In the analysis of qPCR data, the Cq values of replicate reactions is required to differ no more than 0.5 cycles. However, the sampling error that occurs when pipetting a low number of target molecules into the PCR plate is governed by the Poisson distribution. For each Cq and PCR efficiency value we calculated this unavoidable range of Cq values. This range increases with higher Cq values (less target). A decision to exclude replicate reactions based on this expected sampling error avoids bias, prevents unwanted loss of data and increases the statistical power. For a dataset with replicate qPCR measurements of 12 miRNA targets in 834 patients (20,016 reactions) the fraction of excluded measurements decreased from 39% to 7%.In the analysis of qPCR data, the Cq values of replicate reactions is required to differ no more than 0.5 cycles. However, the sampling error that occurs when pipetting a low number of target molecules into the PCR plate is governed by the Poisson distribution. For each Cq and PCR efficiency value we calculated this unavoidable range of Cq values. This range increases with higher Cq values (less target). A decision to exclude replicate reactions based on this expected sampling error avoids bias, prevents unwanted loss of data and increases the statistical power. For a dataset with replicate qPCR measurements of 12 miRNA targets in 834 patients (20,016 reactions) the fraction of excluded measurements decreased from 39% to 7%.
Jan Ruijter, Assistant Professor, University of Amsterdam, The Netherlands
1. Medical BiologyRuijter et al, London, December 2017
Discordance between
replicate qPCR reactions
Jan M Ruijter
Department of Medical Biology
Academic Medical Center
Amsterdam, the Netherlands
2. Medical BiologyRuijter et al, London, December 2017
qPCR data
N N En 0
n
=
number of PCR cycles
start concentration
PCR product
after n cycles
Efficiency (1-2; 2=100%)
20 21 22 23
1
2
2
4
cycles
N
0
1
3
8
etc.
PCR Efficiency = fold increase per cycle
3. Medical BiologyRuijter et al, London, December 2017
0.00000001
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
100
0 5 10 15 20 25 30 35 40
Amplification Curve
Nn
𝑁𝑐 = 𝑁0 𝐸 𝐶
4. Medical BiologyRuijter et al, London, December 2017
0.00000001
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0.1
1
10
100
0 5 10 15 20 25 30 35 40
qPCR Analysis Principle
Nq
Cq
Nn
𝑁𝑐 = 𝑁0 𝐸 𝐶
8. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
Pfaffl (2001)
9. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
Assumes ET = ER = 2
10. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
Assumes ET = ER = 2
Uses ET and ER
11. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
12. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
13. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
0
10
20
30
40
50
60
70
80
432143214321
321
Treatment/Control
target
treatment
Efficiency per target
14. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
0
10
20
30
40
50
60
70
80
432143214321
321
Treatment/Control
target
treatment
Efficiency per target
15. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
Assumes ET = ER = 2
16. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
0
10
20
30
40
50
60
70
80
432143214321
321
Treatment/Control
target
treatment
Efficiency per target
Efficiency = 2 (100%)
17. Medical BiologyRuijter et al, London, December 2017
Efficiency corrected relative quantification
0
10
20
30
40
50
60
70
80
432143214321
321
Treatment/Control
target
treatment
Efficiency per target
Efficiency = 2 (100%)
https://www.qbaseplus.com/knowledge/blog/why-pcr-amplification-efficiency-still-ignored
18. Medical BiologyRuijter et al, London, December 2017
qPCR: Data and Analysis
𝑁𝑐 = 𝑁0 𝐸 𝐶
𝑁0 = Τ𝑁𝑞 𝐸 𝐶 𝑞
19. Medical BiologyRuijter et al, London, December 2017
qPCR: Data and Analysis
𝑁𝑐 = 𝑁0 𝐸 𝐶
𝑁0 = Τ𝑁𝑞 𝐸 𝐶 𝑞
20. Medical BiologyRuijter et al, London, December 2017
Example data
Validation of miRNA biomarkers for heart failure
• 834 patients
• 12 miRNA targets
10008 measurements
• technical duplicates
20016 reactions
1 cycle
difference
between
replicates
21. Medical BiologyRuijter et al, London, December 2017
Discordant Cq values between replicates
Step 57
Examine replicates. All replicates should be
within 0.5 cycles of each other. At low Cq the
tolerance should be lower than at high Cq.
Above cycle 35 the variability will be greater
and quantification may be unreliable.
