Frequency-dependent thalamocortical transfer limits sustained cortical expression of sensory drive

Summary

Frequency-dependent sensory drive must pass through thalamocortical pathways before cortical expression, yet how high-frequency input is transformed into sustained cortical activity remains unclear. We combined 11.7-T fMRI, pathway-specific optogenetic stimulation, tuned-burst thalamic drive, and dynamic causal modeling (DCM) in dexmedetomidine-sedated rats to examine the ventral posterolateral thalamus (VPL)-to-primary somatosensory forelimb cortex (S1FL) pathway. Peripheral forepaw stimulation revealed intermediate-frequency cortical tuning, whereas thalamic nuclei remained responsive at high frequencies. Direct S1FL activation evoked sustained high-frequency blood-oxygen-level-dependent responses, indicating that cortex can express high-frequency drive when upstream relay is bypassed. In contrast, direct VPL activation produced strong thalamic responses but weak, onset-dominated high-frequency cortical responses, and tuned-burst stimulation showed temporal compression before cortical expression. DCM indicated preferential enhancement of VPL-to-S1FL effective connectivity at intermediate frequencies. These findings identify a frequency-dependent bottleneck in VPL-to-S1FL transfer, rather than local cortical responsiveness, as a constraint on sustained cortical expression of temporally dense sensory input.

Graphical abstract

Highlights

• •Cortical BOLD responses peak at intermediate sensory stimulation frequencies
• •Thalamic nuclei remain responsive when high-frequency cortical responses weaken
• •Direct cortical activation preserves sustained high-frequency BOLD responses
• •Cortical frequency tuning reflects pathway-level transfer constraints

Neuroscience; Systems neuroscience; Sensory neuroscience

Introduction

Patterned peripheral sensory stimulation is encoded through the somatosensory system into cortical representations that support tactile perception, sensorimotor control, and adaptive behavior.^1,2 The fidelity of this process has important clinical implications, as sensory disturbances and impaired sensorimotor integration are common features of neurological disorders, including Parkinson’s disease, amyotrophic lateral sclerosis, stroke, and injuries affecting peripheral or central sensory pathways.^3,4,5,6,7 Controlled tactile stimulation provides a tractable way to challenge these circuits under defined input conditions.^7,8,9,10 Among stimulus parameters, frequency is particularly informative because it determines the temporal structure imposed on the somatosensory pathway. Low-frequency inputs can be represented event by event, whereas rapidly repeated inputs require integration or filtering of densely patterned signals.^11,12,13 Varying stimulation frequency therefore offers a useful framework for examining temporal sensory coding and thalamocortical transfer. Although cortical tactile processing has been extensively investigated, the frequency-dependent limits of sensory transmission from the thalamus to the cortex remain poorly defined.^14,15,16

The ventral posterolateral thalamus (VPL)-to-primary somatosensory forelimb cortex (S1FL) pathway provides a tractable circuit for examining how frequency-dependent limits in the thalamic-to-cortical sensory transmission may arise.^14,15,16 VPL receives forepaw-related sensory input and relays it to S1FL, where thalamic drive is incorporated into cortical activity.^14,15 Electrophysiological studies indicate that this pathway does not operate as a simple passive relay. Repetitive thalamic activation can induce short-term synaptic depression, feedforward inhibition can restrict the temporal window for cortical integration, and thalamic relay fidelity can decline at high stimulation frequencies in a brain-state-dependent manner.^14,17,18,19 These mechanisms suggest that thalamocortical communication may depend strongly on input frequency.^14,15 However, how such frequency-dependent transfer is expressed in mesoscale cortical blood-oxygen-level-dependent (BOLD) responses remains unresolved.^8,9,10,16

Rodent forepaw stimulation fMRI provides a mesoscale readout of frequency-dependent somatosensory processing, but existing studies have primarily described a non-monotonic cortical frequency-response profile rather than identifying its circuit origin. S1FL BOLD responses typically peak at intermediate frequencies, around 4–8 Hz, and decline at higher frequencies.^8,9,10,20 This profile raises an unresolved circuit-level question: does high-frequency sensory drive continue to be represented in the thalamus but fail to produce sustained cortical expression, or does the limitation arise earlier or within cortex itself^14,15,16? Peripheral stimulation alone cannot distinguish among these possibilities because it engages peripheral afferents, thalamic nuclei, and cortical circuits simultaneously.^8,9,10,20 Resolving this distinction is important because it determines whether cortical frequency tuning reflects a local cortical property, a change in upstream thalamic representation, or a bottleneck in thalamocortical communication.^21,22,23 Thus, approaches that dissociate cortical responsiveness, thalamic responsiveness, and effective VPL-to-S1FL coupling are needed to interpret frequency-dependent BOLD responses at the circuit level.

We therefore hypothesized that the non-monotonic cortical frequency-response profile may reflect a frequency-dependent constraint in VPL-to-S1FL transfer, rather than a generalized intrinsic inability of cortex or thalamus to respond to high-frequency input.^{14,15,16,17,18,19} Specifically, high-frequency sensory drive may be maintained locally in VPL but fail to generate sustained downstream cortical BOLD responses.^14,15,16 To test this hypothesis, we combined multi-frequency peripheral forepaw stimulation fMRI, circuit-specific optogenetic activation, tuned-burst VPL stimulation, and dynamic causal modeling in dexmedetomidine-sedated rats.^{8,9,10,21,22,23,24} Figure 1 summarizes the overall experimental design. Peripheral stimulation first mapped frequency-response profiles across cortical and thalamic nodes. Direct S1FL activation determined whether cortex can sustain high-frequency BOLD responses when upstream relay is bypassed, whereas direct VPL activation tested whether local thalamic activation is sufficient to drive sustained S1FL activity. Tuned-burst VPL stimulation examined whether temporally dense thalamic input is preserved or reformatted before cortical expression. Finally, dynamic causal modeling assessed whether effective VPL-to-S1FL connectivity is preferentially enhanced at intermediate frequencies. By separating cortical responsiveness, thalamic responsiveness, and thalamocortical coupling, this strategy was designed to determine whether cortical frequency tuning reflects a local response property or a specific bottleneck in the thalamocortical transfer.

Figure 1.

Experimental design for testing frequency-dependent VPL-to-S1FL transfer

(A) Peripheral forepaw stimulation fMRI paradigm used to define frequency-response profiles across the forelimb somatosensory pathway. Bilateral forepaw stimulation was delivered at 1, 4, 8, 25, and 55 Hz using 2-mA, 300-μs electrical pulses in a block design with 15 s stimulation followed by 45 s rest.

(B) Anatomically defined ROIs overlaid on the SIGMA rat brain template, including S1FL (primary somatosensory cortex, forelimb area), S2 (secondary somatosensory cortex), VPL (ventral posterolateral thalamic nucleus), and Po (posterior thalamic nucleus). These regions were used to compare cortical and thalamic frequency-response profiles and to test where sensory drive becomes constrained during VPL-to-S1FL transfer. Coordinates indicate distance from bregma.

Results

Peripheral stimulation dissociates cortical tuning from sustained thalamic responsiveness

We first mapped BOLD activation across the forelimb somatosensory pathway during bilateral forepaw stimulation at 1, 4, 8, 25, and 55 Hz. Representative activation maps showed responses in S1FL, secondary somatosensory cortex (S2), VPL, and posterior thalamic nucleus (Po) (Figure 2; uncorrected, p < 0.001). Activation was spatially and frequency dependent. Cortical responses were strongest and most spatially extensive in S1FL, particularly at 4–8 Hz, whereas thalamic activation in VPL and Po remained prominent at higher frequencies. This pattern suggested that cortical and thalamic nodes may encode stimulation frequency differently, consistent with prior rodent somatosensory fMRI studies.^8,9,10,20,23

Figure 2.

