Molecular classification of tissue from a transformed non-Hogkin’s lymphoma case with unexpected long-time remission

Based on the unexpected long-term remission of a tFL patient on palliative therapy for her first relapse, we obtained informed consent to perform a thorough molecular analysis on her primary and relapsed tumor. The idea was to explore den molecular background in her tumor samples in search for an explanation for her cure following palliation with rituximab (Mabthera^®) and chlorambucil (Leukeran^®).

Additional malignant tissue were available and included, from NHL patients at time of diagnosis before treatment and diagnosis were evaluated according to the Revised European-American Lymphoma Classification [1] and confirmed by two expert hematopathologists at Aalborg University Hospital. Included were patients who had tissue stored in the “Diagnostic Biobank, Department of Hematology, Aalborg” and clinical data, staging, therapy and outcome registered in the National Clinical Quality Database for lymphoma, as approved by the local ethical committee (RetroGene, N-20140099).

RNA from 50 lymphoma samples FL (n = 7), tFL (n = 2; the case) and DLBCL (n = 41) was purified, labeled and hybridized to Affymetrix GeneChip Human Genome U133 Plus 2.0 Arrays and.CEL files generated as described [18] and presented in Additional file 2: Table S1. All were diagnostic samples taken before initiation of treatment. The relapse tFL sample (H385) being the exception, as the patients received R-CHOP treatment for her primary tumor (H302).

The effect of the drugs cyclophosphamide (C), doxorubicin (H), vincristine (O) and rituximab (R) on viable proliferating B-cell lines was measured by an in vitro drug screen strategy as recently described by our group [19‐24]. Sensitivity or resistance towards individual drugs can be predicted by the REGS assignment of probability, which ranges from zero to one, respectively. A website (http://​www.​hemaClass.​org) [25] has been generated to grant access to the classification and prediction tools to all researchers by easy upload of .CEL-file data. The.CEL files from the patient case tumors were analyzed and as build-in reference, our collection of lymphoma samples (RetroGene) performed at our laboratory was chosen in the RMA pre-processing normalization step. We chose the following classification systems: BAGS, R, C, H, O, P and R-CHOP, with the following ranges of non classified/intermediate groups: 0.1–0.9; 0.38–0.54; 0.33–0.55; 0.14–0.9; 0.46–0.62; 0.09–0.93, respectively. The results are downloaded and displayed in Table 3.

DNA from the patient’s two tumor samples was purified as described [26]. The patient’s saliva sample was collected using the Oragene DNA Self-Collection Kit (OG-500) and the DNA purified from non-involved mucosa cells using the Oragene DNA purifier (OG-L2P) following manufactures instructions in protocol PD-PR-006 Issue 3.2 (DNA Genotek Inc, Ottawa, Ontario, Canada).

The purified DNA was labeled and hybridized to Affymetrix SNP6 arrays as described [26]. The .CEL files, generated through AGCC after scanning the arrays, were imported to Partek Genomics Suite Software v. 6.6 (6.14.0828) using interrogating probes only with a pre-background adjustment for GC content and probe sequence. Probe sets wire summarized using allele specific summarization, and normalized against the human Hapmap genome. Amplification and deletions were detected through genomic copy number segmentation using standard settings, in brief: minimum 10 genomic markers, a P value threshold of 0.001, and a signal to noise ratio of 0.3. The diploid copy number range was set to 1.7–2.3.

From the patient’s saliva, primary and relapse tumor samples, standard sequence libraries for studies of point mutations were created from 100 ng of DNA from each sample, following exome capture using the Agilent SureSelect Human All Exon 50 Mb system (Agilent Technologies). Sequencing was performed on a HighSeq 2000 (Illumina, Hinxton, UK) using 76-bp paired-end reads, as described [27, 28].

Raw reads from the sequencer were processed following the Genome Analysis Toolkit (GATK) best practice guidelines [29, 30]. Reads were aligned to the grch37 assembly of the human genome with BWA v0.7.12 [31]. Aligned reads were sorted, converted to BAM format and had PCR duplicates marked using Picard v2.0.1 (http://​broadinstitute.​github.​io/​picard/​). Sequence realignment around INDELs and adjustment of quality scores was done using GATK v.3.5.0 [32]. Discovery of somatic variants in both primary and relapse tumors was performed using MuTect2 [33] incorporating information from dbSNP v138 [34] and COSMICv75 [35]. Somatic variants that passed quality filters were annotated using Oncotator v.1.8.0.0 [36] and finally PHIAL v1.0 [37] was used to score and rank somatic mutations according to clinical relevance and to identify potential targeted therapy for the analyzed patient.

