►
Description
NERSC Data Seminars Series: https://github.com/NERSC/data-seminars
Abstract: Direct physiological observation of subcellular dynamics is now feasible using the Lattice Light-Sheet Microscopy. It is a transformative imaging technology that spans the relevant scales in space and time because of the wide resolution range and large volumetric acquisition capability. This new imaging method allows us to record dynamics at a scale of nanometers and milliseconds, determine their consequences at a scale of microns and hours, and visualize their long-term outcome at a scale of up to several millimeters over several days. I’ll present our past work on combined lattice light sheet microscopy with adaptive optics to achieve (Science, 2018), across large multicellular volumes, noninvasive aberration-free imaging of subcellular dynamics in vivo. Next, I’ll discuss the combination of lattice light-sheet with the physical expansion of samples (Expansion Microscopy) that enables scalable super-resolution volumetric imaging of large tissues (Science, 2019) including the complete fly brain, columns of mouse brain – datasets spanning several hundred terabytes. Finally, I will introduce our next-generation microscope design– dubbed the “Swiss army knife microscope”, which combines at least ten different modes of imaging with integrated light paths. In essence, this new microscope is designed to seamlessly switch between modes of imaging in order to alleviate the tradeoffs related to resolution, speed, invasiveness and imaging depth, which precludes any single optical microscopy to function optimally for a diverse set of biological specimens.
A
So
I'm
sure,
and
from
crd
and
welcome
to
this
week's
c
cs
nurse
data
seminar,
today's
speaker
is
dr
gokulo.
Padilla
and
goku
is
an
assistant
professor
in
the
department
of
molecular
cell
biology
at
uc
berkeley,
and
he
is
also
the
scientific
director
of
the
advanced
bio
imaging
center.
He
did
his
phd
at
uc
riverside
and
did
his
post
doctoral
research
at
harvard
medical,
school
and
boston's
children's
hospital.
A
B
Don't
get
me
wrong.
We've
learned
a
lot
this
way,
but
it's
hard
to
piece
together
how
the
rules
of
the
game
work
from
just
a
couple
of
snapshots
and
same
thing
applies
to
life
and
much
of
what
we
know
about
how
life
works
comes
from
scientists,
scientists
being
able
to
piece
together
from
a
lot
of
different
source,
these
fuzzy
and
fragmented
snapshots,
so
that
they
can
understand
the
rules
of
of
the
fundamental
unit
of
life,
which
is
the
cell.
So
how
can
we
watch
this
particular
cell
to
understand
how
it
works?
B
So
before
I
kind
of
dive
a
lot
more
into
detail,
I
just
want
to
kind
of
give
you
a
intuitive
understanding
of
of
what
it
is
that
is
unique
about
this
imaging
modality.
So
you
can
take
the
cell
on
this
coverslip,
for
example,
and
if
I
want
to
image
any
of
the
dynamics
of
the
cell
on
the
attached
surface
at
the
interface
of
the
coverslip
turf
microscopy
is
probably
the
best
tool.
B
But
if
I
want
to
image
anywhere
other
than
the
bottom
or
the
next
to
the
coverslip.
I
need
to
do
some
sort
of
a
three-dimensional
sectioning
of
this
particular
cell,
and
we
do
this
by
focusing
light
using
a
confocal
scanner
where
it
focuses
a
point
of
light
at
the
depth
of
focus,
and
we
can
move
the
depth
of
focus
up
and
down
to
be
able
to
capture
the
three-dimensional
volume
of
the
cell.
B
And,
of
course,
you
can
multiplex
this
in
something
called
a
spinning
disc,
where
you
have
multiple
pins
of
pinholes
of
light
and
we're
basically
able
to
multiplex
and
speed
up
the
capturing
of
of
this
particular
two-dimensional
section.
So
one
thing
you
should
have
noticed
is
that
for
every
single
two-dimensional
image
we
collect
this
way,
we're
putting
in
light
the
entire
volume
of
the
cell.
So
basically,
if
we're
just
focusing
the
light,
but
the
light
goes
through
the
entire
cell.
B
This
has
been
transformative
for
developmental
biology,
where
you're
able
to
observe
most
of
the
the
high
level
processes
at
the
cellular
level
resolution,
meaning
that
you
can
look
at
where
the
cells
are,
how
they're
moving,
but
it's
very
difficult
to
be
able
to
observe,
what's
happening
in
the
inside
the
cells,
and
this
changed
more
or
less
with
the
lattice
slice
sheet
and
the
way
we
did,
that
is
by
basically
creating
a
super
thin
sheet
of
light.
That's
almost
entirely
confined
within
the
depth
of
focus.
So
you
have
great
contrast.
B
You
basically
don't
waste
any
of
your
photon
budget,
putting
a
light
into
the
into
the
into
the
cells
and
just
to
kind
of
give
you
an
intuitive
sense
as
to
what
this
looks
like.
