编辑 | 新锐恒丰研究院

（植物数量性状CRISPR育种方法）

字幕：right good morning can you hear me well

in the back all right well thank you so

much to the organizers for having me

here this morning it is a pleasure to be

at k-state I guess right away because my

favorite sister lives and works here so

it's been great to see her and also you

might not have known this but my mom was

a k-state graduate student back in the

s and she brought our family along

when I was a six month old infant until

I was three years old so it's nice to be

back in town although I don't really

remember much friend that from that time

I'd like to talk this morning about the

use possible use of CRISPR technology in

breeding for quantitative traits in

plants as dr. Fritz mentioned I've been

thinking a lot about what you might call

us bandwagons in plant breeding a

bandwagon is an idea activity or cause

that becomes increasingly popular or

fashionable as more and more people

adopt it bandwagons may have a negative

connotation but there have been some

wildly successful bandwagons over the

years

hybrid maize for example is an example

of a wildly successful bandwagon when it

was introduced back in the s and so

forth people weren't quite sure whether

this is something that would be feasible

but then as people begins researchers

began to study the bandwagon of hybrid

maize it became adopted more and more

and then the useful aspects of hybrid

mase became incorporated in our body of

knowledge to the extent that none of us

who are living here and now think of

hybrid maize as a bandwagon I've been

fortunate to have to spend my career in

plant breeding and genetics during the

time when many new technologies were

coming to the fore for example dr. Fritz

mentioned RFLPs I remember those days

actually about few months ago I was

looking back into isozymes think about

it I so science because we were doing

work in it from the work it show that we

only need markers in maizefor

genomic selection and I said hey we can

do that with isozymes I soon found you

can't order those chemicals for isozymes

anymore so we're back to the more

- DNA markers if you think of a

bandwagon who this is for QTL mapping

for example there in the there usually

is an initial phase of slow development

followed by a phase of rapid excitement

and funding so in the s for example

QTL mapping was very popular and then it

goes down a little bit because we

realized what the technology can do and

what it cannot do and so we have this

bit of disillusionment until we come to

an equilibrium we know this is a mature

technology we know how to use it and as

we might become discouraged while that

is going on the fortunate thing is that

a new bandwagon comes along and it holds

our attention for a while and so we're

in this phase in our work in which new

technologies come along and we need to

critically evaluate what are the pluses

and minuses of this new technology what

are the things it can deliver and what

are the things that we thought it can

deliver but in hindsight it cannot

deliver I've driven some bandwagons

myself and one of which is genomic

selection or genome-wide selection as we

know as plant breeders and geneticists

you have this basic equation in which

your phenotypic value is C equal to your

genotypic value which we cannot see but

we infer and some random noise or

non-genetic effects what we basically do

in genome-wide selection is that we use

a whole bunch of markers spread

throughout the genome and we estimate an

effect associated with each of those

markers and when we take the sum of

those markers that become that then

becomes a surrogate for our genotypic

value so we're simply saying instead of

inferring genotypic values from

phenotypic data we're trying to infer it

from these joint marker effects and so I

work in maize breeding and genetics

we've been at this whole idea of

genome-wide selection in maize for about

a decade now and in in essence this is

how maize breeding looks like these days

you cross an inbred inbred a with inbred

bee breeders who then typically induce

doubled haploid

from f or f plants because it's a

hybrid crop we need to evaluate the

performance of lines in hybrid

combination rather than the performance

of lines by itself and then those lines

that perform best in hybrid combinatio

for example J and K then become released

as new inbred lines and so we asked the

question during which stages of the

breeding program should we apply a

genome-wide selection and a quick answer

is it will be most efficient if you can

use genome wide selection in those

stages of the breeding program in which

phenotypic selection is ineffective I'll

say that again because that is an

important point for any breeding program

you need to look at your breeding

program and see what are those faces in

the breeding program in which phenotypic

selection is ineffective because that's

where you would get the most bang for

your dollar and so selection for hybrid

yield among f plants is totally

ineffective in maize because of hybrid

vigor the performance of an individual

plant is not indicative of its hybrid

performance plus you're looking at a

single individual f plant so the

question is well if we can do phenotypic

