Fig 1.
Single neurons respond selectively to a few stimuli in the set.
Each row corresponds to data from one unit. The unit in the first row was recorded in the left anterior hippocampus. Units in rows two and three were recorded in the right amygdala. The three units were recorded in three different patients. (A, D, G) Spike shapes are depicted as temperature-scaled density plots. (B, E, H) Bar height in the histogram indicates the strength of neuronal responses to individual stimuli colour-coded according to superordinate semantic category. (C, F, I) Raster plots (stimulus onset: 0 s) depict the six most significant responses (grey rectangles indicate that the spiking pattern is considered as neuronal response to the stimulus; see Neuronal response test section in Materials and methods). Note that the images displayed here are different from the ones we actually used to prevent ostensible copyright infringement. Data and scripts underlying this figure as well as the image files we actually used are deposited here: https://github.com/rebrowski/abstractRepresentationsInMTL.
Fig 2.
Response probabilities (RPs) and selectivity.
(A) RPs (see Calculation of response probabilities section in Materials and methods) in percent across different categories and anatomical regions. Error bars indicate 95% CIs estimated by a subsampling procedure (see Calculation of response probabilities section in Materials and methods). Asterisks on top of a bar indicate a significant deviation of RP from RP of all other categories in a region (Fisher’s exact test, Bonferroni-corrected α = ). (B) Overview of SUs and MUs and their responsiveness to at least one of the stimuli. (C) Cumulative distributions of neuronal response selectivity, that is, the number of response-eliciting images identified per unit, truncated after 35 on the x-axis. Data and scripts underlying this figure are deposited here: https://github.com/rebrowski/abstractRepresentationsInMTL. AM, amygdala; CI, confidence interval; EC, entorhinal cortex; HC, hippocampus; MU, multiunit; nr, nonresponding; PHC, parahippocampal cortex; r, responding; RP, response probability; SU, single unit.
Fig 3.
Abstract semantic features determine representational similarity at the population level.
(A–D) Representational dissimilarity matrices showing the distance between two stimuli quantified as 1 –Pearson’s correlation coefficient R for the response activity of all recorded units. (E–H) Exemplars in two-dimensional space derived from multidimensional scaling of dissimilarity. (I–M) Dendrograms generated from automated hierarchical clustering. See S3 Fig for dendrograms spanning the full page width in each region. Data and scripts underlying this figure are deposited here: https://github.com/rebrowski/abstractRepresentationsInMTL. AM, amygdala; Bi, birds; Cl, clothes; Co, computer; D1, dimension 1; D2, dimension 2; EC, entorhinal cortex; Fl, flowers; Fr, fruit; Fu, furniture; HC, hippocampus; In (green), insects; In (purple), instruments; MF, manmade food; PHC, parahippocampal cortex; WA, wild animals.
Fig 4.
Pattern classifier algorithms learn abstract semantic information.
(A–E) Classifiers were trained to classify the superordinate category from Z scored responses to half of the stimuli per category and tested out of sample on the other half. Classification performance on 100 random divisions of data into training and test set is indicated in box plots (Cohen’s κ). (B–I) Confusion matrices (rows: correct label; columns: predicted label). (F–K) Classifiers were trained on half of the trials per stimulus to predict individual stimulus identity and tested out of sample on the other half of trials. Colour codes extend to maximally 50% (B–E) and 10% (G–K) for display purposes. Values higher than these maxima (for example, squares on the main diagonal) are not resolved in favour of making the patterns in off-diagonal areas more clearly visible. Data and scripts underlying this figure are deposited here: https://github.com/rebrowski/abstractRepresentationsInMTL. AM, amygdala; Bi, birds; Cl, clothes; Co, computer; EC, entorhinal cortex; Fl, flowers; Fr, fruit; Fu, furniture; HC, hippocampus; In, insects; In, instruments; MF, manmade food; PHC, parahippocampal cortex; WA, wild animals.