So
this
is
a
cross-section
of
a
single
beam
inside
of
the
lattice
slidesheet.
A
single
beam
is
called
a
vessel
beam
in
this
case.
B
Of
course,
this
still
doesn't
look
like
a
light
sheet
from
its
cross-section.
So
what
we
do
is
we
just
move
a
mirror
fast
enough
faster
than
how
we
take
a
picture
in
order
to
blur
that
to
create
an
artificial
light
sheet,
and
of
course,
if
you
want
super
resolution
instead
of
sweeping
it
back
and
forth
very
fast,
we
can
take
advantage
by
moving
the
mirror
very
slowly
to
be
able
to
take
advantage
of
modulations
in
the
signal
for
super
resolution,
light
sheet
imaging.
B
These
molecules
tend
to
accumulate
to
self-assemble
into
a
basket
and
at
the
end
of
this
particular
process,
they
basically
bring
in
anything
any
food,
any
reagents,
any
any
molecules
from
outside
the
cell
into
the
cells
by
breaking
free
from
from
the
plasma
membrane,
and
that's
basically
what
we're
observing
in
this
particular
case.
We're
watching
this
particular
process
and
you'll
basically
see
new
spots
appear
and
disappear,
but
now
watch
what
happens
if
I'm
imaging,
at
the
same
ending
signal
to
noise,
same
signal
to
noise
watch
what
happens
basically
within
about
30
stacks.
B
Most
of
my
signal
goes
away
in
my
spinning
disc,
because
the
amount
of
power
that
I'm
putting
in
is
more
than
an
order
of
magnitude
more
in
terms
of
total
light
dose.
That's
been
delivered
to
the
cell
in
the
top
versus
the
bottom
as
a
consequence,
you're
damaging
the
basically
the
fluorescent,
proteins
and
you're,
basically
making
them
go
dark,
so
we're
no
longer
able
to
observe
them
using
a
fluorescence
microscope.
B
Anything
that
is
produced
in
the
cells
needs
to
be
more
or
less
trafficked
and
shipped
to
the
right
place,
so,
basically
we're
looking
at
okay.
How
do
basically,
how
does
the
fedex
work
inside
the
cell,
fedex
and
ups?
So
to
speak?
All
of
the
examples
of
four
examples
I've
talked
about
so
far.
These
are
of
single
cells
on
a
piece
of
glass
and
we've
been
able
to
observe
them
with
great
level
of
detail.
B
If
you
really
want
to
be
able
to
look
at
something
in
extraordinary
detail,
can
we
develop
a
technology?
That's
scalable
in
terms
of
super
resolution,
because
most
super
resolution
techniques
are
are
limited
to
pretty
small
volumes
and
mostly
two-dimensional
in
in
in
many
cases.
So
we
want
to
be
able
to
do
three-dimensional
super
resolution.
That's
scalable
to
not
just
cells
but
to
large
tissues
and
I'll.
B
Not
just
you
know,
collect
these
videos,
but
we
put
in
a
lot
of
effort
in
trying
to
create
tools
to
be
able
to
mine
this
data,
because
it
tells
us
something
super
important
like,
for
example,
when
we're
observing
these
these
eating
processes.
So
to
speak.
So
each
spot
is
more
or
less
an
individual
process
of
how
the
different
regions
of
the
cells
are
eating.
You
see
a
lot
of
heterogeneity
and
how
they
how
they
actually
behave.
B
So
if
I'm
basically
trying
to
quantify
this
we're
looking
at
okay
first,
how
long
do
these
spots
last,
so
you
have
a
huge
heterogeneity
between
you
know
short-lived
guys
versus
long-lived
guys,
and
we
have
very
similar
behaviors
between
the
top
of
the
cell
versus
the
bottom
of
the
cell,
which
is
cool,
but
this
really
tells
us
in
terms
of
how
particles
get
in
this
has
to
do
with
well.
In
this
particular
case,
these
objects
may
be
a
borders,
meaning
that
they
started
to
build,
but
more
or
less.
B
They
had
no
signal
or
no
components
that
got
captured
into
these
vesicles,
so
they
may
have
aborted
so
most
likely
the
short-lived
guys
are
called
abortives
and
the
long-lived
guys
there's
one
of
two
reasons
why
they
last
this
long
one
is
memory,
tension,
meaning
that
the
membrane
is
so
tense
that
it
takes
them
a
lot
of
force
that
they
need
to
generate
in
order
to
create
the
curvature
in
order
to
cause
the
the
vesicle
to
to
close.
The
other
reason
we
see
this
is
when
something
large
gets
stuck
like.
B
For
example,
a
virus
is
hijacking
this
particular
machinery
to
get
into
the
cells.
In
that
case,
the
the
lifetime
of
these
events
increases
as
well,
so
we're
able
to
basically,
you
know,
observe
over
250
000
of
these
events
to
basically
create
like
one
of
the
first
kind
of
robust
and
and
and
and
detail
kind
of
descriptions
of
okay,
how
these
objects
behave
as
a
cohort.
