(bright electronic music) - My next assignment,
which I assigned to myself, was to talk about some issues that arise around the whole editing business. And one of the first things
I'm gonna talk about, and Jennifer introduced this very briefly, is the fact that there are alternatives to the standard
streptococcus pyogenes Cas9. One of the first ones to be appropriated was the staphorias Cas9. This was promoted by Ran in Feng Zhang's lab. And one of the reasons that
they got interested in this was because the pyogenes
Cas9 is a big protein, 1,358 amino acids. I'm sorry, 1,368 amino acids. It's a little cumbersome to work with in some situations. And the staphorias Cas9 is only 1,053 amino acids, 300 amino acids shorter. People have also done a little bit of work with this streptococcus thermophilus Cas9, which is a little bit bigger,
70 or so amino acids bigger than staphorias. The Joung lab isolated CRISPR RNA's and tracer RNA's made themselves a single guide RNA in the way that the Doudna
and Charpentier labs had made a single guide RNA for the strep pyogenes system. They found that you needed a guide length that's a little bit bigger. The pyogenes system works quite well with the 20 nucleotide guide sequence. The staphorias system likes something that's a little bit longer, the guide sequence itself
a little bit longer attached to this framework. Recently Jungsu Kim's group began working with an
even smaller Cas9 protein. This is campylobacter
jejuni, and it's only 984 amino acids. The coding sequence for it is under 3KB. What they showed was that they could make a single guide RNA using the same sort of principles as for the predecessors. And then they looked at guide links and sort of similarly
to the staphorias system guides of 21, 22, 23 nucleotides seemed to work a little bit better than the 20 nucleotide guide sequence just this sequence that directs the Cas9 protein to its target. They also showed that the jejuni Cas9 is quite specific and it can be used in these mouse myotubes by injection into mouse muscle. I'll talk about this again on Wednesday, but one of the reasons for looking for these effective shorter Cas9's, is that a popular vector for gene delivery in human cells and other mammalian cells is the addeno-associated virus and their vector's based on AAV, but they have a limited capacity. And if your Cas9 takes
up less of that capacity, there's plenty of room
for a guide RNA sequence and a promoter driving, it's expression, a promoter
driving expression of Cas9 and then some other bits and pieces all within the packaging limit
of addeno-associated virus and then these are
addeno-associated virus delivery to cells in culture and to living muscle. Jennifer showed a later
version of this slide, mine's from a review. Eugene Koonin gets a lot of mileage when more sequences are accumulated, they make more sequence comparisons and they expand the list
of these CRISPR systems but even as of 2015, they had identified this type five within class two where they're Cas9 like proteins that are called Cpf1. And Cpf1, Jennifer didn't
have time to go over it, although she showed you a
slide that illustrated it. Cpf1 has a bunch of differences from the Cas9 system, which may be useful for some purposes. First of all, there's no tracer RNA. There's a single RNA that's involved, so you don't have to construct a single guide RNA out of
CRISPR and tracer sequences. It's just one RNA. Although the Cas9 system
it's the five-prime-end of the guide RNA, where
the guide sequence lives and it recognizes the target. In the Cpf1, it's the
three-prime-end of the guide RNA that recognizes the target. Instead of making blunt
cuts, it makes staggered cuts that leave a four-base five-prime overhang just like a lot of restriction enzymes. People have considered that
this might have some advantages, although I don't know
of anybody that's taken advantage of those possible advantages. The PAMs for the Cpf1's that I know about are all AT rich rather than GC rich, which has been true for some of the Cas9s and the Cpf1's have been shown to have quite high specificity. This is just some more stuff. This is again from Feng Zhang's lab looking at guide links and you can see anything from 18 to 22 is an effective guide
for in vitro cleavage. There is very little
tolerance of mismatches as you get close to the three-prime-end of the guide sequence. Sorry, to the hand proximal or the five-prime-end of
the guide sequence itself and they also demonstrated
activity in cells, Cpf1's from two different bacteria compared to the strep pyogenes Cas9. This is a very functional
system that people have used. One of the things that
people thought might limit the utility of the original Cas9 system was the fact that you had to have a pair of G's in the genomic sequence next to where the guide would bind. So your target had to be defined by a pair of G's. But now, with some of these
other natural proteins, the Cpf1 proteins that have
these AT rich PAM sequences, the staphorias and campylobacter jejuni have somewhat different guide sequence. I hope everybody's familiar
with the nomenclature that R means purine
and Y means pyrimidine. For example, this jejuni Cas9 recognizes A or G, T or C, then
AC, four base pairs away from the sequence that is represented in the guide sequence of the guide RNA. And then, Keith Joung's lab
has gone about deriving Cas9's from both staphorias and strep pyogenes that recognize somewhat
different sequences. Instead of GG, there's
