Peak height evaluation
The cycle set average peak heights increased as the cycle number increased with 29, 30, and 31 cycles resulting in mean peak heights of 75.0 ± 85.3, 147.1 ± 172.6, and 226.1 ± 298.2 RFU, respectively. A one-way ANOVA (α= 0.01) and post-hoc Tukey’s test (α= 0.05) where p < 0.001 demonstrated statistical significance between the mean peak heights across all cycle groups. The maximum average peak heights (across all loci in a given cycle set) also increase with cycle number – 506RFU, 1825 RFU and 2358 RFU for the 29, 30 and 31 cycle sets, respectively).
Table 1 Average peak heights by locus across 29, 30 and 31 PCR cycles
|
Mean peak height (RFU)
|
Locus
|
29 cycles
|
30 cycles
|
31 cycles
|
AMEL
|
126.1 ± 104.5
|
300.7 ± 385.2
|
340.5 ± 335.9
|
D3S1358
|
115.4 ± 90.3
|
246.2 ± 164.3
|
249.0 ± 271.6
|
D1S1656
|
85.3 ± 75.9
|
226.3 ± 164.3
|
358.2 ± 329.0
|
D2S441
|
61.4 ± 63.9
|
116.6 ± 135.2
|
189.4 ± 211.0
|
D10S1248
|
88.5 ± 73.8
|
145.0 ± 150.0
|
277.4 ± 309.1
|
D13S317
|
55.6 ± 56.2
|
94.9 ± 115.0
|
176.6 ± 288.4
|
Penta E
|
41.6 ± 53.1
|
101.8 ± 148.0
|
122.6 ± 199.5
|
D16S539
|
152.5 ± 130.3
|
279.9 ± 196.5
|
493.9 ± 378.6
|
D18S51
|
142.4 ± 135.7
|
180.8 ± 208.9
|
493.8 ± 570.2
|
D2S1338
|
60.7 ± 83.4
|
154.7 ± 176.5
|
235.7 ± 462.2
|
CSF1PO
|
70.6 ± 90.0
|
113.3 ± 156.1
|
137.5 ± 242.8
|
Penta D
|
55.6 ± 74.8
|
111.3 ± 175.7
|
117.9 ± 194.9
|
TH01
|
50.6 ± 63.2
|
105.8 ± 106.2
|
158.6 ± 148.0
|
vWA
|
53.4 ± 60.6
|
77.0 ± 61.5
|
164.9 ± 176.7
|
D21S11
|
69.3 ± 60.6
|
124.0 ± 116.7
|
161.5 ± 183.0
|
D7S820
|
43.8 ± 46.1
|
100.5 ± 88.5
|
148.2 ± 154.5
|
D5S818
|
58.2 ± 65.4
|
99.3 ± 111.2
|
115.1 ± 151.3
|
TPOX
|
29.8 ± 42.5
|
69.6 ± 116.4
|
127.8 ± 234.2
|
D8S1179
|
50.9 ± 56.8
|
95.7 ± 61.1
|
186.6 ± 150.9
|
D12S391
|
100.5 ± 95.3
|
178.5 ± 153.8
|
266.0 ± 257.8
|
D19S433
|
55.4 ± 49.9
|
150.2 ± 152.4
|
171.7 ± 192.4
|
SE33
|
35.0 ± 41.2
|
59.0 ± 88.0
|
72.2 ± 129.3
|
D22S1045
|
33.3 ± 41.4
|
56.7 ± 80.3
|
75.7 ± 123.1
|
DYS391
|
192.5 ± 160.8
|
511.3 ± 106.6
|
757.6 ± 450.2
|
FGA
|
116.8 ± 100.9
|
234.7 ± 175.1
|
353.5 ± 319.9
|
DYS576
|
103.3 ± 75.0
|
281.8 ± 235.4
|
498.2 ± 285.4
|
DYS570
|
150.6 ± 85.8
|
237.3 ± 109.6
|
401.8 ± 369.5
|
Locus-specific mean peak heights for single cells amplified using 29, 30 and 31 cycles and the PowerPlex Fusion 6c human DNA amplification kit.
A more detailed, locus-specific, analysis of average peak heights shows the same trend, where average peak height (PCR product) increases within each locus as cycle number increases (Table 1, Figure 1 and Additional file 1-Figure 1S). The overall mean peak height derived from the locus-specific average peak heights resulted in mean of 81.4 ± 42.8, 164.9 ± 100.3 and 253.8 ± 162.6 RFU for 29, 30 and 31 cycles respectively. The range (minimum to maximum) and interquartile range also tend to increase as the cycle numbers increase, with ranges of 30-193 RFU, 57-511 RFU and 72-758 RFU for 29, 30 and 31 cycles respectively. These values were calculated based on the presence of data above 10 RFU - these values represent the average peak heights when signal was detected and do not account for dropout.
The distribution of peak heights in the 29 cycle set is consistently tighter than the 30 and 31 cycle groups. Generally, the highest mean peak heights (locus-specific) across the varying cycles are found in the 31-cycle sample set, however this set also displays the widest range of peak heights, thus greater volatility (Figure 2 and Additional file 1 - Figure 1S). Despite this, it was commonplace for the mean peak heights in the 30-cycle set to exhibit higher values than the 31-cycle sample set, despite the maximum in sample peak height being higher at 31 cycles; e.g. CSF1PO, Penta D and SE33 (Figure 2).