B
Now
that
was
a
single
cell
and
a
piece
of
glass,
just
one
cell,
so
goal
is
okay.
We
want
to
go
more
physiological,
and
this
is
the
work
that
got
published
a
couple
of
years
ago
and
we
started
with
a
collaboration
with
david
drewbin's
group
and
and
our
calculus
group
at
at
the
uc
berkeley
in
the
mcb
department,
and
what
they've
generated
are
these
organoid
membranes,
which
are
basically
labeled
in
in
red,
for
the
same
molecules
that
I
was
showing
you
in
the
previous
slide?
B
As
you
can
see,
you
don't
see
those
clear
spots
anymore
and
there's
a
a
reason
for
this,
because
any
of
the
light
that's
kind
of
coming
through,
which
is
the
yellow.
That's
the
excitation
because
of
the
curvature
of
the
sample.
That's
basically,
first
bent
and
second
as
it's
propagating
through
to
the
region
where
we're
imaging,
which
is
the
red
box
there
it's
getting
scrambled.
B
The
light
is
getting
scrambled
anything
you
do
manage
to
excite
you'll
basically
have
the
blue
light
going
back
to
the
detection
objective
that
is
also
getting
scrambled,
but
in
a
slightly
different
way,
because
it's
taking
a
different
path
to
the
objective
and
because
the
yellow
light
bent
relative
to
the
depth
of
focus.
These
two
objectives
are
no
longer
in
focus.
That
means
that
we
need
to
fix
both
the
excitation
scrambling
problem,
the
detection,
light
scrambling
problem
and
the
focus
between
the
two
objectives.
B
So
how
do
we
do
that?
So
I've
shown
you
the
cross-section
of
what
an
ideal
light
sheet
looks
like
this
is
what
we
start
with,
and
we
measure
okay.
How
does
it
aberrate,
so
you
can
see
that
the
light
sheet
is
starting
to
bloom
a
little
bit.
That
means
that
you're
no
longer
putting
in
the
same
flux
of
energy
into
the
same
region.
B
So
we
need
to
correct
this
and
the
way
we
do
that
is,
we
use
something
called
a
shack,
hartman
wayfriend
sensor.
All
that
does
is
it
measures
the
aberration
in
fourier
space,
and
then
we
subtract
that
and
then
inverse
transform
and
put
it
back
into
something
called
a
spatial
light
modulator,
which
is
basically
what
an
lcd
projectors
chips
are
and
we're
able
to
recover
the
lost.
I
guess
recover
from
the
aberration
of
the
the
excitation
side
for
the
detection
side,
it's
slightly
different.
B
What
we
do
is
we
use
something
that
astronomers
have
developed
over
50
years
ago
actually,
and
we
use
something
called
a
guide
star
approach.
So
the
problem
is
from
astronomers
who
figured
this
out.
They
were
trying
to
observe
the
light
from
distant
galaxies,
coming
basically
more
or
less
unaltered
up
until
it
hits
our
upper
atmosphere
in
the
stratosphere
more
or
less
you're,
starting
to
scramble
from
the
the
the
humidity
from
basically,
you
know
all
of
the
the
clouds
and
our
atmosphere
is
basically
scrambling
the
light
causing
it
to
look
pretty
blurry.
B
So
what
they
do
is
basically
they've
been
able
to
generate
something
called
a
guide
star.
So
right
near
the
region
of
the
star
that
they're
trying
to
image,
they
basically
shoot
a
laser
up
from
the
ground-based
telescope
to
create
a
a
guide
star
by
exciting
sodium
atoms
in
our
stratosphere.
So
the
strat,
basically
in
the
same
region
as
to
the
region,
that,
as
as
the
stars
that
they're
trying
to
observe
and
what
they're
doing,
is
they're
basically
seeing
okay.
B
How
does
the
light
from
this
guide
star
come
back
through
the
atmosphere
and
what
they
can
do
is
they
have
this
deformable
mirror
element
that
can
physically
change
its
shape?
To
compensate
for
this,
you
know
aberrated
wavefront,
to
be
able
to
correct
it,
and
we
have
this
like
feedback
loop
system
and
by
doing
this
particular
correction
they're,
basically
able
to
recover
the
loss
resolution
so
they're.
B
The
astronomers
have
been
doing
this
for
for
many
decades
now,
and
what
we've
done
in
the
in
the
past
a
few
years
is
able
to
adapt
this
into
the
light
of
the
slicing
microscope,
and
when
we
do
this,
I'm
sorry
before
I
get
to
that.
So
the
last
problem
is
the
is
the
focus
correction?
So
what
we
do
is
we
shine
light
between.