one that recognizes GAG and another one that recognizes GCG and they made one from staphorias that doesn't care about that G, but only cares about purine and purine T. And this is by in vitro selection. So now you have a variety of PAM sequences so you're less limited
in the genomic targets that you can access. Another thing that people
have done with Cas9, mostly to improve its specificity, is to break it into two pieces. And you can break it
into an N terminal half and a C terminal half and
express them separately. The Douda lab actually
did this in a somewhat more intentional way than just cutting it N and C terminal. You can set it up so that one of the, sorry, so the association
between the N and the C terminals doesn't happen except in the presence of a small molecule or some other initiating condition. This is one, again, this
is from the Joung lab, where rapamycin encourages
the dimerization or the association of the
N and C terminal halves, which then go to the
nucleus and have activity. Now, the activity, the inherent activity, is somewhat compromised compared to the wild type N-type protein, but the specificity is actually quite good because there's nothing going on until you introduce the rapamycin. And I just gave you a list
of some of the other ones, Luke was involved in one of these. There's even a photo inducible dimerization system
that has been developed. So you can look those up on your own. One of the things that people tend to get freaked out about in all of these editing technologies is what's happening at
other places in the genome. What are the off-target effects? The concern about this was
stimulated and enhanced by some papers that
were published in 2013, just a little bit over a year after the original Martin genic Douda Charpentier publication showing what you needed
in order to operate the CRISPR system. This is from Keith Joung's lab. They have a situation where
they've got a single copy GFPG in cells and then they're introducing different guide RNA's
that will cut the GFP and by nonhomologous end joining, ultimately disrupt the
GFP coding sequence, and they're just looking by fluorescent activated analysis what proportion of
cells are now GFP minus. So it's just a disruption assay. They've got three different guide RNAs that have different GC contents that are pretty good when there's a perfect match between the target and the GFPG and the guide sequence. But then as you begin
introducing single based changes, you find that if those
single based changes are near the five-prime-end
of the guide RNA, they have essentially no influence on this or this guide RNA and its effectiveness. Some greater influence on this one. And even when you begin inducing pairs, either adjacent pairs
or more distant pairs of mismatches, you're still getting kind of a lot of cutting with
some of these guide RNAs. Then they went to endogenous genes to look to see what's going on. This is the target. And in the genome of these human cells, there are three different
types of human cells, there are some sequences that are related, but not identical to the intended target. So let's look at U-2 OS cells. They're getting 50%
mutagenesis in U-2 OS cells at the design target. But, they're getting a
similar, almost identical level of mutogenesis at a target that differs by three base pairs and here's one that differs by four that's giving an even
higher level of mutagenesis. So, you know, are we shooting ourselves in the foot or some other more vital organ when we are operating this system. This is from the Feng Zhang
lab showing a similar thing. It's a little difficult to
assimilate this right away, but what they're showing, when there's deep blue, that means a single base change doesn't
interfere with cleavage. And they're using similarly, they're using a system
where they've got guide RNAs that are mismatched to a single location to a genomic target. Single base changes
near the five-prime-end of the guide RNA are
not inhibiting cleavage very much at all. Although, when you get closer to the PAM, it makes a pretty big difference. And this is again from the Joung lab. If you put in more Cas9 and guide RNA, this is using a plasma delivery system, you get more indel formation
at the design target, but you also get more at off target sites and if you compare, you make a specificity calculation design target over secondary target, the specificity goes down, as you put in more of the nuclease. - [Woman] So I assume
that people have shown that that preference for the amino acids, near the, sorry, (mumbles)
to PAM need to be stronger is also mirrored in Cpf1, where the three-prime and
five-prime are flipped? - Yes, but Cpf1 actually, is less tolerant of single base changes at either end than the Cas9 system. But it is true, I think it's at one end. It might have been in a previous slide, I just don't remember
right now how it goes. You can look it up. The harder you try to hit your target, the more you're gonna hit the
secondary targets as well. My view, all the way from the
time those papers came out, little less than four years ago, has been the CRISPR system actually
needs to have room for mismatches. And the reason for that is that, when a virus infects the cell, this cell has incorporated a sequence from a relative of this virus and now is protected
against that relative. The infecting virus this time may not have an identical
genomic sequence. Viral sequences evolve
at a really high rate. And so, you know, even keeping
