Rare Event Detection

1
Rare Event Detection
Albert D. Donnenberg, Ph.D. UNIVERSITY of
PITTSBURGH CANCER INSTITUTE
donnenbergad_at_upmc.edu
Clinical Flow Course 9/05/04
2
Searching for a Needle in a Haystack

• Find the needle
• Determine that it really is a needle
• Make measurements to determine what kind of a
  needle it is

3
Key Elements

• Event frequency
• Inherent property of sample
• Enrichment possible
• Signal to noise ratio
• Minimize noise
• Nonspecific binding (1 mouse Ig)
• Cellular autofluorescence (dump gate, green or
  red excitation, quenching dyes)
• Doublets (ratio of peak height/integral or peak
  height/width)
• Sporadic mechanical or electrical noise (time
  parameter)
• Dead cells (vital dyes)
• Maximize signal
• Best fluorochrome for most critical
  determination
• Optimal antibody concentration

4
Know Your Own Limit (of Detection)

• Limit of detection
• Frequency of false positives in appropriate
  negative control (FMO isotype control, FMO
  isoclonic control, TMer binding of MHC disparate
  cells, known negative sample)
• Calculate upper 95th or 99th percentile of the
  frequency false positive in a series of negative
  controls
• Caution Rare events are log normally
  distributed. Use arithmetic means and you will
  get the wrong answer!

5
Pull the Noise From the Signal

• Dump channel
• Unique location in multiparameter space
• Use the best fluorochrome for the most critical
  measurement
• PE has high quantum efficiency
• Red line used to excite APC and APC tandems
  excites less cellular autofluorescence

6
For a reagents available in several fluorochromes
choose the one with the best signal to noise
ratio for your critical measurement
7
How Many Cells to Acquire

• Short answer The rarer the event the more cells
  required
• Long answer Depends on
• Event frequency
• Tightness of event cluster in multiparameter
  space
• You can determine the number empirically by
  determining the precision of replicate
  determinations
• No matter how many events you acquire, the limit
  of detection is governed by the signal to noise
  ratio

8
Precision of Replicate Determinations
All events in three 5 mL aliquots of leukocyte
depleted platelet product were acquired
Detection of leukocytes in filtered platelet
components Donnenberg et al Transfusion, 2000.
9
Autologous HSCT for SSc

• NIAMS-funded study Auto HSCT in rapidly
  progressing SSc
• CY (2g/m2 over 24 hours), G-CSF (10 mg/kg/day sq)
  mobilization until WBCgt2500
• Apheresis, T-depletion by Isolex 300i positive
  CD34 selection, negative CD3 depletion

10
Single Platform Absolute T-cell Count Day 3
WBC (NOT R1)
CD3 (R2 NOT R1)
CD3 CD45HI (R2 R3 NOT R1)
CAL BEADS (UNGATED)
CD45 VIABILITY (R2 NOT R1)
CD3 VIABILITY (R2 R3 R4 R5 NOT R1)
CD3CD45HI FSINT (R2 R3 R4 NOT R1)
TOTAL CD3 VIABILITY (R2 R3 R4 NOT R1)
0.655 of WBC 7.86 CD3/mL
99.0 WBC Viability
99.3 CD3 Viability
0.2 mL whole product x 6-plicate exhaustively
acquired
C5337.LMD
11
Single Platform Absolute T-cell Count Day 0
WBC (NOT R1)
CD3 (R2 NOT R1)
CD3 CD45HI (R2 R3 NOT R1)
CAL BEADS (UNGATED)
CD45 VIABILITY (R2 NOT R1)
CD3 VIABILITY (R2 R3 R4 R5 NOT R1)
CD3CD45HI FSINT (R2 R3 R4 NOT R1)
TOTAL CD3 VIABILITY (R2 R3 R4 NOT R1)
0.028 of WBC 0.99 CD3/mL
99.7 WBC Viability
100 CD3 Viability
C5256.LMD
12
T-cell Kill Trajectory CY/Flu/ATG
13
Response of healthy A2 control subject to
influenza tetramer (peptide GILGFVFTL)
NOT CD14
NOT dead cells
AND CD3
Dump gate
Cytometry, 41321-328, 2000
14
Reciprocal frequencies (and positive) are log
normally distributed
15
Lower limit of detection defined as log mean of
HLA-A2 negative subjects plus 2 SD 1/7000
A2 Negative
p53 Bulk Line
Reciprocal Frequency
A2 Positive
HN
Normal Control
Reciprocal Frequency
16
TCR Vb Usage in T-cell subsets of HIV infected
subjects
HIV-infected patients (on HAART) recruited from
the Multicenter AIDS Cohort Study, Pittsburgh
center.
17
6 fluorescent parameters on a 5-color instrument
18
Sequential 5-color gating strategy
Ungated
A 14.1
A 14.1
SS LOG
Outcomes
Gate A and F and C
Gate A
Vb 2 4.9
Vb 5.2 1.3
F 97.8
Classifiers
SS LOG
FL2 2/ 5.2
Vb 12 2.4
Naive
FS Lin
FL1 12 / 2
Gate A and F
Gate A and F and E
B 34.5
C 14.6
Vb 5.2 0.2
Vb 2 0.3
FL4 CD27 PC5
FL2 2/ 5.2
Effectors
D 26.7
Vb 12 0.1
FL1 12 / 2
FL3 CD45RA ECD
19
Vb9 Expansion ?
How to define expansions and deletions?
HIV23.1 3/13/02
20
Predictions of the Normal Model
Expansion
Deletion
For our control dataset 2.5 of 24 possible Vb
21 control subjects 12.6 expansions and 12.6
deletions
21
Control Subject VB Usage, Total CD4
25 expansions and 3 deletions!
22
CD4 Percent is Normally Distributed
x
s
Percent CD4
23
Vb usage is log-normally distributed
24
Control Subject VB usage, total CD4
12 expansions and 14 deletions (expected 12.6)
25
Conclusions

• Vb usage is log-normally distributed.
• Failure to use log transformed data results in an
  overestimate of expansions, an underestimate of
  deletions, and loss of power to detect
  significant differences using parametric tests.

