Gene expression profiling of adult acute myeloid leukemia identifies novel biologic clusters for risk classification and outcome prediction
Blood Wilson et al. 108: 685

Supplemental materials for: Wilson et al

1. NPM1 mutation analysis - Materials & Methods:

RNA isolation/cDNA synthesis: Total RNA was isolated from patient samples (either peripheral blood or bone marrow) using Qiagen Mini-RNA kits. One microgram of total RNA was used to generate cDNA in a 20 µL reaction (1X PCR Buffer (50 mM KCl, 10 mM Tris-HCl, pH 8.3, 1.5 mM MgCl₂, 0.01% (w/v) gelatin), 250 µM each dNTP (Invitrogen), 10 mM DTT (Invitrogen), 2.5 µM random hexamers (Applied Biosystems), 1 U RNAsin (Promega), 5 U MMLV-RT (Invitrogen)) with incubations for 10 minutes @ 23°C, 30 minutes @ 42°C followed by 5 minutes @ 95°C. The cDNA was diluted to 50 µL total by the addition of 30 µL 1X TE (10 mM Tris-HCl, pH 8.0, 1 mM EDTA). 2.5 µL of the diluted cDNA (equivalent to 50 ng starting RNA) was used for PCR.

SYBR Green PCR/dissociation analysis: NPM1 forward primer (5´CCA AAG TGG AAG CCA AAT TCA TC 3´) and NPM1 reverse primer (5´ CAT TCT TGG CAA TAG AAC CTG GAC 3´) were designed to generate a 249-bp NPM1 product spanning portions of exons 11 and 12 of the NPM1 gene.* PCR products were generated in the ABI 7900ht as follows: 2.5 µL cDNA + 2.5 µL NPM1 forward primer (20 µM stock) + 2.5 µL NPM1 reverse primer (20 µM stock) + 12.5 µL 2X SYBR Green PCR Master Mix (Applied Biosystems) + 5 µL water was incubated for 2 minutes @ 95°C followed by 40 cycles of 95°C for 15 seconds and 65°C for 1 minute. Dissociation curves were then generated for the range of 65°C to 80°C. The patterns of the dissociation curves were used to select samples that differed from the wild type (see Figure S1).

Gel analysis/hybridization of variant types: 15 µL of each PCR product from the SYBR Green analysis were separated by electrophoresis through a 1.5% agarose (w/v) TBE gel. Fragments were vacuum transferred to a charged Nylon membrane in 0.4N NaOH. The fragments were UV-crosslinked to the membrane and then hybridized to variant A probe (Type A: 5´ GAT CTC TGT CTG GCA GTG GAG G 3´) or a pool of 13 probes for variants B-Q:* (Type B: 5´ GAT CTC TGC ATG GCA GTG GAG G 3´; Type C: 5´GAT CTC TGC GTG GCA GTG GAG 3´; Type D: 5´ GAT CTC TGC CTG GCA GTG GAG 3´; Type E: 5´ CTG GCA GTC TCT TGC CCG G 3´; Type F: 5´ CTG GCA GTC CCT GGA GAG G 3´; Type G: 5´ GAT CTC TGC AGG GCA GTG GAG 3´; Type K: 5´ GAT CTC TGC CGG GCA GTG G 3´; Type L: 5´ GAT CTC TGC CGC CGA GTG G 3´; Type M: GAT CTC TGG CAG AGG ATG GAG G 3´; Type N: GAT CTC TGC CAG GCA GTG GAG 3´; Type O: GAT CTC TGT TTG GCA GTG GAG GAA G 3´; Type P: 5´ GAT CTC TGC TTG GCA GTG GAG G 3´; Type Q: 5´ GAT CTC TGT CGG GCA GTG GAG 3´). All probes were labeled on the 5´ end with 6-FAM. Hybridization was performed for 2 hours at 60°C in 2.5X SSPE, 0.05% SDS at a final probe concentration of 5 nM. Blots were washed 3 × 10 minutes at 60°C in 1X SSPE, 0.05% SDS. Chemiluminescent signal detection (DAKO Corporation) was followed by a 5 minute exposure to film (Kodak Biomax MR). A single specimen with the characteristic dissociation curve for an NPM1 mutation did not hybridize with the above probes. This case was confirmed by sequencing.

* Schnittger S, Schoch C, Kern W, et al. Nucleophosmin gene mutations are predictors of favorable prognosis in acute myelogenous leukemia with a normal karyotype. Blood. 2005;106:3733-3739.