coding sequences the same, you can have redundant codons that would have sequence changes, and so if you were so specific that even a single base pair of mismatch between the CRISPR RNA that was generated from this spacer -- Single mismatch between
this and incoming virus, if that did away with your immunity, you'd be in big trouble. So mismatch tolerance I think is adaptive in the CRISPR system
and we should expect it to have less than perfect specificity. Okay, we're gonna talk
about how much we care, how we measure some of
the off target influences, and what we do about it. How much do we care? Well it depends what you're doing. If you're working in a model organism and you're making new mutants and you're gonna study gene function through this mutagenesis, you can make independent
mutations in the target and study them to see if they
have the same phenotypes. You can clean up the background and a lot of genetic
organisms by out-crossing. And you can complement your
mutant with a wild type gene to make sure that the
phenotype you're studying is do to that mutation
and not to something that happened elsewhere in the genome. If you're working with food organisms, you're concerned about safety. Both livestock and in plants, you typically go through
a very narrow bottleneck, you're essentially cloning the organisms and then expanding them
through sperm donation or seed propagation. And so you can do whole genome sequencing and look to see what else did we hit when we were trying to hit gene X? And when you go to clinical applications, again, we're worried about safety. And when people talk more specifically about some of these clinical applications, you can see that in somatic therapies, particularly ex vivo
modifications of cells that are then going to be
put back into a patient, typically have a limited range of off-target modifications that might come up and bite you. As long as you kind of have those covered, you're probably in pretty good shape. As you get to stem cell therapy, somebody asked the question
earlier about stem cells. Right now I'm more concerned
about the stem cells than I am about the CRISPR
base modifications of them. But that will change over time. In germline therapies,
as Jennifer highlighted, there are lots of reasons to be concerned because you can't go backwards. You're influencing an individual and that individual's
ultimate offspring and you could conceivably try to go back and reverse the change using the genome editing technologies again. If there are multiple
things happening off target, that could be a problem. So how do we assess what's going on at secondary targets? And there are a bunch of ways to do this, and I'm gonna illustrate a few of them. One thing that people will do
is to either make predictions just based on sequence similarities. I think that Jennifer
Listgarten will show you that you can do something
beyond just looking at sequence similarity. Or you can use an in
vitro preference assay, binding preference assay,
something like SELEX. And then there a bunch of things. This is another in vitro method that David Booth's lab came up with. There are a bunch of things you can do that are, actually you're assessing what's going on in cells. One of these, who was
developing Keith Joung's lab is Guide-seq. If you read Keith Joung's papers, you'll know that he develops a lot of very, very useful tools in this area, but none of them is developed without a cool acronym. I think they're probably developing other technologies in their lab, but they discard them if they can't think of a cool acronym. So this is GUIDE-seq, and you have to go to the paper to see what GUIDE stands for. 'Cause I don't remember. It's actually a neat technology. What they reasoned was,
when you make a break, the ends will often go back together and we even know that you can get an N from one break to go together with an N from another break. What if we put in a little piece of double-stranded DNA,
our double-stranded DNA and we just look for end capture of this double-strand DNA by
nonhomologous end joining. And if there's enough of
that around in the cell, instead of the two ends of the break coming back together with each other, they'll capture this oligo in-between. So this is a 34 base pair, a double-stranded ogligonucleotide, it has five-prime phosphates, which help with ligation,
but the ends are protected chemically from degradation, and it doesn't work well
unless you protect those ends. So you're getting blunt
joints for the most part. Now this is in the genome only at sites where a break was made. And so you can use this
as sort of a landing pad for PCR and amplify the sequences that are beside where that oligo got incorporated in the genome. This is just an elaboration of what you do to get this ready for DNA sequencing. When a break has been made, what you should do is you should read out of this oligo tag. It can go in in either orientation, read out of that oligo sequence into genomic sequence in both directions away from the the place where the double stranded break was made. And so this is a way of looking to see where in the genome are
the double strand breaks. And the data from an experiment
like this looks like that. And there will a quiz on this. What I'll do is I'll just
show you one example. Not the best and not the worst example. This is a side in the VEGFA gene. And there is the target sequence up there. And then these are the
number of sequence reads from a GUIDE-seq experiment for the intended target,
