Germline homozygosity and allelic imbalance of HLA-I are common in esophagogastric adenocarcinoma and impair the repertoire of immunogenic peptides.

BACKGROUND: The individual HLA-I genotype is associated with cancer, autoimmune diseases and infections. This study elucidates the role of germline homozygosity or allelic imbalance of HLA-I loci in esophago-gastric adenocarcinoma (EGA) and determines the resulting repertoires of potentially immunogenic peptides. METHODS: HLA genotypes and sequences of either (1) 10 relevant tumor-associated antigens (TAAs) or (2) patient-specific mutation-associated neoantigens (MANAs) were used to predict good-affinity binders using an in silico approach for MHC-binding (www.iedb.org). Imbalanced or lost expression of HLA-I-A/B/C alleles was analyzed by transcriptome sequencing. FluoroSpot assays and TCR sequencing were used to determine peptide-specific T-cell responses. RESULTS: We show that germline homozygosity of HLA-I genes is significantly enriched in EGA patients (n=80) compared with an HLA-matched reference cohort (n=7605). Whereas the overall mutational burden is similar, the repertoire of potentially immunogenic peptides derived from TAAs and MANAs was lower in homozygous patients. Promiscuity of peptides binding to different HLA-I molecules was low for most TAAs and MANAs and in silico modeling of the homozygous to a heterozygous HLA genotype revealed normalized peptide repertoires. Transcriptome sequencing showed imbalanced expression of HLA-I alleles in 75% of heterozygous patients. Out of these, 33% showed complete loss of heterozygosity, whereas 66% had altered expression of only one or two HLA-I molecules. In a FluoroSpot assay, we determined that peptide-specific T-cell responses against NY-ESO-1 are derived from multiple peptides, which often exclusively bind only one HLA-I allele. CONCLUSION: The high frequency of germline homozygosity in EGA patients suggests reduced cancer immunosurveillance leading to an increased cancer risk. Therapeutic targeting of allelic imbalance of HLA-I molecules should be considered in EGA.

Correction to: Identification of TAZ as the essential molecular switch in orchestrating SCLC phenotypic transition and metastasis.

[This corrects the article DOI: 10.1093/nsr/nwab232.].

Somatic rearrangements causing oncogenic ectodomain deletions of FGFR1 in squamous cell lung cancer.

The discovery of frequent 8p11-p12 amplifications in squamous cell lung cancer (SQLC) has fueled hopes that FGFR1, located inside this amplicon, might be a therapeutic target. In a clinical trial, only 11% of patients with 8p11 amplification (detected by FISH) responded to FGFR kinase inhibitor treatment. To understand the mechanism of FGFR1 dependency, we performed deep genomic characterization of 52 SQLCs with 8p11-p12 amplification, including 10 tumors obtained from patients who had been treated with FGFR inhibitors. We discovered somatically altered variants of FGFR1 with deletion of exons 1-8 that resulted from intragenic tail-to-tail rearrangements. These ectodomain-deficient FGFR1 variants (ΔEC-FGFR1) were expressed in the affected tumors and were tumorigenic in both in vitro and in vivo models of lung cancer. Mechanistically, breakage-fusion-bridges were the source of 8p11-p12 amplification, resulting from frequent head-to-head and tail-to-tail rearrangements. Generally, tail-to-tail rearrangements within or in close proximity upstream of FGFR1 were associated with FGFR1 dependency. Thus, the genomic events shaping the architecture of the 8p11-p12 amplicon provide a mechanistic explanation for the emergence of FGFR1-driven SQLC. Specifically, we believe that FGFR1 ectodomain-deficient and FGFR1-centered amplifications caused by tail-to-tail rearrangements are a novel somatic genomic event that might be predictive of therapeutically relevant FGFR1 dependency.

Characterizing Evolutionary Dynamics Reveals Strategies to Exhaust the Spectrum of Subclonal Resistance in EGFR-Mutant Lung Cancer.

