Theta-mediated conceptual reinstatement in vmPFC precedes perceptual reinstatement in ventral visual cortex during memory recall

21 When recalling past episodes, different features of an experience, such as conceptual 22 meaning and perceptual detail, are reconstructed and reinstated across distributed 23 cortical regions. However, c urrent models of human memory remain unclear how these 24 feature-specific reinstatements unfold over time, and whether they interact 25 hierarchically during memory retrieval. Using magnetoencephalography (MEG) and 26 time-resolved encoding-retrieval cross-phase classification analysis, we compared the 27 time course of conceptual and perceptual reinstatement during cued visual recall. 28 Conceptual information decoding in the ventromedial prefrontal cortex (vmPFC) 29 preceded perceptual information decoding in the ventral visual cortex (VVC), and the 30 two were expressed in distinct frequency bands: theta (4–8 Hz) for conceptual and 31 gamma (30–40 Hz) for perceptual information. Cross-correlation and spectral Granger 32 causality analyses revealed that during reinstatement conceptual information in vmPFC 33 predicted and directed perceptual information in VVC, with theta-band information 34 flow predominantly from vmPFC to VVC. In addition, bottom-up connectivity from 35 VVC to vmPFC was expressed in the alpha band. These findings suggest that memory 36 retrieval proceeds from abstract conceptual reconstruction to the reinstatement of 37 perceptual details, mediated by theta oscillatory communication between prefrontal and 38 sensory areas. We propose a hierarchical interactive model in which vmPFC initiates 39 conceptual activation through theta-band top-down signals, while VVC provides 40 alpha-band feedback for evaluating reconstructed perceptual details. 41 42 43 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 3

44 When you recall a beautiful object that fascinated you yesterday, you may first 45 remember that it was a flower and then gradually fill in perceptual details such as its 46 color and shape1–3. This hierarchical reinstatement process during memory recall aligns 47 with theories that describe memory engrams as distributed across multiple, functionally 48 connected cortical regions, with each region corresponding to distinct features or 49 aspects of the memoranda being stored4–6. This organization enables different 50 components of memory, such as perceptual, contextual, and conceptual details, to be 51 reactivated within distributed neural ensembles that are functionally interconnected as 52 part of a unified engram complex7. Current models of memory suggest detailed 53 perceptual features are represented in posterior associative modal neocortical regions, 54 whereas abstract conceptual features are represented in more anterior a-modal regions 55 such as the ventromedial prefrontal cortex (vmPFC)8–11 (but see12). However, current 56 models of human memory are unclear regarding the time-course of memory 57 reinstatement and directionality of interactions between perceptual representations in 58 the posterior neocortex and conceptual representations in the vmPFC during episodic 59 memory retrieval. 60 Cortical reinstatement, which refers to the reactivation of cortical activity patterns 61 present during initial encoding, is considered a core neural mechanism of episodic 62 memory retrieval12–19. Functional MRI studies have shown that recalling visual scenes 63 reactivates the same regions engaged during picture encoding and reinstates the 64 fine-grained spatial patterns observed at encoding, particularly within ventral visual 65 areas (VVC) when the recalled stimuli are complex, real-world pictures14,16,20,21. 66 Moreover, the strength of reinstatement in sensory cortices predicts the vividness of 67 recollected visual memories13,16,20–23. Converging neuroimaging and 68 neuropsychological evidence also suggests that higher-order prefrontal regions, 69 particularly the vmPFC, are involved in the retrieval of conceptual 70 representations9,24–26. Together, these findings suggest that perceptual and conceptual 71 reinstatement reflect distinct yet complementary aspects of memory retrieval. The 72 reconstructive (rather than reduplicative) nature of episodic memory may depend on 73 reducing dimensionality at encoding and expanding memory codes at retrieval 74 (dimensionality transformations)27,28. This suggests that the reconstruction of vivid 75 perceptual details in the posterior cortex entails an expansion of stored compressed 76 conceptual representations in medial prefrontal cortex28,29;therefore, conceptual 77 representations should precede perceptual reinstatement30,31. 78 While functional magnetic resonance imaging (MRI) studies have characterized 79 the location of reinstatement, electrophysiological techniques such as 80 magnetoencephalography (MEG) enables tracking of both where and when 81 reinstatement unfolds with millisecond precision. Encoding and retrieval cross-phase 82 multivoxel pattern analysis (MVPA) can be used to quantify how closely voxel-wise 83 activation patterns during retrieval resemble those recorded during encoding, providing 84 a direct measure of reinstatement of the original memory trace14,20. Combining MEG 85 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 4 recordings with time-resolved source MVPA provides a unique opportunity to examine 86 the temporal dynamics of cortical reinstatement. In the present study, we used this 87 approach to directly compare the time courses of perceptual reinstatement in the VVC 88 and conceptual reinstatement in the vmPFC during memory retrieval. 89 Recent behavioral and EEG evidence supports the idea that conceptual information 90 is accessed earlier than perceptual details during memory retrieval30,31, consistent with 91 the idea of dimensionality expansion27. Specifically, high-level conceptual features can 92 be accessed more rapidly than low-level perceptual features during recall of visual 93 objects, which is the reverse of the sequence typically observed during encoding30,31. 94 Consistent with this behavioral pattern, an EEG study has shown that neural activity 95 associated with conceptual features reemerges earlier than activity linked to perceptual 96 features31. Building on these findings, we hypothesize that during visual memory 97 retrieval, conceptual reinstatement in the vmPFC would occur earlier than perceptual 98 reinstatement in the VVC. 99 Evidence from electrophysiological studies suggests that theta oscillations (4–8 100 Hz) play a key role in episodic memory retrieval32–36. For instance, theta activity has 101 been linked to successful memory recall and reinstatement of past experiences33–35. 