B2: Real-time epigenomic state inference

The epigenome has several distinct layers, each measured by different techniques and contributing different information to cellular identity and state.

DNA methylation — primarily 5-methylcytosine at CpG dinucleotides, with rarer modifications including 5-hydroxymethylcytosine, 6-methyladenine, and others. Methylation is the most stable layer (turnover times of days to weeks), controls long-term gene accessibility, and is measured by bisulfite sequencing or enzymatic alternatives. The most operationally accessible epigenomic mark.

Histone modifications — more than 100 known variants covering acetylation, methylation, phosphorylation, ubiquitination, and others, each at specific residues of specific histone variants (H3K4me3, H3K27ac, H3K9me3, H3K27me3, and many others). These modifications combinatorially regulate chromatin compaction and the recruitment of transcription machinery. Each is measured by chromatin immunoprecipitation (ChIP-seq) with antibodies specific to the modification.

Three-dimensional chromosome organisation — topologically associating domains (TADs) of ~100kb-1Mb, enhancer-promoter contacts, lamina-associated domains. Measured by Hi-C and its variants. Determines which regulatory elements can act on which genes physically. Slowest to change and currently inaccessible in living tissue.

Chromatin accessibility — which regions are physically available for transcription factor binding. Measured by ATAC-seq and DNase-seq. The most directly readable layer in terms of correlating with active gene expression.

For AIHS interpretation, methylation patterns are most useful in the near term: relatively stable, accessible via liquid biopsy (cell-free DNA carries methylation signatures of source cells), and correlating well with cell identity and disease state. Histone modification readouts in vivo are harder but provide more dynamic information about cellular response. Three-dimensional organisation is currently inaccessible in living tissue and remains a longer-horizon problem.

Cell-free DNA (cfDNA) consists of short fragments of DNA released into the bloodstream as cells die through normal turnover, apoptosis, necrosis, or pathological processes. The fragments carry methylation marks from their cell of origin, and the fragmentation pattern itself encodes information about the chromatin state of the source cell.

The mechanism works because methylation patterns are remarkably tissue-specific. Hepatocytes, cardiomyocytes, T cells, and tumour cells each have distinctive methylation signatures at thousands of CpG sites. Computational deconvolution methods identify the tissue (and increasingly the cell type) of origin from these signatures by matching against reference atlases.

Cancer detection is the leading application. The Galleri test (GRAIL) uses targeted methylation panels to detect tumour-specific signatures in cfDNA at sensitivity sufficient for early-stage cancers — the test has clinical use for multi-cancer early detection in patients at elevated risk. Similar approaches are in development for autoimmune disease activity, transplant rejection (donor-derived cfDNA), and progression monitoring of chronic diseases.

Advantages: completely non-invasive (single blood draw), whole-body coverage indirectly, no surgical or imaging infrastructure required, scales to population screening.

Limitations: samples only cells that have released DNA recently, biasing toward dying cells; spatial resolution is poor (you know the tissue, not the location within it); sensitivity depends on the fraction of cfDNA from the cell type of interest, which can be very low for early disease; cannot detect changes in cells that aren't releasing cfDNA.

As a Phase II AIHS capability, cfDNA methylation is the most realistic deployment. The Phase III roadmap target extends this to broader applications. Direct in vivo readout remains a longer-horizon goal.

Direct readout means measuring chromatin state in living cells without removing them from the body. This requires either implantable sensors that detect chromatin modifications in nearby cells, or engineered reporter cells that translate chromatin state into externally-readable signals. Both approaches are immature.

Candidate sensor designs include:

CRISPR-based readers using catalytically dead Cas9 (dCas9) fused to chromatin-modification-binding domains (chromodomains, bromodomains, MBD domains for methylation). When dCas9 brings the reader domain to a specific genomic location, binding to the local chromatin mark produces a detectable signal (typically through proximity-induced fluorescence complementation or split reporter systems).

Engineered transcription factors whose nuclear localisation or transcriptional activity depends on local chromatin state. Synthetic systems based on the LacI-LacO bacterial regulatory pair, repurposed with chromatin-binding domains, can produce reporter expression conditional on chromatin context.

Synthetic biology circuits integrating multiple chromatin signals into reporter output — for instance, AND-gate constructs that fluoresce only when both H3K27me3 and H3K9me3 are present at a target locus, indicating heterochromatic silencing.

Each of these is technology-demonstrated in cell culture but none has been deployed at scale in living animals, let alone humans.

The soonest plausible timeline for clinical deployment of any direct epigenomic readout: 20-25 years for at least some accessible tissues. The dependency on A1 (in vivo single-cell readout) means B2 cannot move faster than the substrate technology it needs.

Until then, indirect inference from transcriptome data (which is the downstream consequence of chromatin state) and cfDNA epigenomic profiling will be the operational substitutes. The Phase III roadmap target is cfDNA-based epigenomic monitoring at scale, with direct in vivo readout deferred to Phase IV.