Peripheral stimulation reveals dissociable cortical and thalamic activation across frequencies

Representative BOLD activation maps from one animal in the peripheral electrical stimulation cohort (n = 12 animals) during bilateral forepaw stimulation at 1, 4, 8, 25, and 55 Hz.

(A) S1FL activation maps.

(B) S2 activation maps.

(C) VPL and Po activation maps. Forepaw stimulation evoked BOLD activation in both cortical and thalamic somatosensory regions. Cortical activation was most prominent and spatially extensive in S1FL, particularly at intermediate frequencies, whereas thalamic activation in VPL and Po remained detectable at higher frequencies. Maps are overlaid on the SIGMA rat brain template and displayed for visualization using an uncorrected threshold of p < 0.001, with the color scale set to t = 5–35. Quantitative group-level ROI analyses are shown in Figure 3. No statistical comparisons are shown in this representative-map figure.

Figures 3A–3H show group-averaged BOLD time courses in S1FL, S2, VPL, and Po across stimulation frequencies; Figures 3I and 3J summarize peak BOLD signal change and full width at half maximum (FWHM); Figures 3K and 3L show latencies to reach 5% and 30% of peak amplitude. Peak BOLD amplitude showed significant effects of region and frequency as well as a strong region × frequency interaction (all p < 0.001; Table S1), indicating that cortical and thalamic regions exhibited different frequency-response profiles. FWHM showed a significant effect of frequency and a region × frequency interaction, whereas the main effect of region was not significant (Table S1).

Figure 3.

Peripheral stimulation shows intermediate-frequency cortical tuning despite preserved high-frequency thalamic responses

Group-level ROI analysis of BOLD responses in cortical and thalamic somatosensory regions during bilateral forepaw stimulation at 1, 4, 8, 25, and 55 Hz.

(A and B) Group-averaged BOLD time courses in S1FL.

(C and D) Group-averaged BOLD time courses in S2.

(E and F) Group-averaged BOLD time courses in VPL.

(G and H) Group-averaged BOLD time courses in Po. Gray shading indicates the 15-s stimulation period. The group-averaged BOLD time courses in panels A–H are presented as mean ± standard deviation (SD) across animals (n = 12).

(I) Peak BOLD signal change across stimulation frequencies and ROIs.

(J) Full width at half maximum (FWHM) of the BOLD response across stimulation frequencies and ROIs.

(K and L) Temporal response characteristics, quantified as the latency to reach 5% (K) and 30% (L) of peak amplitude. Cortical regions showed the largest and most sustained responses at intermediate frequencies, whereas thalamic regions retained high-frequency peak responses and showed earlier response onset.

Data were obtained from the peripheral electrical stimulation cohort (n = 12 animals; each animal is one biological replicate). Peak BOLD signal change and 30%-to-peak latency are represented as mean ± SD; FWHM and 5%-to-peak latency are represented as median (interquartile range [IQR]). Statistical analysis was performed using two-way repeated-measures ANOVA with region and frequency as within-subject factors, followed by within-ROI one-way repeated-measures ANOVA where appropriate. Holm-corrected paired-samples t tests were used for normally distributed measures, and Holm-corrected Wilcoxon signed-rank tests were used for non-normally distributed or discrete measures. ∗ indicates p < 0.05, ∗∗ indicates p < 0.01, ∗∗∗ indicates p < 0.001. Exact statistical results are provided in the main text and Table S1. See also Figures S1–S4 for the full ten-frequency atlas-based ROI profiles.

In S1FL, peak responses varied strongly with frequency and were maximal at 4–8 Hz (4 Hz: 15.4% ± 6.8%; 8 Hz: 17.5% ± 4.7%; frequency effect, p < 0.001). The response at 55 Hz (8.8% ± 1.3%) was larger than that at 25 Hz (2.4% ± 3.6%; p = 0.0005) but remained well below the 8-Hz response. Thus, 25 Hz represented a minimum in the S1FL peak-frequency profile, followed by a secondary increase in peak amplitude at 55 Hz. Importantly, this increase in peak amplitude was not accompanied by recovery of the sustained response observed at 4–8 Hz. FWHM was greatest at 4–8 Hz and was significantly broader at 4 and 8 Hz than at 1, 25, and 55 Hz. Therefore, intermediate-frequency stimulation elicited the strongest and most sustained S1FL responses.

S2 also showed frequency-dependent responses, although its profile differed from S1FL. Peak BOLD responses varied significantly with frequency (p < 0.001), with responses at 4, 8, 25, and 55 Hz exceeding those at 1 Hz. 55-Hz responses were modestly but significantly larger than 25-Hz responses (p = 0.021). Although peak responses at 4 and 8 Hz did not differ significantly from those at 55 Hz (both adjusted p > 0.05), FWHM at 4 and 8 Hz was significantly broader than at 1, 25, and 55 Hz, indicating that 4–8 Hz stimulation elicited more sustained cortical activation despite comparable peak amplitudes.

In contrast to the cortical profile, VPL and Po displayed high-frequency-dominant peak patterns. In VPL, peak amplitude varied significantly with frequency (p < 0.001) and reached its maximum at 55 Hz (5.4% ± 0.2%, mean ± SD). The 55 Hz produced significantly larger responses than 1, 4, and 25 Hz, whereas 55 and 8 Hz did not differ significantly after correction (Table S1). FWHM did not vary significantly across stimulation frequencies in VPL.

Po showed a similar high-frequency-dominant pattern, with maximal activation at 55 Hz (7.4% ± 0.2%; frequency effect, p < 0.001). The 55-Hz response exceeded those at 1 and 25 Hz, whereas differences from 4 to 8 Hz did not survive correction. Although FWHM showed an overall frequency effect in Po, no pairwise difference survived Holm correction. Together, VPL and Po retained strong high-frequency peak responses with largely frequency-invariant response duration, whereas S1FL and S2 showed pronounced temporal broadening at 4–8 Hz.

To quantify how much thalamic activity was expressed as sustained cortical activity, we calculated a peak-normalized sustained cortical expression index (SCEI), defined as the late S1FL response divided by the peak VPL response. SCEI varied significantly across frequencies (Friedman χ²(4) = 40.53, p = 3.36 × 10⁻⁸), with higher values at 4 Hz (3.26 [2.78–4.20]) and 8 Hz (2.22 [2.02–2.71]) than at 25 Hz (0.04 [−0.01–0.13]) or 55 Hz (0.25 [0.19–0.33]; Table S1). The late S1FL/late VPL ratio showed the same intermediate-frequency preference but was more variable at 55 Hz because late VPL responses approached zero in some animals. These results indicate that sustained cortical expression of thalamic activity was maximal at intermediate frequencies and reduced at high frequencies.

Temporal latency analysis further supported an ascending thalamocortical sequence. Both the latency to reach 5% and the latency to reach 30% of peak amplitude showed significant effects of region and frequency and significant region × frequency interactions (all p ≤ 0.001; Figures 3K and 3L; Table S1). Across frequencies, VPL consistently showed the earliest onset. The 5%-to-peak latency was significantly shorter in VPL than in S1FL at 1, 4, 8, and 55 Hz (all p ≤ 0.027), with the same VPL-leading pattern for the 30%-to-peak latency at 4, 8, and 55 Hz (all p ≤ 0.003). Thus, thalamic responses preceded cortical responses across most stimulation frequencies, while cortical and thalamic nodes showed dissociable frequency-response profiles.

