Claims 35
From the paper Persistent hippocampal neural firing and hippocampal-cortical coupling predict verbal working memory load · claims file · checked 2026-10-23: ready
- behavior-accuracy-set-size
Correct response rates were 90.2 +/- 7.8% for IN and 93.1 +/- 4.3% for OUT trials, and the rate of correct responses decreased with set size, from 98.5% (set size 4) to 90.5% (6) and 84.7% (8), a significant effect in a permuted repeated-measures ANOVA over 36 sessions.
The average correct response rates were 90.2 ± 7.8% for IN and 93.1 ± 4.3% for OUT. The rate of correct responses decreased with set size from a set size of 4 (98.5% correct responses) to set sizes of 6 (90.5%) and 8 (84.7%)
Results, Task and behavior, paragraph 2; Figure 1BHow it will be judged
All trials of the 36 sessions. The IN and OUT tolerances are the SD the paper gives; the others are 10%. Open, how percentages are formed; readings (a) the mean of per-session percentages (Fig. 1B shows sessions), (b) trials pooled over sessions. The ANOVA has set size as the within factor and session as the unit (F with 2 and 70 degrees of freedom); its permutation scheme is not given, and the reading is set-size labels shuffled within each session, 4999 permutations (P = 0.0002 as 1/(n + 1), Methods, Statistics). Reproduced when, under one reading, all values are within tolerance, the order 4 above 6 above 8 holds and the ANOVA has P < 0.05.
- behavior-capacity
Across all sessions the working memory capacity (Cowan's K) averaged 6.8 letters.
Across all sessions, the capacity averaged 6.8 [Cowan's K , (correct IN rate + correct OUT rate − 1) × set size]
Results, Task and behavior, paragraph 2How it will be judged
K = (correct IN rate + correct OUT rate - 1) x set size, per session, as the Results define it, averaged over the 36 sessions. Open, which set size and rates enter; readings (a) K at set size 8 from the set size 8 rates, (b) the maximum of K over the three set sizes, (c) the rates pooled over all set sizes multiplied by 8, the largest set size, (d) the mean of K over the three set sizes. Reproduced when one reading is within tolerance.
- behavior-rt-workload
The mean response time of correct trials (1630 trials) increased with workload, by 48 ms per item, a significant effect in a permuted repeated-measures ANOVA.
The mean response time (RT) for the correct trials (1630 trials) increased with workload (48 ms per item; permuted repeated-measures ANOVA, F 2,2007 = 21.10, P = 0.0002; Fig. 1C )
Results, Task and behavior, paragraph 2; Figure 1CHow it will be judged
RT relative to probe onset (Fig. 1C legend), correct trials. Table S2 (Fig. 1c) names the test a permutation ANOVA with n the number of trials (1630) and df = (2, 2007), so the ANOVA is run on trials with set size as factor (one-way, permuted set-size labels). The F has 2007 error degrees of freedom, which does not fit the 1630 trials; its F is reported. Open, how 48 ms per item is obtained; readings (i) the slope of a line fit over all correct trials, (ii) the mean over sessions of per-session slopes, (iii) the slope over the three set-size means, (iv) the difference between set sizes 8 and 4 divided by 4. Open, per-session value for reading (ii), mean RT (the text) or median RT (Fig. 1C). Reproduced when RT rises with set size, the ANOVA has P < 0.05, and under one reading the slope and the trial count are within tolerance. The F value is reported, not judged, since its degrees of freedom cannot be matched to a stated design.
- behavior-rt-correct-incorrect
Correct IN/OUT decisions were made faster than incorrect ones (1.36 +/- 0.64 s against 1.87 +/- 1.21 s).
Correct IN/OUT decisions were made more rapidly than incorrect decisions (1.36 ± 0.64 s versus 1.87 ± 1.21 s, respectively; permutation t test, t = 5.06, P = 5.0 × 10 −5 )
Results, Task and behavior, paragraph 2How it will be judged
RT relative to probe onset. The tolerances are the SDs the paper gives. Readings (a) all trials pooled, mean +/- SD over trials, unpaired permutation t test; (b) per-session means, paired over sessions with at least one error. The permutation t test as in Methods, Statistics (2000 permutations; the printed P implies more, which only moves the minimum P). Reproduced when, under one reading, both means are within tolerance and correct RT is shorter with P < 0.05.
- neurons-recorded
1526 neurons were recorded (51 +/- 35 per session), 676 in hippocampus, 449 in entorhinal cortex and 401 in amygdala.
We recorded the activity of 1526 (51 ± 35 per session) neurons from the hippocampus ( n = 676), entorhinal cortex ( n = 449), and amygdala ( n = 401) across all microelectrodes
Results, Single-unit electrophysiology, paragraph 1How it will be judged
Counts of sorted units over all sessions, a unit seen in several sessions counted once per session (Results, same paragraph). The counts are exact, since they describe the deposited units; a difference is reported with its size. Hippocampus is the hippocampal head and body electrodes (AH and PH in the data descriptor's Table 1), entorhinal cortex EC, amygdala A. Open, the sessions in the per-session mean (1526 over 36 sessions is 42); readings (a) all 36 sessions, (b) sessions with at least one unit.
