Personalized disease signatures through information-theoretic compaction of big cancer data

Transcriptome free energy can serve as a dynamic patient-specific biomarker in acute myeloid leukemia

Acute myeloid leukemia (AML) is prevalent in both adult and pediatric patients. Despite advances in patient categorization, the heterogeneity of AML remains a challenge. Recent studies have explored the use of gene expression data to enhance AML diagnosis and prognosis, however, alternative approaches rooted in physics and chemistry may provide another level of insight into AML transformation. Utilizing publicly available databases, we analyze 884 human and mouse blood and bone marrow samples. We employ a personalized medicine strategy, combining state-transition theory and surprisal analysis, to assess the RNA transcriptome of individual patients. The transcriptome is transformed into physical parameters that represent each sample's steady state and the free energy change (FEC) from that steady state, which is the state with the lowest free energy.We found the transcriptome steady state was invariant across normal and AML samples. FEC, representing active molecular processes, varied significantly between samples and was used to create patient-specific barcodes to characterize the biology of the disease. We discovered that AML samples that were in a transition state had the highest FEC. This disease state may be characterized as the most unstable and hence the most therapeutically targetable since a change in free energy is a thermodynamic requirement for disease progression. We also found that distinct sets of ongoing processes may be at the root of otherwise similar clinical phenotypes, implying that our integrated analysis of transcriptome profiles may facilitate a personalized medicine approach to cure AML and restore a steady state in each patient.

The immunogenic radiation and new players in immunotherapy and targeted therapy for head and neck cancer

Although treatment modalities for head and neck cancer have evolved considerably over the past decades, survival rates have plateaued. The treatment options remained limited to definitive surgery, surgery followed by fractionated radiotherapy with optional chemotherapy, and a definitive combination of fractionated radiotherapy and chemotherapy. Lately, immunotherapy has been introduced as the fourth modality of treatment, mainly administered as a single checkpoint inhibitor for recurrent or metastatic disease. While other regimens and combinations of immunotherapy and targeted therapy are being tested in clinical trials, adapting the appropriate regimens to patients and predicting their outcomes have yet to reach the clinical setting. Radiotherapy is mainly regarded as a means to target cancer cells while minimizing the unwanted peripheral effect. Radiotherapy regimens and fractionation are designed to serve this purpose, while the systemic effect of radiation on the immune response is rarely considered a factor while designing treatment. To bridge this gap, this review will highlight the effect of radiotherapy on the tumor microenvironment locally, and the immune response systemically. We will review the methodology to identify potential targets for therapy in the tumor microenvironment and the scientific basis for combining targeted therapy and radiotherapy. We will describe a current experience in preclinical models to test these combinations and propose how challenges in this realm may be faced. We will review new players in targeted therapy and their utilization to drive immunogenic response against head and neck cancer. We will outline the factors contributing to head and neck cancer heterogeneity and their effect on the response to radiotherapy. We will review in-silico methods to decipher intertumoral and intratumoral heterogeneity and how these algorithms can predict treatment outcomes. We propose that (a) the sequence of surgery, radiotherapy, chemotherapy, and targeted therapy should be designed not only to annul cancer directly, but to prime the immune response. (b) Fractionation of radiotherapy and the extent of the irradiated field should facilitate systemic immunity to develop. (c) New players in targeted therapy should be evaluated in translational studies toward clinical trials. (d) Head and neck cancer treatment should be personalized according to patients and tumor-specific factors.

Overexpressed kinetochore genes are used by cancer cells as genome destabilizers and transformation catalysts

