Fig 1.
Gene selection within cytoarchitectonic neighborhoods of visual cortex.
(A) A schematized version of how cytoarchitectonic areas were defined using an observer-independent algorithm in previously published work (please see [7, 78] for quantification details). Example windows taken from histological stains of human visual cortex, detailing the GLI along the cortical depth. GLIs were measured within these windows that slid along the cortex, and boundaries (black-dotted line) were drawn when GLI profiles changed significantly along several dimensions. The image of the histological section was provided by Evgeniya Kirilina. GLI profiles adapted from [63]. (B) The 13 cROIs delineated on the FreeSurfer cortical average surface. ROIs can be found at: http://www.fz-juelich.de/JüBrain/EN/_node.html. (C) In order to select genes for further analyses, we ran a 1-way ANOVA with cROI group as a factor (Materials and Methods). Histogram of the resulting p-values (negative log-transformed) from the ANOVA shown in gray and blue. Red line denotes the top approximately 1% (200 genes; beyond the red line and shown in blue) of genes for further analyses. Inset illustrates a zoomed-in histogram of the top 1% of genes. See S2 Fig for alternative, PCA-based expression characterization. ANOVA, analysis of variance; cROI, cytoarchitectonic region of interest; FG, fusiform gyrus; GLI, gray-level index hOc, human occipital; PCA, principal component analysis.
Fig 2.
Binary clustering of the top 1% of genes reveals opposed expression gradients in human visual cortex.
(A) Dendrograms showing the algorithmic clustering of the top and bottom 1% of genes. Top 1% of genes cluster predominantly into two groups at the highest level, colored here in pink and green. (B) A 13 × 200 matrix in which each row is an ROI arranged according to numerical order of the area labels (e.g., hOc1, hOc2, the hOc3 cluster, the hOc4 cluster, hOc5, and then the FG cluster) and each column is a gene. Two distinct expression groups are visible: 1) a group with a descending gradient (green) in which genes have the highest expression in posterior visual areas (e.g., hOc1, hOc2, etc.) and 2) an ascending gradient (pink) in which genes show increasing expression levels into ventral temporal visual areas (e.g., FG1, FG2, etc.). Expression levels are normalized to the maximum expression level (reads per kilobase million) across all tissue samples, and warmer colors indicate higher expression levels. To evaluate the significance of the expression matrix, we produced an equally sized hypothesis matrix in which a) columns corresponded to genes and b) column values linearly decreased or increased according to gene gradient membership. A bootstrapping (n = 1,000) approach was taken, shuffling columns of expression to produce a null distribution (orange histogram). Repeating the bootstrap with a subset of cROIs (75% of rows) results in the blue histogram. Solid red line marks the correlation between the original expression matrix and the hypothesis matrix (Pearson r = 0.67, p < 0.001). Red-dotted line denotes the correlation between the expression and hypothesis matrix when hOc1, hOc2, and FG4 were excluded, which is statistically significant, even when evaluated against the orange noise distribution (p = 0.002). Transcription data can be downloaded from the Allen Institute: http://brain-map.org/. cROI, cytoarchitectonic region of interest; FG, fusiform gyrus; hOc, human occipital.
Fig 3.
Average expression levels of the ascending and descending genetic gradients produce a transcriptomic processing hierarchy in visual cortex.
(A) Gene expression levels (z-score normalized) shown across cytoarchitectonic regions for descending-gradient genes. Color shade denotes numerical order of the area labels (e.g., hOc1, hOc2, the hOc3 cluster, the hOc4 cluster, hOc5, and then the FG cluster). (B) Same as A, but for ascending gradient gene expressions. (C) Dendrogram produced from an algorithm that clustered the cROIs based on expression levels within the two gene gradients. The ordering of cROIs along the x-axis in the dendrogram is meaningful because it represents the distance from hOc1 at the level of dendrogram leaves. Consequently, this dendrogram has rooted leaves, and the clusters at this lowest level are not rotatable (Materials and Methods). (D) Distributions of the error values resulting from the model predicting a given tissue sample’s anatomical origin from its expression pattern of the top 200 genes. Goldenrod distribution shows error from the model, while gray distribution reflects error from the noise control in which gene labels were shuffled. The y-axis is probability density function, and x-axis values are the number of hierarchical steps (e.g., V1 to V2 is one step). Transcription data can be downloaded from the Allen Institute website: http://brain-map.org/. cROI, cytoarchitectonic region of interest; FG, fusiform gyrus; hOc, human occipital; hOc d, hOc dorsal; hOc la, hOc lateral anterior; hOc lp, hOc lateral posterior; hOc v, hOc ventral.
Fig 4.
Ascending- and descending-gradient genes mirror inter-regional differences in cortical thickness and myelination.
(A) Violin plots depicting the distribution of T1w/T2w ratio intensity, which is a metric reflective of tissue contrast enhancement associated with myelin content, in blue and cortical thickness in orange across HCP subjects (n = 1,096) in cROIs of the transcriptomic processing hierarchy. (B) Cortical flat maps depicting the average T1w/T2w ratio (top) or cortical thickness (bottom) across HCP subjects. Black-dotted outline highlights visual cortex. (C) Top: correlation plot between the average “myelin” content (T1/T2 ratio) in cROIs and average expression magnitude of genes included within the descending gradient. Bottom: same as above but for the average cortical thickness of a given cROI and expression magnitude of genes included within the ascending gradient. Thickness and myelination data can be downloaded from the HCP: http://www.humanconnectome.org/study/hcp-young-adult/data-releases. cROI, cytoarchitectonic region of interest; FG, fusiform gyrus; HCP, Human Connectome Project; hOc, human occipital; hOc d, hOc dorsal; hOc la, hOc lateral anterior; hOc lp, hOc lateral posterior; hOc v, hOc ventral; T1w, T1-weighted; T2w, T2-weighted.