- spike-sorting-quality
The mean firing rate of all neurons was 2.55 +/- 3.30 Hz; the percentage of interspike intervals below 3 ms was 1.58 +/- 4.18%, and the ratio of the peak of the mean waveform to the SD of the noise was 3.28 +/- 1.56.
The percentage of ISI below 3 ms was 1.58 ± 4.18%, and the ratio of the peak amplitude of the mean waveform to the SD of the noise was 3.28 ± 1.56
Results, Single-unit electrophysiology, paragraph 1; Methods, Spike detection and single-unit identification; fig. S1How it will be judged
Over all 1526 units. The tolerances are the SDs the paper gives. Open, the time base of the firing rate; readings (a) spikes over the total duration of the task trials, (b) over the whole session. Open, the noise SD of the SNR; readings (a) the mean over samples of the deposited waveform SD, (b) the waveform SD at the first samples of the 2 ms window. Reproduced when, under one reading, all three means are within tolerance.
- fig2b-maintenance-neurons-hippocampus
Neurons with higher firing during maintenance than fixation (maintenance neurons) occurred in all three areas, most frequently in hippocampus, where 20.6% of the recorded neurons were maintenance neurons.
We identified maintenance neurons in all recorded areas ( Fig. 2B ); however, they occurred most frequently within the hippocampus (χ 2 1 = 47.30, P = 6.1 × 10 −12 ), where 20.6% of all recorded neurons were maintenance neurons (permutation test against scrambled data, P = 0.002)
Results, Single-unit electrophysiology, paragraph 2; Figure 2B; table S2 (Fig. 2b)How it will be judged
Correct trials. A maintenance neuron has a higher firing rate in maintenance (-3 to 0 s) than in fixation (-6 to -5 s) (fig. S2 legend) by a permutation t test (Results; Methods, Statistics), on per-trial rates, paired and unpaired as named in the opening comment, P < 0.05. Per area, the count is compared with a null from shuffling the maintenance and fixation labels within each trial and rerunning the selection ("permuted labels", Fig. 2 legend; the labels are not named, and the period labels are the only labels of this selection), at both numbers of repetitions. The chi-square has 1 degree of freedom; readings (a) a 2 x 2 table of hippocampus against entorhinal cortex and amygdala pooled, without continuity correction, (b) the same with Yates correction. Reproduced when, under one reading of each open detail, the hippocampal percentage is the highest of the three areas and within tolerance, the chi-square has P < 0.05, and the hippocampal count exceeds the null at P < 0.05; partially reproduced when hippocampus is highest and significant but the percentage is outside tolerance.
- fig2c-load-dependent-maintenance-neurons
During maintenance, 15% of hippocampal maintenance neurons (21 neurons) increased their firing rate with workload (set size 4 against set sizes 6 and 8), a significant percentage against scrambled data; the percentage was not significant in entorhinal cortex or amygdala.
During maintenance, a subset of maintenance neurons in the hippocampus (15%, 21 neurons; permutation test against scrambled data, P = 0.002) increased their firing rates systematically with workload (set size 4 versus set sizes 6 and 8; Fig. 2C )
Results, Single-unit electrophysiology, paragraph 2; Figure 2C; table S2 (Fig. 2c)How it will be judged
Maintenance neurons from fig2b-maintenance-neurons-hippocampus, correct trials, maintenance -3 to 0 s. The test of load dependence is not named; readings (a) a permutation t test of per-trial maintenance rates, set size 4 against 6 and 8 (Methods, Statistics, for two distributions), (b) a cluster-based nonparametric permutation test on the firing rate over the maintenance period, 500 ms bins at 20 ms steps (Fig. 2A legend; Methods, Statistics, for firing rates), each at P < 0.05 with the rate higher for 6 and 8. The null shuffles set size labels across trials and reruns the test ("permuted labels", Fig. 2 legend), at both numbers of repetitions. Reproduced when, under one reading, the hippocampal count is within tolerance and significant at P < 0.05 and neither other area is significant; partially reproduced when hippocampus is significant but the count is outside tolerance or another area is significant too.
- fig3e-probe-neurons-entorhinal
Neurons firing specifically after the probe letter (probe neurons) occurred in all areas, most prominently in entorhinal cortex, where 9% of neurons (41) were probe neurons.