The emergence of resistance to targeted therapies restrains their efficacy. The development of rationally guided drug combinations could overcome this currently insurmountable clinical challenge. However, our limited understanding of the trajectories that drive the outgrowth of resistant clones in cancer cell populations precludes design of drug combinations to forestall resistance. Here, we propose an iterative treatment strategy coupled with genomic profiling and genome-wide CRISPR activation screening to systematically extract and define preexisting resistant subpopulations in an EGFR-driven lung cancer cell line. Integrating these modalities identifies several resistance mechanisms, including activation of YAP/TAZ signaling by WWTR1 amplification, and estimates the associated cellular fitness for mathematical population modeling. These observations led to the development of a combination therapy that eradicated resistant clones in large cancer cell line populations by exhausting the spectrum of genomic resistance mechanisms. However, a small fraction of cancer cells was able to enter a reversible nonproliferative state of drug tolerance. This subpopulation exhibited mesenchymal properties, NRF2 target gene expression, and sensitivity to ferroptotic cell death. Exploiting this induced collateral sensitivity by GPX4 inhibition clears drug-tolerant populations and leads to tumor cell eradication. Overall, this experimental in vitro data and theoretical modeling demonstrate why targeted mono- and dual therapies will likely fail in sufficiently large cancer cell populations to limit long-term efficacy. Our approach is not tied to a particular driver mechanism and can be used to systematically assess and ideally exhaust the resistance landscape for different cancer types to rationally design combination therapies. SIGNIFICANCE: Unraveling the trajectories of preexisting resistant and drug-tolerant persister cells facilitates the rational design of multidrug combination or sequential therapies, presenting an approach to explore for treating EGFR-mutant lung cancer.

Counteracting lineage-specific transcription factor network finely tunes lung adeno-to-squamous transdifferentiation through remodeling tumor immune microenvironment.

Human lung adenosquamous cell carcinoma (LUAS), containing both adenomatous and squamous pathologies, harbors strong plasticity and is significantly associated with poor prognosis. We established an up-to-date comprehensive genomic and transcriptomic landscape of LUAS in 109 Chinese specimens and demonstrated LUAS development via adeno-to-squamous transdifferentiation. Unsupervised transcriptomic clustering and dynamic network biomarker analysis identified an inflammatory subtype as the critical transition stage during LUAS development. Dynamic dysregulation of the counteracting lineage-specific transcription factors (TFs), containing adenomatous TFs NKX2-1 and FOXA2, and squamous TFs TP63 and SOX2, finely tuned the lineage transition via promoting CXCL3/5-mediated neutrophil infiltration. Genomic clustering identified the most malignant subtype featured with STK11-inactivation, and targeting LSD1 through genetic deletion or pharmacological inhibition almost eradicated STK11-deficient lung tumors. These data collectively uncover the comprehensive molecular landscape, oncogenic driver spectrum and therapeutic vulnerability of Chinese LUAS.

Distinct Genetically Determined Origins of Myd88/BCL2-Driven Aggressive Lymphoma Rationalize Targeted Therapeutic Intervention Strategies.

Genomic profiling revealed the identity of at least 5 subtypes of diffuse large B-cell lymphoma (DLBCL), including the MCD/C5 cluster characterized by aberrations in MYD88, BCL2, PRDM1, and/or SPIB. We generated mouse models harboring B cell-specific Prdm1 or Spib aberrations on the background of oncogenic Myd88 and Bcl2 lesions. We deployed whole-exome sequencing, transcriptome, flow-cytometry, and mass cytometry analyses to demonstrate that Prdm1- or Spib-altered lymphomas display molecular features consistent with prememory B cells and light-zone B cells, whereas lymphomas lacking these alterations were enriched for late light-zone and plasmablast-associated gene sets. Consistent with the phenotypic evidence for increased B cell receptor signaling activity in Prdm1-altered lymphomas, we demonstrate that combined BTK/BCL2 inhibition displays therapeutic activity in mice and in five of six relapsed/refractory DLBCL patients. Moreover, Prdm1-altered lymphomas were immunogenic upon transplantation into immuno-competent hosts, displayed an actionable PD-L1 surface expression, and were sensitive to antimurine-CD19-CAR-T cell therapy, in vivo. SIGNIFICANCE: Relapsed/refractory DLBCL remains a major medical challenge, and most of these patients succumb to their disease. Here, we generated mouse models, faithfully recapitulating the biology of MYD88-driven human DLBCL. These models revealed robust preclinical activity of combined BTK/BCL2 inhibition. We confirmed activity of this regimen in pretreated non-GCB-DLBCL patients. See related commentary by Leveille et al., p. 8. This article is highlighted in the In This Issue feature, p. 1.

