Huang, Gaofeng

Background/Objectives: Lung cancer is a devastating disease with the highest mortality rate among cancer types. Over 60% of non-small cell lung cancer (NSCLC) patients, accounting for 87% of lung cancer diagnoses, require radiation therapy. Rapid treatment initiation significantly increases the patient’s survival rate and reduces the mortality rate. Accurate tumor segmentation is a critical step in diagnosing and treating NSCLC. Manual segmentation is time- and labor-consuming and causes delays in treatment initiation. Although many lung nodule detection methods, including deep learning-based models, have been proposed. Most of these methods still have a long-standing problem of high false positives (FPs). Methods: Here, we developed an electronic health record (EHR)-guided lung tumor auto-segmentation called EXACT-Net (EHR-enhanced eXACtitude in Tumor segmentation), where the extracted information from EHRs using a pre-trained large language model (LLM) was used to remove the FPs and keep the TP nodules only. Results: The auto-segmentation model was trained on NSCLC patients’ computed tomography (CT), and the pre-trained LLM was used with the zero-shot learning approach. Our approach resulted in a 250% boost in successful nodule detection using the data from ten NSCLC patients treated in our institution. Conclusions: We demonstrated that combining vision-language information in EXACT-Net multi-modal AI framework greatly enhances the performance of vision only models, paving the road to multimodal AI framework for medical image processing.

Abstract Marginal zone B-cell lymphomas (MZL) represent a sub-group of indolent B-cell lymphomas, normally localized in the “marginal zone” of the secondary lymphoid follicles. Depending on the site of involvement, three distinct sub-types have been identified: extra nodal MZL of MALT type, splenic MZL (SMZL), and nodal MZL (NMZL). MALT lymphoma is the most frequent of the MZL subtypes, representing 70% of MZL and 7% to 8% of non-Hodgkin lymphomas. SMZL and NMZL comprise the remaining 20% and 10% of MZL, respectively, and account for less than 1% of NHL. The relative rarity of these lymphomas and difficulty in distinguishing them from other low-grade B-cell lymphoma subtypes like lymphoplasmacytic lymphomas (LPL), making therapeutic decisions difficult. The aim of this study was to perform a comprehensive high-resolution study to identify distinctive molecular markers in MZL and LPL, which would help to better define these lymphomas at genomic level and distinguish them from other types of low-grade B-cell lymphoma. We analyzed 119 patients, including 22 pulmonary MALT (pMALT), 20 non-pulmonary MALT (npMALT), 34 SMZL, 21 NMZL and 22 LPL (8 Waldenström’s Macroglobulinemia and 14 non-WM) cases. Human Agilent 244A platform was used by array-based comparative genomic hybridization (aCGH) analysis. Overall, 93% of patients have an altered genome. LPL had the highest amount of copy number abnormalities (median of 7.5 per patient; average of 11.7; range: 0–44), followed by npMALT (6.5; 7.6; 0–33), NMZL (5; 8.3; 0–31), SMZL (4; 7.1; 0–56) and pMALT (3.5; 4.3; 1–11). When the overall size of the abnormalities was considered, LPL was also the most affected disease with an average size of 245.15Mb included in copy number changes per patient, followed by NMZL (188.6Mb), npMALT (187.2Mb), SMZL (126.4Mb) and pMALT (109Mb). Deletions were more commonly observed than amplifications but the affected area was significantly smaller (average of 55.2Mb and 118.6Mb included in deletions and amplifications per genome, respectively). Partial or total gains of chromosome 3 and 18 were recurrently identified in all the entities, being the most common abnormalities in npMALT (60% and 50%, respectively) and pMALT (27% each). Trisomy 12 was also ubiquitously observed, ranging from 24% of NMZL to 9% of pMALT. Deletions in 6q were commonly identified in LPL (55%), npMALT (35%) and, in a less extent, in SMZL (12%) and NMZL (5%), but not in pMALT. Biallelic and focal monoallelic deletions at 6q23 clearly delineate a minimal deleted region (MDR) that includes TNFAIP3 as being the target gene of the abnormality. Deletions on 7q were found in SMZL and npMALT, thus delineating two MDR. Interstitial 11q deletions were found in LPL, NMZL and SMZL, but not in MALT. Chromosome 13 abnormalities were monosomies in NMZL and interstitial 13q14 deletions in LPL and SMZL, comprising MIRN15A and MIRN16-1 in the MDR. Partial X chromosome gains were common to all the entities but involving