We identified probe neurons in all areas and, most prominently, in the entorhinal cortex (χ 2 1 = 23.16, P = 1.5 × 10 −6 ; Fig. 3E ), in which 9% of neurons were probe neurons (41 neurons; permutation test against scrambled data, P = 0.001)
Results, Single-unit electrophysiology, paragraph 3; Figure 3E; table S2 (Fig. 3e)How it will be judged
Correct trials. A probe neuron has a higher firing rate after the probe than in encoding (-5 to -3 s) and than in maintenance (-3 to 0 s), each by a permutation t test at P < 0.05 (Results, same paragraph). The probe window is not given; readings (a) 0 to 1 s after probe onset, (b) probe onset to the button press of each trial (the probe is shown until the response, Methods, Task), (c) -500 to 0 ms relative to the button press, the window used for retrieval decoding (Fig. 4F legend). Paired and unpaired as in the opening comment. Null and chi-square readings as in fig2b-maintenance-neurons-hippocampus, with entorhinal cortex against the other two areas. Reproduced when, under one reading, entorhinal cortex has the highest percentage, its count is within tolerance, the chi-square has P < 0.05 and the entorhinal count exceeds the null at P < 0.05.
- fig3-probe-onset-not-movement
Peak firing of probe neurons was higher when aligned to probe onset than when aligned to the button press (5.47 +/- 4.97 Hz against 4.28 +/- 4.37 Hz), so their firing reflects the probe stimulus more than movement preparation.
Aligning the response with the probe stimulus onset resulted in significantly higher peak firing rates (mean for all probe neurons, 5.47 ± 4.97 Hz versus 4.28 ± 4.37 Hz; permutation t test, t 75 = 4.40, P = 0.0005
Results, Single-unit electrophysiology, paragraph 4; Figure 3A and 3BHow it will be judged
All probe neurons of fig3e-probe-neurons-entorhinal (76 by the reported degrees of freedom and by 8 neurons being 10.5%). Per neuron, the maximum of the trial-averaged rate in 200 ms bins at 20 ms steps (Fig. 3 legend) after each alignment; the tolerances are the SDs the paper gives; paired permutation t test over neurons (Methods, Statistics). Open, the window of the maximum; readings (a) 0 to 1 s after the alignment event, (b) 1 s before to 1 s after it. Reproduced when, under one reading, the probe-aligned peak is higher with P < 0.05 and both means are within tolerance; partially reproduced when the direction and test hold but a mean is outside tolerance.
- fig3-probe-neurons-few-task-specific
Only a few probe neurons reflected the decision (8 neurons, 10.5% of probe neurons, IN against OUT), and only one showed workload dependence.
However, only a small number of probe neurons (eight neurons, 10.5% of probe neurons) reflected the subject's decision (IN versus OUT; Fig. 3 , A and B), and only one neuron showed workload dependence
Results, Single-unit electrophysiology, paragraph 4; Figure 3A to 3DHow it will be judged
Probe neurons from fig3e-probe-neurons-entorhinal. IN against OUT, and set size 4 against 6 and 8, by a cluster-based nonparametric permutation test at P < 0.05 on the rate in 200 ms bins at 20 ms steps after probe onset (Fig. 3A and 3C legends). The time range tested is not given; readings (a) 0 to 1 s after probe onset, (b) 0 to 2 s. Reproduced when, under one reading, both counts are within tolerance (a count of one may be 0 to 2, as 10% of one neuron cannot be judged); partially reproduced when one holds.
- fig4a-dpca-workload-separation
The first three demixed principal components of the population firing rates explained 44.7% of the variance and separated set sizes 4, 6 and 8, with three angles of 120 degrees in the dPC2-dPC3 plane; the first 15 components explained 79%.
We formed a three-dimensional space from the first three demixed principal components (dPCs 1, 2, and 3, explaining 44.7% of the variance; Fig. 4A , fig. S3, and movie S1). dPCA distinguished between workloads 4, 6, and 8.
Results, Attractors in neural population dynamics, paragraph 1; Figure 4A; fig. S3How it will be judged
dPCA (Kobak et al. 2016) of all neurons, correct trials, trial-averaged per set size, from spikes in 2 ms bins smoothed with a 1000 ms Gaussian kernel and z-scored to the fixation period (Methods, dPCA of firing rates), with time and set-size marginalizations, components ordered by explained variance. Open, the scale of the z-score; readings (a) mean and SD of the fixation rate, (b) fixation mean and SD over the whole trial. Open, the dPCA regularization; readings (a) the toolbox's cross-validated lambda, (b) none. Angles are those at the centroid between the three set-size points of the dPC2-dPC3 projection, averaged over maintenance. Reproduced when, under one reading, both variances and all three angles are within tolerance; partially reproduced when the set sizes separate (all pairwise maintenance distances exceed those of label-shuffled dPCA runs) but a value is outside tolerance.
- fig4b-speed-lowest-maintenance
In the dPCA state space, the speed of the population trajectory was highest during encoding and lowest during maintenance, lower than in each other trial period, including fixation.
