Search

Abstract Tumor immune microenvironmental alterations occur early in multiple myeloma (MM) development. In this study, we aim to systematically characterize the tumor immune microenvironment (TME) and the tumor-immune interactions from precursor stages, i.e., monoclonal gammopathy of undetermined significance (MGUS) and smoldering MM (SMM), to newly diagnosed MM, comparing these to healthy donors. Using CIBERSORT, mass cytometry (CyTOF), and single-cell RNA sequencing (scRNA-Seq), we examined innate and adaptive immune changes across these stages. We found a decrease in granulocytes in the TME predicts MM outcomes. HLA-DR is reduced in CD16+ monocytes and plasmacytoid dendritic cells, while myeloid dendritic cells show decreased expression of stress and immune-response genes. NK cells and CD8+ T cells shift from a GZMK+ to a GZMB+ cytotoxic phenotype in the TME, with increased inhibitory markers TIM3 and TIGIT. In paired samples, the proportion and gene expression pattern in patient-specific GZMB+CD8+ T cells remain largely unchanged despite MM progression. Our findings provide a comprehensive immune landscape of MM and its precursors, offering insights into therapeutic strategies. Enhancing neutrophil and NK cell cytotoxicity, tumor antigen presentation, and CD8+ T cell versatility in precursor stages may prevent MM progression.

Multiple myeloma (MM) is a clonal plasma cell proliferative malignancy characterized by a debilitating bone disease. Osteolytic destruction, a hallmark of MM, is driven by increased osteoclast number and exacerbated bone resorption, primarily fueled by the excessive production of RANKL, the master regulator of osteoclast formation, within the tumor niche. We previously reported that osteocytes, the most abundant cells in the bone niche, promote tumor progression and support MM bone disease by overproducing RANKL. However, the molecular mechanisms underlying RANKL dysregulation in osteocytes in the context of MM bone disease are not entirely understood. Here, we present evidence that MM-derived CCL3 induces upregulation of RANKL expression in both human and murine osteocytes. Through a combination of in vitro, ex vivo, and in vivo models and clinical data, we demonstrate that genetic or pharmacologic inhibition of CCL3 prevents RANKL upregulation in osteocytes and attenuates the bone loss induced by MM cells. Mechanistic studies revealed that MM-derived CCL3 triggers the secretion of HMGB1 by osteocytes, a process required for osteocytic RANKL upregulation by MM cells. These findings identify a previously unknown CCL3-HMGB1 signaling axis in the MM tumor niche that drives bone resorption by promoting RANKL overproduction in osteocytes.

Abstract Anti-multiple myeloma B cell maturation antigen (BCMA)-specific chimeric antigen receptor (CAR) T-cell therapies represent a promising treatment strategy with high response rates in myeloma. However, durable cures following anti-BCMA CAR-T cell treatment of myeloma are rare. One potential reason is that a small subset of minimal residual myeloma cells seeds relapse. Residual myeloma cells following BCMA-CAR-T-mediated treatment show less-differentiated features and express stem-like genes, including CD24. CD24-positive myeloma cells represent a large fraction of residual myeloma cells after BCMA-CAR-T therapy. In this work, we develop CD24-CAR-T cells and test their ability to eliminate myeloma cells. We find that CD24-CAR-T cells block the CD24-Siglec-10 pathway, thereby enhancing macrophage phagocytic clearance of myeloma cells. Additionally, CD24-CAR-T cells polarize macrophages to a M1-like phenotype. A dual-targeted BCMA-CD24-CAR-T exhibits improved efficacy compared to monospecific BCMA-CAR-T-cell therapy. This work presents an immunotherapeutic approach that targets myeloma cells and promotes tumor cell clearance by macrophages.

