Patients who were diagnosed with gastrointestinal MALT lymphoma through endoscopy or surgery at Dong-A University Hospital were included in the analysis. Patient samples were excluded if they did not match the following criteria: Subjects should have enough FFPE samples for NGS and DNA preparation samples from the FFPE should qualify quality control (QC) criteria. A total of 5 patients (age, 49–72 years; 2 males and 3 females) underwent surgical biopsies. All patients were diagnosed histologically with MALT lymphoma. A total of 4 patients had small intestine lesions, and 1 patient had a stomach lesion. According to the Stage Lugano modification of the Ann Arbor staging system (36), 1 patient was in stage I, 3 patients were in stage III and 1 patient was in stage IV. None of the patients exhibited an increase in lactate dehydrogenase or B symptoms. All patients underwent bone marrow biopsies and no bone marrow involvement was noted. According to the International Prognostic Index (IPI) scoring system (37), 2 patients had an IPI score of 2 and 3 patients had a score of 0. Helicobacter pylori eradication therapy was performed for the patient with gastric MALT lymphoma.

All patients underwent surgery and 4 patients underwent chemotherapy with a regimen of cyclophosphamide, vincristine and prednisolone (CVP) for ≥4-6 cycles. Each cycle of CVP chemotherapy regimens was administered intravenously with cyclophosphamide 750 mg/m², and vincristine 1.4 mg/m² (maximum dose of 2 mg) on first day, and orally with prednisone 100 mg on day 1 to 5. It was repeated every 21 days and treated up to six times. Additionally, the patient with gastric MALT lymphoma underwent radiotherapy prior to chemotherapy with the addition of intravenous Rituximab 375 mg/m² on day 1 to the CVP regimen. Intestinal MALT lymphoma recurred in one patient at 4 years and 6 months following the completion of the initial treatment. All patients were alive at the time of writing (68.3 months following initial treatment; Table I).

The present study was approved by the Institutional Review Board of Dong-A University Hospital, Seo-gu, Busan, Republic of Korea (approval no. DAUHIRB 20-088).

All patients were diagnosed with gastrointestinal MALT lymphoma and underwent surgery to obtain tissue samples. Samples from the patients were obtained at Dong-A University Hospital (Busan, Korea) between January 2011 and December 2014. Immunohistochemistry for gastrointestinal MALT lymphoma diagnosis with LCA, UCHL-1, CD20, CD5, CD10, Bcl-2, Bcl-6, MUM-1, Cyclin D1 and CD23 was performed using formalin-fixed, paraffin-embedded (FFPE) tissues. The requirement for informed consent was waived since the tissues had been donated to the Dong-A University Hospital tissue bank.

FFPE tissue samples were cut into 5-µm thick sections, incubated at 60°C for 30 min, de-paraffinized in three containers of xylene baths at 60°C (3×5 mins), rehydrated with decreasing alcohols (100, 95 and 70%, two washes for 5 mins each) and washed with distilled water two times for 5 mins each. Subsequently, genomic DNA was extracted from the tissue samples using the Maxwell 16 CSC DNA FFPE kit (cat. no. AS1350, Promega Corporation) or the QIAamp DNA FFPE Tissue kit (cat. no. 56404, Qiagen, Inc.), which provides an automated purification of genomic DNA from FFPE tissue samples, thereby maximizing simplicity and convenience. DNA concentration and purity were verified using Nanodrop 8000 UV–Vis spectrometry (Thermo Fisher Scientific, Inc.) and Qubit 2.0 Fluorometry (Thermo Fisher Scientific, Inc.). Degrees of DNA degradation were measured using a 200 TapeStation Instrument (Agilent Technologies GmbH) and quantitative PCR (Agilent Technologies GmbH) (38).

