Longitudinal machine learning uncouples healthy aging factors from chronic disease risks.

To understand human longevity, inherent aging processes must be distinguished from known etiologies leading to age-related chronic diseases. Such deconvolution is difficult to achieve because it requires tracking patients throughout their entire lives. Here, we used machine learning to infer health trajectories over the entire adulthood age range using extrapolation from electronic medical records with partial longitudinal coverage. Using this approach, our model tracked the state of patients who were healthy and free from known chronic disease risk and distinguished individuals with higher or lower longevity potential using a multivariate score. We showed that the model and the markers it uses performed consistently on data from Israeli, British and US populations. For example, mildly low neutrophil counts and alkaline phosphatase levels serve as early indicators of healthy aging that are independent of risk for major chronic diseases. We characterize the heritability and genetic associations of our longevity score and demonstrate at least 1 year of extended lifespan for parents of high-scoring patients compared to matched controls. Longitudinal modeling of healthy individuals is thereby established as a tool for understanding healthy aging and longevity.

Tumor-reactive antibodies evolve from non-binding and autoreactive precursors.

The tumor microenvironment hosts antibody-secreting cells (ASCs) associated with a favorable prognosis in several types of cancer. Patient-derived antibodies have diagnostic and therapeutic potential; yet, it remains unclear how antibodies gain autoreactivity and target tumors. Here, we found that somatic hypermutations (SHMs) promote antibody antitumor reactivity against surface autoantigens in high-grade serous ovarian carcinoma (HGSOC). Patient-derived tumor cells were frequently coated with IgGs. Intratumoral ASCs in HGSOC were both mutated and clonally expanded and produced tumor-reactive antibodies that targeted MMP14, which is abundantly expressed on the tumor cell surface. The reversion of monoclonal antibodies to their germline configuration revealed two types of classes: one dependent on SHMs for tumor binding and a second with germline-encoded autoreactivity. Thus, tumor-reactive autoantibodies are either naturally occurring or evolve through an antigen-driven selection process. These findings highlight the origin and potential applicability of autoantibodies directed at surface antigens for tumor targeting in cancer patients.

Personalized lab test models to quantify disease potentials in healthy individuals.

Standardized lab tests are central for patient evaluation, differential diagnosis and treatment. Interpretation of these data is nevertheless lacking quantitative and personalized metrics. Here we report on the modeling of 2.1 billion lab measurements of 92 different lab tests from 2.8 million adults over a span of 18 years. Following unsupervised filtering of 131 chronic conditions and 5,223 drug-test pairs we performed a virtual survey of lab tests distributions in healthy individuals. Age and sex alone explain less than 10% of the within-normal test variance in 89 out of 92 tests. Personalized models based on patients' history explain 60% of the variance for 17 tests and over 36% for half of the tests. This allows for systematic stratification of the risk for future abnormal test levels and subsequent emerging disease. Multivariate modeling of within-normal lab tests can be readily implemented as a basis for quantitative patient evaluation.

Dynamic and static maintenance of epigenetic memory in pluripotent and somatic cells.

Stable maintenance of gene regulatory programs is essential for normal function in multicellular organisms. Epigenetic mechanisms, and DNA methylation in particular, are hypothesized to facilitate such maintenance by creating cellular memory that can be written during embryonic development and then guide cell-type-specific gene expression. Here we develop new methods for quantitative inference of DNA methylation turnover rates, and show that human embryonic stem cells preserve their epigenetic state by balancing antagonistic processes that add and remove methylation marks rather than by copying epigenetic information from mother to daughter cells. In contrast, somatic cells transmit considerable epigenetic information to progenies. Paradoxically, the persistence of the somatic epigenome makes it more vulnerable to noise, since random epimutations can accumulate to massively perturb the epigenomic ground state. The rate of epigenetic perturbation depends on the genomic context, and, in particular, DNA methylation loss is coupled to late DNA replication dynamics. Epigenetic perturbation is not observed in the pluripotent state, because the rapid turnover-based equilibrium continuously reinforces the canonical state. This dynamic epigenetic equilibrium also explains how the epigenome can be reprogrammed quickly and to near perfection after induced pluripotency.

Epigenetic polymorphism and the stochastic formation of differentially methylated regions in normal and cancerous tissues.

DNA methylation has been comprehensively profiled in normal and cancer cells, but the dynamics that form, maintain and reprogram differentially methylated regions remain enigmatic. Here, we show that methylation patterns within populations of cells from individual somatic tissues are heterogeneous and polymorphic. Using in vitro evolution of immortalized fibroblasts for over 300 generations, we track the dynamics of polymorphic methylation at regions developing significant differential methylation on average. The data indicate that changes in population-averaged methylation occur through a stochastic process that generates a stream of local and uncorrelated methylation aberrations. Despite the stochastic nature of the process, nearly deterministic epigenetic remodeling emerges on average at loci that lose or gain resistance to methylation accumulation. Changes in the susceptibility to methylation accumulation are correlated with changes in histone modification and CTCF occupancy. Characterizing epigenomic polymorphism within cell populations is therefore critical to understanding methylation dynamics in normal and cancer cells.

Primate CpG islands are maintained by heterogeneous evolutionary regimes involving minimal selection.

Mammalian CpG islands are key epigenomic elements that were first characterized experimentally as genomic fractions with low levels of DNA methylation. Currently, CpG islands are defined based on their genomic sequences alone. Here, we develop evolutionary models to show that several distinct evolutionary processes generate and maintain CpG islands. One central evolutionary regime resulting in enriched CpG content is driven by low levels of DNA methylation and consequentially low rates of CpG deamination. Another major force forming CpG islands is biased gene conversion that stabilizes constitutively methylated CpG islands by balancing rapid deamination with CpG fixation. Importantly, evolutionary analysis and population genetics data suggest that selection for high CpG content is not a significant factor contributing to conservation of CpGs in differentially methylated regions. The heterogeneous, but not selective, origins of CpG islands have direct implications for the understanding of DNA methylation patterns in healthy and diseased cells.

DNA methylation programming and reprogramming in primate embryonic stem cells.

DNA methylation is an important epigenetic mechanism, affecting normal development and playing a key role in reprogramming epigenomes during stem cell derivation. Here we report on DNA methylation patterns in native monkey embryonic stem cells (ESCs), fibroblasts, and ESCs generated through somatic cell nuclear transfer (SCNT), identifying and comparing epigenome programming and reprogramming. We characterize hundreds of regions that are hyper- or hypomethylated in fibroblasts compared to native ESCs and show that these are conserved in human cells and tissues. Remarkably, the vast majority of these regions are reprogrammed in SCNT ESCs, leading to almost perfect correlation between the epigenomic profiles of the native and reprogrammed lines. At least 58% of these changes are correlated in cis to transcription changes, Polycomb Repressive Complex-2 occupancy, or binding by the CTCF insulator. We also show that while epigenomic reprogramming is extensive and globally accurate, the efficiency of adding and stripping DNA methylation during reprogramming is regionally variable. In several cases, this variability results in regions that remain methylated in a fibroblast-like pattern even after reprogramming.