Confocal Microscopy List

Re: Building a STORM/FPALM

Classic

List

Threaded

6 messages Options

Andrew York

Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello, my name is Andrew York, I'm a postdoc in Hari Shroff's lab at the
NIH. Possibly relevant to this discussion, we recently released some
open-source software for doing PALM/STORM with 3D superresolution:

code.google.com/p/palm3d

We tried to follow the example of Ricardo Henriques' excellent QuickPALM
software, providing a graphical interface, a step-by-step tutorial, and
sample datasets.

The primary hardware requirement is a precise axial positioner. Existing 2D
PALM/STORM scopes should work with our software, giving 3D superresolution
in thin (<1 micron) or moderately thick samples (up to ~3 microns). For
thicker samples (~10 microns), we use two-photon photoactivation to reduce
background and photobleaching.

Our algorithm uses cross-correlation for 3D localization, which is slower
than QuickPALM's centroid-finding, but gives 3D superresolution with
virtually any PSF. This helps our software work on a wide variety of
hardware - hopefully including yours.

For software and hardware details, see our paper:
http://dx.doi.org/10.1038/nmeth.1571

-Andrew

(This is my first posting; my apologies if this message is inappropriate for
the listserv)

Philippe clemenceau

Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hi Andrew,

How do you correlate depth and psf shape in your software ? are you assuming
there is astigmatism ? or simply a defocus difference as you move along the
z axis ?

What is the z resolution you typically achieve ?

Thanks,

Philippe Clémenceau

Imagine Optic/Axiom Optics
Adaptive Optics for microscopy & Frequency domain FLIM
www.imagine-optic.com, www.axiomoptics.com

-----Message d'origine-----
De : Confocal Microscopy List [mailto:[hidden email]] De
la part de Andrew York
Envoyé : Friday, February 18, 2011 12:32 PM
À : [hidden email]
Objet : Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello, my name is Andrew York, I'm a postdoc in Hari Shroff's lab at the
NIH. Possibly relevant to this discussion, we recently released some
open-source software for doing PALM/STORM with 3D superresolution:

code.google.com/p/palm3d

We tried to follow the example of Ricardo Henriques' excellent QuickPALM
software, providing a graphical interface, a step-by-step tutorial, and
sample datasets.

The primary hardware requirement is a precise axial positioner. Existing 2D
PALM/STORM scopes should work with our software, giving 3D superresolution
in thin (<1 micron) or moderately thick samples (up to ~3 microns). For
thicker samples (~10 microns), we use two-photon photoactivation to reduce
background and photobleaching.

Our algorithm uses cross-correlation for 3D localization, which is slower
than QuickPALM's centroid-finding, but gives 3D superresolution with
virtually any PSF. This helps our software work on a wide variety of
hardware - hopefully including yours.

For software and hardware details, see our paper:
http://dx.doi.org/10.1038/nmeth.1571

-Andrew

(This is my first posting; my apologies if this message is inappropriate for
the listserv)

Andrew York

Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello Philippe,

Thanks for the questions. My answers below.

Q1:
What is the z resolution you typically achieve?
Short answer:
Probably about 100 nm, using genetically expressed proteins in a thick
cell.

Q2:
How do you correlate depth and psf shape in your software? Are you assuming
there is astigmatism? Or simply a defocus difference as you move along the z
axis?
Short answer:
We compare images of single fluorescent molecules to a measured calibration
stack to estimate x, y, and z positions. This approach works with or without
astigmatism.

I hope that addressed your questions.

-Andrew

PS If those short answers were unsatisfying, here's some long answers:
1:
It's very complicated! The supmat of our paper addresses this question, but
here's a summary. Resolution depends on localization precision and density
of fluorescent molecules. Even if we had perfect localization precision,
it's hard to tag a cell with enough fluorescent molecules per cubic
nanometer to get real resolution substantially better than ~50 nm. Our PALM
image resolution is nearly always limited by density of fluorescent
molecules, rather than localization precision.