The speed was highest during encoding, when maintenance neurons ramped up their firing. By contrast, speed was lowest during maintenance (permutation t test against the other trial periods, P = 0.0005)
Results, Attractors in neural population dynamics, paragraph 2; Figure 4BHow it will be judged
The dPCA of fig4a-dpca-workload-separation, first 15 dPCs (Fig. 4 legend). Speed V(t) as Methods equation 2 with step 20 ms, averaged over set sizes and over each period; 200 runs of dPCA on neural trajectories subsampled at 20% with replacement (Methods). Periods fixation -6 to -5 s, encoding -5 to -3 s, maintenance -3 to 0 s; the retrieval window is not given; readings (a) 0 to 1 s after probe onset, (b) 0 to 2 s. Permutation t test of the per-run maintenance speeds against each other period. Open, the number of runs; the Methods give 200 dPCA runs and table S2 (Fig. 4b, 4c) gives n = 2000 "nr of runs"; both are run. Reproduced when, under one reading, maintenance speed is below each other period with P < 0.05 and encoding speed is the highest; partially reproduced when maintenance is lowest but encoding is not highest.
- fig4c-distance-largest-maintenance
The pairwise distance between the trajectories of different set sizes was larger during maintenance than during encoding, and the largest of the trial periods.
Analyzing the pairwise distance between trajectories that corresponded to different set sizes, we found that the distances during maintenance were larger than during encoding (permutation t test, P = 0.0005; Fig. 4C )
Results, Attractors in neural population dynamics, paragraph 2; Figure 4CHow it will be judged
As fig4b-speed-lowest-maintenance, with distance from Methods equation 1 between each of the three pairs of set-size trajectories, averaged over pairs and over each period, 200 or 2000 subsampled runs (as there). Reproduced when, under one reading, maintenance distance exceeds encoding distance with P < 0.05 and is the largest of the four periods; partially reproduced when it exceeds encoding only.
- fig4d-encoding-decodes-correct
During encoding of high-workload trials (set sizes 6 and 8), hippocampal neurons, but not entorhinal or amygdala neurons, predicted whether the subject would later respond correctly, with a median decoding accuracy of 80%.
For the encoding period (−5 to −3 s), we found that the activity of hippocampal neurons, but not neurons in other areas, could be used to decode whether the subject would later reply correctly or incorrectly in the trial. The median decoding accuracy was 80% (permutation test against scrambled data, P = 0.002; Fig. 4D )
Results, Decoding of information during the trial, paragraph 2; Figure 4DHow it will be judged
Pseudopopulation decoding with a support vector machine as in the NDT toolbox (Methods, Neural decoding): spikes in 1 ms bins smoothed in a 1500 ms window, classified at 250 ms steps, 500 resampling runs, trials of set sizes 6 and 8 (Fig. 4D legend), units from sessions with at least five incorrect responses, five cross-validation splits; null from 500 runs with scrambled labels; Bonferroni over 12 tests. Populations hippocampus, entorhinal cortex and amygdala. Open, the smoothing window shape; readings (a) a 1500 ms boxcar (NDT binning), (b) a Gaussian. Open, feature normalization; readings (a) z-score per neuron on training data (NDT's usual preprocessing), (b) none. Open, the SVM kernel; the reading is NDT's default linear kernel. Open, how "median accuracy" is taken; readings (a) median over resampling runs of the accuracy averaged over the 250 ms steps inside -5 to -3 s, (b) median over runs and steps together. Reproduced when, under one reading, hippocampus is above the null at corrected P < 0.05 with median within tolerance and neither other area is significant; partially reproduced when hippocampus holds and another area is significant too.
- fig4e-maintenance-decodes-workload
During maintenance, workload (set size 4 against 6 and 8) could be decoded from hippocampal neurons with a median accuracy of 80%, but not from neurons of the other areas.
For the maintenance period, we found that the workload (set size 4 versus set sizes 6 and 8) of each trial could be decoded from hippocampal neurons but not from neurons in the other areas (permutation test against scrambled data, P = 0.002; Fig. 4E ). The median decoding accuracy was 80%.
Results, Decoding of information during the trial, paragraph 2; Figure 4EHow it will be judged
As fig4d-encoding-decodes-correct, with correct trials, labels set size 4 against 6 and 8, all units, ten cross-validation splits, and the maintenance period -3 to 0 s. Open, class balance; readings (a) equal numbers of trials per class drawn for each neuron (NDT's requirement), (b) all trials. Reproduced when, under one reading, hippocampus is above the null at corrected P < 0.05 with median within tolerance and neither entorhinal cortex nor amygdala is significant; partially reproduced when hippocampus holds and another area is significant too.
- fig4e-workload-decoding-leave-one-subject-out
Excluding any one subject changed the hippocampal workload decoding accuracy to a value in the range 71 to 81%.
To ensure that some subjects did not drive the results, we performed a leave-one-out analysis. The exclusion of any subject changed the decoding accuracy to a value in the range [71 to 81%].
Results, Decoding of information during the trial, paragraph 2How it will be judged
The decoder of fig4e-maintenance-decodes-workload rerun nine times, each without the hippocampal units of one subject. Reproduced when all nine accuracies lie in 71 to 81% widened by 10% of each bound (63.9 to 89.1%), under a reading of fig4e that reproduced.
- fig4e-maintenance-neurons-carry-workload
Workload decoding from hippocampal maintenance neurons alone was little different from decoding from all hippocampal neurons, while excluding the maintenance neurons reduced the accuracy to 71%.
