great question. let’s make “targeting a circuit” concrete, then zoom into what ultrasound physically does to neurons.

#The playbook: how you “target a circuit”

Think “node → projection → effect,” with checks at each step.

1. Pick the circuit node you can actually reach

   • Use MRI (anatomy) + fMRI (function) and often prior causal work to choose a small nucleus or cortical patch that sits on the circuit you care about (e.g., nucleus accumbens for reward/craving; anterior cingulate for pain; central thalamus for arousal). The node is your acoustic bullseye; the circuit is the set of downstream regions you expect to modulate via that node. Human papers that do this include ACC→pain, NAcc→reward learning, thalamus→arousal. (ScienceDirect )

2. Plan the beam through the skull

   • Build a skull model (typically from CT; sometimes MR-based surrogates) that estimates how bone bends and dims the wave at each point. Then a phased array computes per-element phase delays so all waves arrive in-phase at the 3D target (mm-scale focus). This is standard in MR-guided arrays; accuracy improves when you verify the focus in-situ with MRI ARFI (it literally “sees” a tiny displacement at the focus). (ScienceDirect )

3. Choose a dose “shape” for modulation (not ablation)

   • Set frequency (often ~500–700 kHz for skull), pulse duration (e.g., 0.2–10 ms tone bursts), pulse repetition frequency (PRF) (tens–thousands of Hz), and duty cycle (often ≪10% to keep heating negligible). These knobs bias toward excitation or inhibition and affect tolerability (auditory clicks rise with certain PRFs). Studies report mm-scale targeting with negligible heating under neuromodulation settings.

4. Deliver + verify “circuit engagement,” not just “hit the spot”

   • While sonication runs, measure on-target effects: immediate physiology (perfusion, EEG/fMRI connectivity), and behavior (e.g., reduced pain unpleasantness, altered reward-guided choice). You also mask or control for auditory confounds (bone-conducted clicks can mimic effects if you don’t control them). (ScienceDirect , eLife )

5. Adapt per individual (closed loop)

   • Adjust focus and pulse train using ARFI/perfusion/EEG/fMRI feedback until you get the desired signature (e.g., ACC→reduced pain unpleasantness) at safe exposure. This is where “personalization” lives.

#What ultrasound does to neurons (the biophysics, in plain English)

Ultrasound is mechanical energy (tiny, rapid pressure changes). In the low-intensity, pulsed regimes used for neuromodulation, the leading candidates are non-thermal, mechanical mechanisms that tweak a cell’s excitability:

1. It tugs on membranes and proteins → opens mechanosensitive channels

   • Mechanical forces from the wave (via radiation force and micro-scale membrane deformation) can gate mechanosensitive ion channels (e.g., Piezo/TRP/TREK), letting ions—especially Ca²⁺—trickle in. That calcium entry and downstream amplification (via voltage-gated channels) can push neurons toward firing (“excitation”) or, in some patterns, bias inhibition via particular interneurons. Cultured-neuron and slice work supports a primarily mechanical, channel-mediated path without needing heat or cavitation. (Nature , PMC )

2. It modulates membrane capacitance via “intramembrane cavitation” (theory)

   • The NICE/NBLS hypothesis says nanoscopic gas pockets in the lipid bilayer oscillate with ultrasound, changing membrane capacitance at microsecond timescales. Those capacitance oscillations translate into small currents that shift excitability. The theory explains why pulse structure (burst length, PRF) matters and even predicts cell-type selectivity (e.g., short pulses bias low-threshold spiking interneurons, yielding net network inhibition). This is a well-cited mechanistic model, though still debated and hard to measure directly in vivo. (Physical Review Links , PubMed , ResearchGate )

3. It creates tiny pushes in tissue (“radiation force”) that cells feel

   • Even without bubbles, ultrasound exerts a steady microscopic push during each burst. MRI ARFI detects the resulting micrometer-scale displacements—useful for aiming—and those same forces likely strain membranes and the cytoskeleton, modulating channels and synapses.

4. What it’s not (at neuromod doses):

   • Not heating-driven: parameters stay near diagnostic ultrasound limits (very low ISPTA/duty cycle), and studies measure negligible temperature rise in neuromod paradigms.
   • Not inertial cavitation: that’s a damaging, bubble-collapse regime; neuromod avoids it (no microbubbles, low mechanical index). (PMC )

5. Network-level result: “bias the node, ripple the circuit”

   • At the node you targeted (say, NAcc), ultrasound nudges excitability up or down for hundreds of milliseconds to seconds. Through that node’s projections, downstream regions change their firing probability/synchrony, which you can see as shifts in perfusion, connectivity, or behavior (e.g., altered stay/switch choice; reduced pain unpleasantness; vigilance changes after thalamic TUS). (ScienceDirect )

#Why parameters matter (and how “excite vs inhibit” happens)

• Frequency (f): lower (≈500–700 kHz) penetrates skull better; higher shrinks the focal spot but attenuates/heats more.
• Burst length & PRF: short bursts at specific PRFs can favor different channel kinetics or recruit different cell types; theoretical/experimental work shows regimes that excite pyramidal cells vs recruit inhibitory LTS interneurons → net inhibition.
• Duty cycle & intensity: keep duty low to limit heating; intensity indexes (MI, ISPTA) bound safety.
• Excite vs inhibit: depends on the above plus state (ongoing activity), orientation of processes, and cell-type expression of mechanosensitive channels. (This is why closed-loop readouts are valuable.) (PubMed , PMC )

#Important caveat: auditory confounds

Pulsed ultrasound can vibrate skull → bone-conducted sound at the PRF. If you don’t mask or design the PRF thoughtfully, subjects (and animals) can show “effects” that are actually auditory pathway responses, not direct local neuromodulation. Modern studies use masking, sham control, and parameter tweaks to mitigate this and still see site-specific changes. (ScienceDirect , eLife )

#Bottom line (one-liner)

You target a circuit by precisely sonifying a small hub region on that circuit (array beamforming + skull correction + in-situ verification), and the ultrasound’s mechanical micromotions modulate mechanosensitive channels/membrane properties, shifting that node’s excitability so its downstream network changes in predictable ways—confirmed by imaging/behavior readouts. (Nature )

If you want, I can sketch a parameter→effect cheat sheet (typical f, PRF, burst, duty; excite vs inhibit tendencies) and a one-page pipeline for running a circuit-targeted human session (plan → simulate → deliver → verify → adapt), with the key citations next to each step.