Spinal Cord Stimulation: Mechanisms of Action

Evidence status. Mechanisms of tonic, high-frequency, burst, and multiplexed stimulation remain incompletely resolved. Clinical efficacy does not prove a proposed cellular mechanism, and much waveform literature is industry sponsored. Clinical evidence varies by diagnosis, waveform, comparator, and follow-up. Active-comparator trials often favor newer systems, while sham-controlled evidence is limited and mixed; durability and revision or explant burden belong alongside responder rates. See Part IV.

1.Gate Control, and Why the Dorsal Columns

Every account of SCS begins with the gate-control theory of Melzack and Wall, and it is worth stating what they actually proposed, because it is not what we now teach. In the 1965 model the gating element is the substantia gelatinosa cell: large-diameter input excites it, small-diameter input inhibits it, and its output is presynaptic inhibition of the transmission (T) cell that projects to the brain. Increase large-fiber traffic and the gate closes.

Much of that specific circuit did not survive. The substantia gelatinosa turned out to be dominated by excitatory interneurons, the proposed polarity of small-fiber input onto them was wrong, and Wall revised the model himself in 1978. What survived is the conceptual claim: the dorsal horn is a modulated gate rather than a simple relay, and that large-fiber input can suppress nociceptive transmission. Modern accounts put the integrating cell at the wide-dynamic-range (WDR) projection neuron, which sums a wide range of inputs (A-beta touch and pressure, A-delta sharp pain, and C-fiber pain and temperature) and reports stimulus intensity to the brain, and they put the inhibition in GABAergic and glycinergic interneurons acting on that neuron both presynaptically and postsynaptically, with descending brainstem control layered on top. In animal models of neuropathic pain the WDR neuron is sensitized and over-active, and suppressing that hyperexcitability is the leading candidate mechanism for segmental analgesia. Hold gate control as the right idea with the wrong wiring diagram.

The subtlety most worth teaching is that the therapy is indirect. Conventional dorsal epidural stimulation preferentially recruits large myelinated dorsal-root and dorsal-column fibers rather than directly activating small nociceptive afferents at clinical settings, and the resulting volley splits two ways. One branch ascends to sensory cortex and is felt as the paresthesia, which is a marker of recruited sensory pathways but is neither necessary nor sufficient for analgesia. The other branch travels antidromically back down the dorsal-column axon into its segmental collaterals in the dorsal horn. There, in rodent nerve-injury models, stimulation increases GABA release, reduces extracellular glutamate and aspartate, suppresses WDR hyperexcitability, and recruits descending serotonergic, noradrenergic, and cholinergic control from the brainstem; intrathecal baclofen converts some non-responding animals into responders, supporting a GABAergic contribution in that experimental setting, and adenosine and endocannabinoid contributions have also been reported. State the limit plainly: this is rat pharmacology. Nobody has demonstrated GABA release in a human dorsal horn during SCS, and the inhibition involved is probably presynaptic and shunting as much as it is frank hyperpolarization. The relationship among paresthesia, segmental inhibition, and descending modulation is therefore indirect and state dependent, which helps explain why paresthesia-free paradigms can work. Recruitment remains strongly anatomy dependent, but a fixed superficial-depth rule oversimplifies lead position, CSF thickness, fiber diameter, dorsal-root entry, and waveform effects.

2.Tonic (Conventional) Stimulation

Conventional tonic SCS delivers continuous low-frequency pulses (classically on the order of 40–60 Hz) at an amplitude that recruits dorsal-column A-beta fibers to perception, producing a paresthesia. The historical marker of an adequate implant is paresthesia-pain concordance: the tingling must overlap the painful territory. This is not folklore: the classic two-decade Hopkins series (North) identified paresthesia coverage of the painful territory as a significant technical predictor of outcome, and that finding drove decades of technology aimed at maximizing concordant coverage. North’s own later teaching puts the strongest clinical predictor elsewhere, in the degree of relief achieved during the screening trial. The body-region-to-lead-level relationship is mapped in the Barolat paresthesia atlas and used in reverse: the pain location predicts the vertebral level at which to place the lead.

