Sequence analysis of 515 kinase genes in chronic lymphocytic leukemia

The pathogenesis of chronic lymphocytic leukemia (CLL) remains incompletely understood.¹ Although acquired chromosomal aberrations have been demonstrated to influence CLL biology and clinical behavior, it remains unclear what single gene defects other than p53 or ATM mutations cause or contribute to the CLL phenotype.² In particular, recurrent gene mutations that are increasingly found in other hematological malignancies have not yet been identified in CLL. One recent CLL gene re-sequencing study reported the analysis of selected exons of 70 tyrosine kinase genes in 95 CLL patients and reported no somatically acquired mutations.³ Given the frequent identification of stereotypical immunoglobulin receptor genes in CLL, it has been suggested that antigen engagement of the B-cell receptor on CLL cells serves a critical role in CLL cell survival and CLL disease etiology. Further, the reduced expression of del(13q)(14)-resident microRNAs has been implicated in early CLL pathogenesis in a subset of cases.⁴ It is unknown whether CLL is driven by high-frequency recurrent gene mutations in one or a few genes.

To address this question for phosphokinases, we sequenced the coding regions of 515 kinases (for a listing of kinase genes sequenced and kinase family classification see Supplementary Table S1) in DNA from CD19+ sorted cells from 23 CLL cases.⁵ This research was approved by the University of Michigan Institutional Review Board (IRBMED #2004-0962), and written informed consent was obtained from all patients before enrollment. CD19+ and CD3+ cells were purified from CLL samples using FACS as described.⁶ Clinical and molecular characteristics of the CLL cases studied are summarized in Supplementary Table S2. Primers used for sequence analysis of 8308 distinct coding exons from 515 kinase genes were derived from prior sequencing projects.⁷ A summary of kinase gene reference sequences, primer sequences and exon coverage can be found in Supplementary Table S3. Amplicons were sequenced unidirectionally. All mutations were confirmed in independently generated amplicons. A total of 9003 amplicons were considered to be of high enough quality to be scored for mutations. To be considered eligible for scoring at least 50% of the bases in 50% of the samples for a given amplicon had to have a Phred score of 20 and it further had to be judged to be of good quality by visual inspection. A total of 8798 (97.7%) amplicons of the 9003 reported in this study had 20 or more samples that were scored for mutations. Mutations were scored in all 24 samples for 6763 (75%) of the amplicons, and only 12 amplicons had as few as 12 samples that were scored. The average Phred score for all of the bases in all of samples in all of amplicons reported in this study was 54.3.

Six somatically acquired mutations were identified, each occurring once in the kinases WEE1, NEK1, BRAF, KDR, MAP4K3 and TRPM6 (Table 1). Because clinically approved therapeutics that target BRAF are available, we subsequently analyzed all BRAF coding exons in 120 CLL cases and exons 11 and 15 selectively in an additional 130 cases (the sites for the vast majority of BRAF mutations affect amino acid residue 600⁸). Primers to amplify and sequence all coding exons of BRAF and adjacent intronic sequences, including splice junctions, were designed using the primer 3 program (http://frodo.wi.mit.edu/primer3/) and sequence information was generated as described.⁶ Somatic mutations were confirmed using paired patient CD3+/buccal DNA as templates. In total, four BRAF mutations were found, none involving BRAF amino acid residue 600 (Table 1 and Supplementary Table 2).

Table 1.

Listing of kinase gene names, transcript accession ID and mutations for the six mutated kinase genes in CLL

Gene	Transcript accession ID	Coding exon	Tumor	Nucleotide (genomic)	Nucleotide (cDNA)	Amino acid (protein)
BRAF	CCDS5863.1	15	CLL-19	g.chr7: 139906318A>AG	c.1801A>AG	p.K601KE
BRAF	CCDS5863.1	15	CLL-32	g.chr7: 139906318A>AG	c.1801A>AG	p.K601KE
BRAF	CCDS5863.1	15	CLL-50	g.chr7: 139906333G>GC	c.1786G>GC	p.G596GR
BRAF	CCDS5863.1	11	CLL-192	g.chr7: 139934586G>GC	c.1406G>GC	p.G469GA
WEE1	CCDS7800.1	11	CLL-10	g.chr11: 9566645A>AG	c.1861A>AG	p.R621RG
NEK1	NM_012224	9	CLL-29	g.chr4: 170876731C>CA	c.860C>CA	p.P287PH
KDR	CCDS3497.1	30	CLL-52	g.chr4: 55787080C>CG	c.4027C>CG	p.L1343LV
MAP4K3	CCDS1803.1	30	CLL-58	g.chr2: 39397320T>TA	c.2368T>TA	p.C790CS
TRPM6	CCDS6647.1	22	CLL-80	g.chr9: 74627268G>GA	c.2975G>GA	p.G992GE

Abbreviations: cDNA, complementary DNA; CLL, chronic lymphocytic leukemia.

