Discovery of an Accretion Streamer and a Slow Wide-angle Outflow around FUOri...
Cyto 2015 Forensic Flow Cytometry Tutorial
1. Forensic Flow Cytometry:
Troubleshooting Flow Cytometry Data
Pratip K. Chattopadhyay, Ph.D.
ISAC Scholar
ImmunoTechnology Section
Vaccine Research Center, NIH
Jennifer Wilshire, Ph.D.
Assistant Manager
Flow Cytometry Core Facility
Memorial Sloan-Kettering Cancer Center
2. These slides were presented at the CYTO2015 Conference.
They are distributed here for the benefit of the flow cytometry community.
Please do not:
1. Take any material from these presentations (copied or adapted) without
obtaining the express consent of Pratip Chattopadhyay,
pchattop@mail.nih.gov.
2. Modify this material in any way or distribute it outside of this source.
Please do:
1. Send your flow cytometry troubleshooting examples. I’d love to create a
publicly available database of examples like the ones presented here.
2. Send me your questions and comments!
3. Learn and Enjoy!
3. Disclaimer : “Forensic” Flow Cytometry
If you are here to learn about how flow
cytometry is used to solve real crimes
(e.g., murder, robbery, etc.) …
Sorry, we’ll be talking about
other heinous acts:
Poor experimental
Planning,
Execution,
and Analysis.
4. Today
Recognize Problematic Staining Patterns
(See how easy it is to fall prey to criminal cytometry)
Troubleshoot Experiments and Analysis
(Become a crime-stopper)
Guidelines to Improve Cytometry Experiments.
(Learn how to live a virtuous life)
5. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Don’t run the red light.
Spend time planning your experiments properly.”
Start with a robust system for instrument setup & monitoring.
6. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Don’t run the red light.
Spend time planning your experiments properly.”
Start with a robust system for instrument setup & monitoring.
Why bother?
Reliability!
Set Panel
Instruments
Panel
Panel
Panel
PanelSingle
panel
performs
reliably
in
mul2-‐instrument
(center)
study.
Single
instrument
works
for
most
panels
7. Instrument Setup and Monitoring: Voltages
How to set voltages?
Option 1: Just set all unstained signals to 102.
What’s the crime here? (Audience?)
8. Instrument Setup and Monitoring: Voltages
How to set voltages?
Option 1: Just set all unstained signals to 102.
Criminal neglect.
This method ignores:
- dynamic range of PMT,
- linearity of PMT,
- the tenet that signal from dye
should be highest in the detector
designed to read that dye.
9. “Neglecting” Dynamic Range and Linearity of PMTs
Hard to know relative
performance of different
PMTs in your instrument.
CVs of fluorescent signals
may not be optimized,
leading to issues with
resolution and gating.
10. “Neglecting” Dynamic Range and Linearity of PMTs
Hard to know relative
performance of different
PMTs in your instrument.
CVs may not be optimized,
leading to issues with
resolution and gating.
Quality Control
Panel Design
Data Analysis
This complicates…
11. “Neglecting” Dynamic Range and Linearity of PMTs
Hard to know relative
performance of different
PMTs in your instrument.
CVs may not be optimized,
leading to issues with
resolution and gating.
Quality Control
Panel Design
Data Analysis
This complicates…
“I will never know if my PMTs
are working properly.”
“I never resolve dim markers
on the 1st detector off green
laser, so I don’t use Cy7PE.”
“My percent CD4+ never
matches any other site in
multicenter study.”
Leading to criminal confessions:
12. “Neglecting” Dynamic Range and Linearity of PMTs
Is this always a problem?
No… sometimes this will be a victimless crime.
But troubleshooting experiments depends on your
confidence in instrument setup!
Poor instrument setup makes forensic flow harder!
Guideline: Consider linearity and dynamic range
when choosing PMT voltage.
13. Another Rule for Choosing Voltages
Signal from a dye should be strongest in channel set up to detect that dye.
So, signal from Alexa 680 (or R700APC or Cy55APC) should be brightest
when measured off of the red laser detector with 710nm filters in it.
14. Another Rule for Setting Voltages
Signal from a dye should be strongest on channel set up to detect that dye.