Pulmonary cancers across different histotypes share hybrid tuft cell/ionocyte-like molecular features and potentially druggable vulnerabilities.

Tuft cells are chemosensory epithelial cells in the respiratory tract and several other organs. Recent studies revealed tuft cell-like gene expression signatures in some pulmonary adenocarcinomas, squamous cell carcinomas (SQCC), small cell carcinomas (SCLC), and large cell neuroendocrine carcinomas (LCNEC). Identification of their similarities could inform shared druggable vulnerabilities. Clinicopathological features of tuft cell-like (tcl) subsets in various lung cancer histotypes were studied in two independent tumor cohorts using immunohistochemistry (n = 674 and 70). Findings were confirmed, and additional characteristics were explored using public datasets (RNA seq and immunohistochemical data) (n = 555). Drug susceptibilities of tuft cell-like SCLC cell lines were also investigated. By immunohistochemistry, 10-20% of SCLC and LCNEC, and approximately 2% of SQCC expressed POU2F3, the master regulator of tuft cells. These tuft cell-like tumors exhibited "lineage ambiguity" as they co-expressed NCAM1, a marker for neuroendocrine differentiation, and KRT5, a marker for squamous differentiation. In addition, tuft cell-like tumors co-expressed BCL2 and KIT, and tuft cell-like SCLC and LCNEC, but not SQCC, also highly expressed MYC. Data from public datasets confirmed these features and revealed that tuft cell-like SCLC and LCNEC co-clustered on hierarchical clustering. Furthermore, only tuft cell-like subsets among pulmonary cancers significantly expressed FOXI1, the master regulator of ionocytes, suggesting their bidirectional but immature differentiation status. Clinically, tuft cell-like SCLC and LCNEC had a similar prognosis. Experimentally, tuft cell-like SCLC cell lines were susceptible to PARP and BCL2 co-inhibition, indicating synergistic effects. Taken together, pulmonary tuft cell-like cancers maintain histotype-related clinicopathologic characteristics despite overlapping unique molecular features. From a therapeutic perspective, identification of tuft cell-like LCNECs might be crucial given their close kinship with tuft cell-like SCLC.

DSTYK inhibition increases the sensitivity of lung cancer cells to T cell-mediated cytotoxicity.

Lung cancer remains the leading cause of cancer-related death worldwide. We identify DSTYK, a dual serine/threonine and tyrosine non-receptor protein kinase, as a novel actionable target altered in non-small cell lung cancer (NSCLC). We also show DSTYK's association with a lower overall survival (OS) and poorer progression-free survival (PFS) in multiple patient cohorts. Abrogation of DSTYK in lung cancer experimental systems prevents mTOR-dependent cytoprotective autophagy, impairs lysosomal biogenesis and maturation, and induces accumulation of autophagosomes. Moreover, DSTYK inhibition severely affects mitochondrial fitness. We demonstrate in vivo that inhibition of DSTYK sensitizes lung cancer cells to TNF-α-mediated CD8+-killing and immune-resistant lung tumors to anti-PD-1 treatment. Finally, in a series of lung cancer patients, DSTYK copy number gain predicts lack of response to the immunotherapy. In summary, we have uncovered DSTYK as new therapeutic target in lung cancer. Prioritization of this novel target for drug development and clinical testing may expand the percentage of NSCLC patients benefiting from immune-based treatments.

Identification of TAZ as the essential molecular switch in orchestrating SCLC phenotypic transition and metastasis.

Small-cell lung cancer (SCLC) is a recalcitrant cancer characterized by high metastasis. However, the exact cell type contributing to metastasis remains elusive. Using a Rb1 L/L /Trp53 L/L mouse model, we identify the NCAMhiCD44lo/- subpopulation as the SCLC metastasizing cell (SMC), which is progressively transitioned from the non-metastasizing NCAMloCD44hi cell (non-SMC). Integrative chromatin accessibility and gene expression profiling studies reveal the important role of the SWI/SNF complex, and knockout of its central component, Brg1, significantly inhibits such phenotypic transition and metastasis. Mechanistically, TAZ is silenced by the SWI/SNF complex during SCLC malignant progression, and its knockdown promotes SMC transition and metastasis. Importantly, ectopic TAZ expression reversely drives SMC-to-non-SMC transition and alleviates metastasis. Single-cell RNA-sequencing analyses identify SMC as the dominant subpopulation in human SCLC metastasis, and immunostaining data show a positive correlation between TAZ and patient prognosis. These data uncover high SCLC plasticity and identify TAZ as the key molecular switch in orchestrating SCLC phenotypic transition and metastasis.