Proportion of allele and locus dropout
Measures of allele and locus dropout across the varied cycle numbers (29-31) and analytical thresholds (10, 100, and 150 RFU) were used to assess the success of single cell analyses. Complete profile dropout was uncommon, occurring in 3 of 38 samples in the 29 cycle-100 RFU set, 8 of 38 samples when the analytical threshold was increased to 150 RFU, and in only one of 21 samples in the 31-cycle 150 RFU set. Profile dropout was not observed in the 30 and 31-cycle 100RFU and 30-cycle 150 RFU set.
The cycle set mean proportion of allele dropout and locus dropout were consistent with the expected results, where increasing the threshold led to increased proportions of dropout (Table 2). When comparing the cycle set proportion of dropout within threshold groups the mean proportions of allele and locus dropout remain fairly static with slight decreases as cycle number increases (Table 2). An analysis of variance (ANOVA) indicates that the average proportion of allele and locus dropout across cycle numbers at 10RFU (p=0.688 and p=0.458, respectively) and 100 RFU (p=0.129 and p=0.111, respectively) are not significantly different (α = 0.99). In contrast, we observed a significant difference in the cycle number specific dropout proportions using an analytical threshold of 150RFU (p= 0.0096 and p=0.0073, respectively). A post-hoc Tukey test (α=0.05) showed the level of allele dropout observed in the 29 cycle sample set is significantly higher than the 30 and 31 cycle sample sets (p=0.002), whereas the 30 and 31 cycle sets are not significantly different. Similarly, no significant difference observed between the level of locus dropout observed among the 29 and 30 and the 30 and 31 cycle sets (ANOVA α=0.01, p=0.007, Tukey’s test α=0.05, p=0.01).
Table 2 Summary of allele and locus dropout based on cycle number and analytical threshold.
Threshold (RFU)
|
Cycle #
|
n
|
|
|
Allele Dropout
|
|
Locus Dropout
|
Proportion samples w/ full dropout
|
|
Average
|
Range
|
p-value
|
|
Average
|
Range
|
p-value
|
10
|
29
|
38
|
0/38
|
|
0.405 ± 0.223
|
0.044-0.791
|
0.688
|
|
0.257 ± 0.194
|
0.00-0.667
|
0.458
|
30
|
22
|
0/22
|
|
0.369 ± 0.199
|
0.022-0.744
|
|
0.233 ± 0.191
|
0.00-0.667
|
31
|
21
|
0/21
|
|
0.429 ± 0.269
|
0.044-0.954
|
|
0.313 ± 0.277
|
0.00-0.917
|
100
|
29
|
38
|
1/38
|
|
0.675 ± 0.280
|
0.154-1.00
|
0.129
|
|
0.579 ± 0.320
|
0.00-1.00
|
0.111
|
30
|
22
|
0/22
|
|
0.550 ± 0.289
|
0.044-0.977
|
|
0.429 ± 0.309
|
0.044-0.958
|
31
|
21
|
0/21
|
|
0.538 ± 0.294
|
0.044-0.977
|
|
0.430 ± 0.311
|
0.044-0.958
|
150
|
29
|
38
|
7/38
|
|
0.788 ± 0.234
|
0.154-1.00
|
0.0096
|
|
0.712 ± 0.287
|
0.00-1.00
|
0.0073
|
30
|
22
|
0/22
|
|
0.619 ± 0.299
|
0.044-0.977
|
|
0.508 ± 0.330
|
0.00-0.958
|
31
|
21
|
1/21
|
|
0.581 ± 0.300
|
0.044-1.00
|
|
0.474 ± 0.322
|
0.00-1.00
|
Summary of the mean cycle set allele and locus dropout, the range (minimum-to maximum values observed separated by cycle number and the analytical threshold value used to assess the signal) and the p-values of the ANOVA comparing the variance at each threshold across cycle numbers at α= 0.99.
The level of dropout observed in locus-specific sets have a direct relationship with the analytical threshold, a trend that was also observed in the cycle-specific data (where all dropout proportions were pooled and averaged) (Figure 3 and Figure 4). Thus a clear pattern emerges in both the allele and locus dropout analyses. As expected, the proportions of dropout increase as the size of the locus increases (Figure 3, Additional file 1 - Table 3S, Figure 4 and Additional file 1 - Table 4S). Increases in the cycle number generally lead to decreased proportions of allele dropout, which is most apparent in the 150RFU analyses (Figure 3 and Additional file 1 - Table 3S). This trend is less apparent in the 10RFU threshold group, where 18 of the 27 loci (66.7%) show higher levels of allele dropout at 31 cycles compared to the 29 cycle. Whereas the sample sets analyzed at 100 RFU and 150 RFU show this relationship in five and two of the loci, respectively (Figure 3 and Table 3S). Amelogenin and DYS570 share this relationship across all thresholds used.
