Multicolor Flow Cytometry

1
Multicolor Flow Cytometry

• How is it like Astronomy?
• How should we setup instruments?
• How should we set compensation?
• How do we look at data?
• How should we choose reagents?

Alan Stall Vancouver October 2005
2
Overview

• Why Use Multi-color?
• Resolving Positive from Negative
• Factors limiting population resolution
• Background versus Variation (Spread)
• Instrument Set-up and Optimization
• Compensation in a Multi-colored World
• Correcting Fluorescence Spillover
• Analog vs. Digital Compensation
• Tandem Dyes
• Visualizing Compensated Data
• Choosing Reagents for Multi-color Experiments

3
Why Multicolor?

• Cell Lineage
• Subset identity
• Activation status
• Cell Cycle
• Chemokine Production
• Signal Transduction
• Migration/Homing Phenotype

4
Blue/Green, Red and Violet laser dyes

• Current BD Biosciences Immunofluorescent Dyes

5
Multicolor Flow Cytometry the reality

• Chemistry the fluorescent dyes
• Must be bright (high quantum efficiency is
  desirable)
• Minimum spectral overlap (brighter means more
  spill too)
• Straightforward conjugation to antibodies
• Instrumentation
• Multilaser (2 - 4 lasers for 6 - 17 colors)
• 12 to 16 or more parameters
• Digital electronics high resolution ADC (14 bit
  minimum)
• High sensitivity (lt 100 MESF PE is desirable)
• Designing experiments is mostly empirical
• Can take weeks to put together a good 8 color
  panel
• Research gt 625 mAb at Stanford, for example
• Some reagents have more nonspecific binding than
  others

6
Current trends require more colors

• More colors more independent information

16 colors 64X
256 colors 256X
4 colors 32X
Joe Trotter
7
The Big Challenge Signal / Background

• The most basic information to be derived from any
  flow cytometry experiment is
• Whether a cell of interest is positive for a
  given marker
• Experimentally, positive cells are those that can
  be resolved from negative control cells
• The sky is full of stars, invisible by day
  -Henry Wadsworth Longfellow
• Like astronomers, we must deal with background
  contributions as the major challenge in detecting
  dim events.

8
Resolution vs. Background
Negative Population
Positive Population
Negative population has low
background Populations well resolved
Negative population has high
background Populations not resolved
Negative population has low background
high CV (Spread) Populations not resolved
The ability to resolve populations is a function
of both background and spread of the negative
population
9
Review Things That Impact Resolution
Resolution Sensitivity
10
Measure Signal/Background Stain Index1
Goal Normalize the signal to the spread of
background where background may
be autofluorescence, unstained cells, or
compensated cells from another dye dimension.
0
1Stain Index Dave Parks, Stanford
11
Comparing Dim to Background

• Two populations with 10X means may or may not be
  separable.

300 ? 75
30 ? 30
300 ? 185
30 ? 35
12
Spillover Increases Spread
CD8 T Cells No PE Stain
Resolution for a given fluorescence parameter is
decreased by increased spread due to spillover
from other fluorochromes.
Unstained
FITC CD3
FITC CD3 and FITC CD45
This spread is NOT eliminated by Compensation
More colors more spillover
MFI PE
13
So how do we optimize an instrumentfor resolving
dim cells?

• Various methods for establishing PMT gain
• Putting unstained in the 1st decade
• Vulnerable to log scale visual artifacts
  misleading investigator and subjective bias
• I just increase PMT voltage until it looks
  good.
• I lower the PMTs to bring the negative cells in
  the first decade
• Subjective sometime good, sometimes not
• Getting stable signal to background (pos neg)
• Insensitive to population variance
• Useful only for Pulse height (analog)
  measurements
• Achieving a constant CV range using very dim
  particles setup using positives
• Requires Area measurements to correctly determine
  CV
• Probably the best for digital in insures optimal
  resolution of dim from unstained

14
H vs A Pos, Dim and Bkg
300 V
600 V
400 V
500 V
Height
Area
4 Decade Log Scale
15
So how do we optimize an instrumentfor resolving
dim cells?