Cancer cells have an altered transcriptome, which contributes to their abnormal behavior. Many tumors have high levels of kinetochore genes, which play important roles in genome stability. This overexpression could be utilized to destabilize cancer cell genomes, however this has not been proven specifically. We investigated the link between kinetochore gene overexpression, chromosomal number variations (CNVs) and genomic instability. Data on RNA expression and CNV from 12 different cancer types were evaluated using information theory. In all cancer types, we looked at the relationship between RNA expression and CNVs. Kinetochore gene expression was found to be substantially linked with CNV levels. In all cancer types, with the exception of thyroid cancer, highly expressed kinetochore genes were enriched in the most dominant cancer-specific co-expression subnetworks characterizing the largest patient subgroups. Except for thyroid cancer, kinetochore inner protein CENPA was among the transcripts most strongly associated with CNV values in all cancer types studied, with significantly higher expression levels in patients with high CNVs than in patients with low CNVs. CENPA function was investigated further in cell models by transfecting genomically stable (HCT116) and unstable (MCF7 and HT29) cancer cell lines using CENPA overexpression vectors. This overexpression increased the number of abnormal cell divisions in the stable cancer cell line HCT116 and, to a lesser extent, in the unstable cell lines MCF7 and HT29. Overexpression improved anchorage-independent growth properties of all cell lines. Our findings suggest that overexpression of kinetochore genes in general, and CENPA in particular, can cause genomic instability and cancer progression.

Computational quantification and characterization of independently evolving cellular subpopulations within tumors is critical to inhibit anti-cancer therapy resistance

BACKGROUND

Drug resistance continues to be a major limiting factor across diverse anti-cancer therapies. Contributing to the complexity of this challenge is cancer plasticity, in which one cancer subtype switches to another in response to treatment, for example, triple-negative breast cancer (TNBC) to Her2-positive breast cancer. For optimal treatment outcomes, accurate tumor diagnosis and subsequent therapeutic decisions are vital. This study assessed a novel approach to characterize treatment-induced evolutionary changes of distinct tumor cell subpopulations to identify and therapeutically exploit anticancer drug resistance.

METHODS

In this research, an information-theoretic single-cell quantification strategy was developed to provide a high-resolution and individualized assessment of tumor composition for a customized treatment approach. Briefly, this single-cell quantification strategy computes cell barcodes based on at least 100,000 tumor cells from each experiment and reveals a cell-specific signaling signature (CSSS) composed of a set of ongoing processes in each cell.

RESULTS

Using these CSSS-based barcodes, distinct subpopulations evolving within the tumor in response to an outside influence, like anticancer treatments, were revealed and mapped. Barcodes were further applied to assign targeted drug combinations to each individual tumor to optimize tumor response to therapy. The strategy was validated using TNBC models and patient-derived tumors known to switch phenotypes in response to radiotherapy (RT).

CONCLUSIONS

We show that a barcode-guided targeted drug cocktail significantly enhances tumor response to RT and prevents regrowth of once-resistant tumors. The strategy presented herein shows promise in preventing cancer treatment resistance, with significant applicability in clinical use.

Drug-Induced Resistance and Phenotypic Switch in Triple-Negative Breast Cancer Can Be Controlled via Resolution and Targeting of Individualized Signaling Signatures

Simple Summary

Patients with Triple Negative Breast Cancer (TNBC) have a poor prognosis due to high inter-tumor heterogeneity and absence of effective targeted treatments. Through quantification of ongoing processes in each individual with TNBC, we propose an explanation on why certain previously suggested monotherapies, such as anti-EGFR, are not effective. We experimentally demonstrate that monotherapies or drug combinations that are not adjusted accurately to the patient-specific ongoing processes may create an evolutionary pressure on a tumor leading to the emergence of previously undetected or untargeted cellular subpopulations. We show for example that certain TNBC tumors may benefit from therapies targeting estrogen receptors (ER), similarly to ER positive cancers. When untargeted, those tumors may develop large ER positive subpopulations. We propose that anti-TNBC therapy should be accurately tailored to the personalized molecular processes and that incomplete or “wrong” treatments may generate diverse evolutionary routes of TNBC tumors leading to drug resistance.