Therapeutic targeting of the mevalonate-geranylgeranyl diphosphate pathway with statins overcomes chemotherapy resistance in small cell lung cancer.

Small cell lung cancer (SCLC) lacks effective treatments to overcome chemoresistance. Here we established multiple human chemoresistant xenograft models through long-term intermittent chemotherapy, mimicking clinically relevant therapeutic settings. We show that chemoresistant SCLC undergoes metabolic reprogramming relying on the mevalonate (MVA)-geranylgeranyl diphosphate (GGPP) pathway, which can be targeted using clinically approved statins. Mechanistically, statins induce oxidative stress accumulation and apoptosis through the GGPP synthase 1 (GGPS1)-RAB7A-autophagy axis. Statin treatment overcomes both intrinsic and acquired SCLC chemoresistance in vivo across different SCLC PDX models bearing high GGPS1 levels. Moreover, we show that GGPS1 expression is negatively associated with survival in patients with SCLC. Finally, we demonstrate that combined statin and chemotherapy treatment resulted in durable responses in three patients with SCLC who relapsed from first-line chemotherapy. Collectively, these data uncover the MVA-GGPP pathway as a metabolic vulnerability in SCLC and identify statins as a potentially effective treatment to overcome chemoresistance.

CD74-NRG1 Fusions Are Oncogenic In Vivo and Induce Therapeutically Tractable ERBB2:ERBB3 Heterodimerization.

NRG1 fusions are recurrent somatic genome alterations occurring across several tumor types, including invasive mucinous lung adenocarcinomas and pancreatic ductal adenocarcinomas and are potentially actionable genetic alterations in these cancers. We initially discovered CD74-NRG1 as the first NRG1 fusion in lung adenocarcinomas, and many additional fusion partners have since been identified. Here, we present the first CD74-NRG1 transgenic mouse model and provide evidence that ubiquitous expression of the CD74-NRG1 fusion protein in vivo leads to tumor development at high frequency. Furthermore, we show that ERBB2:ERBB3 heterodimerization is a mechanistic event in transformation by CD74-NRG1 binding physically to ERBB3 and that CD74-NRG1-expressing cells proliferate independent of supplemented NRG1 ligand. Thus, NRG1 gene fusions are recurrent driver oncogenes that cause oncogene dependency. Consistent with these findings, patients with NRG1 fusion-positive cancers respond to therapy targeting the ERBB2:ERBB3 receptors.

Two cell line models to study multiorganic metastasis and immunotherapy in lung squamous cell carcinoma.

There is a paucity of adequate mouse models and cell lines available to study lung squamous cell carcinoma (LUSC). We have generated and characterized two models of phenotypically different transplantable LUSC cell lines, i.e. UN-SCC679 and UN-SCC680, derived from A/J mice that had been chemically induced with N-nitroso-tris-chloroethylurea (NTCU). Furthermore, we genetically characterized and compared both LUSC cell lines by performing whole-exome and RNA sequencing. These experiments revealed similar genetic and transcriptomic patterns that may correspond to the classic LUSC human subtype. In addition, we compared the immune landscape generated by both tumor cells lines in vivo and assessed their response to immune checkpoint inhibition. The differences between the two cell lines are a good model for the remarkable heterogeneity of human squamous cell carcinoma. Study of the metastatic potential of these models revealed that both cell lines represent the organotropism of LUSC in humans, i.e. affinity to the brain, bones, liver and adrenal glands. In summary, we have generated valuable cell line tools for LUSC research, which recapitulates the complexity of the human disease.

Clonal dynamics of BRAF-driven drug resistance in EGFR-mutant lung cancer.