Multiple myeloma (MM) remains incurable due to disease relapse and drug resistance. Notch signals from the tumor microenvironment (TME) confer chemoresistance, but the cellular and molecular mechanisms are not entirely understood. Using clinical and transcriptomic datasets, we found that NOTCH3 is upregulated in CD138+ cells from newly diagnosed MM (NDMM) patients compared to healthy individuals and increased in progression/relapsed MM (PRMM) patients. Further, NDMM patients with high NOTCH3 expression exhibited worse responses to Bortezomib (BOR)-based therapies. Cells of the TME, including osteocytes, upregulated NOTCH3 in MM cells and protected them from apoptosis induced by BOR. NOTCH3 activation (NOTCH3OE) in MM cells decreased BOR anti-MM efficacy and its ability to improve survival in in vivo myeloma models. Molecular analyses revealed that NDMM and PRMM patients with high NOTCH3 exhibit CXCL12 upregulation. TME cells upregulated CXCL12 and activated the CXCR4 pathway in MM cells in a NOTCH3-dependent manner. Moreover, genetic or pharmacologic inhibition of CXCL12 in NOTCH3OE MM cells restored sensitivity to BOR regimes in vitro and in human bones bearing NOTCH3OE MM tumors cultured ex vivo. Our clinical and preclinical data unravel a novel NOTCH3-CXCL12 pro-survival signaling axis in the TME and suggest that osteocytes transmit chemoresistance signals to MM cells.

Multiple myeloma (MM) is a malignancy of plasma cells whose antibody secretion creates proteotoxic stress relieved by the N-end rule pathway, a proteolytic system that degrades N-arginylated proteins in the proteasome. When the proteasome is inhibited, protein cargo is alternatively targeted for autophagic degradation by binding to the ZZ-domain of p62/ sequestosome-1. Here, we demonstrate that XRK3F2, a selective ligand for the ZZ-domain, dramatically improved two major responses to the proteasome inhibitor bortezomib (Btz) by increasing: i) killing of human MM cells by stimulating both Btz-mediated apoptosis and necroptosis, a process regulated by p62; and ii) preservation of bone mass by stimulating osteoblast differentiation and inhibiting osteoclastic bone destruction. Co-administration of Btz and XRK3F2 inhibited both branches of the bimodal N-end rule pathway exhibited synergistic anti-MM effects on MM cell lines and CD138+ cells from MM patients, and prevented stromal-mediated MM cell survival. In mice with established human MM, co-administration of Btz and XRK3F2 decreased tumor burden and prevented the progression of MM-induced osteolytic disease by inducing new bone formation more effectively than either single agent alone. The results suggest that p62-ZZ ligands enhance the anti- MM efficacy of proteasome inhibitors and can reduce MM morbidity and mortality by improving bone health.

Abstract Background Oncogenic overexpression of integrin-β7 (ITGB7) in cases of high-risk multiple myeloma (MM) was reported to promote enhanced interactions between neoplastic plasma-B cells and stromal cells to develop cell-adhesion mediated drug resistance. Methods Expression profiles of adhesion related genes were analyzed in a cohort of MM patients containing major IgH translocations or hyperdiploidies (HY), diagnosed at the premalignant monoclonal gammopathy of undetermined significance (MGUS; n = 103), smoldering multiple myeloma; (SMM; n = 190) or MM (MM; n = 53) stage. Differential expression was integrated with loci-specific alterations in DNA-methylation and chromatin marks in MM patients. A CRISPR-based targeted induction of DNA-methylation at the ITGB7 super-enhancer (SE) in MM.1S cells was employed to intersect the impact of cis-regulatory elements on ITGB7 expression. Results ITGB7 was significantly (p

Spatially and temporally resolved data can improve our understanding of evolution and treatment resistance in multiple myeloma (MM). Here, the authors analyse spatial and longitudinal heterogeneity in MM patients using multi-region sequencing, and identify subclones associated with relapse and therapy resistance.