102 Theta rhythms may enable the long-range transfer and integration of mnemonic 103 information between the hippocampus and neocortical regions33,34,37,38, an interaction 104 that is causally linked to vivid re-experiencing of episodic memory39. Previous studies 105 have shown that theta coherence increases between the vmPFC and sensory 106 association areas during retrieval, suggesting that theta oscillations mediate the 107 exchange of information along this prefrontal–sensory pathway33,40–42. Evidence from 108 human and animal studies is consistent with the vmPFC exerting top-down control 109 over reinstatement in the sensory areas, biasing or initiating retrieval to align with 110 prior knowledge and goals43–46. This long-range synchronization is believed to convey 111 mnemonic information across relevant networks, enabling the vmPFC to integrate 112 conceptual representations and guide reinstatement of perceptual details in the VVC. 113 Based on this framework, we hypothesized that theta oscillations would index the 114 interaction between vmPFC and VVC during visual memory recall. 115 To test these hypotheses, we recorded neuromagnetic activity during a 116 cross-modal automatic cued-retrieval paradigm that maximized the time-lock accuracy 117 of item recall and minimized confounding perceptual processing and memory strength. 118 Combining this paradigm with time-resolved multivoxel pattern analysis allowed us to 119 examine the timing, frequency, and direction of conceptual and perceptual 120 reinstatement with high temporal precision. We predicted that 1) conceptual 121 reinstatement in the vmPFC would occur earlier than perceptual reinstatement in the 122 VVC, 2) that conceptual reinstatement in vmPFC would correlate with and predict later 123 perceptual reinstatement in VVC, and 3) that theta oscillations would convey 124 information from vmPFC to VVC during recall. 125 126 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 5

127 Conceptual reinstatement showed an advantage in vmPFC, while perceptual 128 reinstatement showed an advantage in VVC 129 Participants learned four audiovisual pairings that crossed conceptual (architecture vs. 130 plants images; human vs. tool sounds) and perceptual (color vs. black-and-white; low- 131 vs. high-frequency) features (Figure 1). Before MEG recording, they completed 132 extensive training to ensure the sounds would obligatorily and immediately lead to 133 recall of the associated item. During MEG recording, participants completed blocks 134 that began with re-encoding of all four audiovisual stimuli, followed by cued-recall 135 trials in which either a sound cued the recall of its paired picture or a picture cued the 136 recall of its paired sound. Each recalled item was followed by a judgment of vividness. 137 Here, we test the hypothesis of format transformation during visual object episodic 138 reactivation; therefore, we only analyze auditory-cued picture recall trials. 139 First, we compared the reinstatement of conceptual and perceptual features time 140 series separately in vmPFC and VVC. Reinstatement strength was quantified with 141 encoding–retrieval cross-phase classification accuracy at each time point. We trained 142 the conceptual and perceptual classifiers using encoding phase data and tested them 143 using recall phase data (for details, see the Methods section). Higher classification 144 accuracy denotes more robust evidence for reinstatement of a feature type. 145 We found that although both features could be decoded in both regions, 146 reinstatement of conceptual features was stronger than reinstatement of perceptual 147 features in vmPFC from 110 ms to 160 ms (Fig. 2A, /g1868 /g3033/g3050/g3032 < 0.05), while perceptual 148 feature reinstatement was stronger than conceptual feature reinstatement in VVC from 149 190 ms to 230 ms (Fig. 2B, /g1868 /g3033/g3050/g3032 < 0.05). This supports models predicting that 150 conceptual features are more prominently represented in the vmPFC during object 151 recall and perceptual features are more prominently represented in posterior visual 152 association areas8,47. 153 154 Conceptual reinstatement in vmPFC precedes perceptual reinstatement in VVC 155 As shown in Figures 2A and 2B, we observe that the conceptual advantage in vmPFC 156 and the perceptual advantage in VVC appear in distinct time windows. We extracted 157 the peak latency of conceptual reinstatement in vmPFC and the peak latency of 158 perceptual reinstatement in VVC for each participant to directly test whether 159 conceptual reinstatement in vmPFC preceded perceptual reinstatement in VVC. We 160 found that conceptual reinstatement in vmPFC was significantly earlier than perceptual 161 reinstatement in VVC (Fig. 2C, average latency of conceptual reinstatement in vmPFC: 162 165ms, average latency of perceptual reinstatement in VVC: 234ms, paired two-tailed 163 t-test: t(32) = -3.052, p = 0.005). These results support our first hypothesis that 164 conceptual reinstatement in vmPFC precedes perceptual reinstatement in VVC. 165 166 Conceptual advantage in vmPFC is expressed in the theta band, while leads 167 perceptual advantage in VVC is expressed in the gamma band 168 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 6 Next, we asked whether conceptual advantage in vmPFC and perceptual advantage in 169 VVC are expressed in different frequency bands. To examine the frequency-specific 170 effect of the conceptual advantage and perceptual advantage, we first filtered the 171 broadband neural signal into multiple narrowband signals and then performed the same 172 encoding-retrieval cross-phase classification analysis on each narrowband signal (for 173 details, see the Methods section). For the statistical analyses, we averaged across four 174 leading frequency bands (theta: 4–8 Hz, alpha: 8–13 Hz, beta: 13–30 Hz, gamma: 175 30–40 Hz), and compared the time series of conceptual and perceptual reinstatement 176 separately in the vmPFC and VVC across these frequency bands. 177 In the vmPFC, we found that conceptual reinstatement only showed an advantage 178 over perceptual reinstatement in the theta band from 0 ms to 70 ms and from 110 ms to 179 160 ms (Fig. 3A, /g1868 /g3033/g3050/g3032 180 0.05). Meanwhile, in VVC, perceptual reinstatement showed an advantage over 181 conceptual reinstatement in the gamma band from 300 ms to 370 ms (Fig. 3B, /g1868 /g3033/g3050/g3032 0.05). These results 183 indicate that conceptual and perceptual reinstatement are associated with distinct 184 frequency bands, with the conceptual advantage in the vmPFC expressed in the theta 185 band and the perceptual advantage in VVC reflected in the gamma band. 186 187 Conceptual reinstatement in vmPFC directly influenced perceptual reinstatement 188 in VVC 189 Based on the finding that conceptual reinstatement in vmPFC preceded perceptual 190 reinstatement in VVC, we hypothesized that conceptual reinstatement in vmPFC 191 directly influenced perceptual reinstatement in VVC. We performed informational 192 cross-correlation analysis, measuring the similarity between the two time series or 193 signals as a function of the time lag between them from 10 ms to 150 ms in 10 ms steps. 194 This analysis allows for the examination of the directed correlation between two 195 reinstatement time series. 