologous repair products, but in the LIG4
mutants, it was about two-thirds of them, and sometimes quite a bit higher. So, getting rid of one of
the components of the main non-homologous end joining
pathway is a good way to shift the balance. Now, Drosophila and C. Elegans
both are viable and fertile if they don’t have DNA ligase 4. This is not true of mammals. So you can’t just make a
mutant in the background and expect this to work well for you. But people have gone
about using other methods to try to inactivate components
of the end join pathway, and again, I apologize for
how small the print is, but I don’t want you
to take away from this that this is the solution
to all your problems. I’m going to tell you
what’s going on here. This group, this is Ralph
Kunz’s group in Germany. What they did was they
used a reporter system that had been developed in
Seattle called Traffic light. So, Traffic light, the reporter, has a non-functional green
fluorescent protein gene and a non-functional red
fluorescent protein gene. This Venus GFP is
non-functional because it has a deletion in it, and the
RFP gene is non-functional because it doesn’t have
an independent promoter, and it can only be expressed
using a promoter upstream of Venus, upstream of the CRISPR target, and through linkage to one of
these read-through domains. So, if you make a double-strand
break and you provide a donor DNA that can undergo
homologous recombination with Venus, and correct
the deletion in Venus, you’ll now get green fluorescence, and if you get a non-homologous
end joining event that shifts the reading
frame out of Venus into RFP appropriately, a two-base
pair frame shift, then you’ll get red fluorescence. And so you can get an
assessment of how much end joining there’s been
by the red fluorescence, and how much homologous
repair there’s been by looking at green fluorescence
in a FACS experiment. So if you just put in Cas9
and the appropriate guide RNA, you get about five
percent homologous repair, three percent end joining. Now, only a third of
the frame shift events are going to put you
into the correct frame, so this might actually be
closer to 10% end joining, five percent homologous repair, in this particular experiment. And then what they did was they used short hairpin RNAs to inhibit
CU70 and/or DNA ligase 4, two components of the end join pathway, and then they used them together. And what they saw was
the proportion of cells that now had homologous repair products, green fluorescence, went up. They also used a chemical
compound called SCR7 that had been identified
by others by a DNA ligase 4 inhibitor and saw
that the green fluorescence went up, the evidence
of homologous repair. But best of all was co-introducing a vector that expressed two
proteins from adenovirus 4. I didn’t know this until people started, until actually this paper came out, that adenoviruses, which
have a linear genome and don’t want their genome concatenated by cellular proteins or
integrated into the cell genome, all adenoviruses, the ones
that have been studied, have methods, processes, that interfere with
non-homologous end joining. And in the case of adenovirus 4, these two proteins, E1B55k and E4orf6, collaborate to ubiquitinate DNA ligase 4 and target it for degradation. And so expressing these
proteins and cells wiped out end joining and allowed more
homologous repair to occur, similarly to making a Lig4 mutant. And this is just showing
that it’s actually just replotting, replotting
the data from here. Now, why doesn’t everybody do this? Well, first of all, many people have had less than satisfying experience with Scr7. I don’t know for sure why that is. I’ve been told that what you buy as Scr7 doesn’t always have in
it an active compound. And it may be a synthetic
intermediate or a contaminant of some kind that actually has activity. So some people get it to
work, but many people don’t. So, this has not been
resolved, as far as I’m aware. And putting in these
two adenoviral proteins is just, it’s cumbersome, because it means another
vector, two more proteins. It’s just kind of a nuisance. So these approaches have not been adopted, but they seem to be on the right track. Another approach that’s been used, and Jennifer’s lab published
this three years ago, is to try to get the cells
that you’re targeting into the position in the cell cycle where they like doing homologous repair. And that, according to received wisdom, is in the S and G2
phases of the cell cycle. So, they use a variety
of cell cycle inhibitors and then release the
cells from the inhibition and looked at the efficiency of end joining and homologous repair, and just to cut to the chase, what they found in, I think
this was the HEK 293 cells, they found that they got elevated levels of homologous repair from
an oligo nucleotide donor when they had treated the
cells with nocodazole. Now, people at Sangamo BioSciences, even back in the good ol’ ZFN days, used another inhibitor. Nocodazole and vinblastine both inhibit, stall cells in the M
phase of the cell cycle. So, it can be effective,
and there’s another paper that came out quite recently
that emphasizes this. There are issues, a variety of issues. This doesn’t work in all cell types, and cells that are
particularly recalcitrant to doing homologous repair,
sometimes you can’t really bump them up into useful territory. And some of these
inhibitors are not so good for the health of the cells,
so you have to be careful about how you’re approaching
some of these things. But that gives you an idea of
types of approaches people are using when the events you want
are homologous repair events that put in exactly the
sequence you want at your target rather than just getting
end join mutagenesis, over which you have less control. So, I keep a list. I don’t claim to be completely thorough, but as of May this year, I had
found well over 100 organisms whose genomes had been
successfully targeted using one or other of
the editing platforms, and they go all the way
from disease organisms to their vectors, model organisms and various strange marine organisms. There are fish, mammals, lots
and lots of different plants, including quite a number of
food organisms in this list. So, the great thing about
this nuclease-based approach to genome editing is that it’s applicable to essentially any organism. All you have to do is get the nuclease in, which isn’t always easy,
but get the nuclease in, and then these endogenous repair processes will take over and
generate sequences changes at the target in ways I’ve
just been talking about. Okay, and just to give
you a couple of examples of things you’re gonna
see later in the week, you won’t see these particular things, but you’ll see things like it. This is an example that I absolutely love. This is from Caixia Gao’s lab in Beijing. What they did was they said, well, we wanna be able