Direct cortical activation preserves high-frequency cortical responsiveness

The peripheral stimulation results indicated that cortical and thalamic nodes showed dissociable frequency-response profiles, but they did not identify whether high-frequency cortical attenuation reflected local cortical responsiveness or reduced VPL-to-S1FL transfer. To test whether cortex can generate sustained high-frequency BOLD responses when the thalamocortical relay is bypassed, we directly activated S1FL optogenetically (S1FL-ChR2 cohort, n = 6; Figures 4A and 4B).

Figure 4.

Independent cortical and thalamic activation separates local responsiveness from VPL-to-S1FL transfer

(A) Schematic of viral injection sites and optogenetic stimulation paradigms. AAV2/9-CaMKIIα-hChR2(H134R)-mCherry-WPRE was injected into left S1FL or left VPL, and an optical fiber was implanted above the injection site.

(B) Representative activation maps and group-averaged BOLD time courses during direct optogenetic stimulation of S1FL at 1, 8, and 55 Hz (10-ms light pulses, 12 mW at the fiber tip; 15 s ON, 45 s OFF). Direct S1FL activation produced robust local cortical responses, including sustained high-frequency responses.

(C) Representative activation maps and group-averaged local BOLD responses in VPL during direct optogenetic stimulation of VPL at 1, 8, and 55 Hz. Local VPL responses remained robust and increased at high frequency.

(D) Group-averaged downstream BOLD responses in S1FL during VPL stimulation. In contrast to the preserved local VPL response, high-frequency VPL stimulation evoked a weak and onset-dominated cortical response.

Data were obtained from independent S1FL-ChR2 and VPL-ChR2 cohorts (n = 6 animals per group; each animal is one biological replicate). Peak BOLD responses are represented as mean ± SD and were evaluated using one-way repeated-measures ANOVA within each cohort, followed by Holm-corrected paired-samples t tests where appropriate. ∗ indicates p < 0.05, ∗∗ indicates p < 0.01, ∗∗∗ indicates p < 0.001. Exact statistical results, including effect sizes, are provided in the main text. Gray shading indicates the 15-s stimulation period. The activation-map color bars indicate t values (range, 5–9), and anatomical slice coordinates indicate distance from bregma.

Direct S1FL stimulation evoked robust local BOLD responses at all tested frequencies. Peak responses were 3.87% ± 1.64%, 4.47% ± 1.93%, and 3.52% ± 1.64% at 1, 8, and 55 Hz, respectively. The overall effect of stimulation frequency was not significant (one-way repeated-measures ANOVA, F(2,10) = 0.48, p = 0.63, ηp² = 0.09), indicating that local cortical BOLD responses showed no significant omnibus frequency effect when the thalamocortical synapse was bypassed. The prespecified 8-vs-55 Hz comparison showed that the 55-Hz response was modestly smaller than the 8-Hz response (mean difference = 0.96%, 95% confidence interval [CI] [0.60, 1.32], adjusted p = 0.003, dz = 2.77). Critically, the 55-Hz response remained temporally sustained throughout the stimulation period, in contrast to the onset-dominated cortical response observed during high-frequency VPL stimulation. These findings indicate that cortex retains the capacity to generate sustained high-frequency BOLD responses when activated directly.

Direct thalamic activation reveals inefficient high-frequency thalamocortical transfer

We next asked whether high-frequency thalamic drive is effectively transmitted to cortex. In a separate cohort, we stimulated VPL optogenetically and measured both local VPL responses and downstream S1FL responses (VPL-ChR2 cohort, n = 6; Figures 4C and 4D).

Local VPL responses increased with stimulation frequency. Peak VPL responses were 0.93% ± 0.24%, 1.57% ± 0.10%, and 2.24% ± 0.49% at 1, 8, and 55 Hz, respectively, with a significant frequency effect (one-way repeated-measures ANOVA, F(2,10) = 28.59, p < 0.001, ηp² = 0.85). Holm-corrected paired comparisons showed that the VPL response at 55 Hz was significantly larger than that at both 1 and 8 Hz (55 vs. 1 Hz: mean difference = 1.30%, 95% CI [0.94, 1.67], adjusted p < 0.001, dz = 3.73; 55 vs. 8 Hz: mean difference = 0.66%, 95% CI [0.06, 1.27], adjusted p = 0.037, dz = 1.15), and the response at 8 Hz was larger than that at 1 Hz (mean difference = 0.64%, 95% CI [0.34, 0.94], adjusted p = 0.005, dz = 2.25). Thus, VPL responded maximally to high-frequency optogenetic drive.

Downstream S1FL responses during VPL stimulation showed the opposite frequency dependence (Figure 4D). S1FL peak responses were 2.14% ± 1.09%, 9.75% ± 4.72%, and 1.96% ± 0.89% at 1, 8, and 55 Hz, respectively, and varied significantly across stimulation frequencies (one-way repeated-measures ANOVA, F(2,10) = 13.86, p = 0.001, ηp² = 0.73). Holm-corrected paired comparisons showed that 8-Hz VPL stimulation evoked a larger downstream S1FL response than either 1-Hz stimulation (mean difference = 7.61%, 95% CI [1.64, 13.58], adjusted p = 0.044, dz = 1.34) or 55-Hz stimulation (mean difference = 7.78%, 95% CI [3.57, 12.00], adjusted p = 0.015, dz = 1.94), whereas responses at 1 and 55 Hz did not differ significantly (adjusted p = 0.811).

The critical dissociation occurred at 55 Hz: VPL showed a strong local response, but downstream S1FL responses were weak and onset dominated. Together with the direct S1FL results, this pattern places the main constraint downstream of preserved local VPL activation and upstream of sustained S1FL expression. High-frequency thalamic drive is locally represented in VPL but inefficiently expressed in S1FL.

Tuned-burst thalamic stimulation shows temporal compression before cortical expression

To determine how temporally dense thalamic activity is expressed downstream, we used a tuned-burst VPL stimulation paradigm. In this design, 50-Hz pulse trains were embedded within slower carrier cycles of 1, 4, or 8 Hz (Figure 5A). The intraburst high-frequency component was held constant while the carrier envelope varied, allowing us to distinguish effects of local high-frequency thalamic activation from effects of burst timing.

Figure 5.

Tuned-burst VPL stimulation reveals temporal compression before cortical expression

(A) Schematic comparison of conventional and tuned-burst VPL stimulation. Conventional stimulation delivered single 10-ms pulses at the indicated frequency, whereas tuned-burst stimulation delivered five-pulse 50-Hz bursts at carrier frequencies of 1, 4, or 8 Hz, thereby preserving dense intraburst activity while varying the slower temporal envelope. The conventional and tuned-burst paradigms differ in both temporal structure and total pulse density and were not designed to be pulse-number matched.

(B) Group-averaged BOLD time courses in VPL during tuned-burst VPL stimulation. Local VPL responses were present across carrier-frequency conditions, with no significant carrier-frequency effect on peak amplitude or mean stimulation-period response.

(C) Group-averaged BOLD time courses in S1FL during the same tuned-burst stimulation conditions. In contrast to local VPL responses, S1FL responses were dominated by an early transient peak followed by substantial decay during the stimulation block.