We also performed the same analysis using only maintenance neurons; there was little difference in the decoding accuracy between the use of only maintenance neurons and the use of the whole population of hippocampal neurons. Conversely, the exclusion of maintenance neurons reduced the prediction accuracy to 71%.
Results, Decoding of information during the trial, paragraph 2; Figure 4EHow it will be judged
The decoder of fig4e-maintenance-decodes-workload on (i) hippocampal maintenance neurons only and (ii) hippocampal neurons other than maintenance neurons. "Little difference" is not quantified; it is judged with the protocol's default tolerance, the maintenance-neuron median within 8 percentage points (10% of the 80% accuracy) of the all-hippocampus median, with (i) significant against its scrambled null (Fig. 4E legend, "Sufficient decoding was possible"). Reproduced when both parts hold and (ii) is within tolerance and below the all-hippocampus median.
- fig4e-no-decoding-6-vs-8
The decoder did not distinguish trials of set size 6 from set size 8.
The decoder did not distinguish between trials of set sizes 6 and 8, which points to a ceiling effect
Results, Decoding of information during the trial, paragraph 2How it will be judged
A claim of absence. The paper reports no separate test; the decoding test of the Methods is used, as in fig4e-maintenance-decodes-workload, with labels set size 6 against 8, hippocampal neurons, maintenance period, against 500 scrambled-label runs at P < 0.05 (the Bonferroni correction of the Methods covers the 12 tests of Fig. 4D to 4F, which do not include this one). Reproduced when the accuracy is not above the null; not reproduced when it is. The median accuracy is reported either way.
- fig4f-retrieval-decodes-response
During retrieval (500 ms before the button press), neurons in all recorded areas predicted the IN or OUT response; with all neurons the median decoding accuracy reached 90%.
For the retrieval period, we found that neurons from all areas encoded whether the subject would later reply by pressing the IN or OUT button in the trial (permutation test against scrambled data, P = 0.002; Fig. 4F ). The median decoding accuracy reached 90% if firing from all neurons was included
Results, Decoding of information during the trial, paragraph 2; Figure 4FHow it will be judged
As fig4d-encoding-decodes-correct, with correct trials, labels IN against OUT, a 500 ms window ending at the button press (Fig. 4F legend, -500 to 0 ms; Methods, a narrow 500 ms window), ten cross-validation splits, populations hippocampus, entorhinal cortex, amygdala and all neurons. Reproduced when, under one reading, each of the four populations is above its null at corrected P < 0.05 and the all-neuron median is within tolerance; partially reproduced when some but not all areas are significant.
- figs3d-hippocampus-loadings
Ranking the first 200 neurons of each region by their dPCA loading, hippocampal neurons had higher loadings than entorhinal and amygdala neurons at each rank except the first three, where entorhinal neurons had the highest loading.
For each rank, hippocampus (cyan) had a higher loading than entorhinal cortex (magenta) (paired Wilcoxon test P < 10-33) and amygdala (yellow), except for the first three ranks
fig. S3d legendHow it will be judged
The dPCA of fig4a-dpca-workload-separation. The legend names both dPC3 (title) and dPC2 (text); readings (a) dPC2, (b) dPC3. Open, the loading; readings (i) the absolute decoder weight of the neuron, (ii) the signed weight. Per region the 200 largest, sorted; paired Wilcoxon signed-rank test of hippocampus against entorhinal cortex over the 200 ranks. Reproduced when, under one reading, hippocampus is higher with P < 0.05 and higher than amygdala at most ranks.
- fig5-subject1-eeg-power-workload
In subject 1, scalp EEG power during the last 2 s of maintenance was higher for set size 8 than for set size 4 at Pz (8.5 to 16 Hz) and at Fz (8 to 12.5 Hz); this workload-dependent power increase did not appear in the other subjects.
the scalp EEG power increased during maintenance for a high workload (set size 8), with a peak within the 8- to 12-Hz alpha range over parietal cortex ( Fig. 5 , A and B). This workload-dependent EEG power increase extended to frontal electrode Fz ( Fig. 5C ). While this power increase is known from healthy subjects for the same or similar task ( 5 , 6 ), it did not appear in the other subjects of our study group (fig. S5 and table S1)
Results, Electrophysiology of cortico-hippocampal functional coupling, paragraph 2; Figure 5B and 5C; fig. S5; table S1How it will be judged
Last 2 s of maintenance, scalp EEG re-referenced to averaged mastoids, multitaper spectra (+/- 2 Hz smoothing, seven tapers; Methods, Frequency spectra), trials of set size 8 against 4, cluster-based nonparametric permutation test over frequency (P < 0.05). Open, the frequency range tested; readings (a) 4 to 30 Hz (the time-frequency range in Methods), (b) 1 to 40 Hz. Subject 1 reproduced when Pz and Fz each have a significant cluster of higher set size 8 power overlapping 8.5 to 16 Hz and 8 to 12.5 Hz respectively. "Other subjects": table S1 gives EEG workload dependence as no for subjects 2 to 8 and decrease for subject 9, on the channels of fig. S5, which are not named in the text (the legend cites table S7, which does not exist); the reading is the scalp electrode of each subject's maximal PLV (table S1), the columns table S1 sets side by side, with Pz and Fz shown as a variant. Reproduced when subject 1 holds and none of subjects 2 to 9 has a significant increase, as the text says; partially reproduced when subject 1 holds and another subject has an increase.