26
Measuring Vb Repertoire on TMer T-cells
A
G
F
B
V-beta 9
V-beta 18
V-beta 17
V-beta 5.1
FSC Width
SSC
V-beta PE
V-beta PE
V-beta 20
V-beta 16
FSC
FSC
V-beta FITC
V-beta FITC
D
C
Flu TMer APC
CD8 PC7
FSC
CD4 ECD
27
For Multicolor Analysis, Group Parameters as
Classifiers or Outcomes

• Classifiers
• Primary do not branch (e.g. singlets AND CD4 low
  side scatter)
• Secondary branch (e.g. memory, naïve and effector
  populations based on CD45RA and CD27 expression)
• Outcomes
• Measured on each population defined by the
  classifiers
• Color event and backgate to the classifiers

28
Hierarchical Gating Strategy for Multiparameter
Functional Flow
Color-event positive outcomes on classifier
dotplots and exploratory dotplots
Outcomes
CD71 CD25
CD71 CD25
CD71 CD25
CD71 CD25
Naive
C. Mem
Eff Mem
Late Eff
CD45RA x CD27
Secondary Classifier
FSC x LSSc (Live)
CD8 x CD8b (CD8 T cells)
CD45 LSSc (Lymphs)
Primary Classifiers
29
1o Classifiers
Live CD4
DIP d5 CD4
1989.LMD
A
B
40
3
A
B
CD25 PC5
D
C
51
11
C
D
CD71 FITC
2o Classifier
Outcomes
30
Gated on 1o Classifiers
SSc log
CD27 PE
FSc
CD45RA ECD
Color Event the Outcomes
Project onto classifier dotplots
31
The Approach

• CMV high A2.01 donor
• CMV pp65 tetramer added live during acquisition
  (kinetic measurements of tetramer binding)
• Kinetic measurement of calcium flux measured by
  Indo-1 ratio
• Sample maintained at 37 during 30 min
  acquisition
• High resolution immunophenotyping using a
  hierarchical gating strategy
• Beckman-Coulter NextGen prototype cytometer

32
The Panels
33
Kinetics of Tetramer Binding
CD4
CD8
APC CMV TMer
FSc
Time (0 30 min)
CMV pp65 - NLVPMVATV
473.lmd
34
Kinetics of Tetramer Binding
CMV Tetramer
Ionomycin
SDF1
1023
1023
1023
1023
1023
682
682
682
682
682
APC_TMER
APC_TMER
APC_TMER
APC_TMER
APC_TMER
341
341
341
341
341
0
0
0
0
0
341
682
1023
0
341
682
1023
0
341
682
1023
0
341
682
1023
CD8 Tmer
TIME
TIME
TIME
TIME
Scale 1023 30 minutes Running median smooth
CD8 Tmer-
CD4
473.lmd
35
Kinetics of Calcium Flux
473.lmd
36
Kinetics of Calcium Flux
CMV Tetramer
Ionomycin
SDF1
512
512
512
512
512
384
384
384
384
384
RATIO
RATIO
RATIO
RATIO
RATIO
Indo S Fl / Indo L Fl (RATIO)
256
256
256
256
256
128
128
128
128
128
0
0
0
0
0
0
341
682
1023
0
341
682
1023
0
341
682
1023
0
341
682
1023
0
341
682
1023
TIME
TIME
TIME
TIME
TIME
CD8 Tmer
Scale 1023 30 minutes
CD8 Tmer-
CD4
473.lmd
37
Primary Classifiers
SSc log
CD45
CD8
48680.lmd
CD8b
38
Secondary Classifier
Naive
Central Memory
Effector
Effector Memory
48680.lmd
39
Color Eventing Outcomes on Classifier and
Exploratory Parameters
48680.lmd
40
Color Eventing Outcomes on Classifier and
Exploratory Parameters
48680.lmd
41

Color Eventing Outcomes on Classifier and
Exploratory Parameters
CMV Tetramer positive
VB 13.1
VB 5.3
VB 13.6
VB 7.1
VB 3
VB 8
506.lmd
42
Conclusions

• When dealing with multiple parameters it is
  useful to think of parameters as classifiers,
  outcomes, or exploratory
• This facilitates a hierarchical analysis that
  avoids the all possible permutations problem (ie
  4096 distinct parameter combinations in this
  example)
• It is feasible and practical to perform
  functional, kinetic and high resolution
  immunophenotypic studies on populations that are
  considered rare events
• Hongmei Shen and Vera Donnenberg will show how
  these principles can be applied to detection of
  normal and malignant stem cells

43
Acknowledgements

• Vera Donnenberg
• Members of AVDLab past and present
• E Michael Meyer
• Debe Griffin
• Hongmei Shen
• Erin McClelland
• Cassandra Singer
• Thomas Hoffmann
• Sumita Ganguly
• Dawn Betters
• Anita Popovic
• Kit Snow, Todd Lary, Meryl Forman and Tom Franks
  at Beckman Coulter