196 We compared the correlation coefficients at each time lag k between 197 /g1870 /g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g4666/g1863/g4667 (conceptual reinstatement precedes perceptual reinstatement) 198 and /g1870 /g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g4666/g1863/g4667 (perceptual reinstatement precedes conceptual 199 reinstatement) with cluster-based correction. This comparison indicated that 200 /g1870 /g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922 was significantly greater than /g1870 /g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922 in time lags 201 ranging from 70 to 90 ms (Fig. 4A, /g1868 /g3033/g3050/g3032 > 0.05). Next, we extracted the peak 202 conceptual classification accuracy in vmPFC and the peak perceptual classification 203 accuracy in VVC, each identified at subject-specific peak latencies. We found that 204 stronger conceptual reinstatement was significantly linked to later perceptual 205 reinstatement (Fig. 4B, /g1842/g1857/g1853/g1870/g1871/g1867/g1866 /g1855/g1867/g1870/g1870/g1857/g1864/g1853/g1872/g1861/g1867/g1866 /g1870 /g3404 0.36; /g1868 /g3404 0.039 ). These results 206 support our hypothesis that conceptual reinstatement in vmPFC predicts perceptual 207 reinstatement in VVC. 208 209 vmPFC drives VVC activity during retrieval via theta-band oscillations 210 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 7 Lastly, we tested the third hypothesis that theta frequency plays an important role in 211 conveying information from vmPFC to VVC using spectral state-space Granger 212 causality (GC) analysis. We computed the spectral GC estimate from vmPFC to VVC 213 and from VVC to vmPFC across frequencies ranging from 2 to 40 Hz using trial-wise 214 neural activities during cued memory recall. As in the frequency-specific 215 encoding-retrieval cross-phase classification analysis, we also averaged the spectral 216 GC estimates across four main frequency bands (theta: 4–8 Hz, alpha: 8–13 Hz, beta: 217 13–30 Hz, and gamma: 30–40 Hz). Then, we compared the spectral GC estimate from 218 vmPFC to VVC with the spectral GC estimate from VVC to vmPFC within each 219 frequency band. 220 We found that the spectral GC estimate from vmPFC to VVC was significantly 221 stronger than the estimate from VVC to vmPFC in theta neural oscillation (Fig. 5, t(32) 222 = 2.89, p = 0.007). This indicates that information flow is predominantly from vmPFC 223 to VVC in the theta band during cued picture retrieval. These findings support our 224 fourth hypothesis that information flow from vmPFC to VVC is mediated by theta 225 oscillations during picture memory recall. Unexpectedly, we found that the spectral GC 226 estimate from VVC to vmPFC exceeded the spectral GC estimate from vmPFC to VVC 227 in alpha oscillation (Fig. 5, t(32) = -2.71, p = 0.011). In summary, during visual 228 memory retrieval, vmPFC and VVC interact bidirectionally, with coupling from 229 vmPFC to VVC emphasized in the theta band and coupling from VVC to vmPFC 230 emphasized in the alpha band. 231 232 233 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 8

234 Our study provides converging evidence that conceptual and perceptual reinstatement 235 during visual memory recall are temporally, spectrally and spatially dissociable yet 236 functionally and directionally interdependent. Using time-resolved MEG source 237 encoding-retrieval cross-phase classification analysis, we found that conceptual 238 reinstatement in the vmPFC preceded perceptual reinstatement in the VVC, that these 239 reinstatements were expressed in distinct frequency bands (theta oscillation: conceptual 240 reinstatement in the vmPFC, and gamma oscillation: perceptual reinstatement in the 241 VVC, respectively), and that information flowed predominantly from vmPFC to VVC 242 through theta-band oscillations. Together, these findings support the view that memory 243 retrieval proceeds from abstract conceptual reconstruction to the reactivation of 244 perceptual detail, mediated by theta oscillatory communication between vmPFC and 245 sensory areas. 246 The temporal precedence of vmPFC over VVC supports the notion that memory 247 retrieval is initiated by the reinstatement of high-level conceptual information that 248 subsequently guides perceptual reconstruction. This finding directly supports and 249 extends the current models, including Trace Transformation Theory (TTT), which 250 proposes that conceptual and perceptual information are stored as complementary 251 traces in anterior and posterior cortical systems, respectively8,9,47–49. Extending these 252 models, our data demonstrate how different formats of these memory traces reflecting 253 different features are dynamically coordinated in both the time and frequency domains. 254 Furthermore, conceptual reinstatement in vmPFC may thus provide a top-down 255 scaffold that constrains and refines perceptual reconstruction in VVC. This hierarchical 256 temporal structure refines the models of episodic retrieval by showing that conceptual 257 information is encoded as a distinct memory trace early7 in retrieval, rather than 258 extracted over time47,49, and that its reinstatement actively drives the reactivation of 259 perceptual representations rather than passively co-occurring with them. 260 The temporal difference observed in this study is consistent with previous 261 behavioral and EEG findings showing that conceptual features are retrieved faster than 262 perceptual features30,31. This suggests that even within the same object, features at 263 different representational levels vary in their accessibility during retrieval, consistent 264 with models that suggest dimensionality transformation during the encoding and 265 retrieval of episodic memory27,30,31. These different codes are represented in distinct 266 cortical regions, with conceptual information in the anterior vmPFC and perceptual 267 information in the posterior sensory cortices8,9. Here, we show a mechanism by which 268 representations at different hierarchical levels could interact during memory retrieval. 269 Perceptual-to-conceptual transformation is an important process that helps encode 270 and store information into long-term memory27,30,50–52. It can reduce computational 271 load by compressing complex sensory input to simplified storable codes, thus 272 maximizing storage capacity. Dimensionality reduction also allows us to generalize 273 across experiences and integrate new information into existing knowledge 274 structures27,50. Our findings, consistent with previous research30,31, suggest that 275 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 9 memory retrieval operates in the reverse direction of this process, namely as 276 conceptual-to-perceptual reinstatement. The informational cross-correlation results 277 indicated that during recall, abstract conceptual traces are first reactivated within the 278 vmPFC, guiding the reinstatement of perceptual details in the VVC. This interpretation 279 is further supported by the correlation analysis, which shows that stronger conceptual 280 reinstatement in an earlier time window predicted more robust later perceptual 281 reinstatement. Moreover, previous electrophysiological studies with patients who have 282 focal damage to the vmPFC suggest that oscillatory coupling between vmPFC and 283 posterior cortices is disrupted, leading to impaired encoding of episodic memory24 and 284 retrieval of autobiographical information54. Together, these findings provide neural 285 evidence consistent with the TTT, indicating that episodic memory may emerge from a 286 dynamic interplay between conceptual abstraction and perceptual reconstruction across 287 cortical hierarchies. Future studies using methods such as transcranial magnetic 288 stimulation are necessary to determine the causal influence of vmPFC conceptual 289 activity on VVC perceptual reinstatement in episodic memory. 