2. FLT3 mutation analysis
Cases were evaluated for FLT3 ITDs in exon 14 and 15, as previously described,¹ and screened for FLT3 TKD in exon 20 using two methods. The first method used Eco RV digestion of exon 20 PCR product.² Because this method misses mutations in the surrounding codons, all samples were also screened for exon 20 mutations utilizing PCR/SSCP. Novel primers that anneal within the introns surrounding exon 20 were developed: (D20F)5´ TTCGATCTTACAAGTGTGTTCAC3´ and (D20R) 5´TAGCAGCCTCACATTGCC3´. PCR was performed using standard reagents under the following conditions: initial denaturing at 94°C for 5 minutes; 45 cycles at 94°C for 30 seconds, 58°C for 20 seconds, 72°C for 20 seconds, and a final extension step of 72°C for 5 minutes. SSCP analyses were performed at 10°C and 25°C for 5.0 hours and 3.5 hours, respectively.¹ Suspected FLT3 ITDs and TKD mutations were confirmed by sequencing.¹

1. Stirewalt DL, Mashinchi S, Kussick SJ, et al. Novel FLT3 point mutations within exon 14 found in patients with acute myeloid leukaemia. Br J Haematol. 2004;124:481-484.

2. Yamamoto Y, Kiyoi H, Nakano Y, et al. Activating mutation of D835 within the activation loop of FLT3 in human hematologic malignancies. Blood. 2001; 97:2434-2439.

3. VxInsight cluster gene lists derived by Analysis of Variance (ANOVA) with statistical bootstrapping. (See Table S1)
Six genetically stable clusters of AML were identified using the unsupervised force-directed clustering algorithm and visualization program, VxInsight. ANOVA with statistical bootstrapping was used to determine which of the 9,463 genes examined showed the strongest differences between VxInsight groups as described at methodology Web site (http://hsc.unm.edu/crtc/willmanresearch/Pages/UNMHSC_HPC_SNL_Methodology.htm). The P-value represents the estimated fraction of time that a gene was ranked at or above its observed position after tabulation of rankings from 1,000 bootstrap resamplings.

4. VxInsight cluster defined gene lists using Receiver Operator Characteristic (ROC) Curves. (See Table S2)

5. VxInsight cluster defined gene lists using the Genetic Algorithm K-Nearest Neighbor method (GA/KNN). (See Table S3)

6. VxInsight cluster memberships compared to hierarchical cluster memberships for adult AML cohort. (See Table S4)

7. Limited comparison of our gene lists to gene lists published by Valk et al.* (See Table S5)
Similar genes are present in our gene lists (unsupervised learning, ANOVA) and Valk et al gene lists (supervised learning, SAM) despite the use of different Affymetrix GeneChips with different probe sets (U133A versus U95A). The comparison is limited to the 40 most significant genes per cluster (based on Valk data) and to evaluations of clusters with a 1) predominance of monocytic leukemias, and 2) predominance of normal karyotypes.

* Valk PJM, Verhaak RGW, Beijen AM, et al. Prognostically useful gene-expression profiles in acute myeloid leukemia. N Engl J Med. 2004;350:1617-1628.

8. Evaluation of Valk et al.* dataset of 253 adult AMLs for genes predictive of VxInsight cluster membership
The Valk et al. dataset of 253 AML cases, run for gene expression on Affymetrix U133A arrays, was downloaded along with the associated clinical data. The gene expression data and clinical data were loaded into Genespring 7.1 and GE values < 30 were set to 30, the data normalized and relative gene expression ratios were used to create heat maps. The top 50 ANOVA genes for each cluster from UNM AML U95Av2 data were selected to obtain the “best” matching probe set on the U133A array via the NetAffx portal at Affymetrix web site (http://www.affymetrix.com). U133A probe set IDs were then ranked according to ANOVA contrast, redundant genes removed and gene lists were created in Genespring. Heatmap images shown below have the genes examined that predicted VxInsight cluster membership in vertical columns. These genes are ordered based on the cluster for which they showed significant expression (i.e. A, B, C, D and F, cluster E was excluded) and their relative expression compared to the other clusters, either higher (up) or lower (down). The rows represent the gene expression patterns among individual patients in Valk et al dataset, ordered by their cluster number. Gene expression for each individual is compared to the entire patient set and indicated as a spectrum (see legend).

* Valk PJM, Verhaak RGW, Beijen AM, et al. Prognostically useful gene-expression profiles in acute myeloid leukemia. N Engl J Med. 2004;350:1617-1628.

Verhaak RG, Goudswaard CS, van Putten W, et al. Mutations in nucleophosmin (NPM1) in acute myeloid leukemia (AML): association with other gene abnormalities and previously established gene expression signatures and their favorable prognostic significance. Blood. 2005;106:3747-3754.

9. Evaluation of Bullinger et al.* dataset of 119 adult AMLs for genes predictive of VxInsight cluster membership
The Bullinger et al. dataset of 119 AML cases run for gene expression on cDNA arrays was downloaded along with the associated clinical data. The gene expression data and clinical data were loaded into Genespring 7.1; gene expression values were converted from log to linear values and ratios were used to create heat maps. The top 50 predictive ANOVA genes for each cluster (see Tables S1-S6) from UNM AML U95Av2 data were used to obtain the “best” matching genes used by Bullinger et al. on their cDNA array platform. Only 85 of the 300 genes predictive of VxInsight cluster membership were present on the cDNA array platform for evaluation. (See Figure S9)

* Bullinger L, Dohner K, Bair E, et al. Use of gene expression profiling to identify prognostic subclasses in adult acute myeloid leukemia. N Engl J Med. 2004;350:1605-1616.

