Automated image analysis of NSCLC biopsies to predict response to anti-PD-L1 therapy.

BACKGROUND: Immune checkpoint therapies (ICTs) targeting the programmed cell death-1 (PD1)/programmed cell death ligand-1 (PD-L1) pathway have improved outcomes for patients with non-small cell lung cancer (NSCLC), particularly those with high PD-L1 expression. However, the predictive value of manual PD-L1 scoring is imperfect and alternative measures are needed. We report an automated image analysis solution to determine the predictive and prognostic values of the product of PD-L1+ cell and CD8+ tumor infiltrating lymphocyte (TIL) densities (CD8xPD-L1 signature) in baseline tumor biopsies. METHODS: Archival or fresh tumor biopsies were analyzed for PD-L1 and CD8 expression by immunohistochemistry. Samples were collected from 163 patients in Study 1108/NCT01693562, a Phase 1/2 trial to evaluate durvalumab across multiple tumor types, including NSCLC, and a separate cohort of 199 non-ICT- patients. Digital images were automatically scored for PD-L1+ and CD8+ cell densities using customized algorithms applied with Developer XD™ 2.7 software. RESULTS: For patients who received durvalumab, median overall survival (OS) was 21.0 months for CD8xPD-L1 signature-positive patients and 7.8 months for signature-negative patients (p = 0.00002). The CD8xPD-L1 signature provided greater stratification of OS than high densities of CD8+ cells, high densities of PD-L1+ cells, or manually assessed tumor cell PD-L1 expression ≥25%. The CD8xPD-L1 signature did not stratify OS in non-ICT patients, although a high density of CD8+ cells was associated with higher median OS (high: 67 months; low: 39.5 months, p = 0.0009) in this group. CONCLUSIONS: An automated CD8xPD-L1 signature may help to identify NSCLC patients with improved response to durvalumab therapy. Our data also support the prognostic value of CD8+ TILS in NSCLC patients who do not receive ICT. TRIAL REGISTRATION: ClinicalTrials.gov identifier: NCT01693562 . Study code: CD-ON-MEDI4736-1108. Interventional study (ongoing but not currently recruiting). Actual study start date: August 29, 2012. Primary completion date: June 23, 2017 (final data collection date for primary outcome measure).

Nucleosome-driven transcription factor binding and gene regulation.

Elucidating the global function of a transcription factor implies the identification of its target genes and genomic binding sites. The role of chromatin in this context is unclear, but the dominant view is that factors bind preferentially to nucleosome-depleted regions identified as DNaseI-hypersensitive sites (DHS). Here we show by ChIP, MNase, and DNaseI assays followed by deep sequencing that the progesterone receptor (PR) requires nucleosomes for optimal binding and function. In breast cancer cells treated with progestins, we identified 25,000 PR binding sites (PRbs). The majority of these sites encompassed several copies of the hexanucleotide TGTYCY, which is highly abundant in the genome. We found that functional PRbs accumulate around progesterone-induced genes, mainly in enhancers. Most of these sites overlap with DHS but exhibit high nucleosome occupancy. Progestin stimulation results in remodeling of these nucleosomes with displacement of histones H1 and H2A/H2B dimers. Our results strongly suggest that nucleosomes are crucial for PR binding and hormonal gene regulation.

Predictive models of gene regulation from high-throughput epigenomics data.

The epigenetic regulation of gene expression involves multiple factors. The synergistic or antagonistic action of these factors has suggested the existence of an epigenetic code for gene regulation. Highthroughput sequencing (HTS) provides an opportunity to explore this code and to build quantitative models of gene regulation based on epigenetic differences between specific cellular conditions. We describe a new computational framework that facilitates the systematic integration of HTS epigenetic data. Our method relates epigenetic signals to expression by comparing two conditions. We show its effectiveness by building a model that predicts with high accuracy significant expression differences between two cell lines, using epigenetic data from the ENCODE project. Our analyses provide evidence for a degenerate epigenetic code, which involves multiple genic regions. In particular, signal changes at the 1st exon, 1st intron, and downstream of the polyadenylation site are found to associate strongly with expression regulation. Our analyses also show a different epigenetic code for intron-less and intron-containing genes. Our work provides a general methodology to do integrative analysis of epigenetic differences between cellular conditions that can be applied to other studies, like cell differentiation or carcinogenesis.

Pyicos: a versatile toolkit for the analysis of high-throughput sequencing data.