Nolan et al. (2006)
22. Medical BiologyRuijter et al, London, December 2017
Discordant Cq values between replicates
Validation of miRNA biomarkers for heart failure
• 834 patients
• 12 miRNA targets
10008 measurements
• technical duplicates
20016 reactions
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0
200
400
600
800
1000
1200
<25
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
≥40
Frequency
mean Cq per measurement
23. Medical BiologyRuijter et al, London, December 2017
Discordant Cq values between replicates
Validation of miRNA biomarkers for heart failure
• 834 patients
• 12 miRNA targets
10008 measurements
• technical duplicates
20016 reactions
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0
200
400
600
800
1000
1200
<25
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
≥40
Frequency
mean Cq per measurement
Diff Cq > 0.5
24. Medical BiologyRuijter et al, London, December 2017
Discordant Cq values between replicates
Validation of miRNA biomarkers for heart failure
• 834 patients
• 12 miRNA targets
10008 measurements
• technical duplicates
20016 reactions
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0
200
400
600
800
1000
1200
<25
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
≥40
Frequency
mean Cq per measurement
Diff Cq > 0.5
25. Medical BiologyRuijter et al, London, December 2017
Exclusion of discordant replicates
0
2000
4000
6000
8000
10000
<25 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 ≥40
numberofmeasurements
Cq value
Cq <0.5
total n 39% rejected
26. Medical BiologyRuijter et al, London, December 2017
Causes of variation between replicates
Technical replicates differ because of
• Threshold level
• Pipetting error
• Sampling error
27. Medical BiologyRuijter et al, London, December 2017
1
0.1
0.01
0.001
0.0001
Dilution
0.001
0.01
0.1
1
10
10 20 30
Cycle
Fluorescence
5 10 15 20 25 30 35
Cq (mean ± SD per dilution)
Nq
Less Cq variance with high threshold
Ruijter et al. Nucleic Acids Res, 2009
28. Medical BiologyRuijter et al, London, December 2017
Pipetting error in replicates
𝑁𝑞 = 𝑁0 𝐸 𝐶 𝑞
𝐶 𝑞 =
𝐿𝑜𝑔 𝑁𝑞 − 𝐿𝑜𝑔 𝑁0
𝐿𝑜𝑔(𝐸)
𝐶 𝑞,𝑟𝑎𝑛𝑔𝑒 =
𝐿𝑜𝑔(1 + 𝑃) − 𝐿𝑜𝑔(1 − 𝑃)
𝐿𝑜𝑔(𝐸)
Cq range = Cq,low input – Cq,high input
1 − 𝑃 𝑁0
1 + 𝑃 𝑁0
Pipetting
error: P
Cq range is independent of Nq and N0
29. Medical BiologyRuijter et al, London, December 2017
Pipetting error in replicates
30.5
31.0
31.5
32.0
0% 5% 10% 15% 20% 25%
Cq
pipetting error
1 − 𝑃 𝑁0
1 + 𝑃 𝑁0
De Ronde et al. RNA 23: 811-821, 2017
pipetting error of 15% causes a Cq difference of 0.5 cycles
30. Medical BiologyRuijter et al, London, December 2017
Pipetting error in replicates
30.5
31.0
31.5
32.0
0% 5% 10% 15% 20% 25%
Cq
pipetting error
De Ronde et al. RNA 23: 811-821, 2017
1 − 𝑃 𝑁0
1 + 𝑃 𝑁0
pipetting error of 15% causes a Cq difference of 0.5 cycles
NO reason to relax the 0.5 cycles criterion
31. Medical BiologyRuijter et al, London, December 2017
Sampling error
variation source: Poisson effect
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 2 4 6 8 10 12 14 16 18 20
probability
copy number in reaction (N0)
4
2
10 copies in input
32. Medical BiologyRuijter et al, London, December 2017
Sampling error
Sampling error in N0 gives a range in Cq
variation source: Poisson effect
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 2 4 6 8 10 12 14 16 18 20
probability
copy number in reaction (N0)
4
2
10 copies in input
33. Medical BiologyRuijter et al, London, December 2017
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
1.E+05
1.E+06
1.E+07
1.E+08
1.E+09
1.E+10
1.E+11
1.E+12
0 5 10 15 20 25 30 35 40
target
primer
Relation of N0 and Cq
Primer concentration:
1 pmol = 6.1011 molecules
rule of thumb: 10 copies in reaction gives Cq of about 35
exponential
phase ends
at ±35 cycles
10
De Ronde et al. RNA 23: 811-821, 2017
Competition
gives loss of
PCR efficiency
34. Medical BiologyRuijter et al, London, December 2017
For each Cq:
• Calculate N0
• Determine range of N0 because of Poisson effect:
Relation of N0 and Cq
𝑁𝑞 = 𝑁0 𝐸 𝐶 𝑞 𝑁𝑞 = 10𝐸35
𝑁0 = 10𝐸(35−𝐶 𝑞)
copies
1
2
0.025;2𝑁
2
≤ 𝑁0 ≤
1
2
0.975;2𝑁+2
2
rule of thumb: 10 copies in reaction gives Cq of about 35
For each Cq:
• Calculate N0
35. Medical BiologyRuijter et al, London, December 2017
Relation of N0 and Cq
1E-08
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0 5 10
1
2
0.025;2𝑁
2
≤ 𝑁0 ≤
1
2
0.975;2𝑁+2
2
10
103
102
Range of N because of Poisson effect:
36. Medical BiologyRuijter et al, London, December 2017
Relation of N0 and Cq
1E-08
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0 5 10