selection among f plants during which I

mean genome-wide selection among f

plants how much gain can we get and so

the system we've been working at is if I

have I'm trying to predict the

performance of progeny in the cross a by

B what we can do is go back to our

existing data and find all previous

populations that have a as a parent or B

as a parent and then we pull all of

these prior populations for which data

are already existing with a as one

parent and BS one parent as our training

population to predict the performance of

a by B the nice thing about the system

is you have in it you immediately have a

strong relatedness between the training

population and your test population and

so if you have another cross C by D it

means you find all crosses prior

populations with C as a parent and D as

a parent and you make your

prediction equation again and we found

that when we use this system we

basically get about eighty-five percent

of the gains that we would eventually

get from phenotypic selection to put

that in terms of cost genotyping is

about or percent of the Cross of

the cost of multi location phenotyping

so in this system were able to say that

we can get about % of the gains from

phenotypic selection at about to

percent of the cost and so that along

with other studies in other species has

led us through the conclusion basically

that like phenotypic selection geno

genome-wide selection works it doesn't

work all the time in the same way that

phenotypic selection doesn't work all

the time

genome-wide selection doesn't work all

the time either but if we use it on a

routine basis then it works quite well a

poster child or a poster bull in this

case is this number one bull in the land

from a few years ago the bull's name is

badger fluff Fannie Freddie Badger fluff

funny

Fannie Freddie was predicted to be the

number one bull in the land even before

he got one progeny it was based on

genome-wide predictions and sure enough

when his progeny records came it turned

out that he was indeed the top producing

bull in the land so so now that we can

say that about that that bandwagon there

are new technologies that have come

along such as genome editing

specifically and among others the use of

CRISPR technology that has allowed us

that has allowed scientists to make

precise changes in the DNA of known

genes and so as I thought about this how

could we use CRISPR technology in

breeding for different types of traits

I'm wondering whether there is a strong

parallel between the use of CRISPR

technology and QTL mapping in terms of

application for example in QTL mapping

we have found that QTL mapping is most

effective for those types of traits that

are controlled by few relative

if you QTL and for which these QTL have

a major effect things like disease

resistance plant morphology flowering

date and those sorts of traits in this

case the breeding approach becomes one

of introgression either by pedigree

breeding or by backrests crossbreeding

in other words you could think of these

major qtls like lego building blocks and

if i have five or six or ten lego

building blocks I can put these QTL my

LEGO building blocks into a designed

product and that would be my variety

however if I have a very complex trait

such as grain yield which I am most

interested in by definition these traits

we think are controlled by a lot more

QTL and because they are controlled by a

lot more QTL naturally the effects

associated with each quantitative trait

locus would be smaller and because they

are smaller they're harder to identify

and the effects are different are more

difficult to estimate and even if we

were successful in doing so putting all

of these or QTL together into one

genotype quickly it's a sinner

impossibility and that's why we have

resorted to more of a prediction

approach such as genome-wide selection

as opposed to a design approach with

major QTL and so I'm wondering can we

draw that same parallel then with CRISPR

technology if we knew the genes that we

need to edit and we knew the edits that

we need to make then it becomes more of

a design approach but how do we use

CRISPR technology if we have a trait

such as grain yield or lodging or

maturity for which there are many

unknown genes and so therefore even

identifying the genes to edit would be

would be quite quite challenging so

along these lines I've been thinking of

chromosome of genome-wide marker effects

not for the purpose of selection but

first of all for the purpose of getting

to know our germ plasm better and one

way of doing it is to think of

chromosome wide marker effects it's to

think of a chromosome carried by a line

as a package of inherited units not an

individual

package but a package nevertheless

because it is quite often that a whole

chromosome especially with doubled

haploids would be passed intact from a

parent to an offspring because of the

limited recombination so for example if

I have lined one and I have violet

allelic snips and I have lined one with

eight molecular markers and the marker

effects are indicated there blue would

be positive effects and red would be

negative so one would be the negative of

the other so we look at line two it has