Abstract Deciphering genomic architecture is key to identifying novel disease drivers and understanding the mechanisms underlying myeloma initiation and progression. In this work, using the CoMMpass dataset, we show that structural variants (SV) occur in a nonrandom fashion throughout the genome with an increased frequency in the t(4;14), RB1, or TP53 mutated cases and reduced frequency in t(11;14) cases. By mapping sites of chromosomal rearrangements to topologically associated domains and identifying significantly upregulated genes by RNAseq we identify both predicted and novel putative driver genes. These data highlight the heterogeneity of transcriptional dysregulation occurring as a consequence of both the canonical and novel structural variants. Further, it shows that the complex rearrangements chromoplexy, chromothripsis and templated insertions are common in MM with each variant having its own distinct frequency and impact on clinical outcome. Chromothripsis is associated with a significant independent negative impact on clinical outcome in newly diagnosed cases consistent with its use alongside other clinical and genetic risk factors to identify prognosis.

Osteolytic bone disease is a hallmark of multiple myeloma (MM). A significant fraction (~20%) of MM patients do not develop osteolytic lesions (OLs). The molecular basis for the absence of bone disease in MM is not understood. We combined PET-CT and gene expression profiling (GEP) of purified BM CD138+ MM cells from 512 newly diagnosed MM patients to reveal that elevated expression of cystatin M/E (CST6) was significantly associated with the absence of OL in MM. An enzyme-linked immunosorbent assay revealed a strong correlation between CST6 levels in BM serum/plasma and CST6 mRNA expression. Both recombinant CST6 protein and BM serum from patients with high CST6 significantly inhibited the activity of the osteoclast-specific protease cathepsin K and blocked osteoclast differentiation and function. Recombinant CST6 inhibited bone destruction in ex vivo and in vivo myeloma models. Single-cell RNA-Seq showed that CST6 attenuates polarization of monocytes to osteoclast precursors. Furthermore, CST6 protein blocks osteoclast differentiation by suppressing cathepsin-mediated cleavage of NF-κB/p100 and TRAF3 following RANKL stimulation. Secretion by MM cells of CST6, an inhibitor of osteoclast differentiation and function, suppresses osteolytic bone disease in MM and probably other diseases associated with osteoclast-mediated bone loss.

Abstract Background Somatic variant callers are used to find mutations in sequencing data from cancer samples. They are very sensitive and have high recall, but also may produce low precision data with a large proportion of false positives. Further ad hoc filtering is commonly performed after variant calling and before further analysis. Improving the filtering of somatic variants in a reproducible way represents an unmet need. We have developed Filters for Next Generation Sequencing (FiNGS), software written specifically to address these filtering issues. Results Developed and tested using publicly available sequencing data sets, we demonstrate that FiNGS reliably improves upon the precision of default variant caller outputs and performs better than other tools designed for the same task. Conclusions FiNGS provides researchers with a tool to reproducibly filter somatic variants that is simple to both deploy and use, with filters and thresholds that are fully configurable by the user. It ingests and emits standard variant call format (VCF) files and will slot into existing sequencing pipelines. It allows users to develop and implement their own filtering strategies and simple sharing of these with others.

Progression from asymptomatic smoldering multiple myeloma (SMM) to symptomatic Multiple Myeloma occurs at different rates in different patients. Here, the authors report fewer NRAS and FAM46C mutations and adverse translocations in SMM compared to MM, while KRAS mutations are associated with a shorter time to progression.

In multiple myeloma (MM), communication via Notch signaling in the tumor niche stimulates tumor progression and bone destruction. We previously showed that osteocytes activate Notch, increase Notch3 expression, and stimulate proliferation in MM cells. We show here that Notch3 inhibition in MM cells reduced MM proliferation, decreased Rankl expression, and abrogated the ability of MM cells to promote osteoclastogenesis. Further, Notch3 inhibition in MM cells partially prevented the Notch activation and increased proliferation induced by osteocytes, demonstrating that Notch3 mediates MM-osteocyte communication. Consistently, pro-proliferative and pro-osteoclastogenic pathways were upregulated in CD138+ cells from newly diagnosed MM patients with high vs. low NOTCH3 expression. These results show that NOTCH3 signaling in MM cells stimulates proliferation and increases their osteoclastogenic potential. In contrast, Notch2 inhibition did not alter MM cell proliferation or communication with osteocytes. Lastly, mice injected with Notch3 knock-down MM cells had a 50% decrease in tumor burden and a 50% reduction in osteolytic lesions than mice bearing control MM cells. Together, these findings identify Notch3 as a mediator of cell communication among MM cells and between MM cells and osteocytes in the MM tumor niche and warrant future studies to exploit Notch3 as a therapeutic target to treat MM.