B
So
that's
the
difference
between
a
a
conventional
microscope
versus
what
we
can
do
is
now
we're
able
to
correct
for
these
aberrations
using
adaptive
optics,
and
by
doing
so,
we
now
enabled
again
the
quantitative
abilities
of
of
being
able
to
observe
more
complex
samples
and
start
to
kind
of
start
studying
them.
So
this
is
a
complex
sample
in
the
context
that
it's
a
three-dimensional
cell
cluster,
but
it's
still
not
a
living
organism.
B
So
we
took
this
one
step
further.
We
basically
are
able
to
apply
this
to
imaging
zebrafish,
so
zebrafish
is
is
awesome
because
they're
vertebrates
just
like
humans,
but
unlike
humans,
they're,
mostly
transparent.
That
means
that
they
don't
scatter
or
block
the
light,
but
they
do
aberrate,
because
the
refractive
index
of
each
of
the
cells
is
slightly
different
so
that
basically,
that
bends
the
light
in
in
slightly
different
ways.
B
This
is
a
single
cell
that
I
showed
you
early
on
now.
Basically,
you
can
start
to
imagine
our
data
production
rates
and
our
computability
problem
basically
compounds
a
little
bit,
but
before
I
kind
of
dive
deep
into
the
those
problems,
I
want
to
kind
of
give
you
a
flavor
as
to
what
it
is
that
we're
able
to
observe
now
so
now
that
we're
not
just
interested
in
the
cellular
dynamics,
but
we're
more
focused
on
the
subsidiary
dynamics
in
these.
B
So
this
is
a
cool
observation
that
we
were
able
to
make
really
inside
a
living
organism.
That
was
that
was
never
observed
before
so
we
can
also
look
at
not
just
you
know,
diffraction
limited
spots,
which
are
those
little
spots
that
you
were
looking
at
for
the
clathrin,
but
more
complex,
shapes
part
of
the
problem
with
this
is
how
do
you
visualize
this?
B
This
is
like
one
of
the
our
bigger
problems
is
that
this
particular
movie
that
you're
looking
at
is
maybe
about
200,
gigabytes
or
300
gigabytes,
and
there
are
again
each
single
volume
is
relatively
small,
but
we
have
three
different
channels
with
multiple
time
steps
so
to
be
able
to
visualize
this.
What
we've
done
is,
instead
of
looking
at
a
slab
by
slab
basis.
We
now
basically
try
to
augment
the
visualization
by
bringing
in
some
of
the
analysis,
tools
that
we
have
so.
B
So
you
can
see
the
precursor
neuronal
cells
in
the
back.
You
can
see
those
columnar
structures,
the
mitotic
cells
are
these
round
blobs
there
and
the
skin
cells
on
the
top.
Are
these
amorphous
blobs
in
a
second
I'll
turn
off
everything,
except
for
one
cell?
So
we
can
look
at
it
in
a
lot
more
detail,
but
ideally
we
can
basically
do
this
with
any
cell
in
this
particular
tissue
volume.
B
So
as
I
play
the
movie,
you'll
also
see
okay,
how
the
volume
of
the
cell
changes
and
how
the
surface
area
of
the
cell
changes
and
this
particular
tissue
we're
imaging
basically
in
the
brain
of
the
zebrafish
at
about
14
hours,
post
fertilization,
so
more
or
less
the
organelles
are
evenly
distributed
when
they
started,
and
the
cell
was
going
to
do
in
just
a
few
minutes.
So
to
speak.
Is
it's
going
to
divide?
B
B
The
other
cool
thing
is
that
we're
able
to
observe
things
that
were
more
or
less
never
seen
before.
So
this
is
an
example
of
of
one
of
those,
those
examples
where
you're
basically
able
to
look
at
not
just
the
cells,
but
the
behavior
of
this.
So
in
this
case
I'll
show
you
one
example
of
something
called
cascading
divisions
that
we've
never.
You
know
really
been
able
to
see
before,
because
this
is
now
a
product
of
cells
communicating
with
their
neighbor.
B
So
so
these
are
the
types
of
examples
of
of
novel
behaviors
that
we're
able
to
observe
as
a
consequence
of
of
observing
life
in
its
native
state.
So
everything
I've
shown.
You
is
a
single
field
of
view.
I
want
to
now
be
able
to
image
large
fields
of
view
like
as
in
we
want
to
have
the
same
amount
of
sensitivity
resolution,
but
I
want
to
be
able
to
look
at
say
an
entire
organ,
for
example.
B
So,
for
that,
it's
a
more
or
less
a
again,
the
microscopy
challenges
that,
if
you
don't
use
adaptive
optics,
if
I'm
looking
at
the
regions
in
the
the
orange
that
are
zoomed
in
and
or
the
blue,
that
are
zoomed
in
it
kind
of
looks,
more
or
less
fuzzy.
We
can
use
adaptive
optics,
but
we
need
to
now
say
you
know,
only
apply
it
to
a
small
region
which
is
the
the
box
in
orange.