Data were obtained from the VPL-ChR2 cohort (n = 6 animals; each animal is one biological replicate). Peak and mean stimulation-period responses were compared across carrier frequencies using one-way repeated-measures ANOVA. Initial peak responses were compared with late stimulation-period responses using paired-samples t tests with Holm correction across carrier-frequency conditions. Exact statistical results, including effect sizes, are provided in the main text. Data are represented as mean ± SD.

Local VPL responses were present across all tuned-burst conditions and did not differ significantly across carrier frequencies (Figure 5B). VPL peak amplitudes were 1.36% ± 0.89%, 1.50% ± 0.33%, and 2.12% ± 0.80% for the 1-, 4-, and 8-Hz envelope conditions, with no significant effect of carrier frequency (one-way repeated-measures ANOVA, F(2,10) = 1.45, p = 0.281, ηp² = 0.22). Mean VPL responses during the stimulation period also did not differ significantly across carrier frequencies (1 Hz: 0.58% ± 0.62%; 4 Hz: 0.67% ± 0.25%; 8 Hz: 1.11% ± 0.98%; F(2,10) = 0.76, p = 0.493, ηp² = 0.13). Thus, local VPL activation was sustained within each carrier condition and was dominated by the intraburst 50-Hz component rather than by the slower carrier envelope.

S1FL peak amplitudes also showed no significant carrier-frequency dependence (Figure 5C). Peak responses were 2.64% ± 1.21%, 3.33% ± 1.11%, and 3.47% ± 2.07% for the 1-, 4-, and 8-Hz envelope conditions (F(2,10) = 1.32, p = 0.309, ηp² = 0.21). Mean stimulation-period responses in S1FL also did not differ significantly across carrier frequencies (1 Hz: 1.06% ± 0.75%; 4 Hz: 1.24% ± 0.50%; 8 Hz: 1.18% ± 0.87%; F(2,10) = 0.31, p = 0.741, ηp² = 0.06).

Although S1FL response magnitude did not vary significantly with carrier frequency, its temporal profile differed markedly from the local VPL response. Across all tuned-burst conditions, S1FL responses were dominated by an early transient peak followed by substantial decay during the stimulation block. The initial cortical peak was significantly greater than the late stimulation-period response for all carrier frequencies after Holm correction (1 Hz: mean difference = 1.99%, 95% CI [1.14, 2.84], t(5) = 6.01, adjusted p = 0.005, dz = 2.45; 4 Hz: mean difference = 2.42%, 95% CI [1.07, 3.76], t(5) = 4.62, adjusted p = 0.011, dz = 1.89; 8 Hz: mean difference = 2.78%, 95% CI [0.98, 4.59], t(5) = 3.97, adjusted p = 0.011, dz = 1.62). We refer to this response pattern as “temporal compression”, conceptually consistent with previous observations that temporally extended or repeated sensory inputs can produce disproportionately compressed response profiles rather than linear temporal summation.²⁵

These data reveal two separable features of thalamocortical transfer. Within each carrier condition, dense VPL drive was expressed in cortex as a temporally compressed, onset-dominated response. Across carrier conditions, the degree of cortical compression did not differ significantly over the 1–8 Hz carrier range. Notably, the preferential cortical enhancement at 8 Hz observed during peripheral stimulation was not reproduced by tuned-burst VPL stimulation, even though the 8-Hz carrier delivered the same number of bursts per second. Thus, the intermediate-frequency preference during peripheral stimulation may depend on spatiotemporal input patterns that are not reproduced by uniform optogenetic burst activation of VPL.

Effective-connectivity modeling converges on intermediate-frequency VPL-to-S1FL transfer

Finally, we used dynamic causal modeling (DCM)²⁴ to assess whether the frequency-dependent thalamocortical dissociation observed in the BOLD and optogenetic experiments was reflected in effective connectivity. At the intrinsic (A-matrix) level, the model identified a prominent VPL-S1FL connection (Figure 6A), consistent with the ascending organization of the somatosensory pathway.^14,15,16

Figure 6.

Effective-connectivity modeling supports intermediate-frequency enhancement of VPL-to-S1FL transfer

(A) Group-level intrinsic connectivity within the VPL-Po-S1FL-S2 network estimated using DCM, parametric empirical Bayes, and Bayesian model reduction. The intrinsic model identified a prominent baseline connection from VPL to S1FL.

(B–D) Frequency-specific modulatory effects of forepaw stimulation on effective connectivity within the network. Stimulation at 1 Hz negatively modulated Po-to-S1FL coupling, stimulation at 8 Hz selectively enhanced VPL-to-S1FL effective connectivity, and stimulation at 55 Hz produced no significant positive modulation of this pathway. Positive and negative modulatory effects are shown in red and blue, respectively. DCM analyses were performed on the electrical forepaw stimulation dataset used for effective-connectivity modeling (n = 12 animals; each animal is one biological replicate). Connections with posterior probability >0.99 were considered significant.

Frequency-specific modulation of this connection paralleled the experimental findings (Figures 6B–6D). Stimulation at 1 Hz negatively modulated Po-to-S1FL coupling, indicating reduced thalamocortical drive at low frequency. Stimulation at 8 Hz selectively enhanced VPL-to-S1FL effective connectivity (positive modulation, posterior probability >0.99). Stimulation at 55 Hz produced no significant positive modulation. These DCM results provide convergent, model-based evidence that VPL-to-S1FL effective connectivity is frequency-dependently modulated, with maximal enhancement at intermediate frequency and weak or absent enhancement at low and high frequencies.

Because DCM was estimated from BOLD time series, the null modulatory effect at 55 Hz should be interpreted cautiously.²⁴ Reduced BOLD signal-to-noise at high frequency may limit sensitivity for detecting modulatory effects at that frequency. We therefore interpret the DCM results as convergent network-level support for the thalamocortical relay bottleneck identified by the hemodynamic and optogenetic experiments, rather than as standalone proof of reduced synaptic efficacy.

Discussion

As shown in Figure 7, we combined peripheral stimulation fMRI, pathway-specific optogenetic perturbation, tuned-burst VPL stimulation, and dynamic causal modeling to determine how frequency-dependent sensory drive is transformed along the VPL-to-S1FL pathway. The main finding is that high-frequency sensory drive can evoke robust local VPL BOLD responses yet fail to generate sustained cortical BOLD expression in S1FL. Peripheral forepaw stimulation revealed intermediate-frequency cortical tuning despite preserved high-frequency thalamic responses (Figure 7A). Direct S1FL activation showed that cortex can generate sustained high-frequency BOLD responses when upstream relay is bypassed (Figure 7B), whereas direct VPL activation produced strong local thalamic responses but weak, onset-dominated downstream cortical responses (Figure 7C). Together with the tuned-burst and effective-connectivity results (Figure 7D), these findings support a frequency-dependent thalamocortical transfer bottleneck that constrains the sustained cortical expression of temporally dense sensory input.

Figure 7.

Schematic summary of the proposed frequency-dependent thalamocortical transfer model

(A) Peripheral forepaw stimulation revealed maximal sustained S1FL responses at intermediate frequencies (4–8 Hz), whereas VPL responses remained robust at higher frequencies.

(B) Direct optogenetic stimulation of S1FL preserved sustained high-frequency cortical responsiveness, indicating that local cortical responsiveness was not the primary limiting factor.

(C) Direct optogenetic stimulation of VPL produced robust local high-frequency thalamic responses but weak, onset-dominated downstream S1FL responses, supporting a frequency-dependent constraint in VPL-to-S1FL transfer.