The micro array data are deposited at Gene Expression Omnibus in project GSE86622, see Additional file 2: Table S1. The exome sequencing data from the tFL patient samples: non-involved mucosa cells, the lymphoma diagnostic sample and the sample following clinical relaps, are available through the European Genome Phenomena Archive at the European Bioinformatics Institute under accession number EGAD00001002707.

All statistical analyses were performed using R (The R Development Core Team, 2013) version 3.0.1 [38‐41].

The gene expression profiles of the primary and relapse samples from our patient case were analyzed together with 7 FL and 41 DLBCL cases from our clinic in a principal component analysis (PCA) with the results as illustrated in Fig. 1. The cases formed two distinct groups, where the diagnostic and relapse tFL samples were different, in close proximity to the FL and DLBCL groups, respectively.

We performed exome sequencing on the primary and the relapse tumor and identified somatic changes in 687 SNPs and INDELs compared to the constitutional DNA obtained from the patient’s saliva. The primary (H302) and relapse (H385) tumors contained 329 and 358 somatic tumor specific changes, respectively, with an overlap of 127, as shown in Table 1. A thorough review of clinically relevant mutated genes, two (EZH2 and DNMT3A) were unique for the primary tumor and two were unique for the relapse tumor (FBXW7 and FIP1L1). Three of the genes were found in both tumors (NOTCH2, TP53 and EP300), however, NOTCH2 were present at different allelic fractions in the two tumors (Table 2). Of the 37 recurrent gene mutations described by Pasqualucci [8] we recognized mutations in BCL2, HIST1H1E, EP300, TP53 and STAT6 in both tumors, EZH2 in the primary and KDM6B in the relapsed tissue. Overall, the mutation pattern recognized suggests a branched evolution from an unknown common mutated ancestor through the independent acquisition of distinct genetic lesions.

This was also visualized by global SNP6-microarray analysis for copy number variations (CNV) above 100 kb as illustrated in Fig. 2. The diagnostic and relapse samples had various chromosomal regions with identical CNV, illustrated for chromosomes 1, 6, 9, 16, 17 and X. However, many CNVs were unique to the primary tumor—see chromosomes 3, 4, 7, 10, 17 and X, or unique to the relapsed tissue seen in chromosomes 2, 3, 4, 5, 7, 8, 10, 11, 12, 16, 18, 19.

In summary, the diagnostic and relapsed tFL harbor unique mutations and CNVs documenting that the tumors have a common genetic background and the tFL case is a consequence of branched and not direct linear clonal selection and evolution.

The primary and relapse samples were assigned a cell of origin (COO) subtype as described [18] resistance estimate for R-CHOP by REGS [19‐21] classifiers following assignment in http://​www.​hemaClass.​org [25]. This tool provides an easy interface for one-by-one microarray based classification based on our preclinical models for BAGS and REGS. By individual REGS assignment, the primary tumor was predicted to be resistant to the drugs C, H, O and P, but sensitive towards R. In contrast, the relapse tumor had changed drug sensitivity for the alkylating agents C but still resistant towards O and P as well as the very high sensitive towards R, as shown in Table 3.

In summary, the gene signature based subtyping of the two tFL tissue support that they were different although with important overlaps. From a functional perspective both tissues are R sensitive, indicating that the patient was cured by the targeted anti-CD20 therapy.

Sequencing the malignant genome has opened new knowledge to the complexity of malignant lymphoma cell genetics, selection and Darwinian evolution [43, 44]. Multiple studies have analysed paired biopsies from indolent follicular lymphoma and its transformation for differential genetic aberrations associated to the clinical progression. A common finding is the premalignant IGH-BCL2 hybrid transcript expression and several key genetic events that seem to drive the process of progression [5, 6, 8]. In parallel, comparative analysis can also identify the genotypic background and, as in our case, document patterns of branched or linear lymphoma evolution. From published data, we have a long list of recurrent driver mutations from functional screening that illustrates the complexity and how heterogeneous the tFL genomes actually are.

