Targeting MEK in vemurafenib-resistant hairy cell leukemia

Hairy cell leukemia (HCL) is a chronic, incurable B cell malignancy that presents with splenomegaly, bone marrow infiltration, and cytopenias [1]. Long remissions are typically achieved with purine analogs such as cladribine, but most cases will relapse and require further therapy. The discovery of the BRAF V600E mutation in almost all cases of HCL [2] has led to the widespread adoption of the BRAF inhibitor vemurafenib for treatment of patients relapsing after cladribine [3–5]. Impressive responses are reported; however, relapse is inevitable [5, 6] and hematologists are now faced with a growing number of patients with vemurafenib-resistant HCL. The optimal management of these patients remains unclear. The molecular basis of vemurafenib resistance has been extensively investigated in recent years in patients with BRAF mutant solid organ malignancies such as melanoma and colorectal cancer [7]. Resistance to vemurafenib in melanoma frequently results from reactivation of ERK pathway signaling by a variety of genetic mechanisms that include activating mutations of NRAS or KRAS, amplification of mutant BRAF, aberrant splicing of BRAF, and activating mutation of MAP2K1, which encodes the MEK1 protein [7, 8]. ERK-independent mechanisms are less frequent and include activation of PI3K signaling [7]. The predominance of genetic mechanisms converging on ERK reactivation has led to the successful use of dual MEK/ BRAF inhibition in melanoma [9]. In colorectal cancer however, different mechanisms apply; here primary resistance usually results from reduced feedback inhibition upon upstream receptor tyrosine kinase activity leading to restoration of ERK activity [10]. In this scenario, combined BRAF and MEK inhibition has not proved effective [11]. In contrast to our comprehensive understanding in solid organ cancer, very little is known about the mechanistic basis of vemurafenib resistance in HCL. Given the diversity of resistance mechanisms established in other cancers, it is unclear which, if any, of these might predominate in HCL. Two acquired subclonal, activating KRAS mutations were previously reported in a single patient with vemurafenib resistance [5]. Deletions of NF1 and NF2 have been proposed as an alternative mechanism in another case of primary resistance [12]. The use of MEK inhibition has been suggested as a logical therapeutic strategy in patients who have reactivated ERK signaling. However, the use of MEK inhibition has never previously been reported in a patient with HCL and at present there is no consensus on the optimal management of patients relapsing on vemurafenib. A 74-year-old patient with HCL had been treated at our institution with splenectomy, cladribine, and pentostatin. We previously reported his initial response to vemurafenib at a dose of 240 mg twice daily [4]. This dose was lower than used in the initial phase II trial [5], but has since been shown in several reports to be an effective dosing strategy for HCL [3, 13, 14]. Vemurafenib was initially stopped after 58 days; however, this was associated with rapid return of marrow infiltration and thrombocytopenia. Vemurafenib was restarted at the same dose and cytopenias rapidly These authors contributed equally: Rebecca Caeser, Grace Collord.


Library Preparation
For whole genome sequencing, short insert (500bp) genomic libraries were constructed, flowcells prepared and 150 base pair paired-end sequencing clusters generated on the Illumina HiSeq X platform according to Illumina no-PCR library protocols 1 . The average sequence coverage was 52.5X for tumor samples and 30.3X for matched normal sample.
For targeted sequencing we used a custom complementary RNA (cRNA) bait set (SureSelect, Agilent, UK, ELID # 0731661) to enrich for all coding exons of 292 genes implicated in haematological cancers (supplementary table 1). Short insert libraries (150bp) were prepared and sequenced on the Illumina HiSeq 2000 using 75 base paired-end sequencing as per Illumina protocol with 3 samples multiplexed per lane. Mean sequence coverage was 941.7X.

Sequencing data alignment
DNA sequencing reads were aligned to the GRCh 37d5 reference genome using the Burrows-Wheeler transform (BWA); BWA-MEM and BWA-aln were used for whole genome and targeted sequencing, respectively 2,3 . For targeted sequencing, PCR duplicates and reads mapping to regions outside target regions (merged exonic regions + 10bp either side of each exon) were excluded from analysis. Sequencing depth at each base was assessed using Bedtools coverage v2.24.0 4 . Point mutation variants were annotated using VAGrENT 5 according to ENSEMBL version 58.