B
So
if
you
measure
the
aberration
there
and
you
apply
and
you
correct
it-
you're
able
to
now
see
features
that
you
weren't
able
to
before
without
adaptive
optics.
But
if
you
apply
the
incorrect
correction,
it's
possible
to
make
things
much
worse
than
they
were
without
any
correction,
which
is
the
blue
box
in
this
case.
So
it's
important
to
recognize.
When
do
you
need
to
correct,
and
when
do
you
not
need
to
correct?
B
B
So
you
can
see
that
there's
aberrations
that
are
basically
regionally
dependent,
meaning
that,
on
the
left
side
of
the
movie,
you
have
much
worse
aberrations
than
the
right
side
of
the
of
the
particular
volume,
and
this
is
the
excitation
light
sheet.
If
I'm
looking
at
the
same
thing
in
the
detection,
it's
different
because
again
they're
taking
two
separate
light
pads
in
order
to
get
to
where,
where
to
be
able
to
collect
the
image.
B
So
that's
the
reason
we
need
to
break
this
down
into
smaller
chunks
that
we
are
able
to
correct
and
then
stitch
it
back
together.
So
it's
again
another
computational
challenge
after
we've
acquired
the
data.
So
in
this
particular
case
again
looking
at
the
organelle
dynamics.
This
is
the
field
of
view
for
one
such
region
of
interest.
B
B
So,
in
a
second,
what
I'll
do
is
basically
show
you,
the
volume
representation
and
then
take
it
completely
apart,
like
we've
done
in
the
other
data
sets
just
so
that
we
can
kind
of
evaluate
the
morphology
of
each
of
these
cells
in
their
context.
So,
basically
you
can
look
at
the
the
the
retinal
cells,
those
are
the
column,
columnar
cells
and
towards
the
edge
of
the
movie.
On
the
left
side,
you
can
basically
see
the
skin
cells.
Those
are
those
flat
blobs.
B
B
So
you'll
see
these
amorphous
blobs
so
from
these
type
of
data
sets.
What
we're
basically
trying
to
understand
is,
for
example,
how
do
you
differentiate
different
cells
to
know
whether
they're
cells
that
belong
in
the
cornea,
the
cells
that
belong
in
the
retina
or
the
skin?
So
when
we're
looking
at
these
cell
divisions
as
the
cells
are
starting
to
divide,
you'll,
see
that
these
these
retinal
cells,
that
that
are
more
or
less
kind
of
attached
to
both
sides
of
this
particular
apical
and
basal
retina?
B
B
B
In
the
last
example
of
this
particular
data
set.
What
I
want
to
show
you
is
how
we
can
observe
cells
moving
again
in
a
living
organism,
so
here
we're
imaging
cancer,
so
green
cells
are
human
breast
cancer
cells
that
we've
injected
into
the
zebrafish
to
understand
how
they
move
around
to
understand
metastasis
of
cancer.
So
in
this
case
we
can
see
two
examples
of
of
these
cells
in
one
case
on
the
left.
It's
squeezing
through
these
blood
vessels
that
are
in
magenta.
B
So
to
speak.
So
that's
how
they
they
start
to
to
metastasize
from
one
region
of
the
body
to
a
different
region
of
the
body,
and
we
know
that
it's
trying
to
get
out
of
this
particular
blood
vessel,
because
if
you
track
the
surface
area
outside
of
the
blood
vessel,
it
increases
over
a
course
of
say
two
hours
by
almost
50.
It
increases.
B
B
Another
example
we
can
look
at
is
how
does
your
immune
system,
you
know
basically
move
to
to
fight
infections,
so
you
can
see
these
immune
cells
crawling
around
and
what
we've
done
is
we're
trying
to
understand?
Okay,
how
the
immune
cells
are
sensing,
their
environment.
So
what
we've
done
is
we
injected
this
particular
fish
with
a
fluorescently
labeled
sugar,
which
is
dextran
in
this
case
and
you'll
see
that
in
this
one
case
this
particular
cell
doesn't
necessarily
sense
its
environment
but
there's
another
neutrophil.
B
That's
crawling
around
with
more
or
less
you
know:
blue
gems,
in
its
belly,
so
to
speak,
because
that's
constantly
sensing
its
environment
to
understand,
okay,
where
it
needs
to
go
and
then
more
or
less.
This
is
a
neutrophil.
That's
trying
to
understand!
You
know
what
whether
or
not
there
is
a
existential
threat
to
to
this
organism
and
in
this
in
this
again
the
same
field
of
view.
The
amazing
thing
is
that
we're
able
to
observe
about
11,
different
cell
types
and
observing
basically
what's
happening.
B
So
you
can
see
the
immune
cells
crawling.
