Skip to content

GEMINI — a genetically encoded protein-assembly molecular recorder

Speaker: Dingchang Lin (Johns Hopkins University) — hiring PhDs + postdocs; collaboration with the David Baker Lab (University of Washington, Institute for Protein Design) Talk title: "Genetically encoded assembly recorder temporally resolves cellular history" Slot: SCGD26, Jun 12, 2026 — first talk after the host's opening (10:40–11:00 AM EDT) Method: GEMINI ("Granularly Expanding Memory for Intracellular Narrative Integration") Recording: SCGD26 livestream (YouTube) — verify archived link Status: Captured from session transcript

Thesis

A recorder smaller than a "TrueRing"-style device that fits inside a cell, is genetically encoded (so it can be expressed in arbitrarily many cells), and self-assembles into a protein particle whose growth predictably encodes the cell's signaling history for later optical read-out. Analog, not digital: the intensity of each band carries signal amplitude, and band position carries timing.

How it works

Three subunit types co-assemble into a particle (shown growing as a hexagonal structure in live cells):

  1. Blank subunit — drives constant, predictable particle growth (the "ruler" for later temporal decoding).
  2. Reporter subunit — carries a fluorophore; its expression/incorporation is driven by a cell signal (e.g. an NF-κB promoter driving GFP-tagged subunit), so a hyperactive cell deposits a bright band.

  3. Timestamp subunit — incorporation is gated by an external stimulus, correlating wall-clock time to band position.

  4. The particle is a computationally designed self-assembling protein particle (Baker Lab collaboration); the best of many screened variants grows with near-perfect predictability, fit by a simple mathematical growth model.

  5. Timestamping uses HaloTag chemistry — switching small-molecule dye colors creates a color-transition band marking a known time; an earlier linear recorder required fibers to lie parallel to the imaging plane, whereas the isotropic particle reads out identically at any orientation (enables high-throughput tissue read-out).

Key results

  • In vitro decoding: two timestamps 11 h apart bracket a signal; decoded onset shifts with addition time. Temporal accuracy and precision both sub-hour.
  • NF-κB dynamics: decoded TNF-α-driven activation (≈2 h delay vs. timestamp, from transcription/translation lag) and deactivation (TNF-α washout); resolved multiple on/off cycles as distinct bands in single particles.
  • In vivo — tumor xenograft: clonal line injected into immunodeficient mice; LPS-induced systemic inflammation read out via tumor particles; dose-dependent GFP; co-stained CD31 vasculature → higher vascular density correlated with earlier NF-κB peak time and higher intensity. Blood-brain-barrier-permeable dye enabled systemic in vivo timestamping (retro-orbital injection); LPS injected at different days produced cleanly separable onset bands across a 48 h window.
  • Native brain: AAV-delivered particles expressed in cortex/hippocampus with minimal impact on neuronal density, GFAP, calcium dynamics (running-wheel + GCaMP), or behavior (open field, short-term memory). Cellular-resolution pipeline registers each particle to its neuron and scales to whole-brain. Recorded transcription history (tet-on / doxycycline timing) and seizure-induced neural activity; in-brain precision is lower than in vitro/xenograft, attributed to heterogeneity from local AAV injection.

Connections to the corpus

  • Recording-based lineage/temporal tracing — sibling to the DNA/RNA "molecular tape recorder" thread; see Kevin Chao — TimeVault (RNA time-capsule) and Amanda Urke — SynapseSeq (barcode recording of connectivity).
  • Designed proteins — Baker-Lab computationally designed assembly intersects the protein-design axis (cf. EuroBioC 2025 Ferruz keynote on protein design with language models).
  • AACR axis: single-cell + spatial (in situ temporal read-out).

Sources