Longitudinal development of category representations in ventral temporal cortex predicts word and face recognition

Regions in ventral temporal cortex that are involved in visual recognition of categories like words and faces undergo differential development during childhood. However, categories are also represented in distributed responses across high-level visual cortex. How distributed category representations develop and if this development relates to behavioral changes in recognition remains largely unknown. Here, we used functional magnetic resonance imaging to longitudinally measure the development of distributed responses across ventral temporal cortex to 10 categories in school-age children over several years. Our results reveal both strengthening and weakening of category representations with age, which was mainly driven by changes across category-selective voxels. Representations became particularly more distinct for words in the left hemisphere and for faces bilaterally. Critically, distinctiveness for words and faces across category-selective voxels in left and right lateral ventral temporal cortex, respectively, predicted individual children’s word and face recognition performance. These results suggest that the development of distributed representations in ventral temporal cortex has behavioral ramifications and advance our understanding of prolonged cortical development during childhood.

Bars indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in different subsets of voxels of lateral VTC that are matched for variance explained.Maroon bars: a subset of the size of 300 voxels out of the union of voxels in lateral VTC defined in each session.Gray bars: a subset of 300 voxels out of the non-selective voxels of lateral VTC defined in each session.Category-selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).A voxel was defined as selective to a category when t>3. Darker colors: left hemisphere.Lighter colors: right hemisphere.Error bars: 95% CI.If the CI does not cross the y=0 line, the change in distinctiveness is significantly different than 0. Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).Black squares (lh) and gray diamonds (rh) indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in the union of selective voxels in lateral VTC.The x-axis shows the t-value that was used as a threshold to define a voxel as selective.The union of selective voxels was defined by first defining category-selective voxels to all categories and then taking the union of all these voxels.For each category, selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).The results shown for a t-value ≥ 3 (highlighted in yellow) correspond to the main analysis presented in Fig. 2 of the manuscript.Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).Black squares (lh) and gray diamonds (rh) indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in the union of selective voxels in medial VTC.The x-axis shows the t-value that was used as a threshold to define a voxel as selective.The union of selective voxels was defined by first defining category-selective voxels to all categories and then taking the union of all these voxels.For each category, selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).The results shown for a t-value t-value ≥ 3 (highlighted in yellow) correspond to the main analysis presented in Fig. 2  For pOTS-words and mOTS-words data is shown only for the left hemisphere due to low number of participants with ROIs in the right hemisphere.LMM slopes indicating change in distinctiveness per year in category-selective disk ROIs (LMM relating distinctiveness to age, with tSNR as an independent factor, and participant as random effect) for each category.Error bars: 95% confidence interval (CI).If the CI does not cross the y=0 line, the change in distinctiveness is significantly different than 0. Disk-ROIs (radius=10mm) were created for each categoryselective ROI for each participant.To define the disk ROIs, ROIs of all sessions of that contrast in a given participant (i.e., word-selective ROI in session 1, 2 and 3) were used; ROIs had individually been defined on the respective participant's cortical surface.Then, the average center coordinates of the ROIs of all sessions belonging to that participant was computed and the disk ROI was centered on those mean coordinates.This approach was chosen as (i) it ensures that the voxels will show selectivity to a certain category, (ii) it ensures that the number of included voxels is constant across sessions of a participant, (iii) provides an independent definition of the ROI, and (iv) minimizes bias towards a particular session.Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).

Fig
Fig S2.Development of distinctiveness in subsets of the union of selective voxels and the non-selective voxels in lateral VTC that are matched for variance explained.Bars indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in different subsets of voxels of lateral VTC that are matched for variance explained.Maroon bars: a subset of the size of 300 voxels out of the union of voxels in lateral VTC defined in each session.Gray bars: a subset of 300 voxels out of the non-selective voxels of lateral VTC defined in each session.Category-selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).A voxel was defined as selective to a category when t>3. Darker colors: left hemisphere.Lighter colors: right hemisphere.Error bars: 95% CI.If the CI does not cross the y=0 line, the change in distinctiveness is significantly different than 0. Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).
Fig S3.Changes in distinctiveness in the union of selective voxels in lateral VTC are largely stable across different t-values to define category-selective voxels.Black squares (lh) and gray diamonds (rh) indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in the union of selective voxels in lateral VTC.The x-axis shows the t-value that was used as a threshold to define a voxel as selective.The union of selective voxels was defined by first defining category-selective voxels to all categories and then taking the union of all these voxels.For each category, selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).The results shown for a t-value ≥ 3 (highlighted in yellow) correspond to the main analysis presented in Fig.2of the manuscript.Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).

Fig
Fig S4.Changes in distinctiveness in the union of selective voxels in medial VTC are largely stable across different t-values to define category-selective voxels.Black squares (lh) and gray diamonds (rh) indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in the union of selective voxels in medial VTC.The x-axis shows the t-value that was used as a threshold to define a voxel as selective.The union of selective voxels was defined by first defining category-selective voxels to all categories and then taking the union of all these voxels.For each category, selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).The results shown for a t-value t-value ≥ 3 (highlighted in yellow) correspond to the main analysis presented in Fig.2of the manuscript.Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).
Fig S4.Changes in distinctiveness in the union of selective voxels in medial VTC are largely stable across different t-values to define category-selective voxels.Black squares (lh) and gray diamonds (rh) indicate the change in category distinctiveness per year (LMM relating distinctiveness to age and tSNR, with participant as a random effect, n=128 sessions, 29 children) in the union of selective voxels in medial VTC.The x-axis shows the t-value that was used as a threshold to define a voxel as selective.The union of selective voxels was defined by first defining category-selective voxels to all categories and then taking the union of all these voxels.For each category, selectivity was computed by contrasting responses to a category vs. all other categories except the other category from the same domain (e.g., numbers vs. all other categories except words).The results shown for a t-value t-value ≥ 3 (highlighted in yellow) correspond to the main analysis presented in Fig.2of the manuscript.Detailed statistics for this figure can be found in the data repository (see Methods, Data availability).