I suspect you're curious about our localization precision, though.
Localization precision depends on how bright your fluorescent molecules are,
noise sources including background fluorescence, how defocused a molecule
is, the microscope's aberrations, and the quality of your drift correction.
Localization precision is especially hard to measure directly, because
single molecules bleach so quickly. We measured the z-thickness of a thin
layer of PA-mCherry1 on glass to be <150 nm, but this is just a lower bound
since we don't know how flat and smooth the glass coverslip was. We clearly
resolved 50-nm steps of a 100 nm gold fiducial bead, but these beads are
2-4x brighter than a PA-mCherry1 molecule. In one densely-labeled cell, we
saw persistent voids ~50-75 nm thick in z, which is suggestive but perhaps
atypical.

Finally, my experience suggests ~100 nm resolution in z (not precision) is
typical for cells tagged with PA-mCherry1.

2:
Our algorithm works with most PSFs, but some PSFs are better than others. A
pretty good test is to take a calibration stack of a bright bead in 50 nm
steps. Can you distinguish two consecutive images by eye? If not, our
algorithm probably can't either.

Adding astigmatism with a cylindrical lens makes out-of-focus molecules
much easier to see, but makes in-focus molecules slightly harder to see, so
it's a tradeoff. On our scope, the cylindrical lens slightly improves
z-resolution, but slightly degrades x/y resolution. For thin (~1 micron)
samples where most of the molecules are nearly in-focus, we often 3D PALM
without a cylindrical lens.

On Tue, Mar 1, 2011 at 1:05 PM, Philippe Clemenceau <
[hidden email]> wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Hi Andrew,
>
> How do you correlate depth and psf shape in your software ? are you
> assuming
> there is astigmatism ? or simply a defocus difference as you move along the
> z axis ?
>
> What is the z resolution you typically achieve ?
>
> Thanks,
>
> Philippe Clémenceau
>
> Imagine Optic/Axiom Optics
> Adaptive Optics for microscopy & Frequency domain FLIM
> www.imagine-optic.com, www.axiomoptics.com
>
> -----Message d'origine-----
> De : Confocal Microscopy List [mailto:[hidden email]] De
> la part de Andrew York
> Envoyé : Friday, February 18, 2011 12:32 PM
> À : [hidden email]
> Objet : Re: Building a STORM/FPALM
>
> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Hello, my name is Andrew York, I'm a postdoc in Hari Shroff's lab at the
> NIH. Possibly relevant to this discussion, we recently released some
> open-source software for doing PALM/STORM with 3D superresolution:
>
> code.google.com/p/palm3d
>
> We tried to follow the example of Ricardo Henriques' excellent QuickPALM
> software, providing a graphical interface, a step-by-step tutorial, and
> sample datasets.
>
> The primary hardware requirement is a precise axial positioner. Existing 2D
> PALM/STORM scopes should work with our software, giving 3D superresolution
> in thin (<1 micron) or moderately thick samples (up to ~3 microns). For
> thicker samples (~10 microns), we use two-photon photoactivation to reduce
> background and photobleaching.
>
> Our algorithm uses cross-correlation for 3D localization, which is slower
> than QuickPALM's centroid-finding, but gives 3D superresolution with
> virtually any PSF. This helps our software work on a wide variety of
> hardware - hopefully including yours.
>
> For software and hardware details, see our paper:
> http://dx.doi.org/10.1038/nmeth.1571
>
> -Andrew
>
> (This is my first posting; my apologies if this message is inappropriate
> for
> the listserv)
>

David Baddeley

Re: Building a STORM/FPALM

I'm curious as to how you resolve the symmetric ambiguity i
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hi Andrew,

I'm curious as to how you resolve the symmetric ambiguity in an un-modified PSF
(ie the fact that out of focus planes above the PSF look very similar to those
below). I spent quite a while playing with this ~ 3 years ago and eventually
gave up, because although the ambiguity can to some extent be addressed by
leaving some of the spherical abberation uncorrected, the results were never
very satisfactory (even on purely simulated data where the model and actual PSF
were a perfect match). I also suspect that the use of an un-modified PSF might
be quite sensitive to depth-related changes in spherical abberation - have you
experienced this in practice? If these issues aren't as much of a concern as I
had thought, I might be tempted to re-analyse some of my old data.