which is that black square. And a bunch of other genomic sequences. The horrifying thing from this is that the intended target
is only number three in this list. There are two others have three. This has two and this
has three mismatches. That line right there
shows that any nucleotide is acceptable between the
PAM and the guide sequence. That was expected. But these guys are being
captured more frequently by this experiment than
the intended target, which represents only
11% of all the reads. So this again is suggesting that there's a lot of
off-target cleavage going on. Another procedure that's been used, this has been adopted by the Joung lab, although it was invented
by these characters. And this is one where you -- Whereas the guide sequence is cumulative, so the oligo is there, the cleavage re-agent is
there over a period of time and any break that's made
during that period of time is a candidate for picking up the oligo. This is one where the only thing that's going into cells
is the cleavage re-agent. Then you fix the cells and you capture whatever breaks, whatever
the standing level of breaks is at that time. You capture those breaks by ligation to a oligonucleotide, it's
a hairpin oligonucleotide. I think it's this one. You capture with a hairpin oligonucleotide then you shear or cut again. You capture these things
with magnetic beads and then you ligate something
on to the other end. So you've captured,
with this initial oligo, you've captured an end that was made by the cleavage re-agent in cells. And the Joung lab has used this. And these are just data from
the paper where they did that. Where they're showing
levels of sequence reads from that experiment at the design target and some secondary targets. And one of the things they
were trying to demonstrate in this paper is that for
these particular targets, the pyogenes Cas9 seem
to cut secondary sites more frequently than the staphorias Cas9. But that was a way of capturing sites that have been cut by
either of these Cas9 proteins. Capture them in that snapshot. This is a system that was
developed in Fred Alt's lab for a different purpose, but
then repurposed for this. I'm not gonna go through the whole thing, but what they're basically doing is they're using the design target kind of as a GUIDE-seq oligo, to capture cuts that happen
at other sites in the genome. So if you make two cuts in the genome, as I told you, you get
these translocations. One cut finding the other cut and getting a translocation. Well, in fact, that's
what they're depending on. They're depending on these translocations. Here's an example where the intended break was made in a RAG a, inside a RAG1 genome in chromosome 11. And then, when they pulled out everything that was connected to the break at RAG1, they found that there were sequences from elsewhere in the genome and that's what these little
fountains are showing. If they're red, they're
pretty common translocations, if they're just sort of beige, just sort of yellowish, then they're less common translocations. But there are other
places around the genome. So that's identifying secondary sites that have been cut while the primary site was also being cut. This is a procedure that was developed in Jungsu Kim's lab, in Korea. What they do is they identify targets for a particular Cas9 guide RNA complex. By in vitro cleavage. And they drive this really hard. So take total cellular DNA, cut it with Cas9 and a guide RNA and then just randomly sequence all of the molecules in the genome. And for any site that's
susceptible to cleavage, you shouldn't get any reads
that cross that cut site. So you drive it hard, they
pick up the intended site and they pick up secondary sites. Then you take DNA from cells that have been treated with that guide RNA and Cas9 protein, and you should generate
indels in some of the copies of those sites and now the sequence reads will go across instead of all stopping at the break site. You use that to identify targets that Cas9 can cut, and apparently does cut and mutate in cells. So there are all these procedures. Even before GUIDE-seq, people had used a capture of a viral genome an integration defective Lentiviral genome who works by the same principle. Some of less effective. And each of these procedures
has some drawbacks. They all seem to be
pretty good at identifying secondary sites, but as I said, GUIDE-seq and is true of this. IDLV cature, they're cumulative, so you may over emphasize
secondary targets as the cells grow and
you continue cutting. BLESS is just a snapshot, it
very depending on the timing of the snapshot you take relative to the introduction of the nuclease. Digenome-seq requires very deep sequencing and this HTGTS high through-put genome translocation sequencing. That depends on getting
at least two breaks in the cell, so you
have to kind of over cut in order to see anything. They all have some drawbacks and someone should sit down and compare these with dozens and dozens of guide RNAs to see if they all agree. As an alternative to worrying about and measuring these off-target effects, people have also begun working on improving the
specificity of the nucleases so that you have less reason to worry about off-target cleavage. One of the early ways of doing this was to turn Cas9 into a nicking enzyme. Jennifer said that with
these anti CRISPR proteins that interfere with the access of the H and H domain to DNA, that you actually inhibit
cutting of both strands. But if all you do is you make