In the present case, recurrent somatic mutations of BCL2, NOTCH2, TP53 and EP300 were identified in the primary and relapsed tumor, which indicates a common progenitor cell, most likely a transformation from an unknown in situ or premalignant follicular lymphoma [45, 46]. The branched evolution was supported by the presence of gene mutations unique to the primary tFL tumour, e.g. the recurrent genes, EZH2 and DNMT3A that were mutated in 30 and 8% of the reads, but not mutated in the reads from the relapse tFL tumour. On the contrary the relapsed tumor had several specific mutations e.g. in FBXW7 and FIP1L1.

In summary, the data (Figs. 1, 2 and Tables 1, 2) confirms that each of the individual tFL in our patient case has unique genomic profiling. The precision medicine concept argues that genetic heterogeneity is a key-factor for therapeutic failure. This limitation is broadly recognized and represents a considerable challenge, technically and bioinformatically. Understanding the genetic diversity and how it changes in response to interventions, will require deep sequencing and analysis of the genomes of highly selected single cells [47, 48].

Despite the enormous resources spent on developing molecular based cancer classification systems, most of these are still not available in clinical practice. To allow implementation and fast validation of our recent findings in DLBCL [18‐21], we have developed an easily accessible web application that permits other users to assign ABC/GCB, B-cell associated gene signature (BAGS) as well as drug specific resistance gene signature (REGS) on their own datasets. The website called hemaClass.org [25] is a new prospect for easy individual subtyping of malignant B cell diseases; in particular for DLBCL and myeloma.

In summary, the primary diagnosed tFL was classified as a diffuse large B cell lymphoma (DLBCL) of the GCB/centrocytic subtype by “cell of origin BAGS” assignment and R-CHOP resistant by “drug specific REGS” assignment. The relapsed DLBCL was classified as NC/memory subtype and R-CHOP sensitive in support of the branched evolution. Monitoring such functional variations following treatment may identify the mechanisms of molecular drug resistance resulting in clinical relapse.

This case study may be important for future diagnostic phenotyping and implementation of individual targeted drug or specific predictive therapy, however it involves a range of clinical, biological, and statistical limitations to be considered.

First, we cannot trust the conventional classification of poor prognosis associated with tFL, to make a clinical decision at the individual level. However, we expected this to be fulfilled by molecular profiling of drug specific sensitivity and resistance and mutations status for targeted therapy by designer drugs [49]. However, this strategy needs to be prospective validated by selected clinical end point, like level of remission, event free or overall survival. Such studies are ongoing, also in our center including all relapsed patients with haematological malignancies, to be used in the future evidence based strategy for individualized targeted and predicted therapy [49‐53].

Third, this case study has described a frame for future individual evaluation of clinical resistance detected at relapse. However, the individual REGS assignments given in Table 3 showed that the two tFL biopsies are R sensitive and indicate a clinical response to targeted anti-CD20 therapy for both tissues. However, we did not have a drug specific REGS predictor for the alkylating drug Chlorambucil, but the shift in sensitivity for the alkylating drug C may illustrate a potential sensitivity also to Chlorambucil therapy and a clinical impact in combination with R [54, 55]. Ongoing preclinical drug screens do include clinical available drugs and clinical relevant combinations.

Fourth, it has to be stressed that the clinical impact of REGS assignment is documented by retrospective analysis of thousands of patient sample data from several international clinical drug trials [20‐23]. Therefore, we need to await ongoing prospective implementation trials validating the clinical impact of REGS assignment in clinical resistant haematological malignancies, before introducing our “second generation” companion diagnostics for malignant B cell diseases. However, the present case illustrates our research strategy for implementation of individualized care focused on patients with relapse or progression. The specific challenges in this area are the unexplained molecular drug resistance and the undocumented use of drug combinations transferred from the “one size fits all” approach. Key to address this challenge is an understanding of the molecular/genetic profiles of each tumour such that we can tailor therapy appropriately, generate improved and evidence based clinical outcomes, and make the most efficient use of healthcare resources. Predictive companion diagnostics will identify multidrug resistant patients that will be extensively characterized and screened for specific pathway and/or mutations, attempting to validate target therapy by small and limited number of patients [56].