• Various methods for establishing PMT gain
• Putting unstained in the 1st decade
• Vulnerable to log scale visual artifacts
  misleading investigator and subjective bias
• I just increase PMT voltage until it looks
  good.
• I lower the PMTs to bring the negative cells in
  the first decade
• Subjective sometime good, sometimes not
• Getting stable signal to background (pos neg)
• Insensitive to population variance
• Useful only for Pulse height (analog)
  measurements
• Achieving a constant CV range using very dim
  particles setup using positives
• Requires Area measurements to correctly determine
  CV
• Probably the best for digital in insures optimal
  resolution of dim from unstained

16
Classic Setup Error
Day 1
S/N
A
B
C
Day 14
Conclusion
Positive cells are 5x brighter on Day 1 (B) when
compared to Day 14 (D)
D
E
17
So how do we optimize an instrumentfor resolving
dim cells?

• Various methods for establishing PMT gain
• Putting unstained in the 1st decade
• Vulnerable to log scale visual artifacts
  misleading investigator and subjective bias
• I just increase PMT voltage until it looks
  good.
• I lower the PMTs to bring the negative cells in
  the first decade
• Subjective sometime good, sometimes not
• Getting stable signal to background (pos neg)
• Insensitive to population variance
• Useful only for Pulse height (analog)
  measurements
• Achieving a constant CV range using very dim
  particles setup using positives
• Requires Area measurements to correctly determine
  CV
• Probably the best for digital in insures optimal
  resolution of dim from unstained

18
PMT Gain effects Area vs Height

• Height Measured Signal/Background is a function
  of gain use ratio of positive/background to
  optimize gain
• Area Measured Signal/Background is constant
  use CV of dim particles to optimize gain

19
Setting PMTs on an Analog Instrument
S/N Ratio Pos/Neg
FACS Calibur
PMT Voltage

• The optimal PMT voltagesetting for each
  fluorescence channel is the point at which there
  is no increase in the S/N ratio of a negative and
  dim population with increased voltage

20
Pulse Height and Area Differences

• Pulse Height has unsigned integer baseline
  resolution (i.e., no negative numbers) and is
  multiplied by 16 (a 14 bit to 18 bit shift).
• Pulse Area has IEEE float baseline resolution and
  is the sum of the scores. Has an 18 bit range.
• It is important to override underlying noise
  with enough gain to optimize (minimize) the CV
  (pulse area).

21
So how do we optimize an instrumentfor resolving
dim cells?

• Various methods for establishing PMT gain
• Putting unstained in the 1st decade
• Vulnerable to log scale visual artifacts
  misleading investigator and subjective bias
• I just increase PMT voltage until it looks
  good.
• I lower the PMTs to bring the negative cells in
  the first decade
• Subjective sometime good, sometimes not
• Getting stable signal to background (pos neg)
• Insensitive to population variance
• Useful only for Pulse height (analog)
  measurements
• Achieving a constant CV range using very dim
  particles setup using positives
• Requires Area measurements to correctly determine
  CV
• Probably the best for digital in insures optimal
  resolution of dim from unstained

22
PMT Voltage and Instrument Performance
Optimal gains can be predicted by measuring the
CV of dim particles enabled by digital system
and pulse area measurements
23
Optimizing PMT voltage the rules

• Establish Instrument Baseline A sensible PMTV
  value for each detector as a default starting
  point
• Digital Increase PMT voltage until no further
  improvement (decrease) in CV is observed on (dim)
  setup particles
• Analog Increase PMT voltage until no further
  improvement in S/N is observed on positive and
  negative setup particles
• PMT voltage should not be lower than baseline to
  begin characterizing stained cells
• Run stained sample If several photoelectrons/cell
  in stained background (high signal), then PMT
  voltage may be lowered until lower part of
  background approaches noise (can be good
  information if left alone)
• Difficult to determine on analog instruments
• In samples with very bright fluorescence, PMT
  voltage should be lowered to keep brightest on
  scale regardless of where background ultimately
  resides

24
7 Color Family of Curves for Dim CVs
25
Compare two different ways to set
gainAutofluorescence in 1st decade vs CD4,
CD4dim (FACSArray)
CD4 monocytes
CD4 lymphocytes
CD4- lymphocytes
Human blood stained with APC CD4 using the lyse
wash method. For the sample on the left, the
APC gain was set by acquiring an unstained sample
and placing the unstained population in the first
decade. The gain for the right hand sample was
set by placing the CD4 lymphocytes at
approximately channel 50,000.
26
Digital Multicolor Setup the best approach

