Cervical spinal cord stimulation disrupts proprioception yet improves voluntary arm reaching

Only the abstract is available for this paper, so everything below is built on that plus context about the methods it names. I'll flag where the missing details matter most — and they matter a lot here, because the whole value of the paper is in effect sizes and sample size that the abstract omits.

The question that won't die

Does your brain need to feel where your arm is in order to move it accurately?

Intuition says obviously yes. Close your eyes, and you can still touch your nose — that's proprioception, the sense of limb position and movement carried by muscle spindles and joint receptors up the dorsal columns. Take it away and people become clumsy, unable to hold a posture without watching the limb.

Yet the classic experiments told a stranger story. Deafferented monkeys (dorsal rhizotomy) could still make rapid, goal-directed reaches. The famous human case of Ian Waterman, who lost proprioception below the neck, relearned to move — slowly, effortfully, under constant visual control. The lesson many drew: for fast, ballistic movements, the brain runs an internal forward model and largely ignores proprioceptive feedback, because the feedback loop is too slow to help within the ~200 ms of a quick reach anyway. Proprioception, on this view, is for slow correction and posture, not for the ballistic core of reaching.

The problem is that every one of these cases is a chronic loss. By the time you test the animal or the patient, they've had months or years to rebuild strategies around the missing sense. So you can never separate two things:

• what proprioception contributes to movement, versus
• what the nervous system learned to do without it.

That confound is why the debate never closed. You'd want to reach in and switch proprioception off for an afternoon, then back on, in the same person — an acute, reversible lesion. Nobody had a clean way to do that. Until, the authors argue, an unrelated clinical tool handed them one.

The accidental scalpel: cervical spinal cord stimulation

Cervical spinal cord stimulation (SCS) is being tested as a therapy to restore arm and hand function after stroke. Electrodes over the cervical dorsal cord/roots deliver current that recruits large sensory afferents — the very fibers that carry proprioception. In the therapeutic framing, that recruitment boosts excitability of weakened motor circuits and helps paretic patients move.

The insight in this paper is to turn that therapeutic tool into an experimental perturbation. If SCS electrically drives proprioceptive afferents, it isn't delivering clean position information — it's injecting a barrage of synchronous, task-irrelevant activity into the proprioceptive channel. That's noise. So flipping SCS ON should acutely degrade proprioception in a healthy-ish channel, within a single session, with an OFF condition as the within-subject control. No years of compensation to confound it.

That reframing — using a restorative stimulator as a reversible sensory jammer — is the clever move, and it's what makes this a causal test rather than another correlational deafferentation case study.

proprioceptive channel + noise

Feedback-dependent proprioceptive sense posture vs. force WORSE Feedforward + vision rapid reach: smooth, straight, accurate BETTER also: more visual error adaptation

What they report

Same population: individuals with chronic post-stroke hemiparesis, tested with SCS ON versus OFF. Four findings, per the abstract:

1. Proprioceptive perception got worse with SCS ON. This is the sanity check that the perturbation does what they claim — it confirms current is corrupting the proprioceptive channel.

2. Postural stabilization against force perturbations got worse. When something pushes on your arm and you have to resist and hold position, you're running a proprioceptive feedback loop. Degrade the feedback and, as expected, stabilization suffers.

3. Rapid, goal-directed reaching got better — smoother, straighter, more spatially accurate — and this held both with and without vision of the arm.

4. Adaptation to visual errors increased during implicit motor learning.

Findings 1 and 2 point the same way (proprioception down). Findings 3 and 4 point the opposite way for behavior. That's the dissociation, and it's the reason to care.

Why this pattern makes mechanistic sense

The result reads as paradoxical only if you think of proprioception as a single knob for "movement quality." Split movement into two regimes and it hangs together.

Rapid reaching is feedforward. A quick reach is largely committed before proprioceptive feedback can meaningfully steer it. If proprioception during a fast reach mostly contributes noise and drag — especially in a stroke-damaged sensorimotor loop where the feedback may be maladaptive — then quieting or scrambling that channel can clean up the ballistic movement. The finding that the improvement survives with vision removed is the strongest part of the story: it means vision isn't simply substituting for the corrupted proprioception. The feedforward plan itself came out better.