Activation of MAPK signaling via BRAF mutations may limit the activity of EGFR inhibitors in EGFR-mutant lung cancer patients. However, the impact of BRAF mutations on the selection and fitness of emerging resistant clones during anti-EGFR therapy remains elusive. We tracked the evolution of subclonal mutations by whole-exome sequencing and performed clonal analyses of individual metastases during therapy. Complementary functional analyses of polyclonal EGFR-mutant cell pools showed a dose-dependent enrichment of BRAFV600E and a loss of EGFR inhibitor susceptibility. The clones remain stable and become vulnerable to combined EGFR, RAF, and MEK inhibition. Moreover, only osimertinib/trametinib combination treatment, but not monotherapy with either of these drugs, leads to robust tumor shrinkage in EGFR-driven xenograft models harboring BRAFV600E mutations. These data provide insights into the dynamics of clonal evolution of EGFR-mutant tumors and the therapeutic implications of BRAF co-mutations that may facilitate the development of treatment strategies to improve the prognosis of these patients.

Cold and heterogeneous T cell repertoire is associated with copy number aberrations and loss of immune genes in small-cell lung cancer.

Small-cell lung cancer (SCLC) is speculated to harbor complex genomic intratumor heterogeneity (ITH) associated with high recurrence rate and suboptimal response to immunotherapy. Here, using multi-region whole exome/T cell receptor (TCR) sequencing as well as immunohistochemistry, we reveal a rather homogeneous mutational landscape but extremely cold and heterogeneous TCR repertoire in limited-stage SCLC tumors (LS-SCLCs). Compared to localized non-small cell lung cancers, LS-SCLCs have similar predicted neoantigen burden and genomic ITH, but significantly colder and more heterogeneous TCR repertoire associated with higher chromosomal copy number aberration (CNA) burden. Furthermore, copy number loss of IFN-γ pathway genes is frequently observed and positively correlates with CNA burden. Higher mutational burden, higher T cell infiltration and positive PD-L1 expression are associated with longer overall survival (OS), while higher CNA burden is associated with shorter OS in patients with LS-SCLC.

Chromothripsis followed by circular recombination drives oncogene amplification in human cancer.

The mechanisms behind the evolution of complex genomic amplifications in cancer have remained largely unclear. Using whole-genome sequencing data of the pediatric tumor neuroblastoma, we here identified a type of amplification, termed 'seismic amplification', that is characterized by multiple rearrangements and discontinuous copy number levels. Overall, seismic amplifications occurred in 9.9% (274 of 2,756) of cases across 38 cancer types, and were associated with massively increased copy numbers and elevated oncogene expression. Reconstruction of the development of seismic amplification showed a stepwise evolution, starting with a chromothripsis event, followed by formation of circular extrachromosomal DNA that subsequently underwent repetitive rounds of circular recombination. The resulting amplicons persisted as extrachromosomal DNA circles or had reintegrated into the genome in overt tumors. Together, our data indicate that the sequential occurrence of chromothripsis and circular recombination drives oncogene amplification and overexpression in a substantial fraction of human malignancies.

MAPK-pathway inhibition mediates inflammatory reprogramming and sensitizes tumors to targeted activation of innate immunity sensor RIG-I.

Kinase inhibitors suppress the growth of oncogene driven cancer but also enforce the selection of treatment resistant cells that are thought to promote tumor relapse in patients. Here, we report transcriptomic and functional genomics analyses of cells and tumors within their microenvironment across different genotypes that persist during kinase inhibitor treatment. We uncover a conserved, MAPK/IRF1-mediated inflammatory response in tumors that undergo stemness- and senescence-associated reprogramming. In these tumor cells, activation of the innate immunity sensor RIG-I via its agonist IVT4, triggers an interferon and a pro-apoptotic response that synergize with concomitant kinase inhibition. In humanized lung cancer xenografts and a syngeneic Egfr-driven lung cancer model these effects translate into reduction of exhausted CD8+ T cells and robust tumor shrinkage. Overall, the mechanistic understanding of MAPK/IRF1-mediated intratumoral reprogramming may ultimately prolong the efficacy of targeted drugs in genetically defined cancer patients.

Ferroptosis response segregates small cell lung cancer (SCLC) neuroendocrine subtypes.