Amino acid substitutions in WEE1, NEK1, BRAF, KDR, MAP4K3 and TRPM6 were also analyzed using the CHASM algorithm^9,10 to estimate the probability that they impact protein activity in a manner relevant to oncogenicity. We trained an ensemble of decision trees^11,12 (Random Forest) with 3285 likely oncogenic somatic missense mutations from the COSMIC database¹³ and 3300 ‘passenger’ mutations synthetically generated by a computer algorithm to mimic the cancer mutation spectrum. To ensure an unbiased score, 13 unique BRAF amino acid residue substitution mutations were removed from the training set because they occurred at the same position as mutations of interest. For each mutation, the CHASM score is the fraction of trees that assign it to the passenger class; the P-value measures the statistical significance of the score and is corrected for multiple testing (false discovery rate). For CLL, we did not have sufficient data to estimate its spectrum, and we thus used the better-characterized spectrum of colorectal cancer. Using this algorithm, mutations in BRAF and a mutation in TRPM6 were found to be statistically significant as likely driver mutations (Supplementary Table S4). Only the BRAF mutations occurred within the catalytic kinase domain and in codons previously reported as sites of recurrent mutations in solid tumors and lymphomas (COSMIC database cite PMID:20952405); they may have biological roles in the affected CLL cells. Mutations in the remaining kinases did not receive statistically significant driver scores but mutations in all these genes except NEK1 have previously been identified in other tumors (breast, colorectum, pancreas and glioblastoma multi-forme), and WEE1,¹⁴ BRAF¹⁵ and KDR¹⁶ have established roles in cancer biology.

Amino acid substitutions were also analyzed using the program SIFT (Sorting Intolerant from Tolerant; SIFT analysis was performed using the instructions found at http://sift.jcvi.org/). Using this algorithm, mutations in BRAF and TRPM6 scored as ‘Affecting Protein Function’ and thus may have biological roles in the affected CLL cells; mutations in the remaining four kinases scored as ‘Tolerated’.

Our large validation screen had identified four CLL cases with heterozygous BRAF mutations (CLL19 and CLL32, both with K601K/E; CLL50 with G596G/R; and CLL192 with G469G/A). We mapped the three BRAF mutations to an X-ray crystal structure of inactive BRAF (engineered mutant V600E) in complex with the RAF inhibitor BAY43-9006 (PDB ID: 1UWJ). Interestingly, the mutations are close to each other and to the inhibitor in three-dimensional space. All three mutations occur in important functional elements of the protein that are critical to coordinating ligand (ATP, Mg²⁺) interactions and catalytic activity and could potentially impact inhibitor binding (Supplementary Figure 1). No CLL case displayed mutations of the amino acid residue 600, which is the predominant BRAF mutation found across human tumors. Next, N-Ras and K-Ras exons 2 and 3 were sequenced in 234 CLL cases and N-RAS codon 61 mutations were found in 2 cases (CLL155 with Q61Q/R and CLL172 with Q61Q/L).

Using p-ERK immunoblotting of lysates from unstimulated CLL cases (N = 13), we discovered substantially increased basal p-ERK levels in CLL27 (BRAF wild type) and CLL50 and 192 (BRAF mutants G596R¹⁷ and G469A¹⁸) but not in CLL19 or CLL32 (BRAF mutants K601E) compared with the majority of cases with wild-type BRAF,¹⁸ see Supplementary Figure 2. Of note, BRAF mutant K601E has previously been demonstrated to have increased catalytic activity ex vivo, and it therefore remains unsettled from this data what effects BRAF K601E mutants may have on CLL cells.

In summary, our data provide practical information about the mutational state of the CLL kinome with implications for CLL biology/pathogenesis. Given the substantial interest in the CLL research community to identify drivers and modifiers of CLL pathogenesis, these data provide important albeit largely negative information about the mutational state of the kinome in CLL, extending prior negative findings in 70 tyrosine kinase genes in CLL.³ We also identify a small subset of CLL that harbors an activated RAS–BRAF pathway that could be targeted therapeutically. This data should motivate future genome-wide pathogenetic CLL gene discovery efforts to determine whether other, potentially targetable genetic alterations can be found in this disease.