Abstract

Triple-negative breast cancer (TNBC) is an aggressive subgroup of breast cancers which is treated mainly with chemotherapy and radiotherapy. Epidermal growth factor receptor (EGFR) was considered to be frequently expressed in TNBC, and therefore was suggested as a therapeutic target. However, clinical trials of EGFR inhibitors have failed. In this study, we examine the relationship between the patient-specific TNBC network structures and possible mechanisms of resistance to anti-EGFR therapy. Using an information-theoretical analysis of 747 breast tumors from the TCGA dataset, we resolved individualized protein network structures, namely patient-specific signaling signatures (PaSSS) for each tumor. Each PaSSS was characterized by a set of 1–4 altered protein–protein subnetworks. Thirty-one percent of TNBC PaSSSs were found to harbor EGFR as a part of the network and were predicted to benefit from anti-EGFR therapy as long as it is combined with anti-estrogen receptor (ER) therapy. Using a series of single-cell experiments, followed by in vivo support, we show that drug combinations which are not tailored accurately to each PaSSS may generate evolutionary pressure in malignancies leading to an expansion of the previously undetected or untargeted subpopulations, such as ER+ populations. This corresponds to the PaSSS-based predictions suggesting to incorporate anti-ER drugs in certain anti-TNBC treatments. These findings highlight the need to tailor anti-TNBC targeted therapy to each PaSSS to prevent diverse evolutions of TNBC tumors and drug resistance development.

Explant Modeling of the Immune Environment of Head and Neck Cancer

Patients exhibit distinct responses to immunotherapies that are thought to be linked to their tumor immune environment. However, wide variations in outcomes are also observed in patients with matched baseline tumor environments, indicating that the biological response to treatment is not currently predictable using a snapshot analysis. To investigate the relationship between the immune environment of tumors and the biological response to immunotherapies, we characterized four murine head and neck squamous cell carcinoma (HNSCC) models on two genetic backgrounds. Using tumor explants from those models, we identified correlations between the composition of infiltrating immune cells and baseline cytokine profiles prior to treatment. Following treatment with PD-1 blockade, CTLA-4 blockade, or OX40 stimulation, we observed inter-individual variability in the response to therapy between genetically identical animals bearing the same tumor. These distinct biological responses to treatment were not linked to the initial tumor immune environment, meaning that outcome would not be predictable from a baseline analysis of the tumor infiltrates. We similarly performed the explant assay on patient HNSCC tumors and found significant variability between the baseline environment of the tumors and their response to therapy. We propose that tumor explants provide a rapid biological assay to assess response to candidate immunotherapies that may allow matching therapies to individual patient tumors. Further development of explant approaches may allow screening and monitoring of treatment responses in HNSCC.

Overcoming resistance to BRAFV600E inhibition in melanoma by deciphering and targeting personalized protein network alterations

BRAFV600E melanoma patients, despite initially responding to the clinically prescribed anti-BRAFV600E therapy, often relapse, and their tumors develop drug resistance. While it is widely accepted that these tumors are originally driven by the BRAFV600E mutation, they often eventually diverge and become supported by various signaling networks. Therefore, patient-specific altered signaling signatures should be deciphered and treated individually. In this study, we design individualized melanoma combination treatments based on personalized network alterations. Using an information-theoretic approach, we compute high-resolution patient-specific altered signaling signatures. These altered signaling signatures each consist of several co-expressed subnetworks, which should all be targeted to optimally inhibit the entire altered signaling flux. Based on these data, we design smart, personalized drug combinations, often consisting of FDA-approved drugs. We validate our approach in vitro and in vivo showing that individualized drug combinations that are rationally based on patient-specific altered signaling signatures are more efficient than the clinically used anti-BRAFV600E or BRAFV600E/MEK targeted therapy. Furthermore, these drug combinations are highly selective, as a drug combination efficient for one BRAFV600E tumor is significantly less efficient for another, and vice versa. The approach presented herein can be broadly applicable to aid clinicians to rationally design patient-specific anti-melanoma drug combinations.

A novel estimator of the interaction matrix in Graphical Gaussian Model of omics data using the entropy of non-equilibrium systems

MOTIVATION

Inferring the direct relationships between biomolecules from omics datasets is essential for the understanding of biological and disease mechanisms. Gaussian Graphical Model (GGM) provides a fairly simple and accurate representation of these interactions. However, estimation of the associated interaction matrix using data is challenging due to a high number of measured molecules and a low number of samples.

RESULTS

In this article, we use the thermodynamic entropy of the non-equilibrium system of molecules and the data-driven constraints among their expressions to derive an analytic formula for the interaction matrix of Gaussian models. Through a data simulation, we show that our method returns an improved estimation of the interaction matrix. Also, using the developed method, we estimate the interaction matrix associated with plasma proteome and construct the corresponding GGM and show that known NAFLD-related proteins like ADIPOQ, APOC, APOE, DPP4, CAT, GC, HP, CETP, SERPINA1, COLA1, PIGR, IGHD, SAA1 and FCGBP are among the top 15% most interacting proteins of the dataset.