eight chromosome eight markers we add up

the effects of the marker alleles it

carries we say the value of line one on

that chromosome is point two and then

the value of line two on that chromosome

is higher at two and so a student and I

a postdoc and I Andy Thompson have done

this to examine what we call as germ

plasm architecture germ plasm

architecture just means how the alleles

known are unknown for a different

different trader organized or or in the

germ plasm and so in this case here what

we have on on the on the on the x on the

y axis are four different traits and

thesis day down to northern corn leaf

blight and you see the chromosomes

numbered one to ten and on the x axis

are different maze lines about two

hundred seventy one of them and then the

subgroupings population subgroupings b

group misery seventeen group are found

on the top and so this is basically a

heat map it's a heat map if you would of

chromosome white marker effects so if

blue is good and we want blue we could

go down the heat map and look at the

line of interest is there a line for

example that is mostly blue but it had

would have one or two chromosomes in red

and therefore by a chromosome

substitution approach we can come up

with a better line just by substituting

that entire chromosome in that deficient

line where the chromosome from a donor

so again we see this as a way of

organizing

our germ plasm but when I begin to look

at this I go well yes these are

chromosomes but as I said chromosomes

are not indivisible what if we can begin

to divide our chromosomes the way we

want and so therefore this heart can be

back to a slide that I gave at a seminar

about years ago when I wasthinking

of how a car engineer thus does his or

her work and how that differs from our

approach in reading so if you're a car

engineer you do more of a design

approach you take all your knowledge

about cars mechanical engineering

drivers and so forth and you use that

knowledge to come up piece by piece with

a better car however if a breeder were

to design a car the way he or she breeds

plants here's how the breeder would work

a breeder would take two

high-performance cars a Ferrari and a

Porsche would create thousands of cars

that exhibit different Ferrari Porsche

combinations or lamborghini combinations

select the best shuffled car and do that

process again why is that so it's we

could choose the parents to cross the

Ferrari in the poorest Porsche but once

we make that cross between two corn

lines or to wheat lines we cannot really

control what happens there and so

therefore we rely on that natural

shuffling process and we keep on doing

the same thing and selecting the best

shuffle car I think we might be at an

age when we might not have to be limited

to this sort of of

work of going about things because I

think we might be able to start thinking

of what we would call as targeted

recombination targeted recombination

it's the possibility of saying I want a

recombination here and here and here and

here and so how would this work well for

example if I take these two lines and I

have line one it's the same one I had

before with a performance of a point two

and I have line and line and line

one is slightly better than line but

not much better

now we can go and walk through with a

chromosome and say what would be the

value of the progeny on that chromosome

if we can have a recombination at a

certain certain point so if we do that

if we can have a targeted recombination

between markers to and marker three and

recover we recover this particular

doubled haploid we're saying we could

increase performance dramatically in

this example now it's . whatever units

those are compared to the parental

performance of . okay so the question

is well can can we do that

there might be a way of doing that there

was a recent fairly not so recent

anymore but in the May issue of

science there was an article that used

CRISPR technology not for the purpose of

genome editing but for the purpose of

targeted recombination in yeast so this

is yeast it's not corn it's not higher

plants at seized yet they were able to

use targeted recombination successfully

in the sense that you could use the

CRISPR that technology to make

double-stranded breaks and a specific

part of the genome and this the double

double-stranded break would then trigger

a natural mechanism called homologous

response and after a mystery homologous

recombination and after this repair by

homologous recombination the this group

were able to they actually design you

used CRISPR castes to create a mapping

panel as shown on the right side and

this mapping panel had different

recombination points along the different

chromosomes and with this they were able

to map sensitivity to manganese on a

single to a single polymorphic marker on

chromosome so they use thisfor gene

mapping but I'm wondering whether we

could use that same concept not for gene

mapping but actually for targeted

recombination so if you see if you try

to translate how this might work in

crops basically you have CRISPR caste

to induce your double-stranded break you

see you

right to rely on homologous repair hope

it happens there's recombination this is

nearing mitosis by the way not during