Cheers,
David

----- Original Message ----
From: Andrew York <[hidden email]>
To: [hidden email]
Sent: Wed, 2 March, 2011 9:22:49 AM
Subject: Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello Philippe,

Thanks for the questions. My answers below.

Q1:
What is the z resolution you typically achieve?
Short answer:
Probably about 100 nm, using genetically expressed proteins in a thick
cell.

Q2:
How do you correlate depth and psf shape in your software? Are you assuming
there is astigmatism? Or simply a defocus difference as you move along the z
axis?
Short answer:
We compare images of single fluorescent molecules to a measured calibration
stack to estimate x, y, and z positions. This approach works with or without
astigmatism.

I hope that addressed your questions.

-Andrew

PS If those short answers were unsatisfying, here's some long answers:
1:
It's very complicated! The supmat of our paper addresses this question, but
here's a summary. Resolution depends on localization precision and density
of fluorescent molecules. Even if we had perfect localization precision,
it's hard to tag a cell with enough fluorescent molecules per cubic
nanometer to get real resolution substantially better than ~50 nm. Our PALM
image resolution is nearly always limited by density of fluorescent
molecules, rather than localization precision.

I suspect you're curious about our localization precision, though.
Localization precision depends on how bright your fluorescent molecules are,
noise sources including background fluorescence, how defocused a molecule
is, the microscope's aberrations, and the quality of your drift correction.
Localization precision is especially hard to measure directly, because
single molecules bleach so quickly. We measured the z-thickness of a thin
layer of PA-mCherry1 on glass to be <150 nm, but this is just a lower bound
since we don't know how flat and smooth the glass coverslip was. We clearly
resolved 50-nm steps of a 100 nm gold fiducial bead, but these beads are
2-4x brighter than a PA-mCherry1 molecule. In one densely-labeled cell, we
saw persistent voids ~50-75 nm thick in z, which is suggestive but perhaps
atypical.

Finally, my experience suggests ~100 nm resolution in z (not precision) is
typical for cells tagged with PA-mCherry1.

2:
Our algorithm works with most PSFs, but some PSFs are better than others. A
pretty good test is to take a calibration stack of a bright bead in 50 nm
steps. Can you distinguish two consecutive images by eye? If not, our
algorithm probably can't either.

Adding astigmatism with a cylindrical lens makes out-of-focus molecules
much easier to see, but makes in-focus molecules slightly harder to see, so
it's a tradeoff. On our scope, the cylindrical lens slightly improves
z-resolution, but slightly degrades x/y resolution. For thin (~1 micron)
samples where most of the molecules are nearly in-focus, we often 3D PALM
without a cylindrical lens.

On Tue, Mar 1, 2011 at 1:05 PM, Philippe Clemenceau <
[hidden email]> wrote:

Andrew York

Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello David,

What kind of objective were you using? I've only used water-immersion
objectives.

We can't resolve symmetric ambiguities either, so we try to avoid them. I'm
sure our algorithm would do quite poorly with a truly symmetric PSF, but
we've never been lucky enough to face this problem.

Our measured PSFs don't usually look symmetric or ambiguous by eye. I
suspect this is because our water-immersion objective is very sensitive to
tilt:
http://www.ncbi.nlm.nih.gov/pubmed/15369482

I haven't noticed much depth-dependent spherical aberration but maybe I
should look closer. Excessive spherical aberration usually makes our PSF too
dark for one direction of defocus, so we try to reduce this with the
correction collar.