Rarely, immunophenotypically immature B-cell precursor acute lymphoblastic leukemia (BCP-ALL) carries an immunoglobulin- MYC rearrangement (IG-MYC-r). This can result in diagnostic confusion with Burkitt lymphoma/leukemia and use of individualized treatment schedules of unproven efficacy. Here we compare the molecular characteristics of these conditions and investigate historic clinical outcome data. We identified 90 cases registered in a national BCP-ALL clinical trial/registry. When present, diagnostic material underwent cytogenetic, exome, methylome and transcriptome analyses. The outcomes analyzed were 3-year event-free survival and overall survival. IG-MYC-r was identified in diverse cytogenetic backgrounds, co-existing with either established BCP-ALL-specific abnormalities (high hyperdiploidy, n=3; KMT2A-rearrangement, n=6; iAMP21, n=1; BCR-ABL1, n=1); BCL2/BCL6-rearrangements (n=15); or, most commonly, as the only defining feature (n=64). Within this final group, precursor-like V(D)J breakpoints predominated (8/9) and KRAS mutations were common (5/11). DNA methylation identified a cluster of V(D)J-rearranged cases, clearly distinct from Burkitt leukemia/lymphoma. Children with IG-MYC-r within that subgroup had a 3-year event-free survival of 47% and overall survival of 60%, representing a high-risk BCP-ALL. To develop effective management strategies this group of patients must be allowed access to contemporary, minimal residual disease-adapted, prospective clinical trial protocols.

Abstract Background Multiple Myeloma (MM) is a hematological malignancy with genomic heterogeneity and poor survival outcome. Apart from the central role of genetic lesions, epigenetic anomalies have been identified as drivers in the development of the disease. Methods Alterations in the DNA methylome were mapped in 52 newly diagnosed MM (NDMM) patients of six molecular subgroups and matched with loci-specific chromatin marks to define their impact on gene expression. Differential DNA methylation analysis was performed using DMAP with a ≥10% increase (hypermethylation) or decrease (hypomethylation) in NDMM subgroups, compared to control samples, considered significant for all the subsequent analyses with p

In multiple myeloma, disease progresses via seeding to different anatomic sites and clonal expansion. Here, utilising autopsy material, the authors show that systemic seeding accelerates at relapse following treatment.

Abstract Background The study of cancer genomics continually matures as the number of patient samples sequenced increases. As more data is generated, oncogenic drivers for specific cancer types are discovered along with their associated risks. This in turn leads to potential treatment strategies that pave the way to precision medicine. However, significant financial and analytical barriers make it infeasible to sequence the entire genome of every patient. In contrast, targeted sequencing panels give reliable information on relevant portions of the genome at a fiscally responsible cost. Therefore, we have created the Targeted Panel (TarPan) Viewer, a software tool, to investigate this type of data. Results TarPan Viewer helps investigators understand data from targeted sequencing data by displaying the information through a web browser interface. Through this interface, investigators can easily observe copy number changes, mutations, and structural events in cancer samples. The viewer runs in R Shiny with a robust SQLite backend and its input is generated from bioinformatic algorithms reliably described in the literature. Here we show the results from using TarPan Viewer on publicly available follicular lymphoma, breast cancer, and multiple myeloma data. In addition, we have tested and utilized the viewer internally, and this data has been used in high-impact peer-reviewed publications. Conclusions We have designed a flexible, simple to setup viewer that is easily adaptable to any type of cancer targeted sequencing, and has already proven its use in a research laboratory environment. Further, we believe with deeper sequencing and/or more targeted application it could be of use in the clinic in conjunction with an appropriate targeted sequencing panel as a cost-effective diagnostic test, especially in cancers such as acute leukemia or diffuse large B-cell lymphoma that require rapid interventions.