DNA was sheared using Covaris S220 (Covaris, Inc.). Target capture was performed using a SureSelect XT reagent kit, HSQ (Agilent Technologies GmbH) according to the manufacturers protocol. A paired-end sequencing library was constructed using a barcode. Individual ‘barcode’ sequences are added to each DNA fragment during NGS library preparation so that each read can be identified and sorted before the final data analysis (39). After checking for library quality, sequencing was performed on a HiSeq 2500 (Illumina, Inc.) using 100 bp reads. SureSelect XT Human All Exon (version 5; Agilent Technologies, Inc.) was used for target capture for exome sequencing and a TruSeq Nano DNA Sample prep kit (Illumina, Inc.) was used for whole genome sequencing.

Samples were profiled on a HemaSCAN™, a targeted-sequencing platform designed by the LabGenomics in the Samsung Medical Center (81 Irwon-Ro Gangnam-gu. Seoul 06351, Korea). Samsung medical center participated in the panel design for the production of customized panels. The platform performs targeted NGS of cancer-related genes and allows for the detection of a variety of somatic mutations, including clinically actionable mutations. HemaSCAN™ analyzed 426 genes, including essential genes for different blood diseases, as follows: Plasma cell disease [NRAS proto-oncogene, GTPase (NRAS), KRAS and TP53], acute myeloid leukemia [CCAAT enhancer binding protein α, fms related receptor tyrosine kinase 3, Janus kinase 2 (JAK2), KIT proto-oncogene, receptor tyrosine kinase, nucleophosmin 1, disease resistance protein RUN1, TP53, isocitrate dehydrogenase (NADP(+)) 1 and isocitrate dehydrogenase (NADP(+)) 2], acute lymphoblastic leukemia (TP53, RB transcriptional corepressor 1, JAK2, NRAS and IKAROS family zinc finger 1) and malignant lymphoma (MYD88, BRAF and TP53). Additionally, the platform analyzes single nucleotide variations (SNVs), short insertions and deletions (InDels), copy number variations (CNVs) and gene rearrangements in 426 target genes in clinical FFPE specimens. Furthermore, HemaSCAN™ allows researchers and clinicians to include target genes curated from literature by request (Fig. S1).

Tumor purity is the proportion of cancer cells in a tumor sample and is a major confounding factor for analyzing cancer molecular and genomic data with NGS (40). The present study estimated tumor purity via the computational method that is derived from the distribution of variant allelic fractions (VAFs) of somatic single-nucleotide variants (SNVs) within copy-number neutral tumor segments. Tumor purity estimation using a panel sequencing is a more delicate process than using the whole genome due to the limited DNA region in the panel sequencing which could be insufficient for calculation of the genomic alterations (41). First, copy-neutral regions were identified by virtual karyotypes using SNP-based arrays. The minor allele frequencies at known single-nucleotide polymorphisms (SNPs) in the copy-neutral regions were near 0.5. Since only these SNPs were considered, their read densities corresponded to the most prominent peak in the distribution of read coverage at the SNPs, based on the hypothesis that pure polyploidy tumors with 4N in all chromosomes are extremely rare (42). The regions of copy number gain and loss were identified by their adjusted coverage relative to the copy number-neutral regions. Once the copy-neutral, gain and loss regions were delineated, the following formula was used to infer the proportion of each tumor clone: Alternative allele frequency = [PxY+(1-P)]/[PxX+2(1-P)] where, X and Y are the numbers of all and alternative alleles at each group of clustered SNPs in the tumor, respectively, and P is the proportion of tumor clones ranging between 0 and 1. Tumor purity was inferred from the maximum value among the Ps estimated at multiple positions, according to the hypothesis that the largest clone best represents tumor purity (43). Tumor purity <30% was less reliable and was not annotated.