Fig 5.
Ascending and descending gene gradients emerge at different points in human development.
The two gene gradients derived from the top 200 differentially expressed genes in adults (ascending-gradient expression magnitudes shown in pink and descending-gradient magnitudes in green) show differences in their expression levels across development in early (pericalcarine cortex, left points in each line) and late (VLTC, right points in each line) visual cortex. See the following hyperlink for BrainSpan’s atlas detailing ROI definitions: http://www.brainspan.org/static/atlas. At the earliest time point (10 pcws), the ascending gradient (pink) shows an increasing expression profile from early to late visual cortex. However, the gradient genes that will show a descending expression profile (green) in adulthood also show an increasing profile at this stage of development. It is not until 19–24 pcws that the two gene gradients diverge in slope. Error bars reflect standard error across donor samples. Gestational transcription data can be downloaded from the Allen Institute’s BrainSpan project: www.brainspan.org. pcw, postconception week; ROI, region of interest, RPKM, reads per kilobase million; VLTC, Ventrolateral Temporal Cortex.
Fig 6.
Homologous genes in macaques do not form ascending and descending gradients as in human visual cortex.
Lines depicting the slope of transcription magnitude relative to early visual cortex in the macaque monkey for each gene. Green lines (left) are the expression slopes of homologous genes that were identified in humans as belonging to the descending gradient (see upper inset for expression slopes of the same genes in humans from hOc1 to FG4). Pink lines (right) are the expression slopes of genes that form the ascending gradient in humans. Because of sparser tissue sampling in the macaque compared to the human data set, early visual cortex was taken to be any tissue sample residing in either V1 or V2, while those in late visual cortex were taken from area TE, a region thought to be homologous to the FG cytoarchitectonic regions in humans. Solid black lines denote the mean slope. Individual gene names are listed on the right of each plot in descending order of slope (top corresponds to most positive line, bottom corresponds to most negative). Gene names bolded in black show positive slopes in expression magnitude from early to late processing stages. Light gray gene names show negative slopes. NHP transcriptional data can be downloaded here: http://www.blueprintnhpatlas.org/. FG, fusiform gyrus; hOc, human occipital; NHP, nonhuman primate; TE, a high-level visual area in macaque inferotemporal cortex.
Table 1.
Relating cytoarchitectonic areas of the transcriptomic hierarchy to function.
Left: Names of the cROIs from the Jülich Atlas. Middle: fROIs that are located within each cROI. fROI names in standard typeset have been empirically validated, while those fROIs that are italicized have been proposed to be located within a given cROI but have not yet been empirically validated. Right: The general type of visual processing typically associated with the fROIs and cROIs.
Fig 7.
Comparison among hierarchies of areas in human visual cortex generated by transcriptomics, cytoarchitectonics, or functional properties.
Left: connective models of visual regions composing processing hierarchies as measured with either transcription (data from Fig 3B) or cytoarchitecture (a combination of dendrograms produced by Malikovic and colleagues [11] and Lorenz and colleagues [13]). Color along the rainbow spectrum corresponds to the region’s numerical rank according to name (red, orange, yellow, green: hOc1, hOc2, hOc3, hOc4, etc.). hOc5 was not colored because of the fact that only a small number of samples (n = 2) were located within this area. Dotted line denotes where branch from Lorenz and colleagues [13] was added to the dendrogram from Malikovic and colleagues [11]. Right: Linear fits relating pRF size and eccentricity from voxels located in visual field maps (V1 through VO-2), face-selective regions (pFus-faces and mFus-faces), as well as a motion-selective region (hMT+) of the visual hierarchy. Regions are colored according to the cytoarchitectonic region in which they reside based off of previously published work (Table 1). For all regions except mFus-faces, solid lines denote mean line of best fit across 40 subjects from Gomez and colleagues [35, 72] (shaded region is standard error across subjects). For mFus-faces, the solid line and shaded region are the mean and 68% confidence intervals across the three subjects from Kay and colleagues [70]. The slope relating a pRF’s eccentricity to its size increases as one ascends the visual hierarchy. Note that the relationship between pRF size and eccentricity in hMT+ is more similar to pFus-faces and mFus-faces than other visual regions tested in these previously published studies. Because hMT+ is located within hOc5 and pFus-faces is located within FG2, the adjacent positioning of these two regions within the same subcluster in the transcriptomic hierarchy may be related to similar pRF properties between these two regions (see Discussion). FG, fusiform gyrus; hMT+, human motion-selective complex; hOc, human occipital; hOc d, hOc dorsal; hOc la, hOc lateral anterior; hOc lp, hOc lateral posterior; hOc v, hOc ventral; mFus, middle Fusiform; pFus, posterior Fusiform; pRF, population receptive field; VO, ventral-occipital.