different areas: Xq27.3–q28 gains were recurrent events in LPL and NMZL, but not in SMZL and MALT, which were characterized by p arm gains or trisomies. Finally, some recurrent abnormalities were exclusive of specific entities: 1p deletions and 1q gains were only common in NMZL; a focal 2p gain including BCL11A and REL was found in npMALT and partial 8q gains were only identified in LPL. Trisomy 4 was only identified in the subgroup of LPL with WM, but not in the remaining LPL cases or in the remaining diseases. In this study we made a compendium of chromosomal abnormalities present in a group of related diseases, in which the molecular basis remains poorly defined. Using this high-resolution approach we were able to establish recurrent and disease-specific patterns of abnormalities and minimal deleted/amplified regions were redefined. Furthermore, TNFAIP3, a negative regulator of the NF-kB pathway, has been identified as being a commonly affected gene in MALT, LPL and NMZL, highlighting its potential role in pathogenesis.

Abstract Although classified as marginal zone lymphomas under the WHO classification, the molecular relationship between splenic marginal zone lymphoma (SMZL), extranodal marginal zone B-cell lymphoma of mucosa-associated lymphoid tissue (MALT), and nodal marginal zone lymphoma (NMZL) has not been clarified. Furthermore, lymphoplasmacytic lymphoma (LPL) can show clinical morphologic and immunophenotypic overlap with these entities and may present a diagnostic challenge. In this analysis of gene expression data generated using the Affymetrix U133plus 2.0 chip from 32 SMZL, 25 MALTs (GI, salivary gland and lung), 23 NMZL and 25 LPL, we aim to identify the molecular relationship between these pathological entities using unsupervised methods, identify disease specific signatures and markers using supervised methods and also the functional implications of these signatures using a modified gene-set enrichment analysis. Using hierarchical clustering, MALT lymphoma forms a tight cluster regardless of tumor site. In addition, SMZL and NMZL form another large cluster. LPLs are divided into 2 main clusters with occasional samples interspersed amongst the SMZL and NMZL. There is no correlation between the percentage of CD20-positive B-cells, CD3-positive T-cells or CD138- positive plasma cells or tissue origin of the tumor and the way the samples are clustered. However, the separation of the LPLs into 2 major cluster correspond to the presence and absence of underlying Waldenstrom Macroglobulinaemia (WM), suggesting that the genes distinguishing the 2 clusters are potential markers for differentiating WM LPL from non-WM LPL. Next, using supervised analysis we identified a cluster of genes including MMP7, LTF and SFRP2 with high signals specific to MALTs, while other genes including PRDM1 (BLIMP1), XBP1 and TNFRSF17 (BCMA) were specifically over-expressed in LPL. These may therefore represent novel diagnostic marker differentiating these entities. Several of these are further validated at the protein level using immunohistochemistry on a tissue microarray. Consistent with the unsupervised analysis, SMZL and NMZL have little difference and share the over-expression of CD22 and WNT3 among other genes. Clustering of these samples based on the pathways and genesets that are enriched in the individual tumors as compared to their normal tissue counterpart showed a mutually exclusive pattern with significant enrichment of NFKB-related genesets and genes in LPL and MALT, and significant enrichment of B-cell receptor signaling genesets and genes in SMZL and NMZL. Our analysis, for the first time, describes the molecular relationship between these closely related lymphomas. In the process, we identified novel diagnostic markers that may differentiate these conditions and also new insights into molecular pathways that are differentially activated in the different conditions. These may represent potential therapeutic targets