So, signal from Alexa 680 (or R700APC or Cy55APC) should be brightest
when measured off of the red laser detector with 710nm filters in it, even in systems
also measuring Cy55PE. Ensure this by optimizing detector voltages.
15. If voltages across two detectors aren’t optimized…
You may detect more secondary
(spillover) signal from Cy55PE in your
RedLaser/710nm channel, than
primary (Alexa 680).
What happens as a consequence? (Audience?)
16. If voltages across two detectors aren’t optimized…
You may be detect more secondary
(spillover) signal from Cy5PE in your
RedLaser/710nm channel, than
primary (Alexa 680).
What happens as a consequence? (Audience?)
0 10
2
10
3
10
4
10
5
Cy5PE (Secondary)
0
102
103
104
10
5
Alexa680(Primary)
Hint
17. If voltages across two detectors aren’t optimized…
You may be detect more secondary
(spillover) signal from Cy5PE in your
RedLaser/710nm channel, than
primary (Alexa 680).
What happens as a consequence? (Audience?)
0 10
2
10
3
10
4
10
5
Cy5PE (Secondary)
0
102
103
104
10
5
Alexa680(Primary)
Hint Compensation values > 100%.
Will this sink your panels?
Not always…
a multicolor panel can work just fine with
compensations > 100% in some color combinations.
But… troubleshooting becomes harder.
And panel development is more complex.
18. Instrument Setup and Monitoring: Voltages
How to do a better job setting voltages?
Option 1: Just set all unstained signals to 102.
Option 2: Bead-based protocols.
Good practice. Perfetto (Nature Protocols), CS&T (BD)
Provide quantitative measure of instrument performance,
in the form of metrics like Q (sensitivity) and B (resolution).
Beads have limitations, though:
- high intrinsic CV
- loaded with broad-spectrum dyes, not our fluors
- beads and dyes vary by lot.
So, the metrics from these methods are a bit inaccurate.
19. Instrument Setup and Monitoring: Voltages
How to do an even better job setting voltages?
Option 1: Just set all unstained signals to 102.
Option 2: Bead-based protocols.
Option 3: LED Pulser
New device (Jim Wood, Wake) that emits consistent signals to PMT
for accurate measurement of sensitivity and resolution (Q & B).
Reliable method to compare detectors (track them for
QC), or compare instruments.
Can use Q and B values to 1) set voltages properly, 2) predict how a
panel will stain on different instruments, and 3) predict whether a
putative staining panel will work.
More info: CYTO U webinar, Steve Perfetto’s Oral Presentation
20. Review:
Guidelines to Improve Cytometry Experiments
Experiment Planning:
Spend time planning your experiments properly.
1) Set up instrument with bead-based protocols
or LED Pulser.
21. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Spend time planning your experiments properly.”
1) Set up instrument with bead-based protocols
or LED Pulser.
2) Put some effort into panel design!
22. Panel Design: Step 1… Titrate Every Reagent
Titration is fundamental to successful flow and good data.
Don’t rely on the manufacturer’s titer…
… they can’t test all experimental conditions,
and suggested titers may be significantly higher
than what will work just fine in your system.
23. Titration Guidelines
Choose a wide range of concentrations that captures:
Saturation
(where signal does not increase with
more antibody)
Loss of positive staining
(or at least diminished staining).
O/n rt
0 5 10 15 20
Dilution
0
102
10
3
104
105
B515-A:CD8FITCV1002
O/n 37C
0 5 10 15 20
Dilution
O/n 4C
0 5 10 15 20
Dilution
Example of a good titration experiment
(courtesy Margaret Beddall)
24. Titration Guidelines
1) Choose a wide range of concentrations.
2) Perform titrations under the same conditions as
experiment.
If you titrate antibody at room temperature, but need
to stain cells at 37C for one experiment,
Redo the titration at 37C first!
Example of this in case studies later today.
25. Titration Guidelines
1) Choose a wide range of concentrations.
2) Perform titrations under the same conditions as
experiment.
3) Analyze titrations on same cell type as in your study.
Some markers differ in expression level by tissue,
or are expressed across different cell types
at different levels.
Need to account for this by using right sample type, or
including other markers in titration.
33. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Spend time planning your experiments properly.”
1) Set up instrument with bead-based protocols or
LED Pulser.
2) Put some effort into panel design!