We
can
see
how
a
pressure
relief
valve,
which
I'm
not
going
to
talk
about
right
now,
but
how
that's
basically
kind
of
assembling
itself
a
mitotic
endothelial
cell
to
basically,
you
know,
extend
your
your
blood
vessels.
You
have
the
hind
brain
on
the
bottom
to
be
able
to
understand
how
those
neurons
are
are
behaving
and
all
this
in
a
single
field
of
view.
B
That
means
that,
in
with
the
objectives
and
the
numerical
apertures
that
we're
using,
we
have
the
ability
to
look
at
subcellular
structures
inside
of
these
transparent
organisms
and
with
the
right
reagents.
We
really
do
have
closest
single
molecule
sensitivity,
so
that's
more
or
less
the
the
advance
in
microscopy
in
terms
of
the
live
imaging.
B
B
B
B
If
you
use
something
called
a
spinning
disc,
you
more
or
less
bleach
out
most
of
your
volume,
even
before
you're
done
imaging
and
with
the
lattice
we
kind
of
circumvent
both
of
these
problems
because
we
have
low
phototoxicity
and
we
have
high
speed
so
we're
basically
able
to
image
a
a
millimeter
cube
within
you
know
in
in
in
about
a
week
or
so.
So
what
does
this
look
like?
B
So
if
I
zoom
into
one
of
these
regions
with
the
the
neurons,
we're
able
to
basically
observe
a
lot
of
the
the
details
that
we
weren't
able
to
see
before
we
just
diffraction
limited
imaging,
so
in
this
case
we're
able
to
look
at
all
the
compartments
within
this
particular
neuron,
because
we're
labeled
everything
except
for
the
component,
so
we're
segmenting
the
voids
and
we're
able
to
you
know.
Of
course
you
know
quantify
the
the
shapes
of
these
things.
B
You
know
look
at
the
volumes,
the
major
minor
axes,
we're
able
to
understand
okay.
Well,
what
of
these
particular
structures
are
what
com
compartments
so
we're
able
to
identify,
which
ones
are
mitochondria,
which
ones
are
our
lysosomes
and
then
we
can
then
start
to
look
at
different
regions
of
the
of
the
neurons.
So
in
this
case
we're
looking
looking
at
the
dendrites,
the
axons
and
the
somas,
meaning
those
are
the
different
structures
of
the
neuron
and
we
can
look
at
okay
how
these
the
shapes
vary.
B
So
in
this
case
you'll
see
that
in
the
axons.
Basically,
that's
that
transmitting
the
electrical
signals
in
the
in
the
brain.
There
lack
large
mitochondria.
They
only
have
relatively
smaller
mitochondria
compared
to
the
the
soma,
which
is
the
body
of
the
neuron
or
the
dendrites.
The
extensions
of
these
of
these
neurons
and
that's
one
field
of
view.
That
tells
us
you
know
information,
but
it
doesn't
give
us
everything
we
need
to
know.
So
we
need
to
be
able
to
image
much
larger
fields
of
view.
B
So
in
this
case,
we
went
out
to
map
an
entire
cortical
column,
which
is
that
blue
region
in
the
in
the
schematic.
It's
just
that
little
stripe
right
there,
so
we're
basically
imaging
that
particular
volume,
but
it
more
or
less
took
us
about
60,
000,
volumetric
tiles
to
stitch
together.
This
particular
volume
imaging
from
the
surface
all
the
way
to
the
the
white
matter
and
in
that
small
strip.
B
So
this
we
were
able
to
image
again
with
zipper
resolution
and
this
particular
data
set
took
about
17
hours
to
image.
But
we're
able
to
you
know
more
or
less
visualize
the
entire
somatosensory
cortex
column
in
this
particular
case,
and
to
give
you
a
sense
of
what
the
data
looks
like
so
we're
looking
at
the
pre
and
the
post-synaptic
positions
in
this
particular
data
set,
those
are
the
magenta
and
the
blue,
and
I'm
going
to
turn
off
everything.
B
That's
not
strictly
associated
with
the
neurons
that
we're
directly
observing
so
here
you're
only
looking
at
10
of
all
of
the
neurons.
So
that's
the
reason
it
looks
more
or
less
empty
because
we're
only
looking
at
one
out
of
every
10
neurons
that
are
there.
So
the
complexity
is
more
or
less.
You
know
mind-boggling,
to
say
the
least,
but
the
cool
thing
here
is
that
you
can
see
that
the
blue,
the
cyan
and
the
magenta
are
juxtaposed
next
to
each
other.
B
So
those
are
the
pre
and
the
postsynaptic
regions,
and
what
cool
thing
is
is
that
in
this
particular
data
set,
we
were
able
to
measure
the
three
dimensional
distances
between
those
two
markers
to
understand.
Okay,
how
do
you
transmit
the
information
from
one
axon
to
more
or
less
one
of
the
the
spines?
That's
that's
that's
there
and
in
this
one
data
set
we're
able
to
observe
more
than
a
hundred
times
the
data
points
that
were
more
or
less
reported
in
all
of
literature.