Loss of TP53 and RB1 in treatment-naïve small cell lung cancer (SCLC) suggests selective pressure to inactivate cell death pathways prior to therapy. Yet, which of these pathways remain available in treatment-naïve SCLC is unknown. Here, through systemic analysis of cell death pathway availability in treatment-naïve SCLC, we identify non-neuroendocrine (NE) SCLC to be vulnerable to ferroptosis through subtype-specific lipidome remodeling. While NE SCLC is ferroptosis resistant, it acquires selective addiction to the TRX anti-oxidant pathway. In experimental settings of non-NE/NE intratumoral heterogeneity, non-NE or NE populations are selectively depleted by ferroptosis or TRX pathway inhibition, respectively. Preventing subtype plasticity observed under single pathway targeting, combined treatment kills established non-NE and NE tumors in xenografts, genetically engineered mouse models of SCLC and patient-derived cells, and identifies a patient subset with drastically improved overall survival. These findings reveal cell death pathway mining as a means to identify rational combination therapies for SCLC.

A Tuft Cell-Like Signature Is Highly Prevalent in Thymic Squamous Cell Carcinoma and Delineates New Molecular Subsets Among the Major Lung Cancer Histotypes.

INTRODUCTION: In-depth genomic characterization of thymic epithelial tumors (TETs), comprising thymomas and thymic carcinomas (TCs), failed to identify targetable mutations and suggested unique biology of TETs, including KIT expression in most TCs. Recently, tuft cell-like medullary thymic epithelial cells were identified in the murine thymus, and our reanalysis of the published gene expression data revealed that these cells express KIT. In addition, recently, a minor subset of SCLCs with tuft cell-like features was described. METHODS: We interrogated mRNA expression data from our tumor cohorts (N = 60) and publicly available, independent data sets from TETs and NSCLC (N = 1199) for expression of tuft cell genes and KIT. Expression of KIT and of POU2F3 protein, the master regulator of tuft cells, was analyzed in cancer tissue (N = 344) by immunohistochemistry. RESULTS: Normal human thymic tuft cells and most TCs coexpressed KIT and known tuft cell genes, particularly POU2F3 and GFI1B. Unexpectedly, small subsets of tuft cell-like tumors coexpressing POU2F3, GFI1B, and KIT were also identified among pulmonary squamous cell carcinomas, adenocarcinomas, and large cell neuroendocrine carcinoma and clustered together in each histologic cohort. In addition to the tuft cell-like signature, both thymic and lung tuft cell-like carcinomas had distinct genetic, pathologic, and clinical features in each cohort. CONCLUSIONS: We suggest that the tuft cell-like phenotype defines novel subsets of thymic and pulmonary carcinoma. Its high prevalence in thymic squamous cell carcinomas that have no known toxic or viral etiologies suggests a new mechanism of carcinogenesis that may lead to specific drug susceptibilities.

Texture analysis of iodine maps and conventional images for k-nearest neighbor classification of benign and metastatic lung nodules.

BACKGROUND: The purpose of this study was to analyze if the use of texture analysis on spectral detector CT (SDCT)-derived iodine maps (IM) in addition to conventional images (CI) improves lung nodule differentiation, when being applied to a k-nearest neighbor (KNN) classifier. METHODS: 183 cancer patients who underwent contrast-enhanced, venous phase SDCT of the chest were included: 85 patients with 146 benign lung nodules (BLN) confirmed by either prior/follow-up CT or histopathology and 98 patients with 425 lung metastases (LM) verified by histopathology, 18F-FDG-PET-CT or unequivocal change during treatment. Semi-automatic 3D segmentation of BLN/LM was performed, and volumetric HU attenuation and iodine concentration were acquired. For conventional images and iodine maps, average, standard deviation, entropy, kurtosis, mean of the positive pixels (MPP), skewness, uniformity and uniformity of the positive pixels (UPP) within the volumes of interests were calculated. All acquired parameters were transferred to a KNN classifier. RESULTS: Differentiation between BLN and LM was most accurate, when using all CI-derived features combined with the most significant IM-derived feature, entropy (Accuracy:0.87; F1/Dice:0.92). However, differentiation accuracy based on the 4 most powerful CI-derived features performed only slightly inferior (Accuracy:0.84; F1/Dice:0.89, p=0.125). Mono-parametric lung nodule differentiation based on either feature alone (i.e. attenuation or iodine concentration) was poor (AUC=0.65, 0.58, respectively). CONCLUSIONS: First-order texture feature analysis of contrast-enhanced staging SDCT scans of the chest yield accurate differentiation between benign and metastatic lung nodules. In our study cohort, the most powerful iodine map-derived feature slightly, yet insignificantly increased classification accuracy  compared to classification based on conventional image features only.