Two failure modes follow from the anatomy. First, the anatomic midline is not the physiologic midline (the fluoroscopic midline often does not match the cord’s), which is why awake repositioning to center the paresthesia is sometimes needed. Second, if the field is driven too hard or the lead sits too lateral, current spreads to the dorsal root and the patient feels uncomfortable radicular or band-like cramping rather than a smooth paresthesia. And because A-beta recruitment depends on the electrode-to-cord distance, posture changes the dose: patients typically feel stimulation most strongly lying supine, when the dorsal CSF layer is thinnest and the cord sits closest to the lead, and lose effect when upright.

3.High-Frequency 10 kHz: Paresthesia-Free

High-frequency 10 kHz stimulation (HF10 / SENZA) is generally delivered below perception and does not require paresthesia-pain overlap for programming. This is a clinical feature rather than proof of a particular cellular mechanism. Candidate mechanisms include dorsal-column/root, dorsal-horn, glial, and supraspinal effects, and remain unsettled.

The leading hypothesis is a direct effect on the superficial dorsal horn by a field too weak to fire dorsal-column A-beta fibers. Preclinical work (Lee et al., 2020) reported that low-amplitude 10 kHz selectively drives the inhibitory dorsal-horn interneurons (which calm the WDR neuron) without driving the excitatory ones, and that 1 kHz and 5 kHz did not do this, suggesting a frequency-specific effect. Clinical 10 kHz protocols place and program the lead by anatomic landmark (the empirically derived T9–T10 sweet spot) rather than by awake paresthesia mapping, which is the usual explanation for why 10 kHz reaches axial low-back pain that paresthesia-based tonic struggled with. Two honest caveats: the sweet spot was discovered empirically, not predicted from mechanism; and much of the mechanistic narrative comes from manufacturer-affiliated work, with the competing depolarization-block theory unsettled (see the skepticism note below). The pivotal SENZA-RCT (Kapural et al., 2015; 24-month results 2016) reported higher responder rates for 10 kHz than for traditional low-frequency stimulation in chronic back and leg pain: 76.5% versus 49.3% for back pain at 24 months. Read it with its design in mind: it could not be blinded, because one arm feels a paresthesia and the other does not; it was sponsored by the manufacturer of the tested device; and the active comparator was conventional paresthesia-based SCS with programming adjustments, rather than a sham intervention. It established paresthesia-free stimulation as a usable modality. It did not establish an effect size against placebo. Results from sham trials using other kilohertz frequencies should not be presented as direct tests of 10 kHz therapy.

4.Burst Stimulation and the Affective Dimension

Burst stimulation grew from De Ridder’s idea of mimicking the nervous system’s own firing language: the thalamus signals routine, homeostatic information to cortex in a tonic pattern but transmits salient, threatening signals, such as pain, in bursts, which recruit a larger and faster cortical response. The proprietary paradigm delivers five pulses per burst (each ~1 ms wide, ~1 ms apart) repeated 40 times per second, and its distinctive engineering feature is that charge is not recovered after each pulse but accumulates across the burst, with charge balancing only after the complete burst. That property separates this “intrinsic” burst from generic burst patterns that recover charge per pulse.

Its headline claim is dual-pathway action. Like tonic SCS it modulates the lateral (discriminative) spinothalamic pathway (the “where and how intense”). In addition it modulates the medial pathway projecting through medial thalamic nuclei to the limbic system, especially the dorsal anterior cingulate cortex, the pathway thought to carry the affective, suffering, attentional dimension of pain. Two small imaging datasets are cited for this. A source-localized EEG study (De Ridder and Vanneste) compared burst, tonic, and sham and reported dorsal anterior cingulate activity under burst that was not present under tonic or sham. An FDG-PET substudy of SUNBURST reported greater activation under burst than tonic in sensorimotor cortex, dorsal anterior cingulate, and posterior cingulate (that is, greater modulation of both the lateral and the medial pathway rather than selective engagement of the medial one), and it had no sham arm. Both datasets are small, neither has been independently replicated, and both come from investigators affiliated with the waveform’s developer or its manufacturer. The clinical corollary, that burst might help patients in whom the emotional burden of pain dominates, is plausible but not proven, because the pivotal SUNBURST RCT (Deer et al., 2018) was not designed to test it. Two practical footnotes. SUNBURST was a crossover in which each patient served as his or her own control, and many patients could identify which arm they were in by whether they felt a paresthesia, which weakens the blinding; the mean VAS separation between burst and tonic was small even where it reached significance. Burst may be programmed below perception, but paresthesia can occur. Dose and cycling should follow the studied paradigm and device labeling; intermittent protocols cannot be assumed equivalent to continuous delivery.