(D) Tuned-burst VPL stimulation revealed temporally compressed cortical responses, while dynamic causal modeling (DCM) provided complementary network-level evidence of preferential VPL-to-S1FL effective connectivity at intermediate frequencies.

(E) Integrated schematic of the proposed thalamocortical transfer model. Sustained cortical expression is favored at 4–8 Hz, whereas higher-frequency stimulation (25–55 Hz) produces an onset-dominated response with an attenuated sustained component, despite preserved local VPL responsiveness.

Frequency-dependent VPL-to-S1FL transfer shapes cortical tuning

A reduced cortical response to high-frequency sensory stimulation could arise from several levels of the ascending pathway. One possibility is that cortex itself cannot sustain high-frequency activation because of local synaptic adaptation, recurrent inhibition, metabolic constraints, or nonlinear neurovascular responses.^{26,27,28,29,30} Our direct S1FL stimulation experiment argues against this explanation as the primary driver of the observed cortical frequency tuning. When S1FL was activated directly, high-frequency stimulation still evoked a sustained local BOLD response. Thus, the present experiments identify where sustained high-frequency drive becomes constrained at the macroscopic pathway level, but they do not establish the cellular process responsible for this constraint.

The complementary VPL stimulation experiment placed the main constraint downstream of preserved local thalamic activation. Direct VPL activation produced robust local responses that increased with stimulation frequency, but the downstream S1FL response was weak and onset-dominated at 55 Hz. This dissociation indicates that preserved thalamic activation is not sufficient to ensure sustained cortical expression. Rather than reflecting a simple failure of either thalamus or cortex in isolation, the data point to a frequency-dependent constraint in the transfer from VPL activity to sustained S1FL activity. Local cortical and neurovascular factors may still shape the magnitude and temporal profile of the measured BOLD response, but they do not fully explain why high-frequency thalamic drive is poorly expressed downstream.

This interpretation fits with the broader view that thalamus actively transforms sensory information before cortical expression.^{31,32,33,34,35,36} Recent work has shown that vibrotactile frequency information is reformatted along the ascending somatosensory pathway, including a transformation from phase-locked temporal coding to rate-based coding at the level of VPL.¹⁶ Our results extend this framework by identifying a downstream transfer step at which preserved thalamic activation is converted into frequency-dependent cortical expression. At intermediate frequencies, thalamic drive was effectively expressed as sustained cortical BOLD activity. At high frequencies, thalamic activation was expressed mainly as a brief cortical onset response. These observations support a multi-stage view in which tactile frequency information is progressively reformatted before and during thalamocortical transfer.

High-frequency thalamic drive is temporally compressed before cortical expression

The tuned-burst experiments further clarified how temporally dense thalamic activity is expressed downstream. By embedding 50-Hz pulse trains within slower carrier cycles, this paradigm separated local dense thalamic activation from the slower temporal envelope of stimulation. Local VPL responses were present across carrier conditions, whereas S1FL responses were consistently dominated by an early transient followed by decay. Thus, temporal compression was not simply inherited from peripheral input. It emerged during the transformation of dense VPL activation into cortical BOLD activity.

The stronger reduction of S1FL responses during dense tuned-burst stimulation raises the possibility that temporally dense thalamic input undergoes additional synaptic and circuit-level filtering during VPL-to-S1FL transfer. Previous electrophysiological studies have shown that repetitive sensory stimulation can produce stronger adaptation in cortex than in somatosensory thalamus, with activity-dependent depression of thalamocortical synaptic input contributing to this additional cortical attenuation.^17,19 Thalamocortical input also dynamically engages cortical feedforward inhibitory circuitry, which regulates the temporal integration window for cortical responses,¹⁸ while short-term plasticity at individual thalamocortical synapses can vary substantially with recent stimulation history.³⁷ Together, these observations provide a plausible framework in which preserved local thalamic activity is transformed by thalamocortical synaptic dynamics and cortical microcircuit processing before sustained cortical expression. This transformation may also depend on brain state, which regulates thalamic excitability and the relay of temporally patterned sensory inputs.^38,39 The observed responses should also be interpreted in the context of dexmedetomidine sedation, given that anesthetic regimens can alter functional connectivity⁴⁰ and dexmedetomidine can affect both neural and vascular responses in somatosensory cortex.⁴¹

Importantly, this framework does not imply that the same cellular mechanism operates identically in VPL and S1FL. Although the same CaMKIIα-based optogenetic strategy was used here to preferentially activate principal/excitatory neuronal populations in the two regions, thalamic relay neurons and cortical principal neurons occupy distinct synaptic and circuit contexts. Because direct CaMKIIα-based activation of S1FL principal neurons sustained high-frequency BOLD responses, the constraint is unlikely to reflect an intrinsic inability of the targeted cortical principal neurons to follow high-frequency drive; rather, our data favor processes engaged when input arrives through thalamocortical synapses, such as short-term synaptic depression and thalamically recruited feedforward inhibition.^17,18 Future studies should therefore combine matched recordings or perturbations of functionally analogous principal neuronal populations in VPL and S1FL with targeted examination of region-specific circuit elements, particularly cortical inhibitory populations. At the molecular level, targeted pharmacological or genetic manipulation of receptor-mediated excitation and inhibition during repeated stimulation could test whether changes in thalamocortical synaptic efficacy or inhibitory recruitment alter the frequency dependence of VPL-to-S1FL transfer. Such experiments will be needed to determine the relative contributions of thalamocortical short-term synaptic plasticity, cortical excitatory-inhibitory circuit dynamics, and stage-specific neuronal properties to the observed frequency-dependent transfer constraint. Neurovascular nonlinearities may additionally contribute to the measured BOLD response.^26,27,28

The differential high-frequency responsiveness of S1FL and S2 suggests that high-frequency sensory drive may also be distributed through parallel cortical and thalamocortical routes.⁴² During peripheral stimulation, S2 retained stronger high-frequency responses than S1FL. This regional dissociation could reflect differences in thalamic input, including contributions from posterior thalamic nuclei, or convergent corticocortical input from S1FL. Disentangling these possibilities will require pathway-specific perturbation experiments that separately target VPL-to-S1FL transfer, posterior thalamic projections, and corticocortical communication.

Implications for tactile coding and sensory-evoked fMRI

The present findings have implications for how tactile frequency information is represented in cortex. Sustained high-frequency input is not necessarily expressed in cortex as a sustained response of equivalent duration. Instead, cortical activity may preferentially emphasize stimulus onset, transitions, or changes in sensory drive while attenuating redundant steady-state input. This interpretation is consistent with tactile coding frameworks in which frequency information is transformed across peripheral, thalamic, and cortical stages.^{11,12,13,16,36,43,44} In this view, VPL may preserve or reformat temporally dense sensory drive, whereas S1FL expresses the subset of thalamic activity that is effectively transmitted through the thalamocortical pathway.

These results also refine the interpretation of sensory-evoked BOLD fMRI.^{8,9,10,20,26,27,28} A reduced cortical BOLD response at high stimulation frequency does not necessarily indicate that local cortical tissue cannot generate a hemodynamic response.^{21,22,26,27,28} It may instead reflect reduced efficacy of pathway-level transfer from thalamus to cortex. The dissociation observed here, preserved local VPL activation but weak sustained S1FL expression, shows that local activation and downstream cortical expression can be separable. Therefore, stimulation frequency should be viewed not only as a parameter controlling input density but also as a variable that probes the transfer properties of ascending sensory pathways. Consistent with this broader principle, auditory fMRI studies have shown that BOLD responses are sensitive to temporal modulation rate and stimulus bandwidth.^45,46

The DCM results provide complementary network-level support for this interpretation. VPL-to-S1FL effective connectivity was enhanced at 8 Hz, matching the intermediate-frequency transfer optimum inferred from the stimulation experiments. Although DCM cannot provide direct evidence of synaptic efficacy, it nevertheless offers complementary network-level evidence supporting the frequency-dependent modulation of VPL-to-S1FL transfer.^24,47 The weaker S1FL BOLD response at 55 Hz may limit the sensitivity of DCM to detect modulatory effects at this frequency.