- fig5-subject1-hippocampal-power-workload
In subject 1, theta-alpha power (7.5 to 10 Hz) in the left hippocampal iEEG increased with workload during maintenance.
Theta-alpha power (7.5 to 10.0 Hz) in the left hippocampal iEEG increased with workload during maintenance.
Figure 5G and 5H; Results, Electrophysiology of cortico-hippocampal functional coupling, paragraph 2How it will be judged
Subject 1, the left hippocampal bipolar channel of the maximal PLV pair (table S1), last 2 s of maintenance, multitaper spectra as in fig5-subject1-eeg-power-workload, normalized to fixation power (Fig. 5H legend), set size 8 against 4, cluster-based test at P < 0.05. Reproduced when a significant cluster of higher set size 8 power overlaps 7.5 to 10 Hz. Table S1 also lists iEEG workload dependence as increase for subjects 1 and 6 and no for the others; this is reported for all subjects.
- fig5-subject1-neurons-workload
In subject 1, the average firing rate of the five hippocampal maintenance neurons was higher for high workload (set size 8 against 4) during encoding and maintenance.
The average firing rate of maintenance neurons in this subject was higher for a high workload during encoding and maintenance ( n = 5 hippocampal neurons with more than 100 spikes during both encoding and maintenance)
Figure 5I; Results, Electrophysiology of cortico-hippocampal functional coupling, paragraph 2How it will be judged
Subject 1, hippocampal maintenance neurons from fig2b-maintenance-neurons-hippocampus with more than 100 spikes in encoding and in maintenance. Their average rate over time, set size 8 against 4, cluster-based nonparametric permutation test at P < 0.05. Open, the unit of the test; readings (a) trials, with the per-trial rate averaged over the five neurons, (b) neuron-trial pairs. Reproduced when, under one reading, the count is 5 and significant clusters of higher set size 8 rate fall in both encoding and maintenance.
- fig5-subject1-plv-hippocampus-pz
In subject 1, the task increased the phase locking value (PLV) between hippocampal iEEG and scalp EEG at Pz in the alpha band during the last 2 s of maintenance; this PLV was higher for set size 8 than 4 (9.0 to 12.0 Hz) and maximal at occipitoparietal electrodes. The PLV to Fz was also higher for set size 8 (8.5 to 10 Hz).
The WM task enhanced the hippocampal-cortical PLV specifically in the 8–10 Hz alpha range and during the maintenance period ( Fig. 5D and fig. S4). The PLV increased with workload ( Fig. 5E ) and was highest at occipitoparietal electrode sites ( Fig. 5F ).
Figure 5D, 5E and 5F; Results, Electrophysiology of cortico-hippocampal functional coupling, paragraph 3; fig. S4How it will be judged
The PLV method of fig6-plv-all-subjects for subject 1. Reproduced when the maximal PLV pair is the left hippocampus with Pz, its PLV in the last 2 s of maintenance exceeds that in encoding (method of fig6-plv-maintenance-over-encoding, the golden bar of Fig. 5) in bins overlapping 8 to 10 Hz, its workload difference (set size 8 minus 4) exceeds the 95th percentile of the label-shuffled null in bins overlapping 9.0 to 12.0 Hz, the scalp electrode with the highest alpha PLV to that hippocampal channel is parietal or occipital, and the hippocampus-Fz workload difference is significant in bins overlapping 8.5 to 10 Hz; partially reproduced when the Pz parts hold and the Fz part does not.
- fig6-plv-all-subjects
All subjects showed a significant hippocampal-scalp PLV peak during maintenance (set size 8), with individual frequency bands whose lower bound lay in the theta-alpha range and which extended up to 17.5 Hz (median band limits 6.0 to 11.5 Hz).
All subjects showed a significant PLV peak for at least one pair of signals recorded from a scalp electrode and a hippocampal channel (table S1). The frequency band of the maximal significant PLV indicated individual differences
Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 1; Figure 6A to 6H; table S1How it will be judged
Per subject, every pair of a hippocampal bipolar iEEG channel (deepest contacts of each hippocampal macroelectrode) and a scalp electrode. Last 2 s of maintenance, 2-s epochs, set size 8, multitaper spectra (+/- 2 Hz, seven tapers, the spectra of Methods, Frequency spectra, which the PLV section says it uses), PLV as Methods equation 3. Significance per frequency bin against the 95th percentile of 200 PLVs with trials of one channel permuted (one-tailed). The pair and band of the highest significant PLV are kept (table S1); the individual band is the contiguous run of significant bins around the peak. Open, the frequency range searched; readings (a) 4 to 30 Hz, (b) 1 to 40 Hz. Trials as in the opening comment. Reproduced when, under one reading, all nine subjects have a significant pair and both median band limits are within tolerance; partially reproduced when all nine are significant and the medians are not.