meiosis and then you basically need to

recover and regenerate plants that have

the desired recombinations you have are

we here yet absolutely no

the tools the basic tools might be it

might be there but it would require a

complicated workflow to bring this whole

process to fruition a question that we

need to ask though is if if we are to do

this are the gains large enough to

justify investment in this technology

meaning that if this whole process would

lead to only a % or % gain now

there's no point we can get those gains

by using larger population sizes but if

the gains are a lot more substantial

than that say percent or more

then perhaps that's a big carrot it's a

big stick saying yes the games could be

large and therefore if you're a Monsanto

or a pioneer then you might want to

start thinking about developing this

type of technology and so what I've

tried to do and this has been published

in the plant genome journal is try to

determine are the gains for targeted

recombination large enough in maize to

justify the potential development of

this technology and so I looked at two

previous experiments in May so we had at

the University of Minnesota one is a

four by four both by former grad

students and the first one this is the

informated b misery population we

used lines use the markersfrom

patch novels group thanks so much it was

very useful and then the second study

was a mapping panel by Chris Shaffer a

former PhD student we looked at the maze

panel with lines and aboutnearly

nearly , Illumina markers so

basically we used the phenotypic and

marker data estimated genome-wide marker

effects tried to determine where the

combinate where the ideal recombination

point should be and then determine how

much gain would we get if we indeed get

a line at those recombination points and

this is what we found this work here the

wrong button here you go all right I'd

like to focus first on just the first

roll grain yield in kilograms per

hectare this is in the intimated b

misery population for grainyield the

mean grain yield was . or actually

nine eight nearly nine eight point nine

two eight metric tons per hectare if the

breeder were to select the best out of

lines this best out of linesresulting

from random recombination the

the yield of that best out of line

lines is nine point six twometric

tons per hectare compared to the

population mean of eighty nine eighty

two that's about a seven percent gain

. percent gain

now if a breeder were able to induce one

recombination per chromosome at the

right spot on each chromosome one

recombination per chromosome at the

right spot on each chromosome you add up

the gains across the chromosomes because

the chromosomes are independent the

predicted gain is ten point three three

metric tons per hectare . compared

to eight point ninety eight it's about a

percent gain so seven percentgain

with non targeted recombination

percent gain with targeted recombination

that translates to a relative efficiency

using targeted recombination of

percent in other words if we can have

targeted recombination in this

population we could potentially double

the genetic gains for yield I say that

almost nonchalantly I'll say it again we

can double potentially double gains for

yield if we can have targeted

recombination if we can have two

targeted recombination if

two on each chromosome then the games

are even more and you see that fourdifferent

different traits in this

population the relative efficiencies are

also quite high now we need to be

careful in interpreting not only the

relative efficiency but also the amount

of gain because often times the relative

efficiency could be high when the gains

were small take plant height for example

if the gain from non-targeted

recombination is two centimeters and the

gain with targeted recombination is

eight centimeters and that's a four

hundred percent response but eight

centimeters whereas if the gain is you

know ten centimeters versus twenty it's

a percent twice but theamount of

gain is actually larger so we need to

look at both the ratio and the actual

gain now this then is the plot of

targeted recombination for each of the

ten chromosomes for three different

traits so what we're saying is for yield

this triangle here represents the ideal

recombination breakpoint that would

maximize the gains on each chromosome so

it's coded that's chromosome up

to chromosome and so we seehere that

each chromosome has unequal

contributions to the amount of gain that

can be attained the amount of gain was

highest for chromosomes and

it was lowest for chromosome if I if I

recall the numbers right alright so

there is variation among chromosomes

there also is variation among traits so

for example if it's yield here and these

are the targeted recombination

breakpoints for moisture and these are

the targeted recombination breakpoints

and the amount of gain for an

index thatincludes yield moistureand different

lodging traits so a few things to note

we see here that the curve for yield is

a lot smoother than

we have four moisture that shows that

suggests that in this population the the

the favorable alleles for yield are

arranged in larger blocks larger