-Andrew

On Tue, Mar 1, 2011 at 4:28 PM, David Baddeley
<[hidden email]>wrote:

> I'm curious as to how you resolve the symmetric ambiguity i
> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Hi Andrew,
>
> I'm curious as to how you resolve the symmetric ambiguity in an un-modified
> PSF
> (ie the fact that out of focus planes above the PSF look very similar to
> those
> below). I spent quite a while playing with this ~ 3 years ago and
> eventually
> gave up, because although the ambiguity can to some extent be addressed by
> leaving some of the spherical abberation uncorrected, the results were
> never
> very satisfactory (even on purely simulated data where the model and actual
> PSF
> were a perfect match). I also suspect that the use of an un-modified PSF
> might
> be quite sensitive to depth-related changes in spherical abberation - have
> you
> experienced this in practice? If these issues aren't as much of a concern
> as I
> had thought, I might be tempted to re-analyse some of my old data.
>
> Cheers,
> David
>
>
> ----- Original Message ----
> From: Andrew York <[hidden email]>
> To: [hidden email]
> Sent: Wed, 2 March, 2011 9:22:49 AM
> Subject: Re: Building a STORM/FPALM
>
> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> *****
>
> Hello Philippe,
>
> Thanks for the questions. My answers below.
>
> Q1:
> What is the z resolution you typically achieve?
> Short answer:
> Probably about 100 nm, using genetically expressed proteins in a thick
> cell.
>
> Q2:
> How do you correlate depth and psf shape in your software? Are you assuming
> there is astigmatism? Or simply a defocus difference as you move along the
> z
> axis?
> Short answer:
> We compare images of single fluorescent molecules to a measured calibration
> stack to estimate x, y, and z positions. This approach works with or
> without
> astigmatism.
>
> I hope that addressed your questions.
>
> -Andrew
>
> PS If those short answers were unsatisfying, here's some long answers:
> 1:
> It's very complicated! The supmat of our paper addresses this question, but
> here's a summary. Resolution depends on localization precision and density
> of fluorescent molecules. Even if we had perfect localization precision,
> it's hard to tag a cell with enough fluorescent molecules per cubic
> nanometer to get real resolution substantially better than ~50 nm. Our PALM
> image resolution is nearly always limited by density of fluorescent
> molecules, rather than localization precision.
>
> I suspect you're curious about our localization precision, though.
> Localization precision depends on how bright your fluorescent molecules
> are,
> noise sources including background fluorescence, how defocused a molecule
> is, the microscope's aberrations, and the quality of your drift correction.
> Localization precision is especially hard to measure directly, because
> single molecules bleach so quickly. We measured the z-thickness of a thin
> layer of PA-mCherry1 on glass to be <150 nm, but this is just a lower bound
> since we don't know how flat and smooth the glass coverslip was. We clearly
> resolved 50-nm steps of a 100 nm gold fiducial bead, but these beads are
> 2-4x brighter than a PA-mCherry1 molecule. In one densely-labeled cell, we
> saw persistent voids ~50-75 nm thick in z, which is suggestive but perhaps
> atypical.
>
> Finally, my experience suggests ~100 nm resolution in z (not precision) is
> typical for cells tagged with PA-mCherry1.
>
> 2:
> Our algorithm works with most PSFs, but some PSFs are better than others. A
> pretty good test is to take a calibration stack of a bright bead in 50 nm
> steps. Can you distinguish two consecutive images by eye? If not, our
> algorithm probably can't either.
>
> Adding astigmatism with a cylindrical lens makes out-of-focus molecules
> much easier to see, but makes in-focus molecules slightly harder to see, so
> it's a tradeoff. On our scope, the cylindrical lens slightly improves
> z-resolution, but slightly degrades x/y resolution. For thin (~1 micron)
> samples where most of the molecules are nearly in-focus, we often 3D PALM
> without a cylindrical lens.
>
> On Tue, Mar 1, 2011 at 1:05 PM, Philippe Clemenceau <
> [hidden email]> wrote:
>
> > *****
> > To join, leave or search the confocal microscopy listserv, go to:
> > http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> > *****
> >
> > Hi Andrew,
> >
> > How do you correlate depth and psf shape in your software ? are you
> > assuming
> > there is astigmatism ? or simply a defocus difference as you move along
> the
> > z axis ?
> >
> > What is the z resolution you typically achieve ?
> >
> > Thanks,
> >
> > Philippe Clémenceau
> >
> > Imagine Optic/Axiom Optics
> > Adaptive Optics for microscopy & Frequency domain FLIM
> > www.imagine-optic.com, www.axiomoptics.com
> >
> > -----Message d'origine-----
> > De : Confocal Microscopy List [mailto:[hidden email]]
> De
> > la part de Andrew York
> > Envoyé : Friday, February 18, 2011 12:32 PM
> > À : [hidden email]
> > Objet : Re: Building a STORM/FPALM
> >
> > *****
> > To join, leave or search the confocal microscopy listserv, go to:
> > http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> > *****
> >
> > Hello, my name is Andrew York, I'm a postdoc in Hari Shroff's lab at the
> > NIH. Possibly relevant to this discussion, we recently released some
> > open-source software for doing PALM/STORM with 3D superresolution:
> >
> > code.google.com/p/palm3d
> >
> > We tried to follow the example of Ricardo Henriques' excellent QuickPALM
> > software, providing a graphical interface, a step-by-step tutorial, and
> > sample datasets.
> >
> > The primary hardware requirement is a precise axial positioner. Existing
> 2D
> > PALM/STORM scopes should work with our software, giving 3D
> superresolution
> > in thin (<1 micron) or moderately thick samples (up to ~3 microns). For
> > thicker samples (~10 microns), we use two-photon photoactivation to
> reduce
> > background and photobleaching.
> >
> > Our algorithm uses cross-correlation for 3D localization, which is slower
> > than QuickPALM's centroid-finding, but gives 3D superresolution with
> > virtually any PSF. This helps our software work on a wide variety of
> > hardware - hopefully including yours.
> >
> > For software and hardware details, see our paper:
> > http://dx.doi.org/10.1038/nmeth.1571
> >
> > -Andrew
> >
> > (This is my first posting; my apologies if this message is inappropriate
> > for
> > the listserv)
> >
>
>
>
>
>