Paired-end reads were aligned to the human reference genome (hg19 downloaded from http://genome.ucsc.edu) using BWA-MEM (ver.0.7.5) (http://bio-bwa.sourceforge.net/). SAMTOOLS (ver.0.1.18) (http://samtools.sourceforge.net/), GATK (ver.3.1–1) (https://gatk.broadinstitute.org/hc/en-us)and Picard (ver.1.93) (http://picard.sourceforge.net/) were used for file handling, local realignment and removal of duplicate reads, respectively. Base quality scores were recalibrated with GATK BaseRecalibrator, using known SNPs and InDels from SNP database 138 (dbSNP138).

To increase the sensitivity of SNV detection, two published methods, MuTect (version 1.1.4) (44) and LoFreq (ver.0.6.1) (45), were employed using default parameters. Unions of variants identified by the two callers (with high confidence set for MuTect) were used as the candidate set of variants. Small InDels were identified using Pindel (ver. 0.2.4) (46). To identify somatic CNVs, mean read depths at each exon were calculated and normalized to the coverage of target regions in each sample. Normalized read depth was further standardized by dividing the expected coverage for a normal individual (the expected coverage at each exon was taken to be the median of the read depth at that exon across a set of normal individuals). These steps addressed variability in capture efficiencies and GC content at different exons. To infer the correct copy numbers, amplitudes of copy numbers were adjusted based on estimated purity. When the adjusted amplitude of a copy change was >1 or <1 on a log scale, the region concerned was defined as an amplification or a deletion, respectively. Furthermore, most fusions involve intronic breakpoints. To identify fusion using a gene panel, ‘hotspot’ introns that are known to contain most breakpoints were analyzed for a set of clinically relevant fusions.

Variants were classified into three tiers. Tier 1 variants included variants listed as therapeutic targets by the Korean MFDS or the US FDA, and variants reported to be candidates in clinical trials. HemaSCAN™ panel covered all positions of tier 1 variants. Tier 2 variants included any mutation reported in COSMIC (ver. 64) (http://cancer.sanger.ac.uk) and gene fusions involving a target gene and a known Tier 1 partner (in COSMIC) or a novel partner. Tier 3 variants included all other mutations. ‘Actionable’ (that is, potentially responsive to a targeted therapy) variants were defined as Tier 1 variants, while ‘known’ (but not actionable) variants were defined as Tier 2 variants.

The HemaSCAN™ panel (Fig. S1) is a targeted NGS analysis panel that identifies clinically relevant genomic variants in tumor DNA and matches these variants with targeted therapies and/or clinical implications for diagnostic/prognostic purposes. NGS analysis revealed the following genomic variants in the five patients (Figs. 1 and S2): SNVs (Figs. 2 and 3), short InDels (Figs. 2 and 4) and CNVs (Fig. 5). These genomic variants were also reported as annotated, known and novel variants. Variant annotations were used to distinguish ‘real’ variants from sequencing artifacts and to distinguish potentially pathogenic variants from neutral variants (47). Estimated tumor purity in bioinformatics analysis was >90%. HemaSCAN™ analysis results revealed 25 variants in patient 1, 23 variants in patient 2, 51 variants in patient 3, 30 variants in patient 4 and 26 variants in patient 5 (Table II).

Different numbers of SNVs were identified in the five patients: Patient 1, 9; patient 2, 18; patient 3, 14; patient 4, 13; and patient 5, 7 SNVs. There were 6 SNVs [Kelch-like ECH-associated protein 1 (KEAP1), BLM RecQ like helicase, ETS variant transcription factor 6 (ETV6) and heat shock protein 90α family class A member 1] at splicing sites, 3 SNVs [CREB binding protein (CREBBP), TNF receptor superfamily member 14 (TNFRSF14) and RAD50 double strand break repair protein (RAD50)] that caused stop codons and 52 nonsynonymous SNVs that altered protein amino acid sequences. No SNV was identified in the annotated variant. Among the known variants, the same nonsynonymous SNV was present in the SET binding protein 6 (SETBP6) gene in patients 2 and 3 (Fig. 3). The SETBP gene is located on chromosome 18 at 42,643,270 (https://ghr.nlm.nih.gov/gene/SETBP1#conditions). In exon 6 of the SETBP gene, the 4,398th cDNA nucleotide (guanine) was replaced by thymine and the amino acid at position 1,466 of the protein was altered from glutamic acid to aspartic acid (Fig. 3). Among known SNVs, three different SNVs in the ETV6 gene were identified in patients 2 and 3 (Fig. S2). These SNVs were located in the same gene; however, they were at different positions in the 12th chromosome and different exons. Additionally, nucleotide substitutions and amino acid alterations also differed. Hence, they were deemed as different mutations.