Sensory and sensorimotor abnormalities are reported in several neurological disorders, including Parkinson’s disease, amyotrophic lateral sclerosis, and stroke.^3,4,5,6,7 Our findings suggest that reduced sensory-evoked cortical responses should not simply be interpreted as evidence of impaired local cortical responsiveness. It raises the possibility that, in some settings, an abnormal cortical response could reflect altered transmission through upstream thalamocortical pathways rather than a purely local cortical deficit. Frequency-dependent stimulation may therefore provide a useful systems-level probe for distinguishing local cortical responsiveness from pathway-level transfer dysfunction.

Limitations of the study

This study has several limitations. First, the findings localize a frequency-dependent constraint at the macroscopic BOLD level, but they do not identify the cellular mechanism that produces this constraint. The observed temporal compression may involve short-term thalamocortical synaptic depression, feedforward inhibition, state-dependent relay dynamics, neurovascular nonlinearities, or interactions among these processes.^{17,18,26,27,28,30,37,38,39} Electrophysiology, calcium imaging, or cell-type-specific perturbation will be needed to determine how sustained thalamic activation is converted into an onset-dominated cortical response.

Second, all experiments were performed under dexmedetomidine sedation. This regimen preserves robust sensory-evoked hemodynamic responses and provides a stable physiological condition for fMRI measurements. While brain state may influence the quantitative expression of evoked responses, peripheral electrical stimulation engages the canonical ascending somatosensory pathway.^{1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16} Future awake-fMRI or combined electrophysiology-fMRI experiments will be important for determining whether the same frequency-dependent transfer bottleneck operates during active tactile processing.

Third, the perturbation experiments focused on the VPL-to-S1FL pathway. This design allowed us to separate local thalamic responsiveness, local cortical responsiveness, and VPL-to-S1FL transfer, but it does not fully resolve contributions from parallel somatosensory routes. The Po, S2, and corticocortical interactions⁴² may contribute to the preserved high-frequency responses observed outside S1FL. Pathway-specific perturbation of Po-to-cortical and S1FL-to-S2 projections will be needed to determine how high-frequency sensory drive is distributed across the broader somatosensory network.

Fourth, this study did not include behavioral measurements, awake tactile perception experiments, or human psychophysics. The proposed relationship between temporal compression and tactile frequency coding^11,12,13 should therefore be interpreted as a mechanistic hypothesis rather than direct evidence for perception. These limitations do not alter the central conclusion that sustained cortical expression of high-frequency sensory drive is constrained during thalamocortical transfer, but they define the level at which this conclusion should be interpreted.

Finally, only adult male rats were studied. The influence of sex on the frequency-dependent thalamocortical transfer phenotype was therefore not assessed, and the generalizability of these findings to female animals remains to be determined.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Zhentao Zuo (zuozt@ibp.ac.cn).

Materials availability

This study did not generate new unique reagents. Additional information about viral vectors, optogenetic stimulation parameters, and MRI acquisition protocols is available from the lead contact upon reasonable request.

Data and code availability

• •Processed ROI time courses and figure source data are available at Zenodo: https://doi.org/10.5281/zenodo.21068601. Raw MRI data are available from the lead contact upon reasonable request, subject to institutional data-sharing policies.
• •Custom statistical analysis scripts are available at Zenodo: https://doi.org/10.5281/zenodo.21068371.
• •Stimulation timing files and DCM model specifications supporting the findings of this study are available at Zenodo: https://zenodo.org/records/23252869. Additional information is available from the lead contact upon reasonable request.

Acknowledgments

We thank Yuxi Sun and Hanlin Yue for technical assistance and support during the MRI experiments. This work was supported by the Ministry of Science and Technology of China (2023YFF0714204 and 2022ZD0211901) and the Chinese Academy of Sciences (2021091, YSBR-068, and 2024GZL003).

Author contributions

Conceptualization, Z.Y. and Z.Z.; methodology, Z.Y. and Z.Z.; investigation, Z.Y. and H.W.; formal analysis, Z.Y., H.W., and Z.Z.; data interpretation, Z.Y., Y.Z., and Z.Z.; writing – original draft, Z.Y.; writing - review and editing, H.W., Y.Z., and Z.Z.; supervision, Z.Z.; funding acquisition, Z.Z. All authors reviewed and approved the final manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
AAV2/9-CaMKIIα-hChR2(H134R)-mCherry-WPRE	BrainVTA, Wuhan, China	N/A
Chemicals, peptides, and recombinant proteins
Dexmedetomidine (Dexdomitor)	Orion Pharma	CAS: 113775–47–6
Isoflurane	RWD Life Science, Shenzhen, China	CAS: 26675–46–7
Meloxicam	N/A (standard postoperative analgesic)	N/A
Dental cement (Super Bond C&B)	Sun Medical Co., Ltd, Japan	N/A
Deposited data
Processed ROI time courses and figure source data	This paper	Zenodo: https://doi.org/10.5281/zenodo.21068601
Custom statistical analysis scripts	This paper	Zenodo: https://doi.org/10.5281/zenodo.21068371
Stimulation timing files and DCM model specifications	This paper	Zenodo: https://zenodo.org/records/23252869
Raw MRI data	This paper	Available from lead contact upon reasonable request
Experimental models: Organisms/strains
Rat: Adult male Wistar, 8–10 weeks, 302 ± 26 g	Charles River, China	RRID: RGD_13508588
Software and algorithms
SPM12	Wellcome Center for Human Neuroimaging, UCL	https://www.fil.ion.ucl.ac.uk/spm/; RRID: SCR_007037
RABIES v0.4.2	Desrosiers-Grégoire et al.48	https://github.com/CoBrALab/RABIES; RRID: SCR_023722
NORDIC PCA denoising	Moeller et al.49	https://github.com/SteenMoeller/NORDIC
ANTs (Advanced Normalization Tools)	Avants et al.50	https://github.com/ANTsX/ANTs; RRID: SCR_004757
SIGMA rat brain templates and atlases	Barrière et al.51	https://www.nature.com/articles/s41467-019-13575–7; RRID: SCR_025878
SPSS	IBM Corp., Armonk, NY, USA	RRID: SCR_002865
Other
11.7-T Bruker MRI scanner (BioSpec 117/16 USR)	Bruker, Karlsruhe, Germany	N/A
Four-element surface array coil	Bruker, Karlsruhe, Germany	N/A
ISO-FLEX stimulus isolator	AMPI, Jerusalem, Israel	N/A
Arduino UNO trigger system	Arduino S.r.l., Turin, Italy	N/A
473-nm diode-pumped solid-state laser (MGL-F-473 nm)	CNI Laser, Changchun, China	N/A
Optic fiber (200 μm core, NA 0.37)	Inper, Hangzhou, China	N/A
30-gauge needle electrodes	N/A (standard subcutaneous electrode)	N/A
Stereotaxic frame	N/A (standard rodent stereotax)	N/A
Water-circulated heating bed	Bruker, Karlsruhe, Germany	N/A

Experimental model and study participant details

Animals

The animal work in this study was approved by Animal Care and Use Committee of the Academy of Military Medical Sciences (IACUC-DWZX-2021–678). No prospective protocol registration was performed for this animal fMRI study.