Abstract Events mediating transformation from pre-malignant MGUS to MM is currently not well defined. Recurrent genetic abnormalities such as t(11;14), t(4;14), hyperdiploidy and chromosome 13 deletion are already present in MGUS at relatively similar frequency to MM. The unified deregulation of D-type cyclins is also already present in MGUS. Previous GEP studies have revealed little differences between MGUS and MM. A more recent study using a large patient cohort and new generation Affymetrix genechip identifies an MGUS signature but the functional and biological significance underlying this signature is unknown. In the current study, we analyze a cohort of 22 MGUS and 101 MM from the Mayo Clinic with GEP performed on Affymetrix U133A genechip using gene set enrichment analysis, a method that analyze differentially expressed gene between 2 phenotypes of interest in the context of published or curated genesets that represent specific biological, chemical, or molecular perturbation to cells. This method increases the sensitivity of identifying low but significant changes in gene expression. We made further modification to the original method that allows assessment in individual samples rather than the average across a phenotype further increasing the specificity of the output. In this analysis, 313 genesets were significant enriched for genes over-expressed in MM compared to MGUS, representing potential activated pathways that mediate transformation. When MM samples and genesets were clustered using the enrichment score for each genesets and samples, 4 cluster of genesets emerged, one including a number of MYC genesets, one including a number of cell cycle related genesets, one including genesets related to metabolic activity and another including a number of IFN related genesets. Further dissection of these correlated genesets to identify common enriched genes (leading edge genes) led to identification of a MYC core signature, tRNA core signature, Proteosome core signature and metabolic core signature which are highly correlated. From known literature and biology, it is likely that MYC activation leads to downstream activation of protein synthesis (tRNA signature), degradation (proteasome signature), and metabolic pathway (metabolic signature). This is verified by GEP results from in vitro modulation of MYC activation in cell lines. In addition, a cell cycle core signature and IFN core signature was identified. Activation of IFN and MYC core signatures accounts for almost 90% of MM patients. The remaining patients have a metabolic signature without MYC or IFN activation. The activation of the different core signatures is significantly correlated with certain TC classes when assessed by Chi-Square test. IFN activation is significantly correlated with D1 subtypes and negatively correlated with D2 subtypes. MYC activation is negatively correlated with t(11;14). Similar patterns were observed in a validation dataset of 351 MM patients from UAMS (GSE2677) with GEP performed on the U133plus2.0 chip. These results are validated at the protein level using IHC on TMA of the Mayo cohort. The activation of the IFN and MYC pathway may represent predominant mechanism in MGUS to MM progression.

Abstract Abstract 88 Introduction: The biology and genetics of anaplastic large T-cell lymphomas (ALCLs) lacking ALK translocations remain poorly defined. We recently identified recurrent translocations involving the DUSP22-IRF4 locus on 6p25.3 in ALK-negative ALCL (Blood 2011;117:915–9). This translocation is present in about 20% of ALK-negative ALCLs and is absent in other T-cell lymphomas; however, the biologic consequences are unknown. In this study, we utilized gene expression profiling (GEP) to compare T-cell lymphomas with and without this translocation. Methods: cDNA was prepared from total RNA extracted from frozen tissue from 25 T-cell lymphoma patients under a Mayo Clinic IRB-approved protocol as follows: 5 ALK-negative ALCLs with 6p25.3 translocations (2 systemic, 3 cutaneous), 12 ALK-negative ALCLs without 6p25.3 translocations (11 systemic, 3 cutaneous), 5 ALK-positive ALCLs, and 3 CD30-positive peripheral T-cell lymphomas (1 transformed mycosis fungoides and 2 not otherwise specified). Partner loci in cases with 6p25.3 translocations were 7p32.3 in 2, 9p24.3 in 1, and unknown in 2. GEP data (Affymetrix U133 plus 2.0) were analyzed by unsupervised clustering; in addition, a GEP signature was developed using genes with raw values >200 in at least 2 cases with at least 2-fold, statistically significantly different expression between 6p25.3-translocated and non-translocated cases (p<0.05 after Benjamini-Hochberg multiple testing correction). Representative overexpressed genes were confirmed by quantitative real-time (qRT-) PCR. Supervised pathway analysis was performed using signatures derived from other studies or curated from public pathway databases. Results: By unsupervised clustering, the 5 cases with 6p25.3 translocations clustered together (281 probe sets). A GEP signature segregating translocated from non-translocated cases included 283 genes (195 up- and 88 