+ Titration – range of concentrations that works.
34. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Spend time planning your experiments properly.”
1) Set up instrument with bead-based protocols or
LED Pulser.
2) Put some effort into panel design!
+ Titration – range of concentrations that works.
+ Panel Design – Choosing which reagents will
work together, and at what concentrations.
We do this by considering spectral overlap and
spreading error.
35. Review of Panel Design*
www.isac-net.org
Don’t randomly combine antibodies and fluorochromes.
Quick & Easy (Spectral Overlap)
Pick fluorochromes that
don’t overlap.
Qdots and Brilliant Violet
fluorochromes
are particularly
useful for this.
Pair highly overlapping
fluorochromes with markers
that aren’t co-expressed.
Choose bright dyes for dim markers.
The Better Way (Spreading Error)
*See ISAC’s Polychromatic Flow Cytometry Course (2011) and my Tutorials (2010, 2013).
36. Review of Panel Design*
www.isac-net.org
Don’t randomly combine antibodies and fluorochromes.
Quick & Easy (Spectral Overlap)
Pick fluorochromes that
don’t overlap.
Qdots and Brilliant Violet
fluorochromes
are particularly
useful for this.
Pair highly overlapping
fluorochromes with markers
that aren’t co-expressed.
Choose bright dyes for dim markers.
Limitations
*See ISAC’s Polychromatic Flow Cytometry Course (2011) and my Tutorials (2010, 2013).
Multicolor flow will always use dyes that
overlap (and that overlap is no big deal!).
Cells are so heterogeneous that there is
usually some co-expression of markers in
any (interesting) phenotyping experiment.
Reagent brightness doesn’t depend only
on the dye used for conjugation. Antibody
clones differ in signal strength, too.
So, charts and guides ranking dye
brightness may be of limited utility.
37. Review of Panel Design*
www.isac-net.org
Don’t randomly combine antibodies and fluorochromes.
Quick & Easy (Spectral Overlap)
Pick fluorochromes that
don’t overlap.
Qdots and Brilliant Violet
fluorochromes
are particularly
useful for this.
Pair highly overlapping
fluorochromes with markers
that aren’t co-expressed.
Choose bright dyes for dim markers.
The Better Way (Spreading Error)
Characterize spreading
error for the fluorochromes
and instruments you use.
Place dim markers on channels
with low spreading error.
Iteratively add markers to panel to
characterize effect of various
combinations on other channels.
*See ISAC’s Polychromatic Flow Cytometry Course (2011) and my Tutorials (2010, 2013).
38. Why Worry About Spreading Error?
When fluorescence is low, error associated with
photon detection is easily observed.
Particularly high photon counting errors occur
for fluorochromes that give off very few photons (e.g., Cy7APC).
The result: the negative population spreads out and masks dim
populations in neighboring channels.
39. How do you calculate spreading error
www.isac-net.org
(See PC Previous Tutorials for Manual Example.)
40. How do you calculate spreading error
www.isac-net.org
(See PC Previous Tutorials for Manual Example.)
41. How do you calculate spreading error
www.isac-net.org
(See PC Previous Tutorials for Manual Example.)
42. How do use spreading error in panel design?
Save the resulting table.
You can even copy table to spreadsheet and make a heat map.
Dye for Plot vs. Detector for Plot
DyeforPlot
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
Detector for Plot
Spread
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Dye
Detector
43. Dye for Plot vs. Detector for Plot
DyeforPlot
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
Detector for Plot
Spread
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Dye
Detector
Now you can see (FOR ONE PARTICULAR INSTRUMENT) how including Cy5PE
in the panel, induces spreading error into a number of other
detectors off of the green laser. Don’t put dim markers into those detectors.
How do use spreading error in panel design?
44. Dye for Plot vs. Detector for Plot
DyeforPlot
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
Detector for Plot
Spread
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Dye
Detector
You can also see how low the spreading error is off violet laser detectors.
Consider putting dim markers on those channels.
How do use spreading error in panel design?