MOTIVATION: High-throughput sequencing (HTS) has revolutionized gene regulation studies and is now fundamental for the detection of protein-DNA and protein-RNA binding, as well as for measuring RNA expression. With increasing variety and sequencing depth of HTS datasets, the need for more flexible and memory-efficient tools to analyse them is growing. RESULTS: We describe Pyicos, a powerful toolkit for the analysis of mapped reads from diverse HTS experiments: ChIP-Seq, either punctuated or broad signals, CLIP-Seq and RNA-Seq. We prove the effectiveness of Pyicos to select for significant signals and show that its accuracy is comparable and sometimes superior to that of methods specifically designed for each particular type of experiment. Pyicos facilitates the analysis of a variety of HTS datatypes through its flexibility and memory efficiency, providing a useful framework for data integration into models of regulatory genomics. AVAILABILITY: Open-source software, with tutorials and protocol files, is available at http://regulatorygenomics.upf.edu/pyicos or as a Galaxy server at http://regulatorygenomics.upf.edu/galaxy CONTACT: eduardo.eyras@upf.edu SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Structural constraints revealed in consistent nucleosome positions in the genome of S. cerevisiae.

BACKGROUND: Recent advances in the field of high-throughput genomics have rendered possible the performance of genome-scale studies to define the nucleosomal landscapes of eukaryote genomes. Such analyses are aimed towards providing a better understanding of the process of nucleosome positioning, for which several models have been suggested. Nevertheless, questions regarding the sequence constraints of nucleosomal DNA and how they may have been shaped through evolution remain open. In this paper, we analyze in detail different experimental nucleosome datasets with the aim of providing a hypothesis for the emergence of nucleosome-forming sequences. RESULTS: We compared the complete sets of nucleosome positions for the budding yeast (Saccharomyces cerevisiae) as defined in the output of two independent experiments with the use of two different experimental techniques. We found that < 10% of the experimentally defined nucleosome positions were consistently positioned in both datasets. This subset of well-positioned nucleosomes, when compared with the bulk, was shown to have particular properties at both sequence and structural levels. Consistently positioned nucleosomes were also shown to occur preferentially in pairs of dinucleosomes, and to be surprisingly less conserved compared with their adjacent nucleosome-free linkers. CONCLUSION: Our findings may be combined into a hypothesis for the emergence of a weak nucleosome-positioning code. According to this hypothesis, consistent nucleosomes may be partly guided by nearby nucleosome-free regions through statistical positioning. Once established, a set of well-positioned consistent nucleosomes may impose secondary constraints that further shape the structure of the underlying DNA. We were able to capture these constraints through the application of a recently introduced structural property that is related to the symmetry of DNA curvature. Furthermore, we found that both consistently positioned nucleosomes and their adjacent nucleosome-free regions show an increased tendency towards conservation of this structural feature.

Nucleosome positioning as a determinant of exon recognition.

Chromatin structure influences transcription, but its role in subsequent RNA processing is unclear. Here we present analyses of high-throughput data that imply a relationship between nucleosome positioning and exon definition. First, we have found stable nucleosome occupancy within human and Caenorhabditis elegans exons that is stronger in exons with weak splice sites. Conversely, we have found that pseudoexons--intronic sequences that are not included in mRNAs but are flanked by strong splice sites--show nucleosome depletion. Second, the ratio between nucleosome occupancy within and upstream from the exons correlates with exon-inclusion levels. Third, nucleosomes are positioned central to exons rather than proximal to splice sites. These exonic nucleosomal patterns are also observed in non-expressed genes, suggesting that nucleosome marking of exons exists in the absence of transcription. Our analysis provides a framework that contributes to the understanding of splicing on the basis of chromatin architecture.

An evolutionary and structural characterization of mammalian protein complex organization.

BACKGROUND: We have recently released a comprehensive, manually curated database of mammalian protein complexes called CORUM. Combining CORUM with other resources, we assembled a dataset of over 2700 mammalian complexes. The availability of a rich information resource allows us to search for organizational properties concerning these complexes. RESULTS: As the complexity of a protein complex in terms of the number of unique subunits increases, we observed that the number of such complexes and the mean non-synonymous to synonymous substitution ratio of associated genes tend to decrease. Similarly, as the number of different complexes a given protein participates in increases, the number of such proteins and the substitution ratio of the associated gene also tends to decrease. These observations provide evidence relating natural selection and the organization of mammalian complexes. We also observed greater homogeneity in terms of predicted protein isoelectric points, secondary structure and substitution ratio in annotated versus randomly generated complexes. A large proportion of the protein content and interactions in the complexes could be predicted from known binary protein-protein and domain-domain interactions. In particular, we found that large proteins interact preferentially with much smaller proteins. CONCLUSION: We observed similar trends in yeast and other data. Our results support the existence of conserved relations associated with the mammalian protein complexes.