• Rules
• If stained background is very high, gain may be
  lowered until background population approaches
  unstained control
• If brightest positives are too high, the gain
  must be lowered to place them on-scale regardless
  of where background ultimately resides

Once per configuration
Step 1 Characterize Instrument and establish
baseline gains for each color
Single Color Controls FMO controls
27
Compensation

• The single most common source of data error in
• multi-color
• flow cytometry experiments

28
Setting Compensation

• Myths about setting compensation
• You should set your compensation on the same
  tissue you are going to analyze
• Compensation controls should be the same
  intensity as the reagent to be use
• Bright reagents require more compensation than
  dull reagents
• Compensation can be set by eye
• Compensation settings can be saved and used from
  day-to-day
• Improper compensation doesnt affect the data
  very much

29
Fluorochromes emit in other channels
Spillover into other detectors contributes
background to those detectors
Remember the basic assumption of flow analysis
The signal in FL1 the signal from FITC and
only FITC and the signal in FL2 the
signal from PE and only PE.
This is NOT TRUE for the raw data collected in
each FLx parameter! The process by which
each fluorescence channel is corrected for this
spectral spillover is termed Fluorescence
Compensation
30
Basic Principles of Compensation -
This subtraction is done electronically on analog
instruments.
y
Following Compensation
x
FITC FL1 - x PE
PE FL2 - y FITC
FL1 FITC x PE
COMPENSATION
FL2 PE y FITC
PRE
POST
FL2
To get a true measure of the PE signal in FL2 you
have to subtract a percentage of any PE signal
present in the cell. This is compensating the
signal.
FL1 (FITC)
31
Principles of Compensation - Calculations
Calculation of FL1 Spillover into FL2 (FL2 - FL1)
-cells
cells
FL2 Mediancells

FL2
FL1 Mediancells
430
4.7
20.3

FL1 (FITC)
1980
4.1
POST
32
Setting Compensation - Analog

• We correct for dye spillover to align stained
  populations in dye space without bias from
  spectral overlap.
• Analog systems electronically subtract pulses
• All adjustments are made pair-wise FL1- FL2,
  etc.
• All possible adjustments are not available FL2-
  FL4
• Analog systems dampen spread and distort means
  due to errors in compensation circuits and logamp
  nonlinearity
• On analog systems deviations in the log/linear
  amps can adversly affect compensation
• One spillover adjustment may not work for all
  intensity levels

33
Analog Pulse Subtraction Errors1
Fully compensated signals tend to be
systematically over compensated
1 Bigos M, et al, Nine Color Eleven Parameter
Immunophenotyping Using Three Laser Flow
Cytometry, Cytometry 3636-45, 1999.
34
Analog Error Example

• Stain for the Y reagent at high, medium, low
  and blank levels.
• Split each sample and stain half with the X
  reagent and half with mock stain.
• Recombine the corresponding pairs and analyze at
  various compensation settings.
• No compensation value satisfies all tubes of the
  same stain.

35
Software Compensation on Digital Instruments
FITC Spillover calculation AutoCompensation
method (PE 0.83)
Not a subtraction, rather a correction because we
use matrix algebra and compensation coefficients.
Spillover coefficient slope
1.
2.
7
2
3.
10
PEc PE x 1.00209 FITC x -0.25203
4.
36
Analog vs Digital Compensation
37
Laboratories that have both analog digital

• Many labs have FACSCaliburs and one or more
  digital cytometers
• The data will look a little different
• Digital data and software compensated analog data
  are very similar
• Digital data and analog compensated data can look
  different due to analog error
• What to suggest?
• Set the Calibur up normally, but
• Either turn off compensation completely
• Or partially compensate
• Then finish compensation in software

38
Compensation reveals the inherent variation
(spread) in the data
FL3
Using BD CompBeads with different levels of Ab
39
Aligning the means measurement error remains
The linear plot below shows the alignment of
compensated CD20 FITC positive and negative
capture beads in the PE dimension.
The CD20 standard deviation decreases from 192
to 33.4 after compensation yet the CV goes to
1500 because the mean drops from 2010 all the
way to 2.2.
The Coefficient of Variation (CV) is
inappropriate with compensated data. CV 100 x
Std/Mean, so as the mean approaches 0, the CV
approaches 8
40
PE Background Due to FITC Spillover

• Without compensation, the amount of PE MESF
  background contributed by bright CD45-FITC
  staining can be determined (8700 PE MESF).

