Illustration of task and behavioral summary

A. Diagram of the delayed free recall task. B. Schematic of the predicted similarity between items across in the same list (list x) and 2 adjacent lists (list x+1, x+2). Items presented close together in time show the most representational similarity, shown by the drifting reduction in similarity perpendicular to the diagonal. TCM predicts item representations should continue to decay across lists due to increased temporal distance shown as blue transitions to black (bottom left). A boundary-oriented representation would show similarity between boundary items across multiple lists (top right). C. The average proportion of recalled events for each serial position for each participant (n=99). D. The average recall probability (left, n=99, mean±sd: 0.24 ± 0.11) and the average temporal clustering factor (TCF; right, 0.64 ± 0.10) of all study participants. E. The lagged conditional response probability (CRP) is shown for 4 lags in either direction characterizing the contiguity in our data. The error bars in (E) represent 1SEM. Error bars denote ±1SD unless otherwise noted.

Neural drift is enhanced between recalled item pairs within the lateral temporal cortex

A. Schematic of similarity analysis between item pairs in the same list. (sp: serial position; γ map: gamma spectrum, PCA-optimized, time-frequency map) B. Similarity of all item pairs regardless of recall status, averaged by subject, across all electrodes in the LTC. C. Similarity of serial position 1 to each subsequent item in a list contrasting when both items are recalled (yellow) to when both items are not recalled (red). The error bars demonstrate variation in average similarity between participants. The trend lines are for visualization purposes only, the LME used to statistically test the result is described in the results section. Error bars indicate ±1 SEM.

Figure 2—figure supplement 1. Drift rate does not differ between recalled and non-recalled items in the hippocampus and lateral temporal cortex

The first items (SP1) across adjacent lists share a boundary representation.

A. A schematic showing the comparison of SP1 to every item in the adjacent three lists. B. The mean parameter estimate of the predictor for boundary proximity contrasted by region for all item pairs (left; LTC: n=81, mean β=0.020; HC: n=76, mean β=0.022; MPL: n=83, mean β=0.045). The parameter estimate of recalled-only pairs and non-recalled pairs is also shown for the MPL (right; recalled: pink, mean β=0.022; non-recalled: orange, mean β=0.027). C. Correlation of proportion of recalled SP1 by subject to the parameter estimate for boundary proximity (ordinary least squares regress, r=0.41, p=1.1 × 10−4. D. Design of the GLM used to fit adjacent list similarity in the MPL. The response variable (similarity to list x, SP1; gray, top) shown for an example subject with the model prediction (orange, top) overlayed. The predictor variables, boundary proximity (middle) and list distance (pink, bottom), are also shown. E. Mean similarity of all serial positions to the reference item (list x, SP1) regardless of recollection (top). Below, the mean similarity of only instances where both list x, SP1 and the target item are recalled are shown for all serial positions in adjacent lists (list x+1, x+2, x+3; bottom). The error bars demonstrate variation in average similarity between participants. All error bars indicate ±1 SEM. * p < 0.05 Figure 3—figure supplement 1. Serial position 1 does not show boundary representation in the hippocampus and lateral temporal cortex.

Multiple primacy items retain similarity to the boundary representation in the MPL

A. A schematic of showing comparison of the reference and target items (top). The highlighted portion of the matrix demonstrates hypothesized trend in the boundary model presented as a part of 1E (bottom). B. Averaged similarity of only recalled item-pairs comparing the reference items: list x, SP1-3 to each item in list x+1, x+2, x+3. A representation of boundary would be supported by list x, SP2 showing more similarity SP1 in the adjacent lists rather than its own serial position. Alternatively, positional coding, the similarity of the same serial position between adjacent lists, would show greater similarity between list x, SP2 and SP2 in adjacent lists. Note the similarity of list x, SP2 and SP3 to SP1 in list x+3 rather than SP2 and SP3 in list x+3. The confidence interval demonstrates ±1SEM of variation in average similarity between participants. C. Comparison of the similarity between reference items: list x, SP2 and SP3 to the first three serial positions in the adjacent lists. The bar graphs demonstrate the similarity between SP2 and SP3 to the boundary item compared to the same serial position in the corresponding list. The bar colors reflect the magnitude of similarity as shown in (D). D. Same as (B) shown in matrix form for list x, SP1-3 and SP10-12 to demonstrate the lack of boundary similarity of end-list items. The black arrows indicate the first serial position (SP1) of the adjacent lists and emphasize the presence of boundary representation. The similarity of list x, SP2 and SP3 to the boundary is statistically tested in (C) (gray axes). All error bars indicate ±1SEM.

The last item shares similarity to other items at the end of adjacent lists in the MPL.

A. Schematic of the predicted similarity between the last item in a list list x, SP12 to each item in the adjacent lists. B. Mean similarity of all serial positions to the reference item (list x, SP12) regardless of recollection (top). This shows the similarity of end-list items to other end-list items across adjacent lists. Importantly, the last item in each list (SP12) is not similar to SP1 of the adjacent list despite the temporal proximity. The error bars demonstrate ±1SEM of variation in average similarity between participants. C. The mean parameter estimate of the predictor for boundary proximity fit to average similarity data from the MPL for recalled versus non-recalled items (left; recalled: pink, mean β=-0.006; non-recalled: orange, mean β=0.012). Additionally, the mean parameter estimate for all items between regions. (Right; MPL: n=83, mean β=0.011; LTC: n=81, mean β=0.005; HC: n=76, mean β=0.003). D. Same as B for recalled pairs only. All error bars indicate the ±1SEM.

Figure 5—figure supplement 1. End-list items show partial boundary similarity to end-list items across multiple regions

A. Similarity of serial position 1 to each subsequent item in the same list contrasting when both items are recalled (pink) to when both items are not recalled (green) for electrodes in the medial parietal lobe (LME, n=83; βSP=-0.001, CI:[-0.008, -0.003], pSP=5.6 × 10−4; interaction term: SP x recalled items=−9.2 × 10−4, CI:[-0.003, 0.004], pinteraction=0.23). B. Same as (A) except for electrodes in the hippocampus (LME, n=76; βSP=0.004, CI:[-0.007, -0.001], pSP=0.004; interaction term: SP x recalled items=-0.004, [-0.008, 0.000], pinteraction=0.054). All displayed trend lines are for visualization purposes only, the described LME was used to quantify the interaction effect. All error bars indicate ±1SEM.

A. Averaged similarity of all item-pairs comparing list x, SP1 to each item in list x+1, x+2, x+3 for electrodes in the lateral temporal cortex. B. Same as (A) for electrodes in the hippocampus. All error bars indicate ±1SEM.

A. Averaged similarity of all item-pairs comparing list x, SP12 to each item in list x+1, x+2, x+3 for electrodes in the hippocampus. B. Design of the GLM used to fit adjacent list similarity in the MPL for serial position 12. The response variable (similarity to list x, SP12; gray, top) shown for an example subject with the model prediction (orange, top) overlayed. The predictor variables, boundary proximity (middle) and list distance (pink, bottom), are also shown. C. Same as (A) for electrodes in the lateral temporal cortex. All error bars indicate ±1SEM.