BackgroundThe tumor microenvironment (TME) is increasingly appreciated as an important determinant of cancer outcome, including in multiple myeloma (MM). However, most myeloma microenvironment studies have been based on bone marrow (BM) aspirates, which often do not fully reflect the cellular content of BM tissue itself. To address this limitation in myeloma research, we systematically characterized the whole bone marrow (WBM) microenvironment during premalignant, baseline, on treatment, and post-treatment phases.Methods and findingsBetween 2004 and 2019, 998 BM samples were taken from 436 patients with newly diagnosed MM (NDMM) at the University of Arkansas for Medical Sciences in Little Rock, Arkansas, United States of America. These patients were 61% male and 39% female, 89% White, 8% Black, and 3% other/refused, with a mean age of 58 years. Using WBM and matched cluster of differentiation (CD)138-selected tumor gene expression to control for tumor burden, we identified a subgroup of patients with an adverse TME associated with 17 fewer months of progression-free survival (PFS) (95% confidence interval [CI] 5-29, 49-69 versus 70-82 months, χ2 p = 0.001) and 15 fewer months of overall survival (OS; 95% CI -1 to 31, 92-120 versus 113-129 months, χ2 p = 0.036). Using immunohistochemistry-validated computational tools that identify distinct cell types from bulk gene expression, we showed that the adverse outcome was correlated with elevated CD8+ T cell and reduced granulocytic cell proportions. This microenvironment develops during the progression of premalignant to malignant disease and becomes less prevalent after therapy, in which it is associated with improved outcomes. In patients with quantified International Staging System (ISS) stage and 70-gene Prognostic Risk Score (GEP-70) scores, taking the microenvironment into consideration would have identified an additional 40 out of 290 patients (14%, premutation p = 0.001) with significantly worse outcomes (PFS, 95% CI 6-36, 49-73 versus 74-90 months) who were not identified by existing clinical (ISS stage III) and tumor (GEP-70) criteria as high risk. The main limitations of this study are that it relies on computationally identified cell types and that patients were treated with thalidomide rather than current therapies.ConclusionsIn this study, we observe that granulocyte signatures in the MM TME contribute to a more accurate prognosis. This implies that future researchers and clinicians treating patients should quantify TME components, in particular monocytes and granulocytes, which are often ignored in microenvironment studies.

MYC is a widely acting transcription factor and its deregulation is a crucial event in many human cancers. MYC is important biologically and clinically in multiple myeloma, but the mechanisms underlying its dysregulation are poorly understood. We show that MYC rearrangements are present in 36.0% of newly diagnosed myeloma patients, as detected in the largest set of next generation sequencing data to date (n=1,267). Rearrangements were complex and associated with increased expression of MYC and PVT1, but not other genes at 8q24. The highest effect on gene expression was detected in cases where the MYC locus is juxtaposed next to super-enhancers associated with genes such as IGH, IGK, IGL, TXNDC5/BMP6, FAM46C and FOXO3. We identified three hotspots of recombination at 8q24, one of which is enriched for IGH-MYC translocations. Breakpoint analysis indicates primary myeloma rearrangements involving the IGH locus occur through non-homologous end joining, whereas secondary MYC rearrangements occur through microhomology-mediated end joining. This mechanism is different to lymphomas, where non-homologous end joining generates MYC rearrangements. Rearrangements resulted in overexpression of key genes and chromatin immunoprecipitation-sequencing identified that HK2, a member of the glucose metabolism pathway, is directly over-expressed through binding of MYC at its promoter.