Twenty-four adult male Wistar rats (Charles River, China; 8–10 weeks old; 302 ± 26 g) were used. Twelve rats were used for peripheral electrical forepaw stimulation fMRI. For optogenetic experiments, six rats received viral transduction in the left primary somatosensory cortex forelimb region (S1FL) and six rats received viral transduction in the left ventral posterolateral thalamic nucleus (VPL). The tuned-burst stimulation experiments were performed in the VPL-ChR2 cohort.

Animals were housed in individually ventilated cages under controlled temperature (22°C) on a 12-h light/dark cycle, with food and water available ad libitum. Each animal was treated as one independent biological replicate. Sample sizes were based on prior rodent sensory and optogenetic fMRI studies using comparable within-animal repeated-measures designs and the 3Rs principle^{8,9,10,20,21,22,23}; no formal a priori sample-size calculation was performed.

The primary outcome was peak BOLD signal change within anatomically constrained ROIs; secondary outcomes included full width at half maximum (FWHM), onset and rise latencies, mean and late stimulation-period responses, and DCM-derived effective connectivity.

Animals were included if physiological parameters remained stable, image registration and motion quality control were acceptable, and, for optogenetic experiments, viral expression and fiber placement were within the intended target region. These criteria were established before statistical analysis. No animals, fMRI runs, or data points were excluded. Stimulation frequencies were presented in randomized order. Experimenters were aware of surgical targets during surgery and acquisition, whereas preprocessing and ROI extraction followed a standardized atlas-based pipeline.

Method details

Anesthesia and physiological monitoring

Before MRI acquisition, animals were anesthetized with 3% isoflurane and positioned on the MRI cradle. Sedation was then transitioned to dexmedetomidine by intramuscular bolus injection (0.017 mg/kg), while isoflurane was gradually reduced over 10 min and then discontinued. After a 30-min stabilization period, dexmedetomidine was continuously infused intramuscularly at 0.025 mg/kg/h. This protocol was chosen because dexmedetomidine preserves robust sensory-evoked hemodynamic responses in the rodent somatosensory system while minimizing motion and physiological instability.^10,40,41

Body temperature was maintained at 37.0°C–37.5°C using a water-circulated heating bed. Respiration and heart rate were continuously monitored, and the head was secured with ear bars and a bite bar.

Viral transduction and optical fiber implantation

For optogenetic experiments, rats were anesthetized with isoflurane and placed in a stereotaxic frame. To preferentially drive excitatory/principal neurons in the target region,²¹ AAV2/9-CaMKIIα-hChR2(H134R)-mCherry-WPRE (BrainVTA, Wuhan, China) was injected into either S1FL or VPL.

Virus was delivered using a glass micropipette at 25 nL/min. A total volume of 600 nL was injected either into the left S1FL (0–0.2 mm posterior to bregma, 3.5 mm lateral to the midline, 2.0 mm below the dura) or into the left VPL (2.8–3.6 mm posterior to bregma, 3.8 mm lateral to the midline, 6.0 mm below the dura). After injection, the pipette was left in place for at least 10 min to minimize reflux.

An optic fiber (200 μm core diameter, numerical aperture 0.37) was implanted approximately 0.05 mm above the injection site and secured with dental cement. Animals recovered for at least 3 weeks before fMRI. Rats were assigned to either the S1FL-ChR2 or VPL-ChR2 group; each animal received viral injection in only one target region.

After surgery, animals received postoperative analgesia with meloxicam and were monitored daily for body weight, wound condition, mobility, feeding behavior, and signs of distress. Humane endpoints included persistent weight loss, impaired mobility, wound infection, or severe distress. No unexpected adverse events occurred.

At the conclusion of fMRI experiments, mCherry expression and optic fiber tip locations were verified histologically in all animals. Only animals with visible target-region expression and correct fiber placement were included in the analysis.

Peripheral electrical forepaw stimulation

Electrical stimulation was delivered at five representative frequencies (1, 4, 8, 25, and 55 Hz), spanning discrete low-frequency events, the intermediate-frequency range at which cortical BOLD responses typically peak,^8,9,10,20 and the classical perceptual fusion boundary.^11,12,13,16

Stimulation was delivered through two 30-gauge needle electrodes inserted subcutaneously into the plantar skin of both forepaws. The electrodes were driven by an ISO-FLEX stimulus isolator (AMPI, Jerusalem, Israel) and synchronized with the MRI scanner through an Arduino-based trigger system (Arduino S.r.l., Turin, Italy). Stimulation intensity and pulse width were fixed at 2 mA and 300 μs, respectively. Each fMRI run consisted of six stimulation blocks, with 15 s of stimulation followed by 45 s of rest in each block (Figure 1A).

For DCM experiments, a separate paradigm was used in which the left forepaw was stimulated at 1, 8, and 55 Hz within a single session. Each frequency condition was repeated twice, yielding a total of six blocks. This paradigm was applied to all 12 animals of the electrical stimulation cohort.

Optogenetic stimulation of S1FL and VPL

To separate local cortical responsiveness from VPL-to-S1FL transfer, S1FL and VPL were independently activated using optogenetics. Direct S1FL activation bypassed the thalamocortical synapse, whereas direct VPL activation engaged the thalamocortical pathway. The optogenetic experiments were designed to test causal contrasts rather than to remap the entire peripheral frequency-response curve. Therefore, representative low (1 Hz), intermediate (8 Hz), and high (55 Hz) conditions were selected.

Optical stimulation was delivered through the implanted fiber using a 473-nm diode-pumped solid-state laser (CNI Laser, Changchun, China), following established optogenetic fMRI conventions.^21,22,23 Laser power was adjusted to 12 mW at the fiber tip. Trains of square light pulses (10 ms pulse width) were delivered at 1, 8, or 55 Hz. The block design matched that used for peripheral stimulation, with 15 s of stimulation followed by 45 s of rest, repeated six times.

Tuned-burst VPL stimulation

To determine whether temporally dense thalamic activation is converted into sustained cortical expression or instead compressed into an onset-dominated response, a tuned-burst stimulation paradigm was applied in the VPL-ChR2 cohort.

Each burst consisted of five light pulses (10 ms pulse width, 10 ms interpulse interval, 100 ms burst duration) delivered at carrier frequencies of 1, 4, or 8 Hz. Because the intraburst pulse frequency was held constant at 50 Hz, carrier-frequency-dependent differences in cortical responses could therefore be interpreted in relation to the temporal envelope rather than changes in intraburst pulse density.

MRI acquisition

MRI data were acquired on an 11.7-T Bruker scanner (Bruker, Karlsruhe, Germany) at the Institute of Biophysics, Chinese Academy of Sciences, using a four-element surface array coil.

Anatomical images were acquired using T2-TurboRARE sequence: field of view (FOV) = 35 × 20 mm²; echo time (TE)/repetition time (TR) = 30/5700 ms; slice thickness = 0.5 mm, matrix size = 256 × 220. Functional images were acquired using a gradient-echo EPI sequence: slices = 34; FOV = 30 × 18 mm²; TE/TR = 15/1000 ms; flip angle (FA) = 55°; slice thickness = 0.5 mm, matrix size = 100 × 60. Each fMRI run consisted of 370 volumes.