Differential target multiplexed (DTM) stimulation is the fourth waveform named in the evidence-status note above, and it deserves a sentence rather than silence. DTM interleaves pulse trains at different frequencies and pulse widths on different contacts; its stated rationale is that the two targets are different cell populations, neurons and glia, and the supporting mechanistic work is a rodent transcriptomic program from the manufacturer’s own laboratory reporting that DTM shifts glial gene expression more than conventional stimulation does. Treat that as a hypothesis. The human evidence is a multicenter, open-label randomized trial, manufacturer sponsored, reporting better back-pain response than conventional stimulation at 12 months (Fishman et al., 2021). An unblinded trial and a glial mechanism inferred from rat tissue are a thin basis for a clinical preference.

5.Why Open-Loop Stimulation Drifts

Waveform and feedback architecture are different design choices. In a fixed-output open-loop implementation, the device delivers a preset output without measuring neural recruitment. But the distance from a dorsal epidural electrode to the spinal cord is not fixed: it changes with respiration, heartbeat, and especially posture. Because recruitment falls off steeply as the electrode-to-cord distance grows, and the dorsal CSF layer thickens and thins with posture, small movements produce large swings in how much tissue is activated: loss of relief when the cord drifts away (under-stimulation), or unpleasant jolts when it moves closer (over-stimulation). The size and clinical consequences of this variation differ between patients and stimulation paradigms.

6.ECAP-Controlled Closed-Loop Stimulation

The closed-loop solution measures the cord’s own response and corrects for it in real time, the idea of a constant “neural dose.” The evoked compound action potential (ECAP) is the electrical signal generated by the synchronized dorsal-column fibers in response to each stimulus pulse (the same class of signal as the epidural spinal evoked potentials recorded intraoperatively, not the far-field cortical SSEP), and its amplitude reflects synchronized recruitment as well as recording geometry and other physiologic factors. The device stimulates on some contacts and records the ECAP on others (a small neural signal that must be separated from a much larger stimulation artifact), then adjusts the output of every pulse to hold the ECAP within the patient’s therapeutic window, compensating for posture, breathing, and heartbeat as they happen. The EVOKE RCT (Mekhail et al., Lancet Neurology 2020; durability in JAMA Neurology 2022, extended to 36 months in Reg Anesth Pain Med 2024) is the best-controlled trial on this page and the basis for approval: patients were randomized, double-blind, to closed-loop or open-loop stimulation on the same implanted system, the closed-loop arm did better on responder rate, and it held that advantage out to 36 months. Note precisely what was compared. Both arms received active stimulation from the same device, and the trial was sponsored by its manufacturer, so EVOKE supports closed-loop over open-loop control. It does not address whether either beats placebo.

Two distinctions matter. First, this is not the same as accelerometer-based “position-adaptive” stimulation, in which a motion sensor in the generator changes programs by posture: that sensor is far from the cord and too coarse to catch a cough or a fine electrode-to-cord shift, whereas ECAP feedback is measured at the cord itself. Second, and easy to miss, commercial ECAP control requires a measurable evoked response under the system’s own sensing and stimulation scheme. It is currently implemented with dorsal-column-recruiting pulses; that fact should not be turned into a universal claim that every subperception or high-frequency therapy can never be paired with a probe-ECAP strategy.

Examples of FDA-approved ECAP-controlled systems include Saluda’s Evoke (approved in 2022) and Medtronic’s Inceptiv (approved in April 2024), and they are not interchangeable. They differ in lead geometry, sensing scheme, control policy, and which neural-response target the controller is holding. The randomized evidence summarized above is Evoke’s. Do not read it as evidence for any other system, and confirm what each manufacturer’s own pivotal data actually show.