Disruption of the normal splicing patterns of RNA is a major factor in the pathogenesis of a number of diseases. Increasingly research has shown the strong influence that splicing patterns can have on cancer progression. Multiple Myeloma is a molecularly heterogeneous disease classified by the presence of key translocations, gene expression profiles and mutations but the splicing patterns in MM remains largely unexplored. We take a multifaceted approach to define the extent and impact of alternative splicing in MM. We look at the spliceosome component, SF3B1, with hotspot mutations (K700E and K666T/Q) shown to result in an increase in alternative splicing in other cancers. We discovered a number of differentially spliced genes in comparison of the SF3B1 mutant and wild type samples that included, MZB1, DYNLL1, TMEM14C and splicing related genes DHX9, CLASRP, and SNRPE. We identified a broader role for abnormal splicing showing clear differences in the extent of novel splice variants in the different translocation groups. We show that a high number of novel splice loci is associated with adverse survival and an ultra-high risk group. The enumeration of patterns of alternative splicing has the potential to refine MM classification and to aid in the risk stratification of patients.

The emergence of treatment resistant sub-clones is a key feature of relapse in multiple myeloma. Therapeutic attempts to extend remission and prevent relapse include maximizing response and the use of maintenance therapy. We used whole exome sequencing to study the genetics of paired samples taken at presentation and at relapse from 56 newly diagnosed patients, following induction therapy, randomized to receive either lenalidomide maintenance or observation as part of the Myeloma XI trial. Patients included were considered high risk, relapsing within 30 months of maintenance randomization. Patients achieving a complete response had predominantly branching evolutionary patterns leading to relapse, characterized by a greater mutational burden, an altered mutational profile, bi-allelic inactivation of tumor suppressor genes, and acquired structural aberrations. Conversely, in patients achieving a partial response, the evolutionary features were predominantly stable with a similar mutational and structural profile seen at both time points. There were no significant differences between patients relapsing after lenalidomide maintenance versus observation. This study shows that the depth of response is a key determinant of the evolutionary patterns seen at relapse. This trial is registered at clinicaltrials.gov identifier: 01554852.

A Correction to this paper has been published: https://doi.org/10.1038/s41467-021-20978-y.

OBJECTIVES/SPECIFIC AIMS: 1) Determine the mutational landscape, including translocation, mutations and mutational signatures as well as copy number variations of pPCL and identify significant differences to non pPCL MM. 2) Determine whether genetic changes pertinent to pPCL could be explored as therapeutic targets to improve the dismal prognosis of this patient population. METHODS/STUDY POPULATION: Samples from overall 19 pPCL patients that presented to the Myeloma Center, UAMS between 2000-2018 were used for this study. We performed gene expression profiling (GEP; Affymetrix U133 Plus 2.0) of matched circulating peripheral PCs and bone marrow (BM) PCs from 13 patients. Whole exome sequencing (WES) was performed on purified CD138+ PCs from BM aspirates from 19 pPCL patients with a median depth of 61x. CD34+ sorted cells, taken at the time of stem cell harvest from the same 19 patients, were used as controls. Translocations and mutations were called using Manta and Strelka and annotated as previously reported. Copy number was determined by Sequenza. RESULTS/ANTICIPATED RESULTS: 1) GEP from the BM and circulating peripheral PCs showed that the expression patterns of the two samples from each individual clustered together, indicating that circulating PCs and BM PCs in pPCL result from the same clone and are biologically clearly related. 2) The clinical characteristics from the patient cohort used for WES analysis were as follows: median age was 58 years (range 36–77), females accounted for 74% (14/19), an elevated creatinine level was found in 78% (14/18) and an elevated LDH level in 71% (10/14). All patients presented with an ISS stage of III. Median OS of the whole dataset was poor at 22 months, which is consistent with OS from previously reported pPCL cohorts. 3) Primary Immunoglobulin translocations were common and identified in 63% (12/19) of patients, including MAF translocations, which are known to carry high risk in 42% (8/19) of patients [t(14;16), 32% and t(14;20), 10%] followed by t(11;14) (16%) and t(4;14) (10%). Furthermore, 32% (6/19) of patients had at least one MYC translocation, which are known to play a crucial role in disease progression. 4) The mutational burden of pPCL consisted of a median of 98 non-silent mutations per sample, suggesting that the mutational landscape of pPCL is highly complex and harbors more coding mutations than non-pPCL MM. 5) Driver mutations, that previously have been described in non-pPCL MM showed a different prevalence and distribution in pPCL, including KRAS and TP53 with 47% (9/19) and 37% (7/19) affected patients respectively compared to 21% and 5% in non-PCL MM. PIK3CA (5%), PRDM1 (10%), EP300 (10%) and NF1 (10%) were also enriched in the pPCL group compared to previously reported cases in non-pPCL MM. 6) Biallelic inactivation of TP53 – a feature of Double Hit myeloma - was found in 6/19 (32%) samples, indicating a predominance of high risk genomic features compared to non-pPCL MM. Furthermore, analysis of mutational signatures in pPCL showed that aberrant APOBEC activity was highly prevalent only in patients with a MAF translocation, but not in other translocation groups. DISCUSSION/SIGNIFICANCE OF IMPACT: In conclusion we present one of the first WES datasets on pPCL with the largest patient cohort reported to date and show that pPCL is a highly complex disease. The aggressive disease behavior can, at least in part, be explained by a high prevalence of MAF and MYC translocations, TP53 and KRAS mutations as well as bi-allelic inactivation of TP53. It is of interest that only KRAS but not NRAS mutations are highly enriched in pPCL. From all highly prevalent genomic alterations in pPCL, only KRAS mutations offer a potential for already available therapeutically targeting with MEK inhibitors, which should be further explored.