45. Dye for Plot vs. Detector for Plot
DyeforPlot
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
A-BUV395
B-BUV490
C-BUV550
D-BUV680
E-BUV737
F-BUV815
G-BV421
H-BV480
I-BV570
J-BV605
K-BV650
L-BV711
M-BV745
N-BV786
O-BB515
P-BB630
Q-BB667
R-BB700
S-BB796
T-PE
U-CF594PE
V-CY5PE
W-CY55PE
X-CY7PE
Y-APC
Z-R700APC
ZZ-H7APC
Detector for Plot
Spread
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Dye
Detector
This analysis is now EASY to do (using FlowJo),
and can be done with any of the panels you run on your instrument!
How do use spreading error in panel design?
46. Guidelines to Improve Cytometry Experiments
Experiment Planning:
“Spend time planning your experiments properly.”
1) Set up instrument with bead-based protocols or
LED Pulser.
2) Put some effort into panel design!
+ Titration – range of concentrations that works.
+ Panel Design –
Quick and Dirty Spectral Overlap Method
Better Spreading Error Method
47. Guidelines to Improve Cytometry Experiments
Experiment Execution:
Proceed cautiously & follow protocols when staining.
Two examples of vendor protocols I didn’t follow.
(Things you may not be aware of.)
48. Pratip’s Criminal Past (1st Offense)
Ignored manufacturer’s recommendation to stain on ice when
using fix/perm kits for intra-cellular cytokine staining with
surface Qdot antibodies.
49. Pratip’s Criminal Past (1st Offense)
Ignored manufacturer’s recommendation to stain on ice when
using fix/perm kits for intra-cellular cytokine staining.
50. Pratip’s Criminal Past (2nd Offense)
Stained a multicolor panel with many BD and BioLegend
“Brilliant” dyes. BD recommends use of Brilliant Stain Buffer, to
minimize non-specific interactions between reagents.
Non-specific interactions are worst in whole blood lysed
protocols, but can appear when staining ficolled PBMC.
I would see these interactions essentially as slightly
undercompensated staining patterns.
51. Pratip’s Criminal Past (2nd Offense)
Example from BD:
CD8
CD19
Withour Brilliant Stain
Buffer.
52. Pratip’s Criminal Past (2nd Offense)
Example from BD:
CD8
CD19
Without Brilliant Stain
Buffer.
53. Pratip’s Criminal Past (2nd Offense)
Example from BD:
CD8
CD19
Without Brilliant Stain
Buffer.
With Brilliant Stain
Buffer.
54. Guidelines to Improve Cytometry Experiments
Experiment Execution:
Proceed cautiously & follow protocols when staining.
Follow new vendor protocols exactly.
There’s a reason for their recommended conditions.
55. Guidelines to Improve Cytometry Experiments
Analysis of Data:
Use tools that help check data quality.
1) NXN Plot.
56. N X N PlotsFITC
PE
TRPE
Cy5PE
Cy5.5PE
PE
TRPE
Cy5PE
Cy5.5PE
TRPE
Cy5PE
Cy5.5PE
Cy5PE
Cy5.5PE
And so on…
For every marker combination in panel.
57. N X N Plot for Troubleshooting
A rapid means to identify problems.
Over Compensation
Under Compensation
Over Compensation
Transformation/Compensation
58. N X N Plot for Troubleshooting
Let’s look more closely at
Transformation/Compensation
example…
59. Negative and Super-Negative
www.isac-‐net.org
ALer
bi-‐exponen2al
transforma2on,
nega2ve
popula2ons
should
have
a
uniform
distribu2on.
“Super-‐nega2ve”
=
“really
pregnant”
or
“extremely
dead.”
Why
does
this
happen?
0 103
104
105
<R710-A>: CD69
0
10
3
104
10
5
<G780-A>:CD4
CD4-‐Nega2ve
CD4
“Super”
Nega2ve
60. Negative and Super-Negative
www.isac-net.org
After bi-exponential transformation, negative populations should have a
uniform distribution.
“Super-negative” = “really pregnant” or “extremely dead.”
Why does this happen? Bad compensation in another dimension.
0 103
104
105
<R710-A>: CD69
0
10
3
104
10
5
<G780-A>:CD4
0 102
103
104
105
<V800-A>: CD57
0
10
3
104
105
<G780-A>:CD4
61. Negative and Super-Negative
www.isac-net.org
As soon as we fix comps, problem is solved.