to genome edit wheat. Wheat’s a problem, because it’s hexaploid, the standard varieties of wheat. That means you’ve got six copies, two alleles of three homologous
genes that you have to hit. And they said, well, why not? So they had success now with
quite a number of genes, but in this original paper, they targeted a couple of different genes that contribute to, the knockouts give you more oil in the
kernel, I think more oil, and there’s a higher kernel
weight than this particular one which is illustrated down here. The six-allele knockout
gives a shorter plant that’s putting less of its energy into making these very tall stalks and more into making a healthy
and more productive plant. This is an example of work
that’s going on in livestock. This is being done at a company in Minnesota called Recombinetics. What they did was they
genetically turned dairy cattle that normally have horns
into hornless dairy cattle, and they did this by
taking a small piece of DNA from a naturally genetically
hornless breed of Angus beef cattle and inserting it into
the genome of these Holsteins. And this is an animal welfare issue, because if any of you have been on a farm or know something about it, the dairy farmers
typically remove the horns from their animals with saws or loppers or scoops
or something like that so that they don’t harm
each other or the farmer, because they’re kept
in very close quarters, and so by making these
naturally hornless dairy cattle, you’ve eliminated both an animal welfare and a modest economic
issue for dairy farmers. And then you’re going
to hear a good bit more on Friday about applications in medicine, so I apologize for this. This is sort of a cheap shot, right. There’s a one-year-old
kid who had her leukemia, an intractable leukemia, cured, by getting therapeutic T cells that had been modified by TALENs to make them more effective. And there are a few clinical
trials that have been approved and some that are ongoing
that are using one or another of the editing platforms,
so that’s an area that’s just going to continue to expand. So I think that’s all I was
going to say in this first talk. Anybody have questions about these issues? Yeah. – [Audience Member] So, knocking out the two proteins or LIG4, does that have any negative
consequences for cell health? – Yes. What you wanna do is you wanna
have a transient inhibitor of the end join pathway, if you’re going to
approach things that way, and that’s one of the
things that people thought was promising about this SCR7. But if end joining isn’t available, then the cells are more susceptible to things that naturally
produce double-strand breaks, like replication fork stalling, although that’s mostly taken
care of by homologous repair, but other things that
generate breaks in cells, now the cells are more
susceptible to those. They’re not happening at
the place where you put in a nice donor DNA. They’re happening elsewhere
around the genome. So they can have consequences for cells, and what you’d really like
is something you can wash in and wash out during the period
of the editing experiment. Yes. – [Student] I recently saw
a talk with someone saying that when they were
trying to do editing in, I believe it was embryonic stem cells, they were running into the issue that they were artificially
selecting for P53 mutants, because the cell was
trying really, really hard not to have these double-strand breaks. Is that anything you’ve run into? – So, as I understand it,
that happens in IPS cells and in embryonic stem cells naturally, that even without introducing
double-strand break, you know, with your reagents, these cells just growing in culture will naturally generate P53 mutations so that they traverse the
cell cycle more readily, even in the presence of some
of these damaging agents. So it’s a hazard around stem cells that isn’t necessarily connected
to doing genome editing. Yes. – [Audience Member] How can I manage the problem of genome translocation? – The genome translocations
are pretty rare. So, particularly if you
can, if you can clone cells, you can pick cells that have the knockouts you want without translocations. So, if you don’t have the opportunity to go through a cloning step,
then you just have to try to characterize the
population as best you can. But they are rare. You have to go looking hard for them. – [Audience Member] Do donor
DNAs have to be modified? – They don’t have to be. But the yield of the desired
products is typically lower because you can get your
intended sequence in, but then the target gets cut again, and it may be lost for
one reason or another. It’s usually pretty easy. You can make… You can make silent
changes in coding sequences that either knock out the
PAM or that destroy homology to the guide sequence as a
way of preventing recutting. Yes. – [Audience Member] For homologous repair, do you have a feel for
what’s the optimal balance between having your intended
mutation close to the cut site versus having enough homology
on each side of the cut site? – Ah. I eliminated a slide. So, you have to have homology
to one end of the break in your donor, maybe both ends. Where the sequence change is
that you’re trying to introduce depends on what cells and
system you’re working in, and this depends on what
we call conversion tracks. So, it basically is, when
the donor’s being copied into the… Just a second. Hey, Jennifer? Do you wanna get set up? When the donor’s being copied, the copying goes some distance and then quits. We found in Drosophila that
we were getting copying of donor sequences three KB
and more from the cut site. In mammalian cells,
those conversion tracks are really narrow, and you
go more than one or two hundred base pairs from the cut site, and you have a very low probability of getting your sequence
change incorporated. So it depends, depends on the, well, it depends on the
biology of the system. It depends on what the cells you’re working with are capable of doing. (bright instrumental flourish)

6 thoughts on “Dana Carroll: Background on Genome Editing

  1. Very important issue! My blog www.science1984.wordpress.com Visit and share! Thanks! – More health to the world! Best wishes

  2. The thing I'm worried about is that it is a game changer. Hardballers can go around infecting people and have less to worry about with their own hoard. Scientists never seem to pick diseases that affect the slave class like being poisoned by antibiotics and other drugs before birth. The war machine expects us to take it up the butt for 500 years before they'll do anything. Sound familiar?

Leave a Reply

Your email address will not be published. Required fields are marked *