Locus-specific locus dropout behaves in a similar manner across cycle numbers where the 100 and 150 RFU groups tend to decrease as cycle number increases (Figure 4 and Additional file 1 - Table 4S). The relationship of locus dropout and cycle number in the 10 RFU group largely mirrors that of the proportion of allele dropout, where 19 of 27 loci have higher levels of locus dropout at 31 cycles compared to the 29-cycle group (Figure 4 and Additional file 1 - Table 4S). In many cases, both the proportions of locus-specific allele and locus dropout at 30 and 31 cycles are similar, indicating that a stochastic threshold may be reached at 30 cycles, where increasing the cycle number beyond 30 cycles does not increase the information content of the sample.
Measuring success of single cell analysis
Allelic dropout and heterozygote imbalance are inevitable consequences of analyzing low template DNA samples. Therefore, robust characterization (establishing expectations) of the performance of single cell analyses is critical to the success and potential implementation of this method. The RMP was used to evaluate the rarity of the DNA profiles obtained from single cell analysis using 29, 30 and 31 cycles and 100 and 150RFU analytical thresholds. The average RMP for the 100 RFU analysis were 1 in 1.2x1023 ± 7.04x1023, 1 in 1.94x1028 ± 8.9x1028, and 1 in 5.76x1026 ± 1.82x1027, for 29, 30 and 31 cycles, respectively with medians of 2.74x104, 6.36x108 and 8.90x1011. The data analyzed using a 150 RFU analytical threshold yielded RMPs of 1 in 2.4x1018 ± 1.46x1019, 1 in 1.49x1025 ± 5.8x1025, and 1 in 1.83x1024 ± 8.09x1024, for 29, 30, and 31 cycles, respectively (Table 3), with medians of 1.22x102, 8.04x106 and 4.60x109 . Average RMPs decreases when the number of cycles was increased from 29 to 30, whereas a decrease was observed when increasing PCR cycles from 30 to 31. And, as expected, average RMPs increased when the analytical threshold was increased from 100 to 150RFU.
Table 3 A summary of the random match probabilities across varied PCR cycles and analytical thresholds.
Cycle Number
|
AT = 100RFU
|
AT = 150RFU
|
Mean RMP
|
RMP range
|
Mean RMP
|
RMP range
|
29
|
1.20x1023 ± 7.04x1023
|
0 - 4.34x1024
|
2.40x1018 ± 1.46x1019
|
0 - 8.99x1019
|
30
|
1.94x1028 ± 8.90x1028
|
23.5 - 4.18x1029
|
1.49x1025 ± 5.8x1025
|
15.7 - 2.68x1026
|
31
|
5.76x1026 ± 1.82x1027
|
8.87 - 6.10x1027
|
1.83x1024 ± 8.09x1024
|
0 - 3.71x1025
|
The random match probabilities (calculated without theta corrections and the use of the “2p” rule) for profiles generated from single cells amplified with the PowerPlex Fusion 6c amplification kit using 29, 30, and 31 cycles and analyzed using analytical thresholds of 100 and 150RFU.
Inter-Locus comparisons
The relationship among the loci was assessed through (1) a correlation of total allelic product, and (2) the individual locus pairs in obtaining detectable signal across analytical thresholds and cycle number variants. These assessments are kit dependent and therefore represent the amplification dynamics of single cells amplified using the PowerPlex Fusion 6c amplification kit using 29, 30, and 31 PCR cycles.
The average correlation coefficients across loci of the total allelic product generally increased from 29 to 30 cycles and remained consistent between 30 and 31 cycles – 0.52 ± 0.09, 0.61 ± 0.11, 0.62 ± 0.1, for 29, 30 and 31 cycles, respectively (Figure 5). The weakest correlations were observed across the 29-cycle set, whereas the 30 and 31 cycle sets showed similar correlations with the 30-cycle set being most consistent across loci. The relationships between the loci remain largely positive, with only slightly negative values (-0.004) observed between the Amelogenin and D8S1179 loci (and only in the 29 and 31 cycle sample sets). In general, patterns emerge where the total allelic product of smaller loci are correlated with larger loci and large loci are correlated to other large loci. Several individual locus correlations are also particularly noteworthy. D7S820 and D5S818 exhibit high to very high correlation with most loci at 30 and 31 cycles. TPOX remains highly correlated with Penta E, Penta D, and D22S1045 and is consistently correlated at moderate to high with FGA across 29, 30, and 31 cycles. D22S1045 and SE33 are highly to very highly correlated with one another across cycle sets. Penta E is highly correlated to D13S317. vWA is consistently correlated with D8S1179 at low to high levels. D7S820 is very highly correlated with D19S433, D18S51, D16S539 and D1S1656. In contrast, TH01 and D2S441 consistently display a low to moderate correlation across all samples and cycle numbers. D8S1179, D7S820, D19S433, D16S539 and D1S1656 exhibit stepwise increases in correlation with most of the remaining other loci as cycle number is increased. Interestingly, correlation coefficients at D18S51 and vWA are consistently low in the 30-cycle sample set compared to the 29 and 31 cycle sets (Figure 5, Additional file 1 - Figure 5S and Additional file 1 - Table 5S).