0 10
3
10
4
10
5
<R710-A>: CD69
0
10
3
10
4
10
5
<G780-A>:CD4
0 10
2
10
3
10
4
10
5
<V800-A>: CD57
0
10
3
10
4
10
5
<G780-A>:CD4
Sort of. There are still a few super-negative events. Why?
62. Fluorochrome Aggregates
www.isac-net.org
What are they?
High-order complexes of fluorochromes that
non-specifically bind cells.
How do you recognize them?
Random, punctate staining. Brighter than the
larger population of cells that is specifically
staining for the marker.
Why worry about aggregates?
Affect accuracy of subset identification, don’t
phenotype junk!
Effect on transformation and data visualization.
0 10
3
10
4
10
5
<G610-A>: `HLA-DR
0
10
2
10
3
10
4
10
5
<V800-A>:CD57
63. Fluorochrome Aggregates : Solutions
www.isac-net.org
Staining Time
Make staining cocktail and spin at full speed in
microfuge for a few minutes.
Mark cap to indicate where pellet would be,
and transfer cocktail from other side to new
tube.
Analysis Time
Make gates that exclude the aggregates.
Perform transformation after excluding
aggregates.
0 10
3
10
4
10
5
<G610-A>: `HLA-DR
0
10
2
10
3
10
4
10
5
<V800-A>:CD57
64. Guidelines to Improve Cytometry Experiments
Analysis of Data:
Use tools that help check data quality.
1) NXN Plot.
2) FlowClean: gates out problems
in HTS data.
65. FlowClean
As number of parameters increases,
we need to study more samples to make statistically robust conclusions.
High-throughput systems (HTS) for cytometers will help with this.
Quality control of HTS data is challenging.
Cytometer pressure is often disrupted by bubbles, debris, or clogs.
This can alter fluorescent signal, resulting in abnormal populations of cells.
We developed FlowClean because existing methods for finding HTS
problems were subjective, slow, or confusing to use.
Cha$opadhyay
and
Fletez-‐Brant
66. FlowClean
Concept
Most
flow
experiments
aim
to
count
cell
popula2ons
based
on
marker
expression.
Since
the
sample
tube
contains
cells
mixed
together,
the
frequency
of
each
popula2on
should
be
stable
over
the
collec2on
period.
When
it’s
not
stable
(i.e.,
new
popula2ons
appear
and
exis2ng
ones
are
lost)
collec2on/pressure
problems
are
happening.
FlowClean
is
an
automated
tool
to
detect
these.
67. FlowClean
Workflow
FCS File Partition Distributions Generate Cellular Address
#Cells
50%50%
#Cells
50%50%
<V705>
<G560>
0
0
1
1
Partition
Cell <V705> <G560> <B515>
1 0 1 0
2 0 1 0
3 1 1 0
4 0 0 1
Define "Populations"
by matching addresses:!
Population 1!
Population 2!
Population 3
Track Populations Over Time
Transform Population!
Frequencies into CLR
Flag Anamolous Events
Amend FCS File
68. FlowClean
on
Real
Experimental
Data
0
102
10
3
10
4
10
5
0
10
2
10
3
10
4
105
10
4
10
5
<R710-A>:CD38
LSG
RV152 212341-0200_B cel.fcsEvent Count: 164602
Time
03006009001200
Time
101
10
2
103
10
4
Good_vs_Bad
LSG
RV152 212341-0200_B cel.fcs
Event Count: 164602
2 3 4 5
0
102
10
3
10
4
10
5
<R710-A>:CD38
LSG
RV152 212341-0200_B cel.fcsEvent Count: 164602
03006009001200
Time
0
10
2
10
3
10
4
10
5
<R710-A>:CD38
03006009001200
Time
10
1
10
2
10
3
10
4
Good_vs_Bad
LSG
RV152 212341-0200_B cel.fcs
Event Count: 164602
bad
010 2
103
104
105
SSC-A
0
102
10
3
104
105
<R710-A>:CD38
good
RV152 212341-0200_B cel.fcs
010
2
10
3
10
4
10
5
SSC-A
0
102
10
3
10
4
10
5
<R710-A>:CD38
A
B CBad
Good
Time
<R710>CD38
Time
ChangePointMeans
Good_Bad
Time
<R710>CD38
Time
SSC-A
<R710>CD38
• 679
files
from
large
study
(RV152)
• 12-‐14
colors,
HTS
• Each
file
=
seconds,
each
plate
<
5min
69. FlowClean
on
Real
Experimental
Data
12%
88%
Files
Flagged
No
Problems
1 2 3 4 5 6 7 8 9 10 11 12
A 1 111 1 1 111 11 1 111 11 111 1
B 11 11 11 111 11 1 1 1 1 11 1 1
C 1 1 1 11 11 1 1
D
E
F 1 1 1
G 11 1 1 1 1
H 1 1
A
Are
there
any
posi2ons
in
plate
where
problems
are
most
common?