290 We further found that conceptual reinstatement in vmPFC was expressed in the 291 theta band, whereas perceptual reinstatement in VVC was expressed in the gamma 292 band. This frequency dissociation aligns with the proposal that theta and gamma 293 rhythms serve complementary mnemonic roles36,54,55. Previous studies have 294 consistently reported that increased theta power or coherence is associated with 295 successful memory performance33,35,55–57. Because many of these studies employed 296 recognition or associative memory paradigms that emphasize retrieval of gist- or 297 schema-like information, theta oscillations may primarily support the conceptual or 298 relational aspects of memory. In comparison, gamma oscillations are more closely 299 related to fine-grained sensory processing and perceptual encoding58–61. Therefore, 300 reconstructing perceptual details during recall may elicit gamma-band activity. Theta 301 oscillations may index synchronization within a distributed cortical–hippocampal 302 network to reinstate abstract relational structures, whereas gamma oscillations may 303 locally encode and reconstruct fine-grained sensory details. Such a theta-gamma 304 division of labor suggests that conceptual and perceptual aspects of episodic memory 305 are maintained in distinct yet coordinated neural codes. 306 Cross-correlation and Granger causality analyses further revealed that conceptual 307 reinstatement in vmPFC predicted perceptual reinstatement in VVC, with theta-band 308 information flow predominantly from vmPFC to VVC. These results indicate a 309 top–down control mechanism through which vmPFC may orchestrate sensory 310 reactivation by propagating mnemonic predictions or templates that guide the 311 reinstatement of perceptual detail. This interpretation is consistent with intracranial and 312 MEG evidence showing enhanced theta coherence between hippocampus, vmPFC, and 313 sensory cortices during successful recall40–42. Theta oscillations may thus provide the 314 temporal framework that coordinates large-scale reinstatement across conceptual and 315 perceptual systems. 316 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 10 Interestingly, we also observed feedback from VVC to the vmPFC in the alpha 317 band, suggesting that once perceptual details are reinstated, sensory regions may 318 provide feedback to higher-order areas for evaluation and integration. Such alpha-band 319 feedback aligns with models of predictive coding, where alpha synchronization 320 supports feedback signalling of reconstructed sensory evidence62,63. 321 An important limitation of both the present and previous studies is that only a 322 single conceptual dimension (plant versus architecture in the present study, animate 323 versus inanimate in earlier work) and a single perceptual dimension (color versus 324 black-and-white here, photo versus drawing previously) were examined. One possible 325 contributor to the observed retrieval-time difference is the unequal discriminability or 326 representational distance of features within conceptual and perceptual dimensions. For 327 instance, animate and inanimate categories may be more separable in conceptual space 328 than photos and drawings are in perceptual space. However, it is inherently difficult to 329 balance feature distances across conceptual and perceptual domains. Future studies 330 should include multiple conceptual and perceptual dimensions to achieve better 331 generalization and to determine whether retrieval timing differences persist across a 332 broader range of feature contrasts. For example, lower-level perceptual contrasts could 333 include different brightness or contrast levels, whereas higher-level conceptual 334 contrasts could include contextual dimensions (e.g., indoor vs. outdoor items) or 335 functional categories (e.g., tools vs. non-tools). 336 Taken together, we propose a hierarchical interactive model of episodic memory 337 retrieval (Figure 6) in which the vmPFC initiates conceptual reinstatement through 338 theta-mediated top-down signals that guide and constrain the subsequent 339 gamma-mediated reinstatement of perceptual details in the VVC. In turn, the VVC 340 feeds back information to the vmPFC through alpha oscillations, supporting the 341 evaluation of the reconstructed object and the integration of features across 342 representational levels. This temporal and spectral cascade bridges behavioral models 343 of the reverse retrieval hierarchy with neurophysiological mechanisms of large-scale 344 cortical communication30,31. Within this framework, the vmPFC may serve as a central 345 hub that integrates hippocampal outputs and transmits them to sensory cortices, 346 allowing for the reconstruction of vivid episodic experiences. Future research should 347 investigate how the vmPFC and posterior hippocampus interact to influence the 348 reinstatement of perceptual details in posterior sensory regions. 349 350 351 352 353 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 11

354 Participants 355 Thirty-three young adults (22.58 ± 4.54 years old, 18 female) participated in this study. 356 Sample size was determined through a priori power analysis using G*Power version 357 3.1.9.764. Thirty-three participants were sufficient to detect a median effect size (0.5) 358 with 80% power using a paired, two-tailed t-test64. All participants were healthy 359 right-handed individuals with no history of neurological disorder and normal hearing 360 and vision. All participants provided written informed consent approved by the 361 Baycrest Research Ethics Board (Approval No. 24-33). 362 363 Stimuli 364 The stimuli comprised four pictures and four sounds. The four pictures and sounds had 365 two orthogonal features, conceptual (pictures: architecture and plant pictures; sounds: 366 human and tool sounds) and perceptual (pictures: color and black-and-white pictures; 367 sounds: low- and high-frequency sounds). Four pictures and sounds were combined to 368 form four paired audiovisual stimuli. The duration of the audiovisual stimuli was two 369 seconds. Pairings were pseudorandomized for each participant, while the 370 picture-to-sound mapping was fixed. 