1
2
0.025;2𝑁
2
≤ 𝑁0 ≤
1
2
0.975;2𝑁+2
2
10
103
102
Range of N because of Poisson effect:
37. Medical BiologyRuijter et al, London, December 2017
Relation of N0 and Cq
1E-08
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0 5 10
10
103
102
1
2
0.025;2𝑁
2
≤ 𝑁0 ≤
1
2
0.975;2𝑁+2
2
Range of N because of Poisson effect:
38. Medical BiologyRuijter et al, London, December 2017
Relation of N0 and Cq
0.1
1
10
100
1000
10000
0 5 10 15 20 25 30 35 40 45
l
m
u
l
m
u
l
m
u
Nq-Cq
1E-08
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0 5 10
Poisson effect causes an
unavoidable range in Cq values
10
103
102
𝑁𝑞 = 10𝐸35
39. Medical BiologyRuijter et al, London, December 2017
Expected range in Cq values
0.1
1
10
100
1000
10000
0 5 10 15 20 25 30 35 40 45
l
m
u
l
m
u
l
m
u
Nq-Cq
Poisson effect causes an
unavoidable range in Cq values
Acceptable Cq range depends on
PCR efficiency and Cq
1E-08
0.0000001
0.000001
0.00001
0.0001
0.001
0.01
0 5 10
10
103
102
40. Medical BiologyRuijter et al, London, December 2017
Acceptable Cq range
De Ronde et al. RNA 23: 811-821, 2017
41. Medical BiologyRuijter et al, London, December 2017
Application of acceptable Cq range
0
2000
4000
6000
8000
10000
<25 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 ≥40
numberofmeasurements
Cq value
Cq <0.5
total n
61% included
42. Medical BiologyRuijter et al, London, December 2017
Application of acceptable Cq range
0
2000
4000
6000
8000
10000
<25 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 ≥40
numberofmeasurements
Cq value
Cq <0.5
Cq acceptable
total n
leads to rescue of 32% of the reactions
93% included
45. Medical BiologyRuijter et al, London, December 2017
Example data
Validation of miRNA biomarkers for heart failure
• 834 patients
• 12 miRNA targets
10008 measurements
• technical duplicates
20016 reactions
no amplification
no Cq
46. Medical BiologyRuijter et al, London, December 2017
Efficiency-corrected relative quantification
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
no amplification
no Cq
47. Medical BiologyRuijter et al, London, December 2017
Substitution of missing Cq values
Common approach:
Substitute missing Cq with the number of cycles
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
𝐹𝑜𝑙𝑑 = 2−DD 𝐶 𝑞
48. Medical BiologyRuijter et al, London, December 2017
Substitution of missing Cq values
Common approach:
Substitute missing Cq with the number of cycles
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
𝑁𝑞 = 𝑁0 𝐸 𝐶 𝑞 𝑁𝑞 = 10𝐸35
rule of thumb: 10 copies in reaction gives Cq of about 35
𝑁0 = 10𝐸(35−𝐶 𝑞)
copies
49. Medical BiologyRuijter et al, London, December 2017
Substitution of missing Cq values
Common approach:
Substitute missing Cq with the number of cycles
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
𝑁𝑞 = 𝑁0 𝐸 𝐶 𝑞 𝑁𝑞 = 10𝐸35
rule of thumb: 10 copies in reaction gives Cq of about 35
𝑁0 = 10 × 1.8(35−45)
50. Medical BiologyRuijter et al, London, December 2017
Substitution of missing Cq values
Common approach:
Substitute missing Cq with the number of cycles
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
𝑁𝑞 = 𝑁0 𝐸 𝐶 𝑞 𝑁𝑞 = 10𝐸35
𝑁0 = 10 × 1.8(35−45)
= 0.03 copies
rule of thumb: 10 copies in reaction gives Cq of about 35
51. Medical BiologyRuijter et al, London, December 2017
Substitution of missing Cq values
Proposed approach:
Substitute missing Cq with maximum observed Cq +1
per target
𝐹𝑜𝑙𝑑 = 𝐸 𝑅
𝐶 𝑞,𝑅,𝑇𝑟−𝐶 𝑞,𝑅,𝐶
/𝐸 𝑇
𝐶 𝑞,𝑇,𝑇𝑟−𝐶 𝑞,𝑇,𝐶
maximum Cq max Cq + 1
miR-320 35.38 36.38
miR-22-3p 39.32 40.32
𝑁0 = 10 × 1.8(35−40.32)
= 0.41 copies
De Ronde et al. RNA 23: 811-821, 2017
54. Medical BiologyRuijter et al, London, December 2017
Reference / Acknowledgement
De Ronde et al. RNA 23: 811-821, 2017
https://www.qbaseplus.com/knowledge/blog/why-pcr-amplification-efficiency-still-ignored
56. Medical BiologyRuijter et al, London, December 2017
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 2 4 6 8 10 12 14 16 18 20
number in reaction
probability copies in input
10
Limit of quantification (LOQ)
Input in reaction is determined by Poisson distribution
10 copies input results
in CV of ~30%
SD=3.3
Shipley. in PCR Technology: Current Innovations 2013
57. Medical BiologyRuijter et al, London, December 2017
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 2 4 6 8 10 12 14 16 18 20
number in reaction
probability copies in input
10
3
Limit of detection (LOD)
Input in reaction is determined by Poisson distribution
3 copies input will show
amplification in only
95% of the reactions
Shipley. in PCR Technology: Current Innovations 2013