- tables1-plv-per-subject
Table S1 gives for each subject the scalp electrode of maximal PLV, its peak frequency and the maximal PLV (subject 1 Pz 10.5 Hz 0.72; 2 F3 7 Hz 0.40; 3 F3 12 Hz 0.43; 4 F3 14 Hz 0.85; 5 P3 5.5 Hz 0.60; 6 O1 7.5 Hz 0.50; 7 O1 7.5 Hz 0.64; 8 O1 11.5 Hz 0.57; 9 O1 8 Hz 0.69), with the hippocampal electrode in the left hemisphere for subjects 1 to 6 and the right for 7 to 9.
How it will be judged
From fig6-plv-all-subjects. Reproduced when, under one reading, at least seven of the nine subjects have the same scalp electrode, hippocampal hemisphere, and maximal PLV within tolerance; partially reproduced for four to six. Peak frequencies are reported against the table.
- fig6-plv-workload-eight-of-nine
During maintenance, PLV in the individual frequency band was higher for set size 8 than set size 4 in eight of nine subjects (range 5.5 to 16.5 Hz, median band limits 7.8 to 10 Hz).
During maintenance, the PLV was increased for high workload in eight of nine subjects in the individual frequency band (set size 4 versus set size 8; permutation test, P = 0.05; range, 5.5 to 16.5 Hz; median band limits, 7.8 to 10 Hz; Fig. 6 and table S1)
Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 2; Figure 6A to 6H; table S1How it will be judged
The pair of fig6-plv-all-subjects per subject. PLV for set size 8 minus PLV for set size 4, against the 95th percentile of 200 differences with the epochs randomly assigned to the two set sizes (Methods, Phase-locking value). A subject counts when at least one bin inside the individual band is significant. Table S1 lists subject 3 as not significant. Reproduced when, under one reading, eight subjects count (seven or nine is partially reproduced) and both medians and the range (5.5 to 16.5 Hz widened by 10% of each bound) hold.
- fig6-plv-maintenance-over-encoding
The PLV during the last 2 s of maintenance exceeded the PLV during encoding in eight of nine subjects, so hippocampal-cortical connectivity in the theta-alpha range increased specifically during maintenance.
the PLV during maintenance exceeded the PLV during encoding in eight of nine subjects ( Fig. 6 and table S1)
Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 2; Figure 6A to 6H; table S1How it will be judged
The pair of fig6-plv-all-subjects per subject. PLV in the last 2 s of maintenance minus PLV in encoding (-5 to -3 s), against the 95th percentile of 200 differences with epochs randomly assigned to the two periods (Methods). Open, which trials; readings (a) set size 8 trials, where the PLV was selected, (b) all set sizes pooled. A subject counts when a bin in the individual band is significant. Table S1 lists subject 4 as not significant. Reproduced when, under one reading, eight subjects count; partially reproduced for seven or nine.
- fig6-power-unchanged-individual-band
In the majority of subjects, iEEG and EEG power in the individual frequency band did not differ during maintenance, which suggests that the PLV differences came from phase consistency.
In the majority of the subjects, there were no differences in the iEEG or EEG power in the individual frequency band during maintenance, which suggests that the differences in PLV were primarily a result of the differences in phase consistency.
Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 1; table S1How it will be judged
A claim of absence. Table S1 reports it as workload dependence (set size 8 against 4) of EEG and of iEEG power, so that is the comparison: per subject, the channels of the maximal PLV pair, last 2 s of maintenance, multitaper spectra as in fig5-subject1-eeg-power-workload, cluster-based permutation test at P < 0.05 over the individual band. Reproduced when at least five of nine subjects show no significant difference for EEG and at least five for iEEG; not reproduced otherwise. Table S1 has two of nine with an EEG difference (subjects 1 and 9) and two with an iEEG difference (subjects 1 and 6); the per-subject results are reported against it.
- fig6i-plv-posterior-gradient
Across subjects, maximal PLV increased from the anterior to the posterior end of the hippocampal axis (34 electrode pairs, R2 = 0.118), and PLV to posterior sites (y < -20 mm) was higher than to anterior sites.
For the subject group, there was a gradual increase in the PLV from the anterior end to the posterior end of the hippocampal axis during maintenance ( R 2 = 0.118, P = 0.0467) and the PLV to posterior sites ( y < −20 mm) was higher than to anterior sites (permutation t test, t = 2.09, P = 0.0405)
Figure 6I; Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 3How it will be judged
Per subject, the scalp electrode of the maximal PLV pair, paired with every hippocampal bipolar channel; for each, the maximal PLV in the subject's individual band during maintenance, set size 8 (as in fig6-plv-all-subjects), plotted against the MNI y coordinate of the hippocampal contact (Fig. 6I). Open, the coordinate of a bipolar channel; readings (a) the deepest contact, (b) the midpoint of the two contacts. The fit's P from a permutation of PLV values over positions (2000 permutations; table S2 gives 32 degrees of freedom). Reproduced when, under one reading, there are 34 pairs, the slope is positive (PLV higher posteriorly) with P < 0.05 and R2 within tolerance, and posterior sites exceed anterior sites by an unpaired permutation t test at P < 0.05; partially reproduced when the direction and one of the two tests hold.