371 372 Procedure 373 We created a cross-modal automatic-retrieval cued recall paradigm to maximize 374 time-lock accuracy of item recall and to minimize the perception-recall process 375 confound. Prior to the MEG scanning sessions, participants underwent a training phase 376 to familiarize themselves with the four audiovisual stimuli. Each participant completed 377 five training blocks. In each training block, there were twenty-four encoding trials with 378 each audiovisual stimulus presented six times at random. Participants completed eight 379 test trials, including four auditory-cued visual item recall and four visual-cued auditory 380 item recall. In auditory-cued visual item recall, participants were instructed to recall the 381 corresponding picture target as soon as the sound cue was presented, and vice versa for 382 visual-cued auditory item recall. After five blocks, participants could recall the target 383 automatically at cue onset, yielding high time-locking accuracy. After the training 384 phase, participants were instructed to close their eyes and rest for about thirty minutes. 385 In the MEG recording session, participants completed a six-block task. In each 386 block, twenty-four encoding trials with each audiovisual stimulus presented six times 387 were randomly presented at the beginning. Participants were instructed to passively 388 perceive the audiovisual pairs. Audiovisual stimuli last 2 s. Inter-trial intervals (ITI) of 389 encoding trials were 1.0–1.5 s in 0.1 s steps. Then, there were twenty-four 390 auditory-cued visual item recall trials and twenty-four visual-cued auditory item recall 391 trials. Each picture target and sound target was recalled six times in each block. After 392 recalling pictures or sounds, participants provided a vividness rating by pressing the left 393 or right button box (left: vivid; right: not vivid). The interval between cued recall and 394 vividness rating was 1 s. ITI of recall trials was 1.3–1.7 s in 0.2 s steps. Because 395 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 12 comparing auditory and visual memory retrieval was not the aim of the present study, 396 only auditory-cued picture-recall trials were analyzed. 397 After the MEG session, we collected participants’ high-resolution structural MRI 398 data for source localization. 399 400 MEG data acquisition and preprocessing 401 MEG data were recorded using a 275-channel CTF system (VSM MedTech) at a 402 sampling rate of 1200 Hz. Preprocessing was performed in Python using the 403 MNE-Python toolbox65. Continuous data were band-pass filtered between 1 and 120 404 Hz using an FIR filter. To remove artifacts, we first applied independent component 405 analysis to the broadband sensor data. Components reflecting eye blinks, saccades, or 406 cardiac artifacts were identified through visual inspection of component time courses 407 and topographies and then removed. The cleaned data were subsequently epoched 408 around stimulus onsets (−0.5 s to 2.5 s) and baseline-corrected using the −0.2 to −0.02 s 409 pre-stimulus interval. Epochs were down-sampled to 250 Hz and manually inspected to 410 reject residual artifacts. 411 412 Structural data acquisition and source localization 413 Structural MRI data were collected on a Siemens 3T Prisma scanner with a 64-channel 414 head coil. T1-weighted images were acquired using the magnetization-prepared rapid 415 acquisition gradient echo (MPRAGE) sequence (TR = 2000 ms, TE = 2.85 ms, field of 416 view = 256 × 240 mm, voxel size = 0.8 × 0.8 × 0.8 mm). 417 Structural T1-weighted MRIs were processed with FreeSurfer to reconstruct 418 cortical surfaces and define the source space66. MEG–MRI coregistration was 419 performed individually, run by run, by aligning fiducial points to anatomical landmarks 420 (nasion and bilateral preauricular points). A single-shell boundary element model was 421 then generated to model the inner-skull conductivity, from which individual forward 422 solutions were computed for each experimental run in surface space. Linearly 423 constrained minimum variance spatial filters were constructed for each run using 424 empirical data (0.01–2.0 s) and baseline (−0.5 to −0.02 s) covariance matrices. Filters 425 were built with unit-noise-gain normalization and then applied to cleaned MEG epochs 426 to obtain single-trial source time courses. Epochs with head displacements exceeding 427 10 mm were excluded. Head displacement was computed from the circumcenter of 428 three localization coils per epoch. Source estimates were cropped to −0.4–2.2 s, 429 band-pass filtered (1–40 Hz), and down-sampled to 100 Hz. All source data were 430 computed in native space and parcellated according to the Destrieux cortical atlas 431 (aparc.a2009s) for subsequent ROI-based decoding analyses67. 432 433 Encoding-retrieval cross-phase classification analysis 434 We quantified reinstatement between perception and retrieval using a ROI-wise, 435 time-resolved cross-phase decoding analysis on source estimates. Based on our 436 hypotheses, vmPFC ROIs included left and right G_subcallosal, G_orbital, G_rectus, 437 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 13 S_suborbital, G_and_S_cingul-Ant, G_and_S_transv_frontopol, 438 G_and_S_frontomargin and VVC ROIs included left and right 439 G_oc-temp_lat-fusiform, S_oc-temp_lat,S_oc-temp_med_and_Lingual, as defined in 440 the Destrieux (aparc.a2009s) atlas. For each participant, single-trial LCMV source time 441 courses were extracted for each vertex within the ROIs. Within each ROI, we formed 442 feature vectors from vertex-wise source amplitudes at each time point (one vector per 443 trial). Linear support vector machine (SVM) classifiers were trained using data during 444 the encoding phase for each time point. Two kinds of SVM classifiers were trained. 445 Conceptual classifiers determined whether the picture was a plant or an architecture 446 picture, while perceptual classifiers classified whether the picture was a color or 447 black-and-white picture. Then, we examined the conceptual and perceptual 448 reinstatement using conceptual and perceptual classifiers, which were tested on 449 picture-retrieval trials for each time point. Conceptual and perceptual classification 450 accuracy time series were averaged across vmPFC ROIs and VVC ROIs, and time 451 series were temporally smoothed with a Gaussian kernel (FWHM = 25 ms). We 452 interpret higher cross-phase accuracy as stronger reinstatement of category-specific 453 perceptual and conceptual representations during picture retrieval. SVM classification 454 analysis was performed using the scikit-learn toolbox in Python68. 