CUP-AI-Dx: A tool for inferring cancer tissue of origin and molecular subtype using RNA gene-expression data and artificial intelligence.

BACKGROUND: Cancer of unknown primary (CUP), representing approximately 3-5% of all malignancies, is defined as metastatic cancer where a primary site of origin cannot be found despite a standard diagnostic workup. Because knowledge of a patient's primary cancer remains fundamental to their treatment, CUP patients are significantly disadvantaged and most have a poor survival outcome. Developing robust and accessible diagnostic methods for resolving cancer tissue of origin, therefore, has significant value for CUP patients. METHODS: We developed an RNA-based classifier called CUP-AI-Dx that utilizes a 1D Inception convolutional neural network (1D-Inception) model to infer a tumor's primary tissue of origin. CUP-AI-Dx was trained using the transcriptional profiles of 18,217 primary tumours representing 32 cancer types from The Cancer Genome Atlas project (TCGA) and International Cancer Genome Consortium (ICGC). Gene expression data was ordered by gene chromosomal coordinates as input to the 1D-CNN model, and the model utilizes multiple convolutional kernels with different configurations simultaneously to improve generality. The model was optimized through extensive hyperparameter tuning, including different max-pooling layers and dropout settings. For 11 tumour types, we also developed a random forest model that can classify the tumour's molecular subtype according to prior TCGA studies. The optimised CUP-AI-Dx tissue of origin classifier was tested on 394 metastatic samples from 11 tumour types from TCGA and 92 formalin-fixed paraffin-embedded (FFPE) samples representing 18 cancer types from two clinical laboratories. The CUP-AI-Dx molecular subtype was also independently tested on independent ovarian and breast cancer microarray datasets FINDINGS: CUP-AI-Dx identifies the primary site with an overall top-1-accuracy of 98.54% in cross-validation and 96.70% on a test dataset. When applied to two independent clinical-grade RNA-seq datasets generated from two different institutes from the US and Australia, our model predicted the primary site with a top-1-accuracy of 86.96% and 72.46% respectively. INTERPRETATION: The CUP-AI-Dx predicts tumour primary site and molecular subtype with high accuracy and therefore can be used to assist the diagnostic work-up of cancers of unknown primary or uncertain origin using a common and accessible genomics platform. FUNDING: NIH R35 GM133562, NCI P30 CA034196, Victorian Cancer Agency Australia.

The ubiquitous 'cancer mutational signature' 5 occurs specifically in cancers with deleted FHIT alleles.

The FHIT gene is located at the fragile FRA3B locus where activation by carcinogen-induced and endogenous replication stress causes FHIT deletions even in normal cells over a lifetime. Our lab has shown that loss of FHIT expression causes genome instability and provides single-strand DNA substrates for APOBEC3B hypermutation, in line with evidence that FHIT locus deletions occur in many cancers. Based on these biological features, we hypothesized that FHIT loss drives development of COSMIC mutational signature 5 and here provide evidence, including data mining of >6,500 TCGA samples, that FHIT is the cancer-associated gene with copy number alterations correlating most significantly with signature 5 mutation rate. In addition, tissues of Fhit-deficient mice exhibit a mutational signature strongly resembling signature 5 (cosine similarity value = 0.89). We conclude that FHIT loss is a molecular determinant for signature 5 mutations, which occur in all cancer types early in cancer development, are clock-like, and accelerated by carcinogen exposure. Loss of FHIT caretaker function may be a predictive and preventive marker for cancer development.

Exome-wide single-base substitutions in tissues and derived cell lines of the constitutive Fhit knockout mouse.

Loss of expression of Fhit, a tumor suppressor and genome caretaker, occurs in preneoplastic lesions during development of many human cancers. Furthermore, Fhit-deficient mouse models are exquisitely susceptible to carcinogen induction of cancers of the lung and forestomach. Due to absence of Fhit genome caretaker function, cultured cells and tissues of the constitutive Fhit knockout strain develop chromosome aneuploidy and allele copy number gains and losses and we hypothesized that Fhit-deficient cells would also develop point mutations. On analysis of whole exome sequences of Fhit-deficient tissues and cultured cells, we found 300 to >1000 single-base substitutions associated with Fhit loss in the 2% of the genome included in exomes, relative to the C57Bl6 reference genome. The mutation signature is characterized by increased C>T and T>C mutations, similar to the "age at diagnosis" signature identified in human cancers. The Fhit-deficiency mutation signature also resembles a C>T and T>C mutation signature reported for human papillary kidney cancers and a similar signature recently reported for esophageal and bladder cancers, cancers that are frequently Fhit deficient. The increase in T>C mutations in -/- exomes may be due to dNTP imbalance, particularly in thymidine triphosphate, resulting from decreased expression of thymidine kinase 1 in Fhit-deficient cells. Fhit-deficient kidney cells that survived in vitro dimethylbenz(a)anthracene treatment additionally showed increased T>A mutations, a signature generated by treatment with this carcinogen, suggesting that these T>A transversions may be evidence of carcinogen-induced preneoplastic changes.