Posture is feedback. Holding against a perturbation is exactly the regime where you must sense displacement and push back. Corrupt the sense and you can't. So finding 2 is the expected cost, and it cleanly separates the two control modes rather than muddying them.

The adaptation result is the subtle one. Implicit visuomotor adaptation is often modeled as the brain reweighting sensory error sources. When the felt hand position (proprioception) and the seen cursor disagree, the size of your correction depends on how much you trust each channel. Add noise to proprioception and its reliability drops, so — in a Bayesian-cue-combination view — the brain should up-weight vision and adapt more strongly to visual error. That's precisely the direction reported. It's a satisfying consistency check: the same noise injection predicts both worse posture and stronger visual adaptation.

Put together, the abstract's headline — proprioception is not required for rapid goal-directed action — is well-supported in direction. And crucially it's an acute, reversible manipulation, which is the thing the chronic literature could never deliver.

What to be skeptical about

The abstract gives directions, not magnitudes. That's a real limitation for a paper whose entire contribution is the strength of a causal claim.

Sample size is unstated. SCS-in-stroke trials to date have been tiny — the landmark cervical SCS work involved a handful of participants. If this is n = 2–4, the dissociation is a compelling existence proof but not a population claim. Look for the participant count first; the weight you give everything else depends on it.

Effect sizes and variability are unstated. "Improved smoothness, straightness and accuracy" could be large and consistent or marginal and heterogeneous across a few people. Within-subject ON/OFF designs help, but with SCS the ON condition is not blind in any obvious way — participants often feel stimulation, and experimenters know the state. Watch for blinding, sham conditions, and order/carryover controls.

The population is stroke, not neurotypical. A paretic sensorimotor system may have abnormal, even counterproductive, proprioceptive feedback. It's entirely plausible that scrambling bad feedback helps a stroke arm while scrambling good feedback would hurt a healthy one. So "proprioception is not required for rapid movement" is best read, for now, as "corrupting proprioceptive afferents did not prevent — and coincided with better — rapid reaching in chronic stroke." Generalizing to the intact nervous system needs care.

Mechanism vs. therapy is entangled. SCS was designed to help paretic arms. If ON improves reaching partly through its intended excitability boost rather than purely through proprioceptive disruption, then the causal story about proprioception is confounded by the therapeutic effect. The proprioception-degradation measurements (findings 1–2) argue the channel really is corrupted, but disentangling "helped by excitability" from "helped despite lost proprioception" is the key interpretive risk. The full paper's control analyses are where this lives or dies.

Why it matters if it holds

For motor control, this is the cleanest available causal handle on a decades-old question, and it favors a specific model: fast goal-directed movement rides on feedforward internal models and vision, while proprioception governs posture and slow correction and gates multisensory reweighting for learning. That's a more precise claim than "proprioception matters" or "it doesn't."

For BCI and neurostimulation, there's a design lesson hiding here. We usually think of sensory stimulation as adding information — restoring touch, restoring position sense. This paper is a reminder that stimulating an afferent channel can just as easily inject noise, and that the behavioral consequence depends entirely on which control loop consumes that channel. A stimulation protocol that "helps" reaching could be degrading the very sense it appears to engage. For anyone building closed-loop sensory neuroprostheses, knowing that the same input can be simultaneously disruptive (posture) and beneficial (ballistic reach) is a warning against single-number success metrics.

Where to spend your reading time

Go straight to the Methods for participant count and the ON/OFF (and any sham) design, and to the Results tables for effect sizes with variability on the reaching metrics. Then read the adaptation experiment carefully — the increased visual-error learning is the finding that turns a "noise helps a broken arm" story into a principled statement about sensory reweighting, and it's the part most likely to generalize beyond stroke. The abstract's central dissociation is genuinely interesting; whether it's a robust population result or a striking few-subject demonstration is entirely a matter of the numbers this abstract withholds.