10. Comparison of VxInsight gene lists to genes predictive of NPM1 mutations. (See Table S6)

Files in this Data Supplement:

• Table S1 (PDF, 233 KB) -
  Table S1.1. Genes with differential expression patterns between the VxInsight cluster A and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster A members relative to the other AML patients.
  Table S1.2. Genes with significant differential expression patterns between VxInsight cluster B and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster B members relative to the other AML patients.
  Table S1.3. Genes with significant differential expression patterns between the VxInsight cluster C and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster C members relative to the other AML patients.
  Table S1.4. Genes with significant differential expression patterns between the Vx cluster D and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster D members relative to the other AML patients.
  Table S1.5. Genes with significant differential expression patterns between the Vx cluster E and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster E members relative to the other AML patients.
  Table S1.6. Genes with differential expression patterns between the VxInsight cluster F and the rest of the cases. The genes are sorted based on the resulting F-score. The Contrast column defines the degree of gene over-expression (positive value, bolded) or under expression (negative value) in cluster F members relative to the other AML patients.
  
  
• Table S2 (PDF, 204 KB) -
  Table S2.1. Genes with significant differential expression patterns between VxInsight cluster A and the rest of the cases. Group 1 is cluster A (bolded); group 2 includes the remaining clusters.
  Table S2.2. Genes with significant differential expression patterns between the VxInsight cluster B and the rest of the cases. Group 1 is cluster B (bolded); group 2 includes the remaining clusters.
  Table S2.3. Genes with differential expression patterns between the VxInsight cluster C and the rest of the cases. Group 1 is cluster C (bolded); group 2 includes the remaining clusters.
  Table S2.4. Genes with differential expression patterns between the VxInsight cluster D and the rest of the cases. Group 1 is cluster D (bolded); group 2 includes the remaining clusters.
  
  
• Table S3 (PDF, 180 KB) -
  Table S3.1. Genes with differential expression patterns between the VxInsight cluster A and the rest of the cases. The genes are ranked based on the Z score. “High” corresponds to comparative gene overexpression in either cluster A (bolded) or the other clusters (not A).
  Table S3.2. Genes with differential expression patterns between the VxInsight cluster B and the rest of the cases. The genes are ranked based on the Z score. “High” corresponds to comparative gene overexpression in either cluster B (bolded) or the other clusters (not B).
  Table S3.3 Genes with differential expression patterns between the VxInsight cluster C and the rest of the cases. The genes are ranked based on the Z score. “High” corresponds to comparative gene overexpression in either cluster C (bolded) or the other clusters (not C).
  Table S3.4. Genes with significant differential expression patterns between the VxInsight cluster D and the rest of the cases. The genes are ranked based on the Z score. “High” corresponds to comparative gene overexpression in either cluster D (bolded) or the other clusters (not D).
  
  
• Table S4. Frequency of VxInsight cluster memberships versus hierarchical cluster memberships (PDF, 24.9 KB) -
  The value of the adjusted Rand index,* which measures concordance between two cluster membership assignments, is 0.36. This is highly significant at P<.0001 using the Monte Carlo estimation of statistical significance (N=10000 replications).
  * Reference: Yeung KY, Ruzzo WL. Principal component analysis for clustering gene expression data. Bioinformatics 17:763-774, 2001.
  
  
• Table S5 (PDF, 26.7 KB) -
  Table S5.1. Genes with significant expression in both Cluster F and Cluster 5 (Valk et al). These clusters contained a predominance of acute leukemias with monocytic differentiation (FAB M4 and FAB M5). The listed genes represent 23% (9/40) of the top cluster discriminating genes. All genes showed high expression relative to the other clusters of AML.
  Table S5.2. Genes with significant expression in both Cluster A and Cluster 6 (Valk et al). These clusters contained a predominance of normal karyotypes (75% and 88%, respectively). The listed genes showing a similar pattern of expression represent 15% (6/40) of the top discriminating genes. One gene (GPC4) showed disparate expression between clusters (i.e. high versus low).
  
  
• Table S6 (PDF, 40.5 KB) -
  Table S6.1. Comparison of VxInsight gene lists to genes predictive of NPM1 mutations. The top 0.1% (94) genes that predicted each of the VxInsight clusters were compared to genes found to best separated NPMc+ (mutated) from NPMc- (wild type) AML by Alcalay et al.* The following 30 genes overlapped. Only cluster A and cluster F showed the same pattern of expression (high and low) while clusters B, C, D were disparate in expression (high vs low). The numbers represent the VxInsight gene ranking for the given cluster (please refer to the cluster gene lists in section 3 of the Supplement for the gene names or below for cluster A).
  Table S6.2. Top 15 genes predictive of cluster A membership. The highlighted genes represent genes found to be predictive of NPMc+ (mutation) per Alcalay et. al.