Nonsynonymous SNVs were identified in two different Runt-related transcription factor 1 (RUNX1) genes (Fig. 3). One occurred in patients 1, 2 and 5 and the other in patients 2 and 5. Both were identified in exon 6, although nucleotide changes and the resulting amino acid changes differed. In the former, thymine at position 1,189 of cDNA was replaced by guanine, causing a change from serine to alanine at position 397. In the latter, adenine located at position 1,184 of the cDNA sequence was replaced by cytosine, causing a change from glutamic acid to alanine at position 395. SNVs in KEAP1 occurred at the splicing site and were identified in patients 1, 2 and 4. SNVs in CREBBP, TNFRSF1 and RAD50 caused stop codons (Fig. S2).

Among the three variant types, InDels exhibited the lowest frequency and were not identified in the annotated variant. Among the known variants, two single-base deletions were identified in the same marker of proliferation Ki-67 (MKI67) gene; however, these were at different chromosomal positions, resulting in a frameshift deletion (Fig. 4). This deletion occurred at amino acid 1,304 (threonine) of the protein in patients 1 and 3, and at position 2,273 (histidine) in patients 3, 4 and 5. Furthermore, additional frameshift insertions involving known variants were confirmed in the forkhead box O3, mucin 2, oligomeric mucus/gel-forming and BCL10 genes (Fig. S2). Among novel variants, five single-base insertions were identified in piccolo presynaptic cytomatrix protein (PCLO) at different chromosomal positions, resulting in non-frameshift insertions. Non-frameshift mutations occurred at five different positions of the protein: i) At position 2,926 (aspartic acid) in patients 3 and 5; ii) at 375 (glutamine) in patient 3; iii) at 373 (alanine) in patient 5; iv) at 369 (alanine) in patient 5; and v) at position 367 (glycin) in patient 4. Additionally, non-frameshift insertions in nuclear receptor corepressor 2 (NCOR2) were detected in patients 1 and 5.

According to the results, CNVs were the most common variant. The largest number of CNVs was found in patient 3. Two types of CNV were identified: Amplification and deletion. Amplifications were encountered more frequently than deletions. Among the 74 CNVs identified, one was an annotated variant and six were known variants, while the others were novel variants.

Among the novel variants, a CNV in the zinc finger protein 703 (ZNF703) gene was the most common and was identified in patients 1, 2, 3 and 5. A total of 3 patients had CNVs in the AKT1, cyclin dependent kinase inhibitor 2B (CDKN2B), NOTCH1, sphingosine-1-phosphate receptor 2 and WD repeat domain 90 genes, two patients had CNVs in SRY-box transcription factor 2 (SOX2), BTG anti-proliferation factor 2, capicua transcriptional repressor, fibroblast growth factor receptor 3, H1.4 linker histone, cluster member (HIST1H1E), HRas proto-oncogene, GTPase (HRAS), insulin receptor substrate 2, programmed cell death 1, tyrosine kinase 2 and SWI/SNF related, matrix associated, actin dependent regulator of chromatin, subfamily a, member 1 and one patient had a CNV in erb-b2 receptor tyrosine kinase 2 (ERBB2). CNVs were present in oncogenes (AKT1, ERBB2, SOX2 and ZNF703), a proto-oncogene (HRAS) and a tumor-suppressor gene (CDKN2B; Fig. 5).