Fig S6 .
Fig S6.Representational similarities and their development in different subsets of voxels in lateral VTC.(A) Representational similarity matrices (RSM) of left and right lateral VTC for 5-9-year-olds and 13-17-year-olds.One session per child is included per RSM of each age group.(B) Multidimensional scaling (MDS) embeddings for the category representation for the same age groups: 5-9-year-olds (small circles) and 13-17-year-olds (larger circles).(C,D) Same as (A) but for different subsets of voxels; the union of the selective voxels (C) and the nonselective voxels (D).The color scale is the same across (A,C,D).

Fig S7 .
Fig S7.Representational similarities in different subsets of voxels in medial VTC.(A) Representational similarity matrices (RSM) of left and right medial VTC for 5-9-year-olds and 13-17-year-olds.One session per child is included per RSM of each age group.(B) Multidimensional scaling (MDS) embeddings for the category representation for the same age groups: 5-9-year-olds (small circles) and 13-17-year-olds (larger circles).(C,D) Same as (A) but for different subsets of voxels; the union of the selective voxels (C) and the nonselective voxels (D).The color scale is the same across (A,C,D).

Fig
Fig S9.Reading and face recognition performance are reliable across tests (A) Reading performance on pseudowords (x-axis) predicts reading performance on real words (y-axis), n=70 sessions, 26 children.(B) Face recognition performance in the Cambridge face recognition task (CFMT) with images of adults faces (x-axis) predicts performance on the CFMT test with images of child faces (y-axis), n=86 sessions, 28 children.Dashed line indicates chance level in the CFMT.Participants are coded by color.Gray line: Linear mixed model (LMM) prediction.Shaded gray: 95% confidence interval of the slope (CI).
Fig S10.No significant link between behavior and the number of word-and face-selective voxels or the size of word-or face-selective ROIs after correcting for age.(A) Linear mixed model (LMM) with random slopes and intercepts relating reading performance of pseudowords (Woodcock Reading Mastery Test, WRMT) to the number of voxels selective to pseudowords in anatomically defined left lateral VTC ROIs.Selective voxels are defined by a t-value > 3 (contrast: pseudowords vs. all other categories except numbers).(B) Same as (A) but for volume of the left word-selective pOTS-words.(C) LMM with random slopes and intercepts relating performance in the Cambridge Face Memory Test (CFMT) to the number of voxels selective to adult faces in anatomically defined right lateral VTC ROIs.Selective voxels are defined by a t-value > 3 (contrast: adult faces vs. all other categories except child faces).The LMM shows a significant relation between face recognition performance and the number of selective voxels (model parameters and statistics in the figure).The effect of distinctiveness is no longer significant once age is added to the model: LMM with factors age and number of selective voxels: βnr of Voxels=0.0065,t(79)=1.16,p=0.25; βage=3.78,t(79)=7.48,p<0.001.(D) Same as (C) but for the number of voxels in the face-selective ROI pFus-faces in the right hemisphere.The LMM shows a significant effect of the number of selective voxels (see parameters indicated in the figure).This effect is no longer significant once age is added to the model: LMM with factors age and number of selective voxels: βnr of Voxels=0.01,t(56)=1.22,p=0.23; βage=3.81,t(56)=5.94p<0.001.In all panels, model parameters and stats are indicated in the bottom.Each dot is a session; Dots are colored by participant.Colored lines: individual slopes and intercepts.Thick gray line: LMM prediction not including age in the model.Shaded gray: 95% CI.

Table 1 .
Parameters and statistics of LMMs on changes in distinctiveness in lateral VTC ROIs per year.Related to Fig 1. FDR-corrected p-values are reported to adjust for multiple comparisons.

Table 2 .
Parameters and statistics of LMMs on changes in distinctiveness in medial VTC ROIs per year.Related to Fig 1. FDR-corrected p-values are reported to adjust for multiple comparisons.

Table 3 .
Parameters and statistics of LMMs on changes in distinctiveness in the union of selective voxels in lateral VTC ROIs per year.Related to Fig 2. FDR-corrected p-values are reported to adjust for multiple comparisons.

Table 4 .
Parameters and statistics of LMMs on changes in distinctiveness in the non-selective voxels in lateral VTC ROIs per year.Related to Fig 2. FDR-corrected p-values are reported to adjust for multiple comparisons.

Table 5 .
Parameters and statistics of LMMs on changes in distinctiveness in the union of selective voxels in medial VTC ROIs per year.Related to Fig 2. FDR-corrected pvalues are reported to adjust for multiple comparisons.

Table 6 .
Parameters and statistics of LMMs on changes in distinctiveness in the non-selective voxels in medial VTC ROIs per year.Related to Fig 2. FDR-corrected p-values are reported to adjust for multiple comparisons.