David Baddeley

Re: Building a STORM/FPALM

I started out with a 1.4 NA oil (Leica), but any recent wor
*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hi Andrew,

I started out with a 1.4 NA oil (Leica), but any recent work has been using a
1.47 NA Nikon TIRF objective (with correction collar). It's super at collecting
light, but due to it's high NA, also quite sensitive to abberations and in
practice we also never have a truely symetrical PSF. I should probably note that
we're using a glycerol/pbs/glucose mixture with a refractive index ~1.46 and not
aqueous media.

My initial attempts were using an analytical model for the widefield PSF, which
fitted pretty well to the simulated data (using the same model) but wasn't
particularly good for the actual PSF. At the time I didn't put to much effort
into finding a better approximation because I couldn't satisfactorily get
around the ambiguity in simulation (even with the addition of reasonable
ammounts of added spherical abberation). I've now got pretty robust code to fit
an interpolated measured PSF and should probably give it another shot (I tend to
prefer fitting to centroid/correlation based measures as, given a realistic
noise model, you get better results as well as error estimates).

thanks,
David

----- Original Message ----
From: Andrew York <[hidden email]>
To: [hidden email]
Sent: Wed, 2 March, 2011 11:49:39 AM
Subject: Re: Building a STORM/FPALM

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
*****

Hello David,

What kind of objective were you using? I've only used water-immersion
objectives.

We can't resolve symmetric ambiguities either, so we try to avoid them. I'm
sure our algorithm would do quite poorly with a truly symmetric PSF, but
we've never been lucky enough to face this problem.

Our measured PSFs don't usually look symmetric or ambiguous by eye. I
suspect this is because our water-immersion objective is very sensitive to
tilt:
http://www.ncbi.nlm.nih.gov/pubmed/15369482

I haven't noticed much depth-dependent spherical aberration but maybe I
should look closer. Excessive spherical aberration usually makes our PSF too
dark for one direction of defocus, so we try to reduce this with the
correction collar.

-Andrew

On Tue, Mar 1, 2011 at 4:28 PM, David Baddeley
<[hidden email]>wrote: