A1: In vivo single-cell readout

Methods to profile gene and protein expression of individual cells in a living patient without removing tissue.

Technical risk Moderate

Time horizon 10–20 years

Bucket A · Diagnostics

Bottleneck

Sensor delivery and signal transduction through tissue

Current state of the science

Current single-cell methods are destructive. scRNA-seq, CITE-seq, spatial transcriptomics — all require dissociating tissue or freezing thin sections. The information they produce is rich (tens of thousands of cells profiled in parallel) but it's a snapshot of a sample, not a real-time measurement of a patient.

The closest current approximations to in vivo single-cell readout are intravital microscopy (which lets researchers watch fluorescently-labelled cells in living animals through optical windows) and engineered reporter cells that broadcast their state through detectable signals. Neither scales to clinical use yet.

Technical pathway

Three plausible technical pathways are visible from current research. The first is implantable microsensor arrays — chips small enough to sit between cells and report on local protein or RNA concentrations via wireless transmission. Prototype devices exist for glucose and a handful of metabolites; extending to thousands of targets is the engineering challenge.

The second is engineered reporter cells distributed throughout the body that broadcast their internal state — gene expression, stress markers, damage indicators — via detectable optical or chemical signals. This is essentially using engineered biology as a distributed diagnostic substrate, and it parallels how the protomolecule is imagined to work, minus the autonomy.

The third is advances in non-invasive optical methods that can read cellular state through tissue. Diffuse optical tomography and Raman spectroscopy have shown progress here but are limited by tissue scattering at depth.

What is blocking it

The fundamental challenge is signal-to-noise. Living tissue is dense, optically scattering, and full of background biological signals. Getting clean readouts of individual cells from inside a body — without surgery, without contrast agents that themselves perturb the system, and with the resolution and speed needed for AIHS-grade diagnosis — is a problem that doesn't yet have a clear technical winner among the candidate approaches.

Research ecosystem

Active research communities include intravital microscopy groups, synthetic biology programmes working on biosensor cells, neural interface labs developing implantable sensor arrays, and quantum-sensing groups working on NV-diamond magnetometry for cellular signals. The NIH SPARC program and similar bioelectronic medicine initiatives have funded relevant infrastructure.

Research Questions

Open questions from the field

Detail-level questions a researcher landing on this page would likely ask. Expand each to see the answer.

Why can't existing intravital microscopy techniques scale to AIHS-grade cellular readout?

Intravital microscopy operates through implanted optical windows — cranial windows for brain imaging, abdominal windows for visceral organs, dorsal skin chambers for vasculature. Within those windows, multi-photon imaging extends usable depth to roughly 1mm in scattering tissue, with longer-wavelength three-photon variants pushing slightly further. Beyond that, tissue scattering and absorption attenuate signal exponentially.

For AIHS, which requires routine cellular readout in deep organs (liver, kidney, pancreas, deep brain structures) without surgical window preparation, the optical approach hits a hard physical limit set by the scattering properties of biological tissue. The mean free path for visible-spectrum photons in mammalian soft tissue is on the order of 100 micrometres; even with sophisticated reconstruction, signal collapses past ~1mm.

Photo-toxicity from prolonged imaging — accumulating damage during multi-hour sessions even at "safe" intensities — is a separate constraint that already affects long-duration imaging research. Together these mean A1 cannot be solved by incremental improvements to fluorescence imaging; it needs either a different physical modality (acoustic, magnetic, electrochemical) or an architectural shift to engineered reporter biology distributed in the tissue itself.

What are the candidate signal modalities and what specifically limits each?

Fluorescence imaging is brightest and most molecularly specific but limited to ~1mm in scattering tissue. Raman spectroscopy provides label-free molecular fingerprinting and reaches deeper, but spontaneous Raman signals are orders of magnitude weaker than fluorescence at equivalent specificity. Surface-enhanced Raman (SERS) partially closes the gap when nanoparticle contrast agents are introduced.

Photoacoustic imaging combines optical absorption with ultrasonic detection — light excites absorbers, which emit thermoacoustic waves that ultrasound can detect through centimetres of tissue. Resolution degrades with depth (roughly proportionally to the imaging depth divided by ultrasound frequency). Magnetic resonance with hyperpolarised agents (carbon-13 pyruvate, xenon-129) penetrates the whole body but has poor sensitivity per cell — typically requires aggregates of thousands of cells to detect.

Implanted electronic sensors avoid the depth problem entirely but face power, biocompatibility, density, and biofouling constraints (covered in A3). Engineered reporter cells produce detectable molecules but solving the inverse problem of localising the source from observed signal is itself unresolved.

The realistic AIHS architecture is composite — different modalities for different tissue accessibility regimes — rather than a single technique covering all cellular readout needs.

How would engineered reporter cells actually be deployed, and what limits their useful lifetime in the body?

Reporter cells would typically be derived from the patient (autologous) to avoid immune rejection, engineered ex vivo with the reporter circuit, and reintroduced via routes paralleling those used for CAR-T therapy — intravenous infusion for circulatory or tissue-homing cells, intrathecal delivery for CNS, intratumoural injection for solid tumours, possibly direct organ injection where local readout is needed.

Useful lifetime depends on cell type and immune response. Engineered T cells can persist for months to years; engineered macrophages and dendritic cells turn over faster. The dominant lifetime constraint is acquisition of immune recognition — even autologous cells eventually trigger a response against their engineered transgenic proteins, particularly with strong constitutive reporter expression. Lower-expression inducible systems extend persistence but at the cost of signal sensitivity.

Position tracking is a separate unsolved problem: knowing what a reporter is reporting on requires knowing where it is. Approaches under investigation include co-delivery of MRI-detectable contrast agents (iron oxide nanoparticles, gadolinium chelates), genetically encoded MRI contrast (LRP1, ferritin overexpression), and ultrasound-detectable gas vesicle reporters. None are yet at clinical maturity. Without solved localisation, reporter cells provide population-level state information rather than the spatially-resolved readout AIHS specifies.

Going Deeper

Key papers & programmes

Starting points for a researcher who wants to go beyond this page. Programmes and references relevant to the discussion above, with where each leads.

The BRAIN InitiativeprogramUS programme funding in-vivo neural recording and imaging tools — the closest large-scale effort to the deep-tissue cellular readout A1 requires.
Intravital multiphoton microscopy literaturepaperThe body of work establishing the ~1mm depth ceiling and photo-toxicity limits that bound the optical approach.
Genetically encoded MRI reporters (LRP1, ferritin)resourceResearch line attempting to make cell state readable at depth via MRI rather than light — a candidate route past the scattering limit.
Gas-vesicle ultrasound reporters (Shapiro lab, Caltech)labAcoustic biomolecules that let engineered cells be localised and read by ultrasound, addressing the reporter-cell position problem.

Falsifiable Signposts

What would change our assessment

Raises

A non-optical modality demonstrates routine single-cell-resolution readout in a deep organ (e.g. liver or pancreas) in a living animal. This would break the depth ceiling that currently makes A1 a genuine gap rather than an engineering task.

Raises

Engineered reporter cells with solved in-vivo localisation persist for months without immune clearance. Solving the position-tracking and lifetime problems together would make the reporter-cell route viable.

Lowers

Reporter-cell immunogenicity proves irreducible — every sufficiently expressive reporter construct eventually triggers clearance. This would push A1 toward hardware sensors and away from the biological-reporter route.

Upstream dependencies

No upstream dependencies — this advance can be pursued independently.

This advance unlocks

A2 · Whole-body molecular mapping A5 · Distributed pathogen and damage detection B2 · Real-time epigenomic state inference

Referenced from across the study

Bucket Bucket A — Second, it is the substrate that makes every other diagnostic capability useful: without an integrated state representation, the outputs of A1 , A2 , A3 , and A5 are heterogeneous data streams that must be interpreted manually.
Bucket Bucket B — …in a living patient without destroying samples requires the in-vivo readout capability of A1 — there is no way to infer epigenomic state in vivo without the means to read cellular state in vivo.…
Phase Phase I — A1 (in vivo single-cell readout) remains at proof-of-concept in animal models — Phase I produces the first chronic in vivo cellular readouts, in mice and non-human primates, demonstrating one or more candidate approaches.
Phase Phase II — Bucket A · Diagnostic priorities A1 (in vivo single-cell readout) moves from animal proof-of-concept to first human clinical demonstrations.
Phase Phase III — A1 (in vivo single-cell readout) extends to deep internal tissues.
Topic Atlas — Horizon ⇅ Depends on A1 In vivo single-cell readout Reading cellular state at depth in a living patient.
Topic Glossary — The data is destructive — the cells must be killed to read their state — which is why doing this in vivo remains an unsolved problem (see A1 ).
Topic Integration — B2 (epigenomic state inference) requires A1 and A2 .
Topic Roadmap — In vivo single-cell imaging proofs of concept in animal models ( A1 ).
Advance A2 — …r distributed engineered reporters (a different technology with its own constraints — see A1 ).…
Advance A5 — This is where A5 connects directly to A1 (in vivo cellular readout for localising the source of DAMP signals) and A2 (whole-body mapping for tracking damage propagation).
Advance B2 — The dependency on A1 (in vivo single-cell readout) means B2 cannot move faster than the substrate technology it needs.
Advance C4 — …chievable in principle through the in vivo readout and whole-body mapping capabilities of A1 and A2 , plus advanced connectivity imaging (diffusion tractography and its successors).…
Advance C7 — That kind of targeted, tunable metabolic control is well beyond current capability and depends on the same in-vivo cellular-targeting and readout capabilities ( A1 , C2 ) that the rest of AIHS needs.

Continue

Continue · Bucket A

Whole-body molecular mapping

The diagnostic equivalent of spatial transcriptomics applied to an entire intact patient. Currently impossible at any meaningful resolution.

Back · Bucket A overview

Diagnostics

Return to the bucket overview with all required advances.

Back to hub

AIHS study overview

Return to the main study page.