# Karel Svoboda — Neural Dynamics Review
## Overall assessment
**Greenlight: maybe — the question is important, but I want to flag a dynamics gap that could weaken the result if unaddressed.**
The proposal asks whether static anatomical connectivity predicts degenerative vulnerability. That is an important question, and the Allen Connectivity Atlas is the right tool to anchor it. But as someone who thinks about neural dynamics, I want to push on whether anatomy alone is sufficient, or whether you need to know something about activity patterns.
## The dynamics concern
Here is the core issue: connectivity is a structural scaffold, but **vulnerability may be driven by activity, not anatomy**. A circuit with high convergent input that is chronically active may be vulnerable because of metabolic load, calcium stress, or synaptic fatigue — not because of the wiring per se. A circuit with the same convergent input that is mostly silent may be fine.
This means:
1. If you find a connectivity-stress correlation, it could be confounded by activity differences.
2. If you don't find it, it could be because connectivity alone doesn't capture the relevant variable (activity).
### My suggested mitigation
You don't need to solve this in the pilot, but you should **measure basal activity** in at least a subset of the predicted-vulnerable vs. predicted-resilient circuits. Two practical options:
- **c-Fos immunohistochemistry** at baseline (no stimulation) in 5xFAD vs. wild-type at 2 and 3 months. This gives you a coarse but scalable readout of which circuits are hyperactive.
- **Calcium imaging** in a small cohort using head-mounted miniscopes (e.g., Inscopix) in retrosplenial cortex (predicted vulnerable) vs. primary visual cortex (predicted resilient) during free behavior.
If the connectivity-stress correlation holds after controlling for basal activity, the result is much stronger. If it disappears, you've learned something equally important: that activity, not anatomy, drives vulnerability.
## What I like
- The CCF alignment is correct and essential — I agree with Hongkui on this.
- The INTACT approach for nuclear isolation is well-validated and compatible with the Cre lines.
- The proposal to use three connectivity metrics (convergent input, input diversity, reciprocity) is smart because it decomposes what aspect of connectivity matters.
- The plaque-distance control is critical and well-designed.
## Practical feasibility notes
- **Timeline**: 3 months is tight for 72 snRNA-seq libraries + MERFISH. I would suggest 4 months or dropping MERFISH to a planned follow-on.
- **Cell-type resolution**: The two-Cre-line panel (Tlx3 + Pvalb) is sensible. If you add more lines, do it in a second phase.
- **Sample size**: n=4 per genotype per timepoint is the minimum for detecting moderate effects. Consider n=5 to guard against attrition in the 5xFAD cohort.
## Feasibility score
I would score this **0.55 to 0.6 feasible** for a 3-month pilot — achievable but ambitious. The main risk is not the biology but the throughput: snRNA-seq library prep and sequencing for 72 samples in 3 months requires dedicated technician time and core facility scheduling.
## Bottom line
The question is worth asking, and the Allen resources make it uniquely possible to ask it. But I want to see basal activity measurements added so that the connectivity-prediction claim can be distinguished from an activity-prediction claim. If you add c-Fos at minimum, I'd move to a full greenlight.