B
So
we're
able
to
look
at
about
25,
000
thousand
of
these
objects
and
be
able
to
more
or
less
extract
out
their
different.
I
mean
their
distances
and
then
we're
able
to
now
understand
a
subset
of
how
they
kind
of
you
know
differ
in
different
in
different
regions
of
the
in
different
regions
of
the
brain.
B
So
this
is
again
a
small
section
of
an
entire
mouse
brain,
but
that's
not
giving
us
the
entire
picture.
So
what
we
decided
to
do
is
go
smaller
in
terms
of
the
organism,
so
here
we're
imaging
flies
and
the
fly
brain
that
we're
imaging
here
is
roughly
the
same
size
as
the
volume
that
I've
that
I've
imaged
in
the
in
the
mouse
brain.
B
But
the
cool
thing
is
that
we're
able
to
capture
the
entire
fly
brain
so
in
this
case
we're
imaging
the
entire
fly
brain,
but
we're
only
looking
at
the
dopaminergic
neurons
and
we're
looking
at
the
synapses
that
are
associated
with
these
dopaminergic
neurons.
And
the
amazing
thing
about
this
technology
is
that
people
have
tried
to.
You
know
image
the
synaptic
positions
in
the
fly
brain
of
very
specific
regions.
So
this
is
called
the
mushroom
body
which
is
responsible
for
memory
and
learning
in
flies.
B
It
took
them
about
a
year
and
a
half
to
do
the
three-dimensional
em
imaging
of
this,
and
then
it
took
them
an
additional
ten
thousand
human
hours
to
be
able
to
annotate.
These
data
sets
to
be
able
to
display
the
positions
of
every
single
synapse
in
this
particular
region
of
the
brain.
So
that's
roughly
about
two
years
of
work
to
map
one
one
hundredth
of
the
flybring.
B
So
if
the
question
is
to
map
the
synaptic
positions,
we
can
have
molecular
markers
to
be
able
to
image
them.
But
what
we
do
is
before
we
know
whether
it
works
in
our
case
or
not.
We
mapped
in
this
particular
really
dense
region,
what
the
distribution
of
these
synapses
and
we
find
that
it's
more
or
less
about
250
nanometers,
the
mode
of
these
of
this
distribution,
and
we
have
better
than
about
60
nanometer
resolution.
So
we
really
are
able
to
capture
almost
all
of
these
and
differentiate
them
as
different
synaptic
positions.
B
So
by
using
again
our
combination
of
expansion
and
lattice
we're
able
to
now
not
just
map
just
in
a
small
region,
but
we
were
able
to
map
synapses
in
the
entire
flybring.
So
that's
about
40
million
synapses
that
have
been
counted
directly
from
this
fly
brain
and
we
were
able
to
do
this
in
more
or
less
under
a
week.
So
to
speak
and
the
cool
thing
is
we're
not
just
able
to
do.
B
B
B
So
with
that
I'm
going
to
stop,
and
I
do
want
to
kind
of
go
over
some
of
the
challenges
and
the
technical
parts
in
terms
of
the
data.
But
before
I
do
everything
I've
shown,
you
has
been
an
amazing
teamwork
from
basically,
you
know
like,
like
several
different
groups,
so
from
from
my
postdoctoral
work.
It
was
with
tom
kirkhausen,
but
I
worked
very
closely
with
eric
betzig
and
his
group
at
genellium
sung.
B
Lee
was
the
one
who
first
built
the
prototype
for
the
the
adaptive
optics
lightest
light
sheet.
Dan
milkey
is
our
more
or
less
our
god
when
it
comes
to
control
software
to
be
able
to
work,
the
instruments
and
push
to
you
know
push
them
to
their
limits,
and
we
work
with
all
of
the
zebrafish
work
with
the
medicine
lab
at
harvard
medical
school.
Particularly
ian
is
the
one
who's
generated,
all
of
the
the
zebrafish,
the
actual
reagents
that
I've
I've
shown
you
in
ed
boyden's
lab
at
mit.
B
We
work
with
them
on
the
combination
of
this
of
the
lattice
light
sheet,
with
expansion
microscopy
now,
specifically
with
with
show
and
ring,
and
the
next
basically
group,
the
stephen
softballs
group
they've
been
our
computational
stitching
pipeline
gods,
so
to
speak.
They
were
basically
able
to
generate
these
these
java
tools
to
be
able
to
more
or
less
stitch,
literally
tens
or
hundreds
of
terabytes
of
of
data
using
the
genealogy
clusters.
B
So
with
that,
I'm
gonna
kind
of
tell
you
what
we're
up
to
next
and
what
our
challenges
are.
So,
in
the
context
of
super
resolution
everything
I've
shown,
you
was
what
we're
calling
last
gen
expansion.