contiguous blocks for yield than they

are for moisture which is a lot more

jagged second is that if we are to use

recombination targeted recombination we

need to determine beforehand how we

would weight each trait because right

now with with non-targeted recombination

a breeder can just not have any index or

not have any preconceived notion of how

much weight to give to each trade and

just select what the best ones are at

the end of the process if we are to do

targeted recombination we need to

specify beforehand the weights to each

trade because that would drive where the

targeted recombination points would be

next is that this the graph here shows

the targeted recombination break points

for plant height when one line nd is

crossed to several different lines and

we see here that the targeted

recombination points on the same

chromosome chromosome do varyeven if

we use the same parent so again so the a

point here is that this needs to be made

on a case-by-case or a cross by cross

basis what are some considerations that

we would need to have as we as we think

about using this technology well first

of all are there caveats the predictions

need to be effective the prediction

accuracy needs to be high meaning so

that you would have confidence in the

estimates of marker effects and and you

would have confidence in the targeted

recombination breakpoints second is that

the targeted recombination break

points would be only in thosegenomic regions

where recombination already takes place

because the marker effects depend on the

recombination that takes place so if

there are spots in the genome called

spots where this there is no

recombination that would not be a target

that recombination

break point from this process at the

same time if you have that tool for

targeted recombination it's one possible

way to to study what the effect of

recombination is on these existing cold

spots you might need to prioritize

chromosomes but particularly if you have

a lot of chromosomes and the technology

is ineffective as we showed some

chromosomes have a larger contribution

than others so it might be in this maze

example for yield I might want to target

recombination of chromosomes and

maybe a fourth one and then in the

remaining chromosomes just to rely on

random recombination and segregation and

see and see what we get

and lastly although I started out with

CRISPR and I'm now beginning to think

that we actually don't need CRISPR to do

targeted recombination for some species

for that for some species in which

markers are available in doubled

haploids are available just by the

aggressive use of molecular markers and

doubled haploids that we might be able

to create pretty much a the line that

carries a targeted recombination on

several if not nearly all of the

chromosomes just to illustrate this is

from a study published by a student Josh

sleeper we found that in doubled

haploids

then this is maize again with

chromosomes the mean number of

chromosomes of recombinations per

chromosome it's about . it does vary

from on chromosome which which

would be the largest and chromosome

has actually a lot fewer recombinations

at least in this population and so a

possible approach would be first step

one is we develop a library of we

develop a library of lines that would

have a target and recombination on

chromosome another strainwith a

target and recombination on up to

and then the next step wouldbe just

by breeding and back Rossing to assemble

these into one line and that is possible

because the there is a fairly good freak

see of a chromosome passed intact from

parent to offspring so for example

chromosome two nearly fifty percent of

the time it's passed intact from parent

off spring so building that library by

introgression would be a possibility and

so for this perhaps some even some

operate operations research type of

approach would be needed to see what's

the fastest way towards this product

given given were where we stand now if

you're working on sugarcane with

chromosomes or so then it's probably not

gonna work so you you gotta do something

else

so with that just to summarize what I've

shared with you this morning

I think a next bandwagon could be the

use of CRISPR not for gene genome

editing but for targeted recombination

the point being that with genome wide

marker effects we can identify what the

recombination points would be for each

trait we could potentially with this

technology potentially double the rates

of yield gains in case you didn't get

that the previous point is a big deal

especially as we're looking at feeding

an increasingly large population by

and I think I'm beginning to think that

we we probably might be able to

accomplish this at least in

maize orspecies with fewer chromosomes like

barley with seven not with CRISPR but

with a straight breeding methodology

however much work is needed and so I'm I

hope to be working in this field in the

next several years with that thank you

so much for your time and attention if

there is time for questions I'd be

pleased to try to answer them this

morning

okay any questions Rex that was a

absolutely fabulous idea I really like

that that's could be a game-changer as

you say now of course that's the