455 In the group analysis, we exported conceptual and perceptual reinstatement time 456 series for each participant from the cross-phase classification analysis. The analyzed 457 time window was from 0 to 400 ms. Conceptual and perceptual reinstatement strength 458 were contrasted for each time point within the time window using paired t-tests 459 separately for vmPFC and VVC. Cluster-based permutation testing was performed to 460 control for multiple comparisons (cluster-forming p < 0.05; α = 0.05; 2,000 461 permutations). The results were visualized using the seaborn library69. Because we used 462 an automatic cued-recall paradigm, participants heard sounds during both encoding and 463 retrieval, and these trials served as the training and testing sets for the classification 464 analysis. Importantly, the results are unlikely to reflect auditory perception, because 465 sound–picture pairings were randomized across participants. The classification was 466 categorical. Therefore, any given sound item was paired with different perceptual or 467 conceptual picture categories across participants. As a result, group-level differences in 468 classification performance reflect differences in perceptual or conceptual picture 469 category processing rather than sound category processing. 470 471 Frequency-specific encoding-retrieval cross-phase classification analysis 472 We filtered the broadband source time series from 2 to 40 Hz in 0.5 Hz steps. Then, the 473 same procedure as the encoding-retrieval cross-phase classification analysis was 474 performed on narrowband time series at each frequency. The averaged vmPFC and 475 VVC frequency-specific reinstatement time series were further averaged across four 476 frequency bands: theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and gamma (30–40 477 Hz). The same group analysis comparing conceptual and perceptual reinstatement with 478 cluster-based correction was performed for each frequency band. 479 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 14 480 Conceptual and perceptual reinstatement latency comparison 481 Conceptual and perceptual reinstatement latencies for each participant were defined as 482 the peak latencies within the 0–400 ms range. Paired t-tests were performed to compare 483 the peak latencies between conceptual and perceptual reinstatement. 484 485 Informational cross-correlation analysis 486 We performed informational cross-correlation analysis to test lead–lag relationships 487 between conceptual and perceptual reinstatement. We computed lagged 488 cross-correlations on individual ROI-averaged conceptual reinstatement time series in 489 the vmPFC and perceptual reinstatement time series in the VVC, using 15 lags (10–150 490 ms; with 10 ms steps at 100 Hz). A directionality index D(k) was defined at each lag k 491 as the difference between /g1870 /g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g4666/g1863/g4667 and /g1870 /g2926/g2915/g2928/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g1372/g2913/g2925/g2924/g2913/g2915/g2926/g2930/g2931/g2911/g2922/g4666/g1863/g4667 . 492 Group inference on D(k) used a two-tailed t-test with permutation testing across 493 subjects with cluster-based correction over the lag dimension (cluster-forming p < 0.05; 494 cluster α = 0.05; 2,000 permutations). 495 496 Spectral state-space Granger causality analysis 497 We tested directed interactions between vmPFC and VVC during retrieval using 498 spectral state-space Granger causality on time-reversed single-trial source estimates 499 during picture retrieval70. Within each ROI, vertex activity was summarized using the 500 mne.extract_label_time_course function with “pca_flip” mode. Spectral time-reversed 501 GC was computed using vmPFC→ VVC and VVC→ vmPFC index sets 502 (mne_connectivity.spectral_connectivity_epochs). For each participant, we obtained 503 GC spectra (vmPFC→ VVC and VVC→ vmPFC) across 2 to 40 Hz. Similar to the 504 frequency-specific cross-phase classification analysis, we also averaged GC estimates 505 across four frequency bands: theta (4–8 Hz), alpha (8–13 Hz), beta (13–30 Hz), and 506 gamma (30–40 Hz). Then, we performed a paired t-test between GC (vmPFC→ VVC) 507 and GC (VVC→ vmPFC) to examine whether the information flow during picture 508 retrieval is from vmPFC to VVC or vice versa. 509 510 511 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 15 Acknowledgments 512 Funding: This work was supported by 513 Competing interests: The authors declare that they have no competing financial 514 interests. 515 Data and materials availability: The paper and/or the supplementary materials 516 contain all the data needed to evaluate the conclusions. The behavioral and fMRI data 517 that support this study's findings will be available on OSF. 518 519 520 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 16

521 1. V ogt, S. & Magnussen, S. Expertise in Pictorial Perception: Eye-Movement Patterns and 522 Visual Memory in Artists and Laymen. Perception 36, 91–100 (2007). 523 2. Wing, E. A., Burles, F., Ryan, J. D. & Gilboa, A. The structure of prior knowledge enhances 524 memory in experts by reducing interference. Proc. Natl. Acad. Sci. U.S.A. 119, e2204172119 525 (2022). 526 3. Konkle, T., Brady, T. F., Alvarez, G. A. & Oliva, A. Conceptual distinctiveness supports 527 detailed visual long-term memory for real-world objects. Journal of Experimental Psychology: 528 General 139, 558–578 (2010). 529 4. Josselyn, S. A., Köhler, S. & Frankland, P. W. Finding the engram. Nat Rev Neurosci 16, 530 521–534 (2015). 531 5. Josselyn, S. A. & Tonegawa, S. Memory engrams: Recalling the past and imagining the future. 532 Science 367, eaaw4325 (2020). 533 6. Teyler, T. J. & DiScenna, P. The hippocampal memory indexing theory. Behavioral 534 Neuroscience 100, 147–154 (1986). 535 7. Hebscher, M., Wing, E., Ryan, J. & Gilboa, A. Rapid Cortical Plasticity Supports Long-Term 536 Memory Formation. Trends in Cognitive Sciences 23, 989–1002 (2019). 537 8. Moscovitch, M., Cabeza, R., Winocur, G. & Nadel, L. Episodic Memory and Beyond: The 538 Hippocampus and Neocortex in Transformation. Annu. Rev. Psychol. 67, 105–134 (2016). 539 9. Sekeres, M. J., Winocur, G. & Moscovitch, M. The hippocampus and related neocortical 540 structures in memory transformation. Neuroscience Letters 680, 39–53 (2018). 541 10. Mesulam, M. M. From sensation to cognition. Brain 121, 1013–1052 (1998). 542 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 17 11. Ranganath, C. & Ritchey, M. Two cortical systems for memory-guided behaviour. Nat Rev 543 Neurosci 13, 713–726 (2012). 544 12. Bone, M. B., Ahmad, F. & Buchsbaum, B. R. Feature-specific neural reactivation during 545 episodic memory. Nat Commun 11, 1945 (2020). 546 13. Dijkstra, N., Bosch, S. E. & Gerven, M. A. J. van. Vividness of Visual Imagery Depends on 547 the Neural Overlap with Perception in Visual Areas. J. Neurosci. 37, 1367–1373 (2017). 548 14. Xue, G. The Neural Representations Underlying Human Episodic Memory. Trends in 549 Cognitive Sciences 22, 544–561 (2018). 550 15. Wheeler, M. E., Petersen, S. E. & Buckner, R. L. Memory’s echo: Vivid remembering 551 reactivates sensory-specific cortex. Proceedings of the National Academy of Sciences 97, 552 11125–11129 (2000). 553 16. Ritchey, M., Wing, E. A., LaBar, K. S. & Cabeza, R. Neural Similarity Between Encoding 554 and Retrieval is Related to Memory Via Hippocampal Interactions. Cerebral Cortex 23, 555 2818–2828 (2013). 