It
took
us
about
three
days
to
be
able
to
image
the
whole
fly
brain,
but
with
what
we
built
at
berkeley
today
with
the
current
technology,
what
we're
calling
it
where
it
used
to
be
called
next-gen.
But
now
it's
it's
operational.
B
So
these
instruments
typically
are
not
on
24
is
because
we
have
no
way
of
dealing
with
the
data,
so
we
basically
have
to
do
surgical
experiments
which
really
prohibit
our
ability
to
explore
and
more
or
less
kind
of
you
know
try
to
to
go
in
without
a
hypothesis
and
then
to
use
the
data
to
generate
a
hypothesis.
So
that's
a
problem.
The
second
problem
is
computational
tools
and
the
challenge
there
is
that
the
diversity
of
biological
samples
that
go
in
to
a
fluorescence
microscope
is
is
immense,
meaning
that
the
type
of
sample
is
different.
B
The
levels
of
signals
are
different.
The
features
that
we're
trying
to
identify
are
very
dependent
on
what
is
tagged
and
labeled,
so
that
basically
makes
generalization
more
or
less
challenging
and
generalizing
it.
Broad
enough
is
also
extremely
difficult
and
then
the
future,
more
or
less
in
terms
of
breakthroughs,
really
require
interdisciplinary
teams.
So
the
only
reason
I've
been
successful.
The
past
few
years
is
because
I
was
able
to
work
with
an
amazing
group
of
interdisciplinary
experts.
So
I
had
virologists
cell
biologists
software
engineers.
B
That's
of
interest
to
to
folks
that
want
to
work
with
us
and
to
give
you
an
example:
we've
had
collaborations
where
we
just
give
the
data,
because
you
know
they're
trying
to
build
a
new
infrastructure
for
how
you
more
or
less
store
or
read
or
write.
The
data
we've
had
collaborations,
where
they're
trying
to
build
new
tools
to
be
able
to
segment
different
things,
because
you
know
these
are
the
levels
of
details
that
were
never.
I
mean
these
type
of
data
sets
never
existed
before.
B
B
So
to
give
you
context
of
of
data
scales,
the
last
two
projects.
Sorry,
I'm
going
to
skip
this
part,
the
last
two
projects
that
we
worked
on,
which
is
the
adaptive
optics
and
the
and
the
expansion
lattice.
That
was
about
three
quarters
of
a
petabyte,
and
this
is
literally
just
one
group
of
people
with
cell
phone
collaborators,
but
about
four
different
people
working
on
on
both
of
these
projects
more
or
less
to
push
them
from
the
start
to
the
end.
B
A
B
B
I
feel
like
building
a
mausoleum
inside
of
barker,
and
since
that
you
can
see
the
the
line
width
of
the
the
network
connection
going
from
from
barker
to
to
the
berkeley
it
core
and
that's
about
one
gig,
but
I
think
they
recently
upgraded
10
gig
our
data
outputs
from
the
ao
swac
that
I'm
calling
over
there.
Those
are
the
microscopes
they
generate
data
at
close
to.
B
We
had
the
luxury
of
having
genelec
cluster,
which
you
know
is
like
5000
nodes,
with
literally
only
10
people
using
it,
including
us.
So
that
was
basically
what
facilitated
that
and
sorry.
I
want
to
skip
a
little
bit
ahead,
just
to
kind
of
talk
about
our
main
problems.
So
in
terms
of
the
I
mean
not
including
the
the
control
software
I'm
just
now
talking
about
what
comes
out
of
the
microscope
and
what
we're
trying
to
mine.
B
There's
these
commercial
and
visualization
tools
that
are
kind
of
exist,
but
not
really
designed
for
the
scale
of
data,
and
that
is
one
challenge
the
open
source
tools
they
work
as
a
solution,
but
most
of
it.
What
we
do
is
python,
matlab,
c
plus
plus,
and
the
unfortunate
thing
is
that
this
is
not
biologist
friendly,
meaning
that
we
can
write
a
tool
but-
and
we
can
put
that
on
github.
B
But
you
know
most
of
that
more
or
less
those
tools
have
been
kind
of
developed
with
a
particular
data
set
in
mind
and
when
they
you
know,
others
want
to
reuse
them.
They
may
not
have
the
expertise
to
be
able
to
take
that
and
and
modify
it
as
their.
You
know
to
fit
their
needs,
and
the
last
is:
how
do
we
do
scalable
and
automated
ai
solutions
for
this
type
of
data,
meaning
that
we
have
incredible
amounts
of
data
sets?
B
We
can
start
to
generate
labeled
data
sets
for
different
types
of
problems
with
different.
You
know
snr's
with
different
to
basically
mimic
the
diversity
of
samples
that
can
come
up
with,
and
this
is
something
that
I
think
we're
ex
super
excited
to
be
at
berkeley
to
explore,
with,
with
collaborations
on
how
to
kind
of
get
this
working.