Point mutations
Single base somatic substitutions were called using an in-house version of CaVEMan v1.11.2 (Cancer Variants through Expectation Maximization) 6 . CaVEMan compares sequencing reads from tumor and matched normal samples and uses a naïve Bayesian model and expectationmaximization approach to calculate the probability of a somatic variant at each base (https://github.com/cancerit/CaVEMan). Post-processing filters required that the following criteria were met to call a somatic substitution: 1. At least a third of the reads calling the variant had a base quality of 25 or higher. 2. If coverage of the mutant allele was less than 8, at least one mutant allele was detected in the first 2/3 of the read. 3. Less than 5% of the mutant alleles with base quality ≥ 15 were found in the matched normal. 4. Not all mutant alleles reported in the second half of the read. 5. Mean mapping quality of the mutant allele reads was ≥ 21. 6. Mutation does not fall in a simple repeat or centromeric region. 7. Position does not fall within a germline insertion or deletion. 8. Variant is not reported by ≥ 3 reads in more than one percent of samples in a panel of approximately 400 unmatched normal samples. 9. A minimum 2 reads in each direction reporting the mutant allele. 10. At least 10-fold coverage at the mutant allele locus. 11. Minimum variant allele fraction 5%. 12. No insertion or deletion called within a read length (150bp) of the putative substitution. 13. No soft-clipped reads reporting the mutant allele. 14. Median BWA alignment score of the reads reporting the mutant allele ≥ 140.
Small insertions and deletions were sought using an in-house version of Pindel v2.2.2 7 (https://github.com/cancerit/cgpPindel). Post-processing filters required that the following criteria were met for a variant to be called: All mutations meeting potential driver criteria were manually inspected in the Jbrowse genome viewer 8 prior to being discarded.

Rearrangements
Structural variants were called using a bespoke algorithm, BRASS v12.3.1 (https://github.com/cancerit/BRASS). Only breakpoints that could be validated by reconstruction at a base pair resolution are reported.

Copy number variants
The ascatNGS algorithm (v4.0.1) 9 as used to estimate tumor purity and ploidy and to construct copy number profiles.

Variant calling from deep targeted sequencing
Somatic single nucleotide variants (SNVs) were called using deepSNV, an algorithm developed for detecting subclonal mutations in deep sequencing experiments (https://github.com/gerstung-lab/deepSNV) 10 . Only reads with minimum nucleotide and mapping quality of 25 and 40, respectively, were considered. This algorithm models the error rate at individual loci using information from multiple unrelated samples. Allele counts at detected mutations were generated using an in-house script (https://github.com/cancerit/alleleCount) and manually inspected in the Jbrowse genome browser 8 .
Small insertions and deletions (indels) were sought using an in-house version of Pindel as described in Methods 12.1.2 complemented with t the aforementioned deepSNV algorithm in order to increase sensitivity for indels present at low VAF. VAF correction was performed using an in-house script (https://github.com/cancerit/vafCorrect ).

Curation of oncogenic variants
We considered variants as potential drivers or RAF inhibitor resistance events if they met any of the following criteria: 1. Variants in tumor suppressor genes were considered if they occurred within 3 amino acids of a recurrent hotspot mutation or were deleterious (nonsense, essential splice site or frameshift indel, disruptive rearrangement breakpoints or focal (<1 Mb) homozygous deletions). 2. Mutations in oncogenes were considered potential driver events if they were located at previously reported canonical hot spots (point mutations) or involved focal (<1Mb) amplification (>5 copies) of the intact gene. 3. Non-synonymous mutations in genes involved in the RAF pathway or previously implicated in RAF inhibitor resistance 11 .

Targeted Amplicon Sequencing
The genomic sequences spanning the KRAS, BRAF and MAP2K1 mutations identified by whole genome and deep targeted sequencing were amplified by PCR in duplicate (Supplementary Table 3), then routinely purified, barcoded and pooled for Illumina MiSeq sequencing 12 . The data analyses and variant calls were carried out as described previously 13 , then curated by manual inspection of sequencing reads. The variants that appeared in both replicates and were above the assay sensitivity threshold (>0.7% VAF, defined by mean + 3SD of background noise from non-neoplastic samples) were retained.

Phospho-ERK Immunohistochemistry
Immunostaining for phospho-ERK was performed on fixed, decalcified trephine biopsy material using pERK antibody from Cell Signaling Technology (Rabbit D13.14.4E, #4370) at 1:400 dilution with antigen retrieval using HIER for 20 minutes and detected using Bond™ Polymer Refine Detection.

Months from start of cobimetinib Method
Method used for sequencing -Whole Genome (WGS), Targeted pulldown or targeted amplicon DNA Source Source of cells used for DNA extraction, CD19 purification of peripheral blood (PB) or Bone marrow (BM) or total BM.

Marrow Burden
Product of marrow cellularity and tumor infiltrate