It is challenging to cluster cancer patients of a certain histopathological type into molecular subtypes of clinical importance and identify gene signatures directly relevant to the subtypes. Current clustering approaches have inherent limitations, which prevent them from gauging the subtle heterogeneity of the molecular subtypes. In this paper we present a new framework: SPARCoC (Sparse-CoClust), which is based on a novel Common-background and Sparse-foreground Decomposition (CSD) model and the Maximum Block Improvement (MBI) co-clustering technique. SPARCoC has clear advantages compared with widely-used alternative approaches: hierarchical clustering (Hclust) and nonnegative matrix factorization (NMF). We apply SPARCoC to the study of lung adenocarcinoma (ADCA), an extremely heterogeneous histological type, and a significant challenge for molecular subtyping. For testing and verification, we use high quality gene expression profiling data of lung ADCA patients, and identify prognostic gene signatures which could cluster patients into subgroups that are significantly different in their overall survival (with p-values < 0.05). Our results are only based on gene expression profiling data analysis, without incorporating any other feature selection or clinical information; we are able to replicate our findings with completely independent datasets. SPARCoC is broadly applicable to large-scale genomic data to empower pattern discovery and cancer gene identification.

Identifying clinically relevant subtypes of a cancer using gene expression data is a challenging and important problem in medicine, and is a necessary premise to provide specific and efficient treatments for patients of different subtypes. Matrix factorization provides a solution by finding checkerboard patterns in the matrices of gene expression data. In the context of gene expression profiles of cancer patients, these checkerboard patterns correspond to genes that are up- or down-regulated in patients with particular cancer subtypes. Recently, a new matrix factorization framework for biclustering called Maximum Block Improvement (MBI) is proposed; however, it still suffers several problems when applied to cancer gene expression data analysis. In this study, we developed many effective strategies to improve MBI and designed a new program called enhanced MBI (eMBI), which is more effective and efficient to identify cancer subtypes. Our tests on several gene expression profiling datasets of cancer patients consistently indicate that eMBI achieves significant improvements in comparison with MBI, in terms of cancer subtype prediction accuracy, robustness, and running time. In addition, the performance of eMBI is much better than another widely used matrix factorization method called nonnegative matrix factorization (NMF) and the method of hierarchical clustering, which is often the first choice of clinical analysts in practice.