• 2 Color Data

CD45- FITC
CD45 FITC only
41
Effect of Spillover on Double Stained Cells
Compensated data CD45 FITC makes dim CD4,CD45
difficult to distinguish due to FITC
spillover into PE and resultant spread
CD45 PerCP allows same dim CD4CD45 cells to
be distinguished from background (Little
spillover into PE)
42
Compensation
Setting up Contols for Compensation
In a perfect world compensation controls should
be run for every fluorochrome for every experiment
43
4 Simple Rules for Setting Compensation

• 1. The fluorescence spectrum ( Spillover) of the
  compensation control reagent should be identical
  to the reagent used in the experiment
• a. Critical for Tandem reagents where different
  reagents can have different spillovers
• b. Even similar fluorochromes like FITC / Alexa
  488 or APC/ Cy5 should be compensated for
  separately
• 2. The negative and positive populations must
  have the same autofluorescence
• a. Critical when using stained cells as
  compensation controls
• i. i.e. compare CD3 lymphocytes and CD3-
  lymphocytes
• ii. dont use CD3 lymphocytes and CD3- monocytes
• 3. The positive population should be as bright as
  possible
• a. Insures values in spillover channel are above
  background
• b. Provides greater accuracy for calculating low
  (but real) spillovers
• FITC -gt PE-Cy7 (1-2) or APC -gt PE-Cy7
• 4. Take enough events to get statically accurate
  numbers
• a. 10,000 events gives a 1 SD
• b. Provides greater accuracy for calculating low
  (but real) spillovers

44
Tandem Dye Reagents

• Tandems - PE-Cy5, PE-Cy7, PerCP-Cy5.5

Typically, the chemical coupling of a protein
fluorochrome (e.g. PE) which acts as the donor
and a small organic fluorochrome (e.g. TR) which
acts as the acceptor
PE-Cy7
45
Tandem Compensation Different lots have
different spillovers
Comparison of FL2- FL3 Compensation required
for 4 different PE-TR conjugates
46
Compensation Tandem Dyes
Lot-to-Lot Differences in spectral spillover
REQUIRES that there is a compenstion control tube
for EVERY tandem reagent in an experiment This
is possible with AutoComp in Diva software
47
4 Simple Rules for Setting Compensation

• 1. The fluorescence spectrum ( Spillover) of the
  compensation control reagent should be identical
  to the reagent used in the experiment
• a. Critical for Tandem reagents where different
  reagents can have different spillovers
• b. Even similar fluorochromes like FITC / Alexa
  488 or APC/ Cy5 should be compensated for
  separately
• 2. The negative and positive populations must
  have the same autofluorescence
• a. Critical when using stained cells as
  compensation controls
• i. i.e. compare CD3 lymphocytes and CD3-
  lymphocytes
• ii. dont use CD3 lymphocytes and CD3- monocytes
• 3. The positive population should be as bright as
  possible
• a. Insures values in spillover channel are above
  background
• b. Provides greater accuracy for calculating low
  (but real) spillovers
• FITC -gt PE-Cy7 (1-2) or APC -gt PE-Cy7
• 4. Take enough events to get statically accurate
  numbers
• a. 10,000 events gives a 1 SD
• b. Provides greater accuracy for calculating low
  (but real) spillovers