556 17. Bartlett, F. C. & Kintsch, W. Remembering: A Study in Experimental and Social Psychology . 557 (Cambridge University Press, 1995). doi:10.1017/CBO9780511759185. 558 18. Schacter, D. L. & Addis, D. R. The cognitive neuroscience of constructive memory: 559 remembering the past and imagining the future. Philosophical Transactions of the Royal 560 Society B: Biological Sciences 362, 773–786 (2007). 561 19. Schacter, D. L., Norman, K. A. & Koutstaal, W. The cognitive neuroscience of constructive 562 memory. in False-memory creation in children and adults: Theory, research, and 563 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 18 implications 129–168 (Lawrence Erlbaum Associates Publishers, Mahwah, NJ, US, 2000). 564 20. Zhang, L., Alain, C. & Buchsbaum, B. R. Differential reliance on sensory reinstatement and 565 internally transformed representation during vivid retrieval of visual and auditory episodes. 566 Preprint at https://doi.org/10.1101/2025.06.16.659408 (2025). 567 21. Gordon, A. M., Rissman, J., Kiani, R. & Wagner, A. D. Cortical Reinstatement Mediates the 568 Relationship Between Content-Specific Encoding Activity and Subsequent Recollection 569 Decisions. Cereb Cortex 24, 3350–3364 (2014). 570 22. Barker, R. M., St-Laurent, M. & Buchsbaum, B. R. Neural reactivation and judgements of 571 vividness reveal separable contributions to mnemonic representation. NeuroImage 255, 572 119205 (2022). 573 23. Heinbockel, H., Wagner, A. D. & Schwabe, L. Post-retrieval stress impairs subsequent 574 memory depending on hippocampal memory trace reinstatement during reactivation. Science 575 Advances 10, eadm7504 (2024). 576 24. Bowman, C. R. & Zeithamova, D. Abstract Memory Representations in the Ventromedial 577 Prefrontal Cortex and Hippocampus Support Concept Generalization. J. Neurosci. 38, 578 2605–2614 (2018). 579 25. Ghosh, V. E., Moscovitch, M., Colella, B. M. & Gilboa, A. Schema Representation in Patients 580 with Ventromedial PFC Lesions. J. Neurosci. 34, 12057–12070 (2014). 581 26. Giuliano, A. E., Bonasia, K., Ghosh, V . E., Moscovitch, M. & Gilboa, A. Differential 582 Influence of Ventromedial Prefrontal Cortex Lesions on Neural Representations of Schema 583 and Semantic Category Knowledge. J Cogn Neurosci 33, 1928–1955 (2021). 584 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 19 27. Kerrén, C., Reznik, D., Doeller, C. F. & Griffiths, B. J. Exploring the role of dimensionality 585 transformation in episodic memory. Trends in Cognitive Sciences 29, 614–626 (2025). 586 28. Mack, M. L., Preston, A. R. & Love, B. C. Ventromedial prefrontal cortex compression 587 during concept learning. Nat Commun 11, 46 (2020). 588 29. Bein, O. & Niv, Y . Schemas, reinforcement learning and the medial prefrontal cortex. Nat. 589 Rev. Neurosci. 26, 141–157 (2025). 590 30. Lifanov, J., Linde-Domingo, J. & Wimber, M. Feature-specific reaction times reveal a 591 semanticisation of memories over time and with repeated remembering. Nat Commun 12, 592 3177 (2021). 593 31. Linde-Domingo, J., Treder, M. S., Kerrén, C. & Wimber, M. Evidence that neural information 594 flow is reversed between object perception and object reconstruction from memory. Nat 595 Commun 10, 179 (2019). 596 32. Guth, T. A. et al. Theta-phase locking of single neurons during human spatial memory. Nat 597 Commun 16, 7402 (2025). 598 33. Herweg, N. A., Solomon, E. A. & Kahana, M. J. Theta Oscillations in Human Memory. 599 Trends in Cognitive Sciences 24, 208–227 (2020). 600 34. Kota, S., Rugg, M. D. & Lega, B. C. Hippocampal Theta Oscillations Support Successful 601 Associative Memory Formation. J. Neurosci. 40, 9507–9518 (2020). 602 35. Rudoler, J. H., Herweg, N. A. & Kahana, M. J. Hippocampal Theta and Episodic Memory. J. 603 Neurosci. 43, 613–620 (2023). 604 36. Ursino, M. & Pirazzini, G. Theta–gamma coupling as a ubiquitous brain mechanism: 605 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 20 implications for memory, attention, dreaming, imagination, and consciousness. Current 606 Opinion in Behavioral Sciences 59, 101433 (2024). 607 37. Anderson, K. L., Rajagovindan, R., Ghacibeh, G. A., Meador, K. J. & Ding, M. Theta 608 Oscillations Mediate Interaction between Prefrontal Cortex and Medial Temporal Lobe in 609 Human Memory. Cereb Cortex 20, 1604–1612 (2010). 610 38. Backus, A. R., Schoffelen, J.-M., Szebényi, S., Hanslmayr, S. & Doeller, C. F. 611 Hippocampal-Prefrontal Theta Oscillations Support Memory Integration. Current Biology 26, 612 450–457 (2016). 613 39. Hebscher, M., Meltzer, J. A. & Gilboa, A. A causal role for the precuneus in network-wide 614 theta and gamma oscillatory activity during complex memory retrieval. eLife 8, e43114 615 (2019). 616 4 0. B ar r y , D. N., B ar n e s , G . R. , C l ar k , I . A . & Ma gu i r e , E. A . T h e Ne ur al Dy na mi cs of Nov e l 617 Scene Imagery. J. Neurosci. 39, 4375–4386 (2019). 618 41. Fuentemilla, L., Barnes, G. R., Düzel, E. & Levine, B. Theta oscillations orchestrate medial 619 temporal lobe and neocortex in remembering autobiographical memories. NeuroImage 85, 620 730–737 (2014). 621 42. Kaplan, R. et al. Medial prefrontal theta phase coupling during spatial memory retrieval. 622 Hippocampus 24, 656–665 (2014). 623 43. Barron, H. C. et al. Unmasking Latent Inhibitory Connections in Human Cortex to Reveal 624 Dormant Cortical Memories. Neuron 90, 191–203 (2016). 625 44. Tomita, H., Ohbayashi, M., Nakahara, K., Hasegawa, I. & Miyashita, Y. Top-down signal 626 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 21 from prefrontal cortex in executive control of memory retrieval. Nature 401, 699–703 (1999). 627 45. Guise, K. G. & Shapiro, M. L. Medial Prefrontal Cortex Reduces Memory Interference by 628 Modifying Hippocampal Encoding. Neuron 94, 183-192.e8 (2017). 629 46. McCormick, C., Barry, D. N., Jafarian, A., Barnes, G. R. & Maguire, E. A. vmPFC Drives 630 Hippocampal Processing during Autobiographical Memory Recall Regardless of Remoteness. 631 Cereb Cortex 30, 5972–5987 (2020). 632 47. Robin, J. & Moscovitch, M. Details, gist and schema: Hippocampal–neocortical interactions 633 underlying recent and remote episodic and spatial memory. Current Opinion in Behavioral 634 Sciences 17, 114–123 (2017). 635 48. Mesulam, M.-M. Behavioral Neuroanatomy Large-Scale Networks, Association Cortex, 636 Frontal Syndromes, the Limbic System, and Hemispheric Specializations. in Principles of 637 Behavioral and Cogn itive Neurology (ed. Mesulam, M.-M.) 0 (Oxford University Press, 638 2000). doi:10.1093/oso/9780195134759.003.0001. 639 49. Winocur, G. & Moscovitch, M. Memory Transformation and Systems Consolidation. Journal 640 of the International Neuropsychological Society 17, 766–780 (2011). 641 50. Favila, S. E., Lee, H. & Kuhl, B. A. Transforming the Concept of Memory Reactivation. 642 Trends in Neurosciences 43, 939–950 (2020). 643 51. Gloede, M. E. & Gregg, M. K. The fidelity of visual and auditory memory. Psychon Bull Rev 644 26, 1325–1332 (2019). 