Preprocessing and first-level GLM analysis

Functional images were denoised using the NORDIC pipeline⁴⁹ and manually reoriented in SPM12. Preprocessing was performed using RABIES (v0.4.2)⁴⁸ and included N4 bias field correction⁵² and rigid-body motion correction using ANTs.⁵⁰ Functional images were then nonlinearly registered to the SIGMA rat brain template,⁵¹ resampled to 0.15-mm isotropic resolution, and spatially smoothed with a 0.3-mm FWHM Gaussian kernel. First-level analysis was performed in SPM12 using a general linear model (GLM).

ROI definition and BOLD time-course analysis

ROIs were anatomically constrained using the SIGMA atlas⁵¹ (Figure 1B). Within each atlas-defined region, a small sphere (0.5-mm radius) was centered on the group-level activation peak across all stimulation frequencies, constrained within the anatomical ROI. Because the group-level peak was derived from the same data used for time-course extraction, this procedure may modestly inflate reported effect sizes⁵³; a sensitivity analysis using purely anatomical ROI definitions (atlas-region centroids without peak selection) was therefore performed and confirmed the same qualitative frequency-tuning patterns in all key regions (Figures S1–S4).

Percent signal change was calculated relative to the 10-s prestimulus baseline. For each stimulation frequency, peak BOLD signal change (%) was defined as the maximum percent signal change during the stimulation-evoked response window. FWHM was calculated as the temporal width of the response above half-maximum amplitude. Because of the TR = 1 s sampling interval, FWHM values are discrete and were treated as such in all analyses.

Peripheral-stimulation ROI analyses were based on the electrical stimulation cohort (n = 12). Optogenetic ROI analyses were based on the S1FL-ChR2 or VPL-ChR2 cohorts (n = 6 per cohort). BOLD time courses were averaged across runs within each animal.

Late response was defined as the mean percent signal change during the final 5 s of the 15-s stimulation period. To quantify sustained downstream cortical expression of thalamic activity during peripheral stimulation, we calculated a peak-normalized sustained cortical expression index (SCEI) for each animal and frequency as the late S1FL response divided by the peak VPL response. The late S1FL/late VPL ratio was examined as a sensitivity measure.

Regional differences in frequency-response profiles were assessed using two-way repeated-measures ANOVA with Region and Frequency as within-subject factors. When a significant Region × Frequency interaction was detected, frequency effects within each ROI were examined using one-way repeated-measures ANOVA. For the optogenetic cohorts (n = 6), pairwise comparisons were performed using paired-samples t tests. Where the Shapiro-Wilk test indicated departure from normality, results were verified to be robust to the choice of parametric and non-parametric tests. All statistical decisions followed the framework detailed under Quantification and statistical analysis.

For optogenetic experiments, ROI analyses were centered on the stimulation target or the downstream cortical region. For the S1FL-ChR2 cohort, local S1FL responses were compared across 1, 8, and 55 Hz stimulation. For the VPL-ChR2 cohort, local VPL responses and downstream S1FL responses were analyzed separately. The temporal profile of the block-averaged BOLD response was quantified by comparing the initial peak with the late stimulation-period response. For direct S1FL stimulation, the 8-vs-55 Hz comparison was treated as a prespecified within-animal comparison because it directly tested whether cortex retained high-frequency responsiveness when the ascending relay was bypassed.

For tuned-burst stimulation, peak responses and mean responses during the 15-s stimulation period were quantified for VPL and S1FL. The initial response peak was defined as the maximum BOLD response during the stimulation period, and the late stimulation-period response was defined as the mean signal during the final 5 s of stimulation. To evaluate whether cortical responses were dominated by an early transient component, the initial response peak was compared with the late stimulation-period response.

Temporal dynamics characterization

To characterize temporal response profiles, block-averaged BOLD time courses were fitted with a double gamma-variate function^54,55:

$$f\left(t\right)=a{\left(t/\left(p\times q\right)\right)}^p{e}^{p-t/q}-b{\left(t/\left(r\times s\right)\right)}^r{e}^{r-t/s}$$

where t is the time variable, and a, b, p, q, r and s are the parameters for the double gamma variate function. All fitted curves were normalized to their peak amplitude, and latencies to reach 5% and 30% of the peak were calculated as measures of response onset and early rise, respectively.

Dynamic causal modeling (DCM)

Effective connectivity was estimated using DCM for fMRI as implemented in SPM12.²⁴ Four volumes of interest (VOIs) were included: VPL, Po, S1FL, and S2 (secondary somatosensory cortex), anatomically guided by the SIGMA atlas.⁵¹ For each VOI, the first eigenvariate of the BOLD time series was extracted from voxels within a 0.5-mm-radius sphere centered on the group-level activation peak.

A fully connected bilinear DCM was specified for each subject. Intrinsic connections between all regions were modeled in the A matrix. Electrical stimulation was modeled as a driving input entering VPL in the C matrix. Stimulation frequency (1, 8, and 55 Hz) was modeled as a modulatory input on all connections in the B matrix. The fully connected model served as a parent model; unnecessary parameters were pruned at the group level using Bayesian model reduction. Group-level effects were assessed using parametric empirical Bayes (PEB), Bayesian model reduction (BMR), and group-level Bayesian model selection.^56,57

Quantification and statistical analysis

ROI-based analyses and frequentist statistical tests were performed in SPSS (IBM Corp., Armonk, NY, USA).

Normality of all continuous outcome measures was assessed at the level of each Region × Frequency cell using the Shapiro-Wilk test. Measures conforming to a normal distribution are summarized as mean ± SD and analyzed with parametric tests; measures deviating from normality, or inherently discrete (FWHM), are summarized as median [IQR] and analyzed with non-parametric equivalents (Wilcoxon signed-rank test for paired comparisons).

For repeated-measures ANOVAs, sphericity was assessed using Mauchly’s test, and Greenhouse–Geisser correction was applied when sphericity was violated. Effect sizes are reported as partial eta-squared (ηp²) for ANOVA effects and as Cohen’s dz with 95% confidence intervals (parametric) or the rank-biserial correlation (rrb; non-parametric) for paired comparisons.

For non-parametric one-way repeated-measures comparisons across more than two conditions (SCEI analysis), the Friedman test was used, followed by Holm-corrected Wilcoxon signed-rank tests for post hoc comparisons. No widely accepted non-parametric equivalent exists for two-factor repeated-measures designs; therefore, two-way repeated-measures ANOVA was used as the omnibus test for both parametric and discrete measures, with post hoc comparisons following the scheme appropriate to each measure.

Statistical significance was set at p < 0.05 (two-tailed). Multiple-comparison correction was applied using the Holm step-down procedure within each family of hypothesis tests, where a family was defined as all post hoc paired comparisons following a significant omnibus ANOVA within a given ROI × experiment combination (e.g., the six pairwise comparisons among five frequencies for S1FL peak amplitude in the peripheral stimulation experiment). This experiment-wise correction controls the familywise error rate within each set of comparisons but does not control the false discovery rate across all tests reported in the paper. In figures, statistical significance is denoted by asterisks: ∗p < 0.05, ∗∗p < 0.01, and ∗∗∗p < 0.001. Where asterisks represent post hoc pairwise comparisons, these thresholds refer to two-tailed Holm-adjusted p values. The specific statistical tests corresponding to each annotation are described in the relevant figure legends.

For DCM analyses, connections with posterior probability >0.99 were considered significant, corresponding to a Bayesian threshold that provides strong evidence without additional frequentist correction.