7.The Evidence Problem, and Why It Is Not the Mechanism Problem

Almost every trial named above compares one active waveform with another. Very few compare stimulation with a credible placebo, and the placebo-controlled literature is the weakest part of the field. That is a different defect from mechanistic uncertainty, and a more consequential one at the point of consent: a mechanism can be wrong and the therapy still work, but if the effect size against placebo is unknown, nobody can say how much of the relief is the stimulation.

Perruchoud and colleagues randomized patients already stabilized on tonic SCS to subperception 5 kHz stimulation or sham in a double-blind crossover and found no difference (Neuromodulation, 2013). Hara and colleagues randomized 50 patients with chronic radicular pain after lumbar spine surgery to burst stimulation or placebo stimulation in a blinded crossover, alternating 3-month periods, and found no significant difference in Oswestry disability: a mean between-group difference of −1.3 points (95% CI −3.9 to 1.3), P = .32 (JAMA, 2022). A 2023 Cochrane review of SCS for low back pain (Traeger et al.) found moderate-certainty evidence of little or no benefit over placebo at six months for low back pain, based on one 50-person study, and no placebo-controlled evidence at 12 months or longer. That scope should not be generalized to every SCS indication or confused with the duration of all included trials.

The sham literature is not uniformly negative: Al-Kaisy and colleagues randomized 24 patients in a short crossover study and found greater pain reduction with 5882 Hz than with sham, while lower frequencies did not separate. This was not a 10 kHz trial. Hara tested a four-pulse burst implementation on a different manufacturer’s generator; its result should not be represented as a direct sham test of the five-pulse proprietary BurstDR waveform. Small samples, carryover, blinding, and waveform differences constrain both positive and negative findings.

Durability is a separate problem. In Kemler’s randomized CRPS trial, the advantage of stimulation plus physical therapy over physical therapy alone was no longer present at five years. In a propensity-matched insurance-claims cohort (Dhruva et al., JAMA Neurology 2023; 1260 stimulator recipients matched to 6300 patients on conventional management), stimulation was not associated with lower chronic opioid use (the odds were modestly higher in the first year and similar in the second), total costs were roughly $39,000 higher in the first year, and 22.1% of implanted patients had the device removed or revised within two years. Published series put explant rates in a broadly similar range over 2 to 5 years, and the reason most often recorded is loss of analgesic effect rather than infection or hardware failure.

None of this says the therapy does not work, and each of these papers has been contested in print, the Cochrane review and the claims analysis particularly hard, on trial selection, cohort definition, and outcome choice. What it says is that the size of the true treatment effect, and how much of it is nonspecific, are unresolved. A trainee should be able to state that as fluently as the mechanism, and the surgeon consenting a patient should say it out loud.

8.Platforms, Battery, and MRI

Once the mechanism is understood, system selection becomes a rational match between the pain problem, the patient’s practical life, and what each platform does. Representative manufacturers offer different waveform and feedback capabilities. The differences that actually change a surgeon’s recommendation are fewer than the marketing suggests: the battery model, the waveform philosophy, closed-loop and DRG capability, and MRI access.

Platform names, indications, and capabilities change faster than comparative trials. This table is orientation, not a ranking; verify current FDA labeling and the complete implanted system before recommending a product.

The selection factors that matter most are these. Rechargeable versus primary cell. Rechargeable gives a smaller can and longer service life but imposes a daily-to-periodic charging commitment (a real adherence and dexterity issue, and a trial-blinding confounder), while a primary cell removes charging but means an eventual replacement operation. Waveform and capability. Does the case call for paresthesia-free sub-perception (10 kHz, BurstDR, FAST), physiologic closed-loop (for example, Evoke or Inceptiv), or DRG targeting (Abbott)? MRI access. Verify whether the specific system, leads, and extensions are conditional for the field strength and body region the patient will need, remembering that many implanted patients will need MRI during the life of the device and that MR Conditional means safe only under the specified conditions. Lead type. Percutaneous versus paddle (the latter placed surgically, often with lower migration).