Replicative Stress and the FHIT Gene: Roles in Tumor Suppression, Genome Stability and Prevention of Carcinogenesis.

The fragile FHIT gene, encompassing the chromosomal fragile site FRA3B, is an early target of DNA damage in precancerous cells. While vulnerable to DNA damage itself, FHIT protein expression is essential to protect from DNA damage-induced cancer initiation and progression by modulating genome stability, oxidative stress and levels of accumulating DNA damage. Thus, FHIT, whose expression is lost or reduced in many human cancers, is a tumor suppressor and genome caretaker whose loss initiates genome instability in preneoplastic lesions. Ongoing studies are seeking more detailed understanding of the role of FHIT in the cellular response to oxidative damage. This review discusses the relationship between FHIT, reactive oxygen species production, and DNA damage in the context of cancer initiation and progression.

Fhit deficiency-induced global genome instability promotes mutation and clonal expansion.

Loss of Fhit expression, encoded at chromosome fragile site FRA3B, leads to increased replication stress, genome instability and accumulation of genetic alterations. We have proposed that Fhit is a genome 'caretaker' whose loss initiates genome instability in preneoplastic lesions. We have characterized allele copy number alterations and expression changes observed in Fhit-deficient cells in conjunction with alterations in cellular proliferation and exome mutations, using cells from mouse embryo fibroblasts (MEFs), mouse kidney, early and late after establishment in culture, and in response to carcinogen treatment. Fhit (-/-) MEFs escape senescence to become immortal more rapidly than Fhit (+/+) MEFs; -/- MEFs and kidney cultures show allele losses and gains, while +/+ derived cells show few genomic alterations. Striking alterations in expression of p53, p21, Mcl1 and active caspase 3 occurred in mouse kidney -/- cells during progressive tissue culture passage. To define genomic changes associated with preneoplastic changes in vivo, exome DNAs were sequenced for +/+ and -/- liver tissue after treatment of mice with the carcinogen, 7,12-dimethylbenz[a]anthracene, and for +/+ and -/- kidney cells treated in vitro with this carcinogen. The -/- exome DNAs, in comparison with +/+ DNA, showed small insertions, deletions and point mutations in more genes, some likely related to preneoplastic changes. Thus, Fhit loss provides a 'mutator' phenotype, a cellular environment in which mild genome instability permits clonal expansion, through proliferative advantage and escape from apoptosis, in response to pressures to survive.

PP2A-activating drugs selectively eradicate TKI-resistant chronic myeloid leukemic stem cells.

The success of tyrosine kinase inhibitors (TKIs) in treating chronic myeloid leukemia (CML) depends on the requirement for BCR-ABL1 kinase activity in CML progenitors. However, CML quiescent HSCs are TKI resistant and represent a BCR-ABL1 kinase-independent disease reservoir. Here we have shown that persistence of leukemic HSCs in BM requires inhibition of the tumor suppressor protein phosphatase 2A (PP2A) and expression--but not activity--of the BCR-ABL1 oncogene. Examination of HSCs from CML patients and healthy individuals revealed that PP2A activity was suppressed in CML compared with normal HSCs. TKI-resistant CML quiescent HSCs showed increased levels of BCR-ABL1, but very low kinase activity. BCR-ABL1 expression, but not kinase function, was required for recruitment of JAK2, activation of a JAK2/ß-catenin survival/self-renewal pathway, and inhibition of PP2A. PP2A-activating drugs (PADs) markedly reduced survival and self-renewal of CML quiescent HSCs, but not normal quiescent HSCs, through BCR-ABL1 kinase-independent and PP2A-mediated inhibition of JAK2 and ß-catenin. This led to suppression of human leukemic, but not normal, HSC/progenitor survival in BM xenografts and interference with long-term maintenance of BCR-ABL1-positive HSCs in serial transplantation assays. Targeting the JAK2/PP2A/ß-catenin network in quiescent HSCs with PADs (e.g., FTY720) has the potential to treat TKI-refractory CML and relieve lifelong patient dependence on TKIs.