48
4 Simple Rules for Setting Compensation

• 1. The fluorescence spectrum ( Spillover) of the
  compensation control reagent should be identical
  to the reagent used in the experiment
• a. Critical for Tandem reagents where different
  reagents can have different spillovers
• b. Even similar fluorochromes like FITC / Alexa
  488 or APC/ Cy5 should be compensated for
  separately
• 2. The negative and positive populations must
  have the same autofluorescence
• a. Critical when using stained cells as
  compensation controls
• i. i.e. compare CD3 lymphocytes and CD3-
  lymphocytes
• ii. dont use CD3 lymphocytes and CD3- monocytes
• 3. The positive population should be as bright as
  possible
• a. Insures values in spillover channel are above
  background
• b. Provides greater accuracy for calculating low
  (but real) spillovers
• FITC -gt PE-Cy7 (1-2) or APC -gt PE-Cy7
• 4. Take enough events to get statically accurate
  numbers
• a. 10,000 events gives a 1 SD
• b. Provides greater accuracy for calculating low
  (but real) spillovers

49
Why Bright Compensation Controls
Estimating a low spillover fluorescence
accurately is impossible if the level
fluorescence into the spillover channel is near
or at background levels in that channel
From Mario Roederer - NIH
50
Types of Compensation Controls

• Beads stained with dyes similar to a known
  fluorochrome
• Very Bad- small differences in emission spectra
  can result in large compensation errors (Rule 1)
• Beads coated with a known fluorochrome
• CaliBrite / 7-color Beads coated with FITC, PE,
  APC, APC-Cy7, etc.
• Advantages
• Easy to use automated set-up and compensation
• Limitations-
• Spectra (especially tandems) may be slightly
  different than a conjugated mAb
• Best used with pre-defined clinical cocktails
  with pre-tested reagents
• Limited number of fluorochrome-conjugated beads

51
Types of Compensation Controls

• Cells stained with a fluorochrome-conjugated mAb
• Advantages
• Best match of spectra
• Can be used for any fluorochrome
• Disadvantage
• Have to stain cells
• Assume identical autofluorescence of and - pop.
  (Rule 2)
• Positive stained cells may be dull or few in
  number (Rules 3 4)
• Beads which capture a defined amount of
  conjugated mAb
• BDTM Anti-mouse/Rat IgG CompBeads
• Advantages
• All of the above
• Disadvantages
• None of the above

52
Setting-up a Flow Cytometer - Compensation
BD Anti-Mouse CompBeads
FITC
PE
FL3
APC
APC-Cy7
mAb
53
Summary

• Spillover from other fluorescence channels will
  increase the background and spread (CV) of a
  given fluorescence parameter.
• Compensation can remove the background component
  but NOT the increased spread
• Software (digital) compensation is more accurate
  than electronic (analog) compensation
• Where possible compensate analog data off-line
  with software
• Correct Compensation Controls are the key to a
  good Multi-color experiment
• Run Compensation controls for EVERY parameter in
  an experiment
• Run individual compensation controls for EACH
  tandem reagent
• In general Ig-capture beads are the best controls

54
Analyzing Data
Visualizing Compensated Data
55
Full Log Display Visualization Artifacts
8 modeled populations 2 of which are double
positive
Difficult with low autofluorescence and
compensation because of high spillover (22) of X
into Y, low spillover (3) of Y into X causes
high background of X into Y on single positive
bright X population, which inflicts significant
data spread after compensation.
56
Full Log Display Visualization Artifacts
8 modeled populations 2 of which are double
positive
57
FMO (Fluorescence Minus One)

• Compensated data exhibits spread
• Bright single positives may change threshold
  levels between dim and background in other
  dimensions
• Unstained and/or isotype controls are NOT useful
  for determining threshold over background
• The best control is one stained with all reagents
  except the one of interest

58
Use Fluorecence Minus One (FMO) Controls to
Identify Positives
PBMC were stained as shown in a 4-color
experiment. Compensation was properly set for
all spillovers
Unstained Control
Fully Stained
FITC
None
CD3
PE
None
CD4
Cy5PE
None
CD8
Cy7PE
None
CD45RO
5
10
4
10
3
10
2
10
1
10
PE
0
10
0
1
2
3
4
0
1
2
3
4
0
1
2
3
4
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
FITC
Courtesy Mario Roederer
59
The Biexponential Scale
Standard Log Scale
60
The Biexponential Scale
The Biexponential transformed display shows
aligned populations in the APC-Cy7 dimension
Antibody capture beads stained with 3 levels of
an APC reagent
61
CD20 FITC Capture Beads
Biexponential display reveals proper compensation
better than log
2 over
Correct
62
Transformation Comparison
FlowJo 6.1.1 Transformed
DiVa 4.1 Transformed
63
Visualizing / Analyzing Data