Perhaps
at
early
wells,
sugges2ng
need
for
beder
cleaning
of
HTS
equipment.
70. FlowClean
Summary
• Developed
a
computa2onal
tool
that
makes
flow
data
more
powerful,
by
elimina2ng
ar2facts
and
noise.
• We
proved
it
worked
in
a
large
dataset,
not
on
a
“toy”
case,
as
is
oLen
done
for
new
algorithms.
• We
showed
that
it’s
unique
features
are
necessary,
classical/
simple
methods
don’t
work.
• We
compared
it
to
exis2ng
algorithms,
and
objec2vely
showed
that
it
was
beder.
Available
on
BioConductor
and
GenePadern
(J.
Spidlin).
71. Guidelines to Improve Cytometry Experiments
Analysis of Data:
Use tools that help check data quality.
1) NXN Plot.
2) FlowClean: gates out problems
in HTS data.
3) Concatenation of all study files.
72. Concatenation for QC
When performing a large study, with many samples,
covering many weeks of experiments, you need good ways
to detect experimental or instrument issues and flag them.
One method:
Add a parameter to each FCS file that indicates sample
number. (Numbers should be consecutive: 1, 2, 3…)
Export a randomly selected subset events from each fcs file
into a single “concatenated” file.
74. Guidelines to Improve Cytometry Experiments
Planning:
Stop
to
Do
Setup
and
Design
ExecuEon:
CauEon
with
Reagent
Protocols
Analysis:
Use
Tools
that
Drive
Data
Quality
76. Crimes Against Flow Cytometry
Flow Cytometry Unit
Punk E.
CytometriskiFLOW CYTOMETRY UNIT
(FCU)
Suspect
DescripEon
Talented
graduate
student,
with
some
flow
cytometry
experience.
Indicted
on
five
counts,
described
herein.
77. Count 1: Filing False Police Report
Suspect
visited
FCU
claiming
that
CD25
staining
“totally
different”
across
two
panels.
Even
though
he
was
“100%
sure”
that
an2body
was
added
to
both
tubes
at
same
concentra2on,
and
that
same
reagent
was
used.
79. Count 1: Filing False Police Report
www.isac-‐net.org
Panel
1
Panel
2
Bright
Dim
Nega2ve
“The
padern
is
totally
different,”
said
suspect.
Upon
interroga2on,
suspect
claimed
there
were
no
differences
in
staining
or
analysis.
“I
don’t
know
what
the
#$@#%
is
happening!”
he
exclaimed.
81. How Do These Plots Differ?
Panel
1
Panel
2
The
axes
differ…
Compensa2on
and/or
transforma2on
must
be
different.
82. How Do These Plots Differ?
Panel
1
Panel
2
Lower
resolu2on
plot
(bigger
dot/per
event)
shows
lots
of
“junk”
in
Panel
2.
Fluorochrome
aggregates.
83. Panel
1
Panel
2
Count 1: Filing False Police Report
When
fluorochrome
aggregates
are
gated
out,
and
then
the
data
is
transformed…
paderns
are
very
similar.
Bright
Dim
Nega2ve
84. Count 2: Unauthorized Entry
Searching
the
suspect’s
FCS
files,
FCU
inves2gators
discover
unusual
popula2ons.
We
oLen
see
three
popula2ons
(bright,
dim,
and
nega2ve)
in
flow
cytometry
staining,
but
four
is
unusual.
Also,
highly
correlated
expression
(diagonals)
are
unusual.