a single amino acid change in the protein, so that the H and H domain, for example, is not cut, the other active site can cut the guide strand and it
can cut the guide strand wherever in the genome Cas9 has to bind. So if you provide two guide RNAs, targeted to opposite strands, pretty close to each other, the cell looks at this as if
it was a double-strand break. A frank double-strand break. But, what you've done is
you've squared the specificity of the system, because now
both of these guide RNAs have to bind in order to make the nick. This improves specificity, and remarkably, these guide RNAs combine the sequences that are up to 100 base pairs apart in the genome and still
have pretty good efficacy. So there's a paper from
George Church's lab that shows that you can
get end-joined mutogenesis and hemology dependent repair using these double-nicking systems. This is a paper from Fung Zhung's lab showing essentially the same thing. And one of the things
they've done is that they've shown that with the double-nicking, that your specificity is better. That the ratio of mutogenesis at defined off targets -- On-target to defined off-targets gets better with the double-nicking that's in the shaded bars here than it is with the double-strand break inducing wild type Cas9 protein. And this is just showing that you can go some distance apart. They looked at the
separation between these. You can go some distance apart and still get pretty good efficacy. I should emphasize that single nicks are not very effective. Nancy Maizels at the
University of Washington has shown that you can get some efficacy out of a single nick, but it's not as good as double-nicking. This is looking at homologous repair. I guess all of these are
looking at homologous repair. And when you're doing double-nicking, generating a five-prime overhang. So you have a nick on one
strand, nick on the other strand, if it's five-prime-ends
that are overhanging between the nicks, that works
better for homologous repair than the other way around. And I can tell you why I
think that is, if you ask me. Another thing that Jungsu Kim found was that if you just put
a couple of extra G's on the five-prime-end of the guide RNA, that don't match the target, it actually improves the
specificity a little bit. This is just data from his paper and I can tell you in a
minute why I think that works. In Keith Joung's lab, said, well, what if we shorten the
length of the guide sequence? We just take a single guide RNA instead of making the match to
the genome, 20 base pairs, we make it 19 or 18 or 17. It turns out for many of these guide RNAs, going to 17 or 18 base pairs, doesn't reduce the indel frequency, but it can still support homologous repair, but it
does improve the specificity. This is showing for
single base mismatches, full-length guide RNAs
versus truncated guide RNAs, and this is for two-base mismatches. And you can see that whereas this guide RNA isn't very sensitive to single-base mismatches, even
when you use a shorter one. But this one is very much more sensitive to single-base mismatches
and this one is quite a bit more sensitive. And even this one is much more sensitive to two-base mismatches. They also went and they did GUIDE-seq, whereas intitially at
this particular target, the design target was third best and represented at 11% of the reads. Now, with the 17-base guide RNA, it's moved into first place and
represents 30% of the reads. It's improving things. I had talked a little bit about
donors that take advantage of the displaces strand. And Jennifer talked
about that a little bit. In Slaymaker, in Fung Zhung's lab, said, well, part of the energy
of binding of Cas9 and the guide RNA to the target comes through non specific
ionic interactions between positive charges on the protein and negative charges on
this displace strand. So if there's excess energy
at the design target, that means the secondary sites that are only partially matched now have enough energy
to get bound and cut. What if we reduce the
energy of interaction to the displace strand? And so there was no crystal
structure at that time that actually showed where
the displaced strand is. Now there is one, or at
least that structure. But they thought they could identify positively charged residues out here. They were good candidates based on the existing crystal structures. And so they mutated them one at a time. And looked to see what's
the efficiency of cleavage at the design target and a couple of identified secondary targets. And they found ones that seemed to cut and make indels in the design target quite efficiently. As efficiently as the wild type, but significantly reduced the cleavage at these secondary sites. And they then began
combining these mutations. They're substitutions of alanine, uncharged amino acid for
positively charged lysines or argons and they eventually came up with enhanced streptococcus
pyogenes Cas9 1.1 that has three of these substitutions and shows very high cleavage at the design target and very little at the secondary sites. So that's a way to enhance specificity. And this is just showing the specificity. Here's 1.1 out here that is and also here, is quite sensitive. Remember blue indicates insensitivity, white indicates sensitivity. It's much more sensitive to
mismatches than the wild type and varies to single mismatches and very sensitive to
these double mismatches. Keith Joung's lab had the same idean, but they went for the
other strand of the target. They went for the RNA-DNA hybrid and there they actually
had a crystal structure that showed the contacts between the protein
and the DNA-RNA hybrid. So they identified some residues they thought they could mutate that would reduce the binding energy between Cas9 guide RNA and the target and they made mutations in those and similarly to what
the Joung lab had shown, they found increase specificity by combining a number of these mutations. These are again, alanine substitutions. This is arginine,
asparagine, and glutamine in this particular case. They did Guideseek on these. They didn't actually on
this paper do Guideseek on the same target I showed you earlier, but they did Guideseek on
a bunch of other targets and show that. Here's one where the wild type showed pretty good cleavage
at the design target but lots of capture of sequences
elsewhere in the genome and this high fidelity one
version of Cas9 protein picked up no sequences
from these other sites using Guideseek. And there are a bunch of
them where the specificity was very greatly enhanced. So you've got all of these methods to improve specificity. I think that the truncated guides and I think it's also true of these five-prime GG extensions and these mutations in basic or polar residues in the protein may all work by eliminating
that excess affinity. So if you bind the target more tightly than you absolutely need to. You've got this excess affinity, which means the secondary targets, although they have some
mismatches that reduce affinity, they've still got that extra affinity from other interactions. If that's true, then
the on target affinity is going to depend on the
sequence of the target and the guide sequence. So for very AT rich targets, you may not be able to employ some of these affinity reducing measures, you'd be better off going a wild type. But if you go to very GC rich target, where the RNA/DNA hybrid
energy is still really high, you may want to employ some
of these other approaches. As far as I've seen, people who are doing sort of routine use of the system, have not started to employ these high fidelity modifications. They're still kind of going
with the basic system. But as they get more
finicky, they may need to. It's also true, as I think I said before, that the Cpf1 homologue seem to have higher inherence specificity. So that's another thing you can do. This is just showing, this is comparing well, there are a variety
of things going on here. This is from Jungsu Kim's lab. One of the things that's showing is the off-target cleavage
can be very much lower. I'm looking for one where
they actually had -- The comparison that they have done with pyogenes Cas9 has
to do with doing a lot of sequencing of the cut sites and making a sequence logo for one
of the preferences of the Cpf1s and Cas9. And Cas9, obviously can tolerate some substitutions,
particularly far from the PAM, whereas the Cpf1's
again, far from the PAM, they tolerate more. But these are much more
well defined sequence logos indicating higher specificity. This is an interesting observation. They set up a situation where these two Cpf1's from
two different bacteria and SP Cas9 actually
cut the same sequence. Just based on the disposition of the PAMs. And what they found was that
even between the two Cpf1s, the indel signature
was somewhat different. This Cpf1 preferred
making a 15-based deletion with a minus six-base and
a three-base deletion, whereas this one cutting the same target, preferred the six-based deletion and had a 16-based deletion
as the secondary preference and little or none of
the three-based deletion. So there's something about the proteins, not just the
underlying sequence, but something about the proteins that's influencing the indel outcome and then the pyogenes Cas9
cutting in the same site had different preference. And as I'd said before, if you do the experiment multiple times, you get the same preference
for the same protein and guide RNA. This is GUIDE-seq looking at Cpf1 homologs and again, you're seeing situations where they're getting very
little capture of breaks at secondary sites in the genome. One of the Cpf1's gives
a few other sequences. This one's giving essentially none and the reason that
there's just a single line in many of these, they've
captured nothing but the designed target, the Cpf1 cleavage. If you're worried about specificity, then you have things you can do about it. First you can design a new guide RNA, that's what people tend to do, initially. You can use paired nicking, you can use this mismatched GG extension, you can make the truncated guides. We actually, that one I
was here working at IGI, we made a 17 or 18 based guide sequence that worked very well. You can use these Cas9 mutants from the Zhung lab or the Joung lab Cpf1. I'm gonna mention on Wednesday, why using RNP delivery has some advantages for specificity over some of
the other delivery methods. There are a lot of things you can do to enhance this specificity. Since this lecture was called editing issues, I'm gonna do something like what Jennifer did very briefly. I'm gonna emphasize a
couple of different thing from what she talked about. She was talking about the applications that themselves and how they would influence society. What I'm gonna do is I'm gonna talk about two issues that I think
are quite important and these may come up again
on Friday, I don't know. One is whether you're talking about making changes, enhancements of