down-regulated). Key features of this signature included altered expression of histone-modifying genes (e.g. EZH2, SUZ12, MLLT3, MLL; HDAC4) and up-regulation of cancer-testis antigen (CTA) genes (e.g. SSX4, TEX14, MAGEA12, SSX1, CTAG2). The signature included down-regulation of DUSP22 but not of other genes on 6p25 in the signature. Additional overexpressed genes were validated by qRT-PCR (relative expression: TEX14, 2.9, p=0.01, t test; MAL, 6.5, p=0.003; CCR8, 7.6, p=0.0004). Supervised pathway analysis demonstrated a high proliferation signature in cases with 6p25.3 translocations (p=0.006 vs. untranslocated cases). Conclusions: ALK-negative ALCLs with translocations involving the DUSP22-IRF4 locus on 6p25.3 have a distinct gene expression signature characterized by alterations of histone-modifying genes and CTA genes (which typically are regulated epigenetically, either through histone modifications or changes in methylation). This signature appears independent of anatomic site (systemic vs. cutaneous) and partner locus (7p32.3 vs. other). DUSP22, encoding a dual-specificity phosphatase involved in modulating mitogen-activated protein kinase (MAPK) signaling, is significantly down-regulated in translocated cases; however, pathway analysis did not show significant alterations in MAPK-associated genes. Other known genes in the region of the breakpoint on 6p25 were not affected substantially. These findings suggest the possibility of an unrecognized function of DUSP22. Contributions of the varying partner loci remain unclear. The distinct GEP signature associated with 6p25.3 translocations provides strong support for their biologic relevance. Finally, the genes involved in the histone-modifying signature show overlap with histone-modifying genes mutated in B-cell lymphomas, suggesting potential commonalities in the biology and possibly targeted treatment of B- and T-cell lymphomas. Disclosures: No relevant conflicts of interest to declare.

Abstract Abstract 3931 Poster Board III-867 Extranodal nasal-type Natural Killer/T-cell lymphoma (NKTCL) is a distinct clinicopathologic entity most commonly affecting Asians and Central and South Americans, and characterized by a clonal proliferation of NK or T cells with a cytotoxic phenotype. There is a strong association with Epstein-Barr Virus. The tumor is aggressive with patient usually surviving short duration even with chemotherapy. We performed the first comprehensive genome-wide gene expression profiling (GEP) of extranodal nasal-type Natural Killer/T-cell lymphoma (NKTCL) using formalin-fixed paraffin embedded (FFPE) tissue (n=25) and NK cell lines (n=5) and compared the results to the GEP of normal NK cells using the Illumina DASL whole genome array, with the objective of understanding the oncogenic pathways involved in the pathogenesis of NKTCL and to identify potential therapeutic targets. Quantitative-PCR validation of the GEP findings revealed over-expression of candidate genes BIRC5 (survivin), EZH2 and STMN1 in NK cell lines compared to normal NK cells, consistent with the GEP data. We then extracted a list of genes that are differentially expressed between NKTCL and normal NK cells and tissue controls. We than subjected this list of genes to pathway and network analysis using Metacore. This revealed a significant enrichment for cell cycle related genes and pathways. Furthermore, the network analysis results demonstrated a pro-proliferative and anti-apoptotic phenotype in NKTCL characterized by activation of Myc and nuclear factor kappa B (NF-KB), and deregulation of p53. This was further corroborated using immunohistochemistry on tissue microarray of NKTCL (n=33, including all cases with GEP performed). We observed a significant percentage of NKTCL showing overexpression for c-Myc (45.4%), p53 (87.9%) and NF-KB p50 (67.7%) on immunohistochemistry. Notably, overexpression of survivin was observed in 97% of the cases. Based on our findings, we propose a model of NKTCL pathogenesis where deregulation of p53 together with activation of MYC and NF-KB, possibly driven by EBV LMP-1, result in the cumulative upregulation of survivin. When KHYG and NKYS cell lines were treated with a compound IDR E804 which inhibited survivin, there is significant inhibition of cell growth as assess by MTS assay and induction of apoptosis as measured using Annexin V staining by flow cytometry. This suggests that compounds inhibiting survivin may be a potentially useful novel therapeutic approach in NKTCL. Disclosures: No relevant conflicts of interest to declare.