- fig6i-no-left-right-difference
There was no evidence that the PLV differed between left and right hippocampus.
There was no evidence that the PLV differed between the left and right hippocampus.
Results, Cortico-hippocampal low-frequency coupling in the group of subjects, paragraph 3; Figure 6IHow it will be judged
A claim of absence. The paper names no test here; Methods, Statistics, give the permutation t test for comparing two distributions, which is used: the 34 maximal PLV values of fig6i-plv-posterior-gradient, left against right channels, unpaired, P < 0.05. Reproduced when the test is not significant; not reproduced when it is. The difference of means is reported either way.
- soz-plv-no-difference
The increase in PLV did not differ significantly between hippocampal channels inside and outside the seizure onset zone.
When comparing hippocampal channels within and outside the seizure onset zone (SOZ) (table S1), the increase in the PLV was not significantly different between the non-SOZ electrodes and SOZ electrodes.
Results, Recordings in the epileptogenic zone, paragraph 1; table S1How it will be judged
A claim of absence, with the permutation t test of Methods, Statistics, unpaired, P < 0.05. SOZ membership from the deposited SOZ electrode list (table S1 marks the maximal-PLV hippocampal channel as SOZ in subjects 3 and 4). Open, which increase; readings (a) set size 8 minus set size 4, (b) maintenance minus encoding, each the maximum in the individual band. Open, which channels; readings (a) all hippocampal channels paired with the maximal-PLV scalp electrode, as in Fig. 6I, (b) the nine maximal-PLV pairs. Reproduced when no reading finds a significant difference; not reproduced when a reading does. Effect sizes are reported for every reading.
- hfo-rates-unaffected
The rates of clinically relevant high-frequency oscillations were not affected by performing the task.
The rates of clinically relevant HFOs were not affected by the performance of the task.
Results, Recordings in the epileptogenic zone, paragraph 1How it will be judged
A claim of absence for which the paper names neither the HFO detector nor the test nor the comparison. It cites reference 27 (Fedele et al. 2017, Sci Rep 7, 13836) for the detector; that paper is not under paper/ and was not read. The comparison is not given; the readings the text allows are (a) per-channel HFO rates in the task periods (encoding, maintenance) against fixation, (b) rates in trials of set size 8 against set size 4. The test is then the permutation t test of Methods, Statistics, at P < 0.05, the paper's general test for two distributions. Reproduced when no reading finds a significant difference in any channel group (SOZ, non-SOZ); not reproduced when one does. Without the detector of reference 27 (a question to a reader) the claim is blocked as methods_underspecified.
- sfc-maintenance-over-encoding
Maintenance neurons showed significant spike-field coherence to the local field potential, but not to the iEEG or scalp EEG, and for all frequency bins the percentage of neurons with spike-field coherence was higher during maintenance than during encoding.
We found significant spike-field coherence to the LFP but not to the iEEG or the scalp EEG. For all frequency bins, the percentage of neurons with spike-field coherence was higher during maintenance than during encoding (fig. S6).
Results, Spike-field coherence, paragraph 1; fig. S6b; Figure 5JHow it will be judged
Maintenance neurons of fig2b-maintenance-neurons-hippocampus with at least 100 spikes in both periods (fig. S6b legend gives N = 94). Pairwise phase consistency, the P-hat-2 estimator of Vinck et al. 2012 (FieldTrip ppc2), between each neuron's spikes and the LFP of the same microwire; maintenance is the last 2 s (fig. S6a), encoding -5 to -3 s. Significance per frequency bin against the 95th percentile of 200 PPCs with the spike trials permuted (Methods, Spike-field coherence). Open, the frequency range; readings (a) 4 to 30 Hz, (b) 1 to 40 Hz. Open, the spectral estimate around each spike; the reading is FieldTrip's spike-triggered spectrum with a window of a fixed number of cycles, and the cycle number is a further open detail, run at 3 and 5 cycles. Reproduced when, under one reading, the maintenance percentage exceeds the encoding percentage in every frequency bin and the percentages for iEEG and scalp EEG are no higher than the 5% expected by chance (the paper does not say how "not to the iEEG or the scalp EEG" was judged; this is the reading of its per-bin 95th percentile threshold); partially reproduced when the maintenance percentage is higher in most bins. The LFP of the microwires is not among the data listed by the data descriptor; if the dataset lacks it, the LFP parts are blocked and the iEEG and scalp EEG parts are run.
Pipeline
- Paper
- done (2 of 2)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none