85. Count 2: Unauthorized Entry
0 10
2
10
3
10
4
10
5
SSC-A
0
50K
100K
150K
200K
250K
FSC-A
Original
lymphocyte
gate
Diagonal
popula2on
overlaid
Inves2ga2ve
process
called
“backga2ng.”
The
diagonal
popula2on
is
not
distributed
throughout
lymphocyte
gate.
Cells
entering
lymphocyte
gate
from
border,
where
SSC
is
larger.
86. Count 2: Unauthorized Entry
0 10
2
10
3
10
4
10
5
SSC-A
0
50K
100K
150K
200K
250K
FSC-A
Solu2on:
A
2ghter
lymphocyte
gate
that
excludes
border
dwellers.
New
lymph
gate
overlaid
on
old
gate.
Diagonal
popula2on
from
old
gate.
New,
2ghter
lymph
gate.
88. Count 3: T.W.I.
T
.
W
.
I.
“Titra2on
Without
Intracellular
(staining).”
1:20
dilu2on
of
an2body
1:20
dilu2on
of
an2body
Suspect
asks,
“why
don’t
they
look
the
same?”
89. Count 3: T.W.I.
T
.
W
.
I.
“Titra2on
Without
Intracellular
(staining).”
1:20
dilu2on
of
an2body
Intracellular
1:20
dilu2on
of
an2body
Surface
Suspect
admits
one
added
before
fix/perm,
another
aLer.
Titrate
under
same
condi2ons
as
experiment.
90. Count 4: Felony Failure to Wash
FCU
inves2gators
found
staining
that
didn’t
make
biological
sense.
Chadopadhyay,
et
al.
Perforin
always
expressed
with
Granzyme
B,
but
Granzyme
B
can
be
expressed
without
Perforin.
What
had
the
suspect
done
to
the
Granzyme
B+
Perforin-‐
cells?
91. Count 4: Felony Failure to Wash
During
a
vigorous
interroga2on,
suspect
broke
down
and
admided
that
he
had
forgoden
to
add
GrzB
an2body
during
staining.
Tried
to
cover
up
crime
by
adding
an2body
at
the
cytometer,
but
forgot
to
wash
it
out.
Moral:
Inves2gate
paderns
that
aren’t
consistent
with
biology.
Technical
mistakes
can
lead
to
non-‐specific
binding.
92. Count 5:
Misrepresentation without Titration
At
trial,
defense
adorney
submided
evidence
claiming
that
“Pra2p’s
an2body
conjugates
never
work
reliably,”
hoping
to
undermine
the
case
against
his
client.
93. Count 5:
Misrepresentation without Titration
Witnesses
report,
however,
that
defendant
“is
always
rushing,”
and
“seems
to
finish
staining
way
before
anyone
else
in
the
lab.”
Surveillance
footage
revealed
that
suspect
stained
at
4C
for
5
minutes.
FCU
2tra2ons
showed
that
reagent
was
fine.
Titra2on
under
same
condiEons
as
experiment
shows
that
staining
will
work
even
at
4C,
but
you
need
much
more
an2body.
0 5 10 15 20
Dilution
0
102
103
104
105
B515-A:CD8FITC
0
102
103
104
105
B515-A:CD8FITC
2' 37C
2' 4C
0 5
5' 4C
5' 37C
15 20
tion 15' rt
0 5 10 15 20
Dilution5' rt
0 5 10 15 20
Dilution 2hrs rt
0 5 10 15 20
Dilution 6hrs rt
0 5
5
minutes
4C
15
minutes
RT
Dilu6on
94. Summary
Resolving Problems in Flow Cytometry Data
• is easier when instrument, reagents, and staining
procedures are optimized.
• requires examination of ALL parameters (hidden
problems can exist with certain marker combinations that
aren’t present in gating tree).
• is based on pattern recognition, with some patterns
indicating a very easy to diagnose problem.
www.isac-net.org
95. Welcome to FCU, Now Here Are Your Cases
“Report of missing CCR7+ central memory cells.”
CD45RA
FITC
CCR7
A700
Mul2color
Panel
Percentage of CD45RA- CCR7+ cells is too
low in multicolor panel.
CD45RA
FITC
CCR7
PE
2-‐color
Test
What happened?
96. Welcome to FCU, Now Here Are Your Cases
“Report of missing CCR7+ central memory cells.”