Abstract Abstract 834 Multiple myeloma (MM) is an incurable bone marrow cancer. Events mediating transformation from the pre-malignant monoclonal gammopathy of undetermined significance (MGUS) to MM is unknown. We analyzed 2 gene expression datasets generated on the Affymetrix U133 platform. The test set consisted of 22 MGUS and 101 MM (GEO Accession GSE6477) and the validation set 50 MGUS and 351 MM (GEO Accession GSE2658 and GSE5900). The gene expression profiles of MM were compared to MGUS using gene-set enrichment analysis. Genes over-expressed in MM were enriched for cell cycle, proliferation and MYC activation gene-sets. We dissected the relationship between MYC and cell cycle, and identified a MYC activation signature dissociated from proliferation. We validated our MYC signature in publicly available mouse and human cell line gene expression dataset (GEO Accession GSE3151 and GSE3158 respectively), showing specific expression of our MYC signature in cell lines forced to expressed MYC but not when over-expressing other oncogene such as E2F, HER2. In these analyses, we noted that tumors with RAS mutation also consistently express this signature. Applying this signature to the test dataset, we showed that MYC is activated in 60% of myeloma but none of MGUS. This pattern is reproduced in an independent validation dataset. In order to validate the hypothesis that RAS mutation may also lead to the MYC activation signature, we correlated RAS mutation with MYC activation as measured by a MYC index calculated from the median expression of genes that constitute the MYC activation signature, and showed that almost all cases with RAS mutation have a high MYC index. Together with samples with very high expression of MYC mRNA corresponding to those with MYC translocations, these 2 mechanisms account for 67% of cases with MYC activation. To further confirm MYC activation, we performed immunohistochemistry using a validated MYC antibody together with CD138 double-staining. We can clearly identify CD138 positive plasma cells with and without nuclear MYC expression and found that nuclear expression of MYC, a marker of MYC activation, correlated strongly with the MYC signature, therefore providing clear evidence that MYC activation is present in majority of newly diagnosed MM but not in MGUS. MYC activation is not very well correlated with proliferation as assessed by the plasma cell labeling index. Among newly diagnosed myeloma patients with plasma cell labeling index less than 1, those with nuclear MYC expression have significantly shorter survival than those without (Median survival 77.7 months versus 37.9 months, log-rank p-value 0.04). Multiple pathways converge on MYC activation, which is a common transforming event in MM associated with poor prognosis. MYC nuclear staining by IHC can be a useful clinical surrogate. Targeting MYC can be a chemoprevention strategy. Disclosures: Fonseca: Amgen: Consultancy; Medtronic: Consultancy; Celgene: Consultancy; Bristol Mayor Squibs: Consultancy; Genzyme: Consultancy; Otsuka: Consultancy.

Key PointsDUSP22-rearranged ALCLs belong to a distinct subset of ALCLs lacking activated STAT3. DUSP22-rearranged ALCLs have a unique molecular signature characterized by DNA hypomethylation and an immunogenic phenotype.