• Summary
• Log plots of highly compensated data can be
  visually misleading many edge effects
• Biexponential plots provide the best
  visualization of compensated data
• Easier to detect errors in compensation
• Easier to see spread in populations set correct
  gates
• Quadrant gates are of limited use in multi-color
  experiments
• Use bent quadrant gates or individual gates

64
Designing Multi-color Experiments
Choosing Reagents
65
Factors in Selecting Reagents

• Intensity of the Fluorochrome
• Stain index
• Density of the specificity on the cell
• Autofluorescence of the cell in the channel
• Lower in red channels
• Spillover of other fluorochromes into the channel
• Spillover increases spread
• Adding colors generally adds background and
  usually complicates population resolution
  spread increases
• APC ?PE, PerCP-Cy5.5, Alexa 700, APC-Cy7, AmCyan
• PE ?FITC, PE-Alexa610, PE-Cy7, AmCyan

66
Multicolor Design more colors

• An antibody/dye combination that marginally
  allows discrimination of positives/negatives in a
  single color assay is unlikely to contribute
  anything helpful in a multicolor experiment.
• Which is more important, a marker being turned
  on/off or an increase from 1000 to 2000?
• Multicolor experiments help identify relative
  changes in populations
• These are often changes in the expressing a
  marker
• Other times the change is in the expression level
• Sometimes both and level of expression change

67
Which Fluorochrome For Which Specificity?

• For most experiments the parameters are either
  Classifying or Experimental
• Classifying e.g. CD4, CD8- define
  sub-populations
• Experimental Unknown or changing expression
  levels
• In general Experimental parameters need the best
  low level resolution (especially if expression
  level is unknown)
• Use fluorochromes with brightest intensity (Stain
  Index)
• Use fluorochromes with minimal spillover from
  other channels
• Total spillover is a function of the sum of the
  spillover x intensity from all other channels
• Defining parameters need sufficient brightness to
  resolve the population
• In general use bright fluorochromes for low
  density antigens (e.g. CD25) and dim
  fluorochromes for high density antigens (CD45,
  CD4)
• Keep in mind how much will fluorescence from
  these reagents contribute to background spread in
  the experimental channels

68
BDB-Pharmingen CD4 Reagent Brightness
CD4 ranked by Stain Index
100
90
80
70
S/N
60
SI
50
40
AF
30
20
10
0
PE
FITC
APC
PerCP
A647
A610
A488
A700
PE-Cy5
PE-Cy7
PacBlue
AmCyan
APC-Cy7
PerCP-Cy5.5
69
How bright is a marker? How to compare?
70
Spiilover of Fluorochromes FACS Canto
Spillover Column into Row (calculated using
7-color beads and FACS Canto software
Note Few fluorochromes spill into FITC
All fluorochromes spill into PerCP
Before designing a multi-color experiment KNOW
ALL the spillovers for the fluorescence
parameters to be used in your experiment
71
Example CD4 PerCP-Cy5.5 Spread
PE-Cy7
APC
Alexa 700
APC-Cy7
Raw
6.3X
6.6X
1.9X
1.9X
Compensated
If using CD4-PerCP-Cy5.5. the dont use PE-Cy7 or
Alexa700 for a parameter requiring low level
resolution
72
Summary

• Background from spillover and other sources
  remains a challenge in multicolor experiments
• Pulse Area is more accurate than Pulse Height
• BD systems have an 18 bit range (max value), but
  the resolution is that of a 32 bit IEEE float
• Digital systems may be easily optimized by
  evaluating the behavior of very dim particles
• The biexponential (and other) transformations
  render immunofluorescence data much better than
  log
• Normalized signal to background, the Stain Index,
  provides a good way to compare the brightness of
  a stained reagent over different backgrounds, and
  labeled with different stains

73
Acknowledgements

• Joe Trotter
• Bob Hoffman
• Dennis Sasaki
• Tiffany Clarke
• Holden Maecker
• Mario Roederer
• David Parks

74
6 Color Example Murine LN

• CD44 FITC, CD122 PE, B220 PE-Cy5.5, CD4 PE-Cy7
• CD25 APC, CD8 APC-Cy7