645 52. Clarke, A., Taylor, K. I., Devereux, B., Randall, B. & Tyler, L. K. From Perception to 646 Conception: How Meaningful Objects Are Processed over Time. Cereb Cortex 23, 187–197 647 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 22 (2013). 648 53. Gilboa, A. & Moscovitch, M. Ventromedial prefrontal cortex generates pre-stimulus theta 649 coherence desynchronization: A schema instantiation hypothesis. Cortex 87, 16–30 (2017). 650 54. Griffiths, B. J. & Jensen, O. Gamma oscillations and episodic memory. Trends in 651 Neurosciences 46, 832–846 (2023). 652 55. Mysin, I. & Shubina, L. From mechanisms to functions: The role of theta and gamma 653 coherence in the intrahippocampal circuits. Hippocampus 32, 342–358 (2022). 654 56. Min Park, Y ., Park, J., Young Kim, I., Koo Kang, J. & Pyo Jang, D. Interhemispheric theta 655 coherence in the hippocampus for successful object-location memory in human intracranial 656 encephalography. Neuroscience Letters 786, 136769 (2022). 657 57. Nyhus, E. & Curran, T. Functional role of gamma and theta oscillations in episodic memory. 658 Neuroscience & Biobehavioral Reviews 34, 1023–1035 (2010). 659 58. Welle, C. G. & Contreras, D. Sensory-driven and spontaneous gamma oscillations engage 660 distinct cortical circuitry. J Neurophysiol 115, 1821–1835 (2016). 661 59. Wilming, N., Murphy, P. R., Meyniel, F. & Donner, T. H. Large-scale dynamics of perceptual 662 decision information across human cortex. Nat Commun 11, 5109 (2020). 663 60. Lizarazu, M., Lallier, M. & Molinaro, N. Phase−amplitude coupling between theta and 664 gamma oscillations adapts to speech rate. Annals of the New York Academy of Sciences 1453, 665 140–152 (2019). 666 61. Busch, N. A., Dubois, J. & VanRullen, R. The Phase of Ongoing EEG Oscillations Predicts 667 Visual Perception. J. Neurosci. 29, 7869–7876 (2009). 668 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 23 62. Michalareas, G. et al. Alpha-Beta and Gamma Rhythms Subserve Feedback and Feedforward 669 Influences among Human Visual Cortical Areas. Neuron 89, 384–397 (2016). 670 63. Bastos, A. M. et al. Canonical Microcircuits for Predictive Coding. Neuron 76, 695–711 671 (2012). 672 64. Faul, F., Erdfelder, E., Lang, A.-G. & Buchner, A. G*Power 3: a flexible statistical power 673 analysis program for the social, behavioral, and biomedical sciences. Behav Res Methods 39, 674 175–191 (2007). 675 65. Larson, E. et al. MNE-Python. Zenodo https://doi.org/10.5281/ZENODO.592483 (2025). 676 66. Fischl, B. FreeSurfer. NeuroImage 62, 774–781 (2012). 677 67. Destrieux, C., Fischl, B., Dale, A. & Halgren, E. Automatic parcellation of human cortical 678 gyri and sulci using standard anatomical nomenclature. NeuroImage 53, 1–15 (2010). 679 68. Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 12, 680 2825–2830 (2011). 681 69. Waskom, M. seaborn: statistical data visualization. JOSS 6, 3021 (2021). 682 70. Winkler, I., Panknin, D., Bartz, D., Muller, K.-R. & Haufe, S. Validity of Time Reversal for 683 Testing Granger Causality. IEEE Trans. Signal Process. 64, 2746–2760 (2016). 684 685 686 687 688 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 24 689 Figure 1. Experimental paradigm and analysis pipeline. (A) Four picture stimuli 690 used in the experiment were organized along two orthogonal dimensions: a perceptual 691 dimension (color vs. black-and-white) and a conceptual dimension (architecture vs. 692 plant). (B) MEG run structure. Each run began with twenty-four encoding trials in 693 which participants viewed intact audiovisual pairs. This was followed by twenty-four 694 cued-recall trials, including auditory-cued picture retrieval and picture-cued sound 695 retrieval. (C) After MEG data acquisition, sensor-level data were source-localized 696 using an LCMV beamformer. We then extracted trial-wise time series from vmPFC and 697 VVC regions of interest. Encoding–retrieval MVPA was performed at each time point 698 to quantify perceptual and conceptual reinstatement. vmPFC, ventromedial prefrontal 699 cortex; VVC, ventral visual cortex; MVPA, multivariate pattern analysis. 700 701 702 703 704 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 25 705 Figure 2. Conceptual reinstatement in vmPFC precedes perceptual reinstatement 706 in VVC. (A). vmPFC showed stronger conceptual reinstatement than perceptual 707 reinstatement from 110-160 ms (/g1868 /g3033/g3050/g3032 < 0.05). (B) VVC showed stronger perceptual 708 reinstatement than conceptual feature reinstatement from 190-230 ms (/g1868 /g3033/g3050/g3032 < 0.05). 709 (C) The onset latency of conceptual reinstatement in vmPFC was significantly earlier 710 than the onset of perceptual reinstatement in VVC (/g1868 /g3033/g3050/g3032 < 0.05). vmPFC, 711 ventromedial prefrontal cortex; VVC, ventral visual cortex; **, p < 0.01. 712 713 714 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 26 715 Figure 3. Conceptual advantage in vmPFC expressed in theta band while leads 716 perceptual advantage in VVC expressed in gamma band. (A). In vmPFC, 717 conceptual reinstatement was significantly stronger than perceptual reinstatement in 718 the theta band during the two time windows: 0–70 ms and 110–160 ms ( < 0.05). 719 (B) In contrast, VVC showed significantly stronger perceptual than conceptual 720 reinstatement in the gamma band from 300–370 ms ( < 0.05). vmPFC, 721 ventromedial prefrontal cortex; VVC, ventral visual cortex. 722 723 724 725 726 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 27 727 Figure 4. Conceptual reinstatement in vmPFC directly influenced perceptual 728 reinstatement in VVC. (A) vmPFC conceptual reinstatement exerted a direct 729 influence on VVC perceptual reinstatement during the 70–90 ms time window. (B). 730 Stronger conceptual reinstatement could significantly predict later perceptual 731 reinstatement. vmPFC, ventromedial prefrontal cortex; VVC, ventral visual cortex. 732 733 734 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 28 735 Figure 5. Bi-directional connectivity between vmPFC and VVC. vmPFC drives 736 VVC activity via theta-band top-down influences during retrieval, whereas VVC 737 provides bottom-up feedback to vmPFC through alpha-band connectivity. vmPFC, 738 ventromedial prefrontal cortex; VVC, ventral visual cortex; *, p < 0.05; **, p < 0.01. 739 740 741 742 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint 29 743 Figure 6. Hierarchical interactive model of episodic memory retrieval. During 744 episodic retrieval, conceptual information is first reactivated in vmPFC, which provides 745 high-level, schema-like constraints. This conceptual signal guides and constrains 746 perceptual reinstatement in VVC via top-down theta-band oscillatory interactions. 747 Meanwhile, VVC sends bottom-up alpha-band feedback to vmPFC, supplying 748 perceptual evidence for further evaluation and integration. Together, these bidirectional 749 interactions support the hierarchical reconstruction of vivid episodic memories. 750 vmPFC, ventromedial prefrontal cortex; VVC, ventral visual cortex. 751 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted December 20, 2025. ; https://doi.org/10.64898/2025.12.17.695049doi: bioRxiv preprint