agrucultural organisms, or changes in humans to fix them, you have to decide at the
outset who needs to be fixed. What are you gonna change? What people are you gonna tell need to be fixed 'cause they aren't right. And then once you've decided what to fix, how are you gonna distribute that fix to the organisms, people
who need the most. Even in the realm of agriculture, I showed you an example of modifications and Rita showed you an
example of dehorning cattle. What other changes should be made and are they changes that would benefit farmers in Wisconsin, or are they changes that should benefit farmers
in Sub-Saharan, Africa. On the human side, some
of you may know this book by Andrew Solomon, it's
called Far From the Tree and it talks about people who are unusual in various ways. And one of the groups he talks about are people of short stature. Have genetic causes for being very short. And he makes the point that many people who were born with short stature wouldn't change that, even if they could. They're comfortable with who they are and in many cases, wouldn't even want a genetic change that would
make their children tall instead of short. So you have to be careful who you identify as needing to be fixed. In the realm of distribution, when I was here on
sabbatical two years ago, we worked on a project to fix the mutation in sickle cell disease. Well here we are, we're sitting up here in Northern California and
there are sickle cell patients in Oakland, maybe a few in Berkeley, but the vast majority of
sickle cell trait carriers and disease sufferers
are in tropical regions in Africa and Asia and
if we develop this fancy, expensive molecular technology, how do we get it to the people, the huge numbers of
people who really need it? So these are issues that come up sort of post technology. What are you gonna do with these tools that you now have? So I just wanted to highlight that and that I didn't want to show. So questions about Cas9 variance and off-target issues? Yes? - [Man] If I remember right, for prop modification maybe to start from a single (mumbles). On genome sequencing you can, otherwise people should use it all. - Particularly for the
agricultural organisms, plants and animals, the
phenotype is the key, right? You want to test the product and not care as much about the method by which the product was generated. So you can do whole genome sequencing. There's a disadvantage to
doing whole genome sequencing, there's no chance that the genome sequence of one of those bulls I showed you earlier is identical to his parent. It's just not gonna be true. Because of the number of cell divisions and the mutations that accumulate just during self proliferation. So there are gonna be differences and if you insist that
there be no differences, you'd never do anything. Whole genome sequencing can reveal things that are pretty common
and look like things that could have been generated
by off-target cleavage, but you have to be
careful about what you do. How many people notice this paper that appeared in Nature
Methods a little while ago that said, "Unanticipated, high levels of "off-target mutogenesis." You should ignore that
paper, absolutely completely! Maybe those of you who haven't seen it, it's in Nature Methods, June maybe? May or June. Very recent. The first author is Schaffer. So if you want to look it up, I can give you the reference to it later, if you're interested. And maybe we should have a quiz later. People need to tell me why
you should ignore this paper. What's wrong with this paper? Anyway, you should ignore it, nothing it says is true. Yeah. - [Man] I was just wondering, with the Cas9 variance, if
you look at the CRISPR rates and the species, do you see,
like the guide sequences longer or shorter? If that was like the origin if adjusting the guide sequence or do they
just (mumbles) force that? - Did who group force it? - [Man] People who study the variance. Did they feel like, well, I'm gonna try ... - So, the CRISPR RNAs are
developed from a long transcript. And when that transcript is processed, you don't get the whole
of what we call spacer in the final CRISPR RNA,
so you can't just look at the size of the spacers
and from first principles decide how much is going to appear in the ultimate CRISPR RNA, so they could've gone and looked at the natural CRISPR
RNAs, but it was easier to look at this in a more technological setting. It's also true that the repeat sequence part of that is eliminated during processing. - [Woman] So because we know our sequence of our guide RNA, theoretically, bio (mumbling) you
should be able to predict what the off-target sequencing would be. Do we find that to be the case when we compare it to
the empirical results of of say, GUIDE-seq? - There is an imperfect overlap between predicted and
detected secondary targets. I don't know that anybody really
knows the reason for that. I guess it's on Wednesday
that Jennifer Listgarten is gonna be here and
she's developed some tools for predicting good targets. You can maybe ask her that question. But it us surprising that sometimes a single mismatch doesn't appear as frequently in the GUIDE-seq study or one of these other studies as a two or three base mismatch. And I went and I tried to do something looking trying to use
energetics to predict what targets would be more susceptible to mismatches than others. There was an imperfect overlap of that and experiment also. So I don't think anybody really has a great way of rationalizing that. (bright electronic music)