AVAILABILITY AND IMPLEMENTATION

The supplementary materials can be found in the following URL: http://dynamic-proteome.utmb.edu/PrecisionMatrixEstimater/PrecisionMatrixEstimater.aspx.

SUPPLEMENTARY INFORMATION

Supplementary data are available at Bioinformatics online.

Artificial intelligence in cancer research: learning at different levels of data granularity

From genome-scale experimental studies to imaging data, behavioral footprints, and longitudinal healthcare records, the convergence of big data in cancer research and the advances in Artificial Intelligence (AI) is paving the way to develop a systems view of cancer. Nevertheless, this biomedical area is largely characterized by the co-existence of big data and small data resources, highlighting the need for a deeper investigation about the crosstalk between different levels of data granularity, including varied sample sizes, labels, data types, and other data descriptors. This review introduces the current challenges, limitations, and solutions of AI in the heterogeneous landscape of data granularity in cancer research. Such a variety of cancer molecular and clinical data calls for advancing the interoperability among AI approaches, with particular emphasis on the synergy between discriminative and generative models that we discuss in this work with several examples of techniques and applications.

Exploring Alzheimer's Disease Molecular Variability via Calculation of Personalized Transcriptional Signatures

Despite huge investments and major efforts to develop remedies for Alzheimer’s disease (AD) in the past decades, AD remains incurable. While evidence for molecular and phenotypic variability in AD have been accumulating, AD research still heavily relies on the search for AD-specific genetic/protein biomarkers that are expected to exhibit repetitive patterns throughout all patients. Thus, the classification of AD patients to different categories is expected to set the basis for the development of therapies that will be beneficial for subpopulations of patients. Here we explore the molecular heterogeneity among a large cohort of AD and non-demented brain samples, aiming to address the question whether AD-specific molecular biomarkers can progress our understanding of the disease and advance the development of anti-AD therapeutics. We studied 951 brain samples, obtained from up to 17 brain regions of 85 AD patients and 22 non-demented subjects. Utilizing an information-theoretic approach, we deciphered the brain sample-specific structures of altered transcriptional networks. Our in-depth analysis revealed that 7 subnetworks were repetitive in the 737 diseased and 214 non-demented brain samples. Each sample was characterized by a subset consisting of ~1–3 subnetworks out of 7, generating 52 distinct altered transcriptional signatures that characterized the 951 samples. We show that 30 different altered transcriptional signatures characterized solely AD samples and were not found in any of the non-demented samples. In contrast, the rest of the signatures characterized different subsets of sample types, demonstrating the high molecular variability and complexity of gene expression in AD. Importantly, different AD patients exhibiting similar expression levels of AD biomarkers harbored distinct altered transcriptional networks. Our results emphasize the need to expand the biomarker-based stratification to patient-specific transcriptional signature identification for improved AD diagnosis and for the development of subclass-specific future treatment.

Translating Cancer Molecular Variability into Personalized Information Using Bulk and Single Cell Approaches

Cancer research is striving toward new frontiers of assigning the correct personalized drug(s) to a given patient. However, extensive tumor heterogeneity poses a major obstacle. Tumors of the same type often respond differently to therapy, due to patient-specific molecular aberrations and/or untargeted tumor subpopulations. It is frequently not possible to determine a priori which patients will respond to a certain therapy or how an efficient patient-specific combined therapy should be designed. Large-scale datasets have been growing at an accelerated pace and various technologies and analytical tools for single cell and bulk level analyses are being developed to extract significant individualized signals from such heterogeneous data. However, personalized therapies that dramatically alter the course of the disease remain scarce, and most tumors still respond poorly to medical care. In this review, the basic concepts of bulk and single cell approaches are discussed, as well as their emerging role in individualized designs of drug therapies, including the advantages and limitations of their applications in personalized medicine.

