1. Robustness of Amplicon Deep Sequencing Underlines Its Utility in Clinical Applications 
Vera Grossmann, Andreas Roller, Hans-Ulrich Klein, Sandra Weissmann, Wolfgang Kern, Claudia Haferlach, Martin Dugas, Torsten Haferlach, Susanne Schnittger, Alexander Kohlmann 
The Journal of Molecular Diagnostics 
Volume 15, Issue 4, Pages (July 2013)
DOI: /j.jmoldx
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

2. Figure 1
Technical precision of the sequencing work flow. Two samples with known RUNX1 mutations and varying mutation load at diagnosis were analyzed in multiple replicates using either the same MID or different MIDs. Top panel: Measured variance for the same MID using seven replicates on two different Roche 454 GS FLX systems, the combination of the two Roche 454 GS FLX systems (14 replicates), and the measured variance for 12 replicates using different MIDs on the same sequencing lane. Bottom panel: Results showing the mean and variance of the obtained mutation loads.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

3. Figure 2
Determination of sequencing error rates on a Roche 454 GS FLX instrument. A: Nucleotide-specific error rate for a given nucleotide. B: Estimated number of specific substitution events for each possible nucleotide exchange. C: Distribution of potential variants. A confusion matrix showing the specific number for each nucleotide is given in Supplemental Table S2.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

4. Figure 3
A: Serial dilution data of PCR amplicon deep sequencing using the Titanium chemistry standard format. For this exemplary p.Ser67Asn mutation in RUNX1, the number of reads (coverage) is given on the right y axis, representing data from five distinct sequencing lanes for each of the serial dilution data points. The mutation load of the p.Ser67Asn (c.200G>A) alteration in exon 3 is given on the left y axis. In its undiluted form, the mutation was detected with 48.4% of sequencing reads (886-fold coverage). In detail, the mutation load decreased from 48.4% to 23.0% (1:2 dilution; 588-fold coverage), 10.0% (1:4 dilution; 452-fold coverage), 4.9% (1:8 dilution; 514-fold coverage), and 2.5% (1:16 dilution; 490-fold coverage), respectively. By fitting a regression line to these data, we obtained a slope β of −1.085 (95% CI, ±0.069), which indicates a near-perfect decrease of the mutation load of the p.Ser67Asn alteration. B: Results of the Titanium assay dilution experiment for nine samples with RUNX1 mutations. The y axis shows the relative frequencies of the mutations in logarithmic scale, and the x axis indicates the dilution grade. In addition to the single measurements (gray points), the regression line of the linear model fitted across all samples (red line) is also shown. Accounting for all nine dilution experiments, the average slope was β = −1.000 (95% CI, −1.046 to −0.955) on the log2 scale.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

5. Figure 4
A: Comparison of standard PCR versus microfluidics-format PCR data. For this exemplary KRAS mutation, the number of reads (coverage) is given on the right y axis. The mutation load of the p.Gly13Asp (c.38G>A) alteration in exon 3 is given on the left y axis. In its initial analysis using a two-primer standard PCR, the mutation had been detected with a mutation load of 4.98% (823-fold coverage). Similar results were obtained when the same genomic DNA had been reanalyzed in duplicate using a four-primer design microfluidics format: 4.92% (1016-fold coverage) and 4.97% (966-fold coverage), respectively. B: Bland-Altman plot of data from the standard PCR versus microfluidics-format PCR comparison. The mean relative variant frequency was plotted against the difference in the relative variant frequency between standard PCR and microfluidics-format PCR. The dotted lines show the upper and lower limits of agreement. The dashed line shows the mean difference between standard PCR and microfluidics-format PCR. Of 36 data points, 35 lie within 2 SDs of the mean. Some representative cases have been exemplified by coloring.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

6. Figure 5
A: Serial dilution deep sequencing data of amplicons generated in a microfluidics-format PCR. For this exemplary EZH2 mutation, the number of reads (coverage) is given on the right y axis, representing data from two distinct sequencing lanes for each of the serial dilution data points. The mutation load of the p.Thr344Ala (c.1030A>G) alteration in exon 3 is given on the left y axis. In its undiluted stage, the mutation was detected with 51.52% of sequencing reads (658-fold coverage). We fitted a regression line to the data and obtained a slope β of −1.024 (95% CI, ±0.066), which indicates a near-perfect decrease of the mutation load in this dilution experiment. In detail, the mutation load decreased from 51.52% to 25.60% (1:2 dilution: 25.26% at 859-fold coverage and 25.93% at 850-fold coverage), 12.08% (1:4 dilution: 12.21% at 786-fold coverage and 11.94% at 720-fold coverage), 6.81% (1:8 dilution: 6.29% at 700-fold coverage and 7.32% at 765-fold coverage), and 2.89% (1:16 dilution: 2.77% at 793-fold coverage and 3.01% at 679-fold coverage), respectively. B: Results of the Fluidigm assay dilution experiment for all nine samples. The y axis shows the relative frequencies of the mutations in logarithmic scale, and the x axis gives the dilution grade. In addition to the single measurements (gray points), the regression line of the linear model fitted across all nine samples (red line) is also shown. The average slope was β = −0.979 (95% CI, −1.012 to −0.938) on the log2 scale per dilution step.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

7. Figure 6
Detection of a clinically relevant TP53 mutation. In this patient, a series of six available time points was analyzed in triplicate by deep sequencing. Applying amplicon deep sequencing using the 454 technology, a p.Met40GlyfsX2 mutation was detected with 41.7% mutation load after 602 days from initial diagnosis. This mutation further increased within 6 days to 51.1%. After treatment, the mutation load decreased to 36.3% and 26.9% after 11 and 13 months of treatment, respectively. The relative variant frequency is given on the y axis. The x axis shows the number of days from the initial measurement. The six serial measurements are evenly distributed on the x axis. The dashed line indicates the detection level for Sanger sequencing. More important, the mutation was observable in the first two time points (mutation load of 1.76% and 2.60%, respectively) by deep sequencing but not by Sanger sequencing. The same mutations are detected with higher coverage using the sequencing-by-synthesis chemistry on an MiSeq instrument (Illumina). Additional data are included through concomitant abnormalities, as detected by an alternative laboratory method (FISH; ie, a 6q21 deletion and a TP53 deletion).
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions

8. Figure 7
Serial analyses of changing RUNX1 subclones in AML. At diagnosis, we identified a point mutation (p.Val103Phe) in exon 4 with a mutation load of 57% (454-amplicon deep sequencing). After treatment, the mutation load decreased to lower than 1% at the first follow-up measurement. During the second, third, and fourth follow-up measurements, no mutation load was detected for mutation p.Val103Phe (left bar at each time point). However, a new mutation (p.Asn112_Tyr1113insPro) was observed, with a mutation load of 14% at the third follow-up measurement in exon 4 (right bar at each time point). At the last measurement, an increase of the mutation load to 42% was detected. The time line is given on the x axis. The number of reads (coverage) is given on the right y axis. The mutation load of the two distinct exon 4 alterations is given on the left y axis.
The Journal of Molecular Diagnostics  , DOI: ( /j.jmoldx )
Copyright © 2013 American Society for Investigative Pathology and the Association for Molecular Pathology Terms and Conditions