CD45RA
FITC
CCR7
A700
Mul2color
Panel
Percentage of CD45RA- CCR7+ cells is too
low in multicolor panel.
CD45RA
FITC
CCR7
PE
2-‐color
Test
What happened?
1) Titration of A700 reagent?
2) Spreading error from
another marker in panel?
3) Poor reagent choice for
CCR7?
97. Welcome to FCU, Now Here Are Your Cases
“Domestic dispute, disagreement about whether new antibody works.”
10
0
10
1
10
2
10
3
10
0
10
1
10
2
10
3
Compensated
New
AnEbody
CD4
Cy5PE
Who’s right?
*
Compensated
data
98. Welcome to FCU, Now Here Are Your Cases
“Domestic dispute, disagreement about whether new antibody works.”
10
0
10
1
10
2
10
3
10
0
10
1
10
2
10
3
Compensated
New
AnEbody
CD4
Cy5PE
Who’s right?
*
Compensated
data
Events piling up
on axes
Spreading of
negative
distribution
“Holes” in distribution:
appear to separate
uniform populations
99. Welcome to FCU, Now Here Are Your Cases
“Domestic dispute, disagreement about whether new antibody works.”
10
0
10
1
10
2
10
3
10
0
10
1
10
2
10
3
Compensated
New
AnEbody
CD4
Cy5PE
Who’s right?
*
Compensated
data
Events piling up
on axes
Spreading of
negative
distribution
“Holes” in distribution:
appear to separate
uniform populations
100. Welcome to FCU, Now Here Are Your Cases
“Fraud: Vendor sold me a kit for making cell lines from single sorted cells.
Most of the cells didn’t grown into lines. Vendor claims user error.”
What could have gone wrong?
Full Panel
101. Welcome to FCU, Now Here Are Your Cases
“Fraud: Vendor sold me a kit for making cell lines from single sorted cells.
Most of the cells didn’t grown into lines. Vendor claims user error.”
What could have gone wrong?
1) Not really CMV+, just
fluorochrome aggregates.
2) No viability marker, all/most cells
are dead.
Full Panel
102. Welcome to FCU, Now Here Are Your Cases
“Fraud: Vendor sold me a kit for making cell lines from single sorted cells.
Most of the cells didn’t grown into lines. Vendor claims user error.”
What could have gone wrong?
1) Not really CMV+, just
fluorochrome aggregates.
2) No viability marker, all/most cells
are dead.
Full Panel
Viability
Gate
103. Welcome to FCU, Now Here Are Your Cases
“Serial criminal on the loose.
Find the crimes…”
Experimental
Stain
CD4 A488
CD3 PE
CD8 Cy7PE
DAPI
Lymphs
(Stained
today)
Beads
No stain
1. Set
voltages
Beads
CD3 FITC
Lymphs
CD3 PE
DAPI
Beads
CD4 Cy7PE
Beads
DAPI
Single
Colors
Universal
Nega2ve
Beads
No stain
(Stained
last
week
and
fixed)
2.
Compensa2on
Lymphs
No stain
3.
Ga2ng
104. Welcome to FCU, Now Here Are Your Cases
“Serial criminal on the loose.
Find the crimes…”
Experimental
Stain
CD4 A488
CD3 PE
CD8 Cy7PE
DAPI
Lymphs
(Stained
today)
Beads
No stain
1. Set
voltages
Beads
CD3 FITC
Lymphs
CD3 PE
DAPI
Beads
CD4 Cy7PE
Beads
DAPI
Single
Colors
Universal
Nega2ve
Beads
No stain
(Stained
last
week
and
fixed)
2.
Compensa2on
Lymphs
No stain
3.
Ga2ng
105. Review
Recognize Problematic Staining Patterns
Troubleshoot Experiments and Analysis
Characteristic patterns, tools to identify trouble,
strategies to troubleshoot
Guidelines to Improve Cytometry Experiments.
Stop and Setup, Cautiously Experiment, Drive New Tools for QC
106.
Thank
you!
Help
us
compile
more
troubleshoo2ng
examples!
If
you
have
good
troubleshoo2ng
examples,
please
email:
pchadop@mail.nih.gov
Perhaps
we
can
have
a
public,
anonymous
repository
for
such
data
through
FlowRepository
or
CytoU.