reaction I should have at the beginning

of the bandwagon I guess but I really

that that's very exciting I liked what

you're doing in terms of looking at

these historical data sets what we've

also seen though in corn breeding is

that there have been one key in bread

comes along and it changes the whole

thing do those contain rare

recombination events consistent with

this hypothesis yeah that's a great

question I've been thinking about that

what makes what makes a line or a hybrid

a superstar if you would it's this

what's special and it might be that if

we're talking about it could potentially

double gains maybe that line just

happened to have some of those are close

to those that I would contribute to that

being just head and shoulders above the

rest of its sister lines in the same

cross it's a great point and then if I

could ask for your feedback on a comment

so at one point I was talking to a

breeding company and I suggested that as

they start genotyping all their lines

that they save those lines with rare

recombination events in anticipation

that someday they might be useful any

thoughts about that

I think I I think I think that's a good

point

save it saving those who especially with

the rail recombination events of events

I think that's an excellent point it's

it's linkage is a it's an interesting

thing because linkage of course as we

know it can release hidden genetic

variation but it can also preserve good

good blocks of alleles and so there's

been a long debate in in corn breeding

do we want to induce do we want more

recombination or do we want less

recombination I think an answer is well

we want recombination where it's helpful

and we don't want recombination where

it's not helpful and I think we can

begin to start thinking of that question

and then save whatever recombinant

survey recombinants we

that we might come across more questions

please okay excuse me thank you as we

know their big difference between mass

and weight we works on weight it's more

complicated so what small concentrations

for weight so what wheat has pairs is

all right so yes let go with it's

more difficult than Lego and if you

think of chromosomes as lego building

blocks so it might be the case where you

would need to in a species like wheat or

sugarcane or large numbers of

chromosomes you might need to prioritize

your chromosomes and and focus on those

a nice thing though is if you're trying

to build an ideal genotype if you would

with the recombinations and if you're

doing that on a piece by piece basis

your progress is not all in the end that

as you're building accumulating those

recombinants you're already making you

hope progress and so you could track

that and see if are you indeed making

progress during that time so again it's

you think of it that the way the way we

work now is we use molecular markers and

and phenotypic data to select which ones

are the best out of what happens

randomly and now we're shifting it to

say well we're using this information to

actually try to select the best

recombinations and again yeah there are

challenges in in in species with large

numbers of chromosomes if you're in

mosquito with three chromosomes it

should be a piece of cake okay more

questions we have enough time

yeah my question is how this new

technology is applicable to the orphan

crops which they didn't have the luxury

of genome sequence information available

data so yes so obviously for this

technology to work you need to have

cheap and abundant molecular markers and

you need to have good phenotypic data if

you have if you don't have both of those

then you cannot use this technology and

so I don't have a quick answer for how

it would work in an orphan species but

the fact of the matter is and the

reality is you need both phenotypic data

and marker data to do this great talk a

question about your modeling and the

gain made so have you thought about

modeling out then if you were able to

select that directory combination that

you wanted what would be the effect on

antagonistic traits such as maturity

moisture plant height loss of test

weight things like that yes and so so

that that's when you could do that so so

you could say if I'm going to select for

just yield you also have the marker

effects for the associate to trade so I

say okay if I have a recombinant

recombination event on this between

markers and for yield you could

easily ask the question how would that

affect moisture and how would that

affect test weight and as I showed in

that graph that the recombination points

differ from trade to trait and therefore

you need to a priori have some idea of

how you're going to weight your traits

because that's going to affect where youwould like to target so it's so you

can't sit back anymore and say I'm going

to wait for the results to come before I

give weights to each trade it's more you

need to do that earlier on the process

before the data come in any more

questions

if not let's thank dr. Bernardo one more

time so much so this is a time for our

morning break but before we go we have a

few announcements there are restrooms in

the east and west side of the building

and at this point we also start

（根据软件自动整理的，难免有误）

（以下为广告）