Identification of Platform-Independent Diagnostic Biomarker Panel for Hepatocellular Carcinoma Using Large-Scale Transcriptomics Data

The high mortality rate of hepatocellular carcinoma (HCC) is primarily due to its late diagnosis. In the past, numerous attempts have been made to design genetic biomarkers for the identification of HCC; unfortunately, most of the studies are based on small datasets obtained from a specific platform or lack reasonable validation performance on the external datasets. In order to identify a universal expression-based diagnostic biomarker panel for HCC that can be applicable across multiple platforms, we have employed large-scale transcriptomic profiling datasets containing a total of 2,316 HCC and 1,665 non-tumorous tissue samples. These samples were obtained from 30 studies generated by mainly four types of profiling techniques (Affymetrix, Illumina, Agilent, and High-throughput sequencing), which are implemented in a wide range of platforms. Firstly, we scrutinized overlapping 26 genes that are differentially expressed in numerous datasets. Subsequently, we identified a panel of three genes (FCN3, CLEC1B, and PRC1) as HCC biomarker using different feature selection techniques. Three-genes-based HCC biomarker identified HCC samples in training/validation datasets with an accuracy between 93 and 98%, Area Under Receiver Operating Characteristic curve (AUROC) in a range of 0.97 to 1.0. A reasonable performance, i.e., AUROC 0.91–0.96 achieved on validation dataset containing peripheral blood mononuclear cells, concurred their non-invasive utility. Furthermore, the prognostic potential of these genes was evaluated on TCGA-LIHC and GSE14520 cohorts using univariate survival analysis. This analysis revealed that these genes are prognostic indicators for various types of the survivals of HCC patients (e.g., Overall Survival, Progression-Free Survival, Disease-Free Survival). These genes significantly stratified high-risk and low-risk HCC patients (p-value <0.05). In conclusion, we identified a universal platform-independent three-genes-based biomarker that can predict HCC patients with high precision and also possess significant prognostic potential. Eventually, we developed a web server HCCpred based on the above study to facilitate scientific community (http://webs.iiitd.edu.in/raghava/hccpred/).

Decoding cancer heterogeneity: studying patient-specific signaling signatures towards personalized cancer therapy

The past years have witnessed a rapid increase in the amount of large-scale tumor datasets. The challenge has now become to find a way to obtain useful information from these masses of data that will allow to determine which combination of FDA-approved drugs is best suited to treat the specific tumor. Various statistical analyses are being developed to extract significant signals from cancer datasets. However, tumors are still being assigned to pre-defined categories (breast luminal A, triple negative, etc.), conceptually contradicting the vast heterogeneity that is known to exist among tumors, and likely overlooking unique tumors that must be addressed and treated individually. We present herein an approach based on information theory that, rather than searches for what makes a tumor similar to other tumors, addresses tumors individually and unbiasedly, and impartially decodes the critical patient-specific molecular network reorganization in every tumor.

Methods: Using a large dataset obtained from ~3500 tumors of 11 types we decipher the altered protein network structure in each tumor, namely the patient-specific signaling signature. Each signature can harbor several altered protein subnetworks. We suggest that simultaneous targeting of central proteins from every altered subnetwork is essential to efficiently disturb the altered signaling in each tumor. We experimentally validate our ability to dissect sample-specific signaling signatures and to rationally design personalized drug combinations.

Results: We unraveled a surprisingly simple order that underlies the extreme apparent complexity of tumor tissues, demonstrating that only 17 altered protein subnetworks characterize ~3500 tumors of 11 types. Each tumor was described by a specific subset of 1-4 subnetworks out of 17, i.e. a tumor-specific altered signaling signature. We show that the majority of tumor-specific signaling signatures are extremely rare, and are shared by only 5 tumors or less, supporting a personalized, comprehensive study of tumors in order to design the optimal combination therapy for every patient. We validate the results by confirming that the processes identified in the 11 original cancer types characterize patients harboring a different cancer type as well. We show experimentally, using different cancer cell lines, that the individualized combination therapies predicted by us achieved higher rates of killing than the clinically prescribed treatments.

Conclusions: We present a new strategy to deal with the inter-tumor heterogeneity and to break down the high complexity of cancer systems into simple, easy to crack, patient-specific signaling signatures that guide the rational design of personalized drug therapies.