Abstract
The neural circuits that support human cognition are a topic of enduring interest. Yet, there are limited tools available to map brain circuits in the human and nonhuman primate brain. We harnessed high-resolution diffusion MR tractography, anatomic, and transcriptomic data from individuals of either sex to investigate the evolution and development of frontal cortex circuitry. We applied machine learning to RNA sequencing data to find corresponding ages between humans and macaques and to compare the development of circuits across species. We transcriptionally defined neural circuits by testing for associations between gene expression and white matter maturation. We then considered transcriptional and structural growth to test whether frontal cortex circuit maturation is unusually extended in humans relative to other species. We also considered gene expression and high-resolution diffusion MR tractography of adult brains to test for cross-species variation in frontal cortex circuits. We found that frontal cortex circuitry development is extended in primates, and concomitant with an expansion in corticocortical pathways compared with mice in adulthood. Importantly, we found that these parameters varied relatively little across humans and studied primates. These data identify a surprising collection of conserved features in frontal cortex circuits across humans and Old World monkeys. Our work demonstrates that integrating transcriptional and structural data across temporal dimensions is a robust approach to trace the evolution of brain pathways in primates.
SIGNIFICANCE STATEMENT Diffusion MR tractography is an exciting method to explore pathways, but there are uncertainties in the accuracy of reconstructed tracts. We broaden the repertoire of toolkits to enhance our ability to trace human brain pathways from diffusion MR tractography. Our integrative approach finds corresponding ages across species and transcriptionally defines neural circuits. We used this information to test for variation in circuit maturation across species and found a surprising constellation of similar features in frontal cortex neural circuits across humans and primates. Integrating across scales of biological organization expands the repertoire of tools available to study pathways in primates, which opens new avenues to study pathways in health and diseases of the human brain.
Introduction
The human frontal cortex (FC) is larger than that of many other species. The claim that the FC, and particularly the prefrontal cortex (PFC) white matter, is unusually large in humans compared with other primates is not without controversy (Semendeferi et al., 2002; Carlén, 2017; Donahue et al., 2018). The expansion of the PFC white matter points to possible modifications in neural circuits across species. However, there are limitations in the tools available to study human brain pathways (Charvet, 2020). At the microscale, RNA sequencing from bulk samples and single cells offers an unprecedented perspective on the classification of cell types and transcriptional profiles. These data, however, lack key information about the structure of circuits (Luo et al., 2017; Nguyen et al., 2017; Saunders et al., 2018; Zhu et al., 2018; Krienen et al., 2020). At the macroscale, diffusion MR tractography offers an exquisite three-dimensional perspective of pathways across the brain (McNab et al., 2009; Takahashi et al., 2012; Wedeen et al., 2012; Edlow et al., 2019; Charvet et al., 2020; Nolan et al., 2021), which makes diffusion MR tractography an exciting tool to explore pathways. The human brain scans we used in this study are of unprecedentedly high resolution for the study of primate brain pathways (Edlow et al., 2019; Nolan et al., 2021), yet the termination sites of tracts can be imprecise. Consequently, results from tractography necessitate validation (Jones et al., 2013; Maier-Hein et al., 2017). We harnessed transcriptomic and structural data across temporal scales to trace the evolution of pathways in the primate brain (Zeng et al., 2012; Krienen et al., 2016; Hendy et al., 2020).
Neuronal populations across the depth of the cortex can be used to predict connectivity patterns (Barbas, 1986; Barbas and Rempel-Clower, 1997). For example, the number of excitatory layer III FC neurons, which preferentially form long-range cross-cortical projections, and the relative number of corticocortical pathways are increased in humans relative to mice (Charvet et al., 2017, 2019; Luo et al., 2017). Several genes, such as some supragranular-enriched (SE) genes, identify long-range projecting (LRP) neurons (e.g., NEFH, VAMP1, SCN4B) in layers II–III and V–VI. However, a major hurdle in linking transcription with tractography is the lack of adequate markers to transcriptionally define neurons that project over long distances (Krienen et al., 2016; Charvet et al., 2019). To overcome this hurdle, we aligned transcriptional and structural variation over the course of human development to transcriptionally define neurons that project through the white matter. We focused our attention on the evolution of corticocortical pathways given major reported differences between humans and mice (Charvet et al., 2017; Luo et al., 2017; Charvet et al., 2019).
Developmental timelines vary tremendously across species, with human development extended compared with many other model systems (Clancy et al., 2001; Charvet, 2020, 2021). We lack appropriate norming procedures to compare brain development across species. We applied machine learning to RNA sequencing data to align ages in humans and macaques. We then tested whether FC maturation is unusually extended in humans relative to macaques after accounting for variation in developmental schedules (Finlay and Darlington, 1995; Clancy et al., 2001; Charvet, 2021). These data reveal that many transcriptional and structural features of FC circuits, previously presumed to be unique to humans, are in fact shared with macaques.
Materials and Methods
We discuss how we found corresponding ages across species and how we tested for cross-species modifications in FC circuitry. All statistics were performed with the programming language R. Ages were expressed in days after conception.
Transcriptional and structural variation to infer corresponding ages.
We used 104 time points to find corresponding ages between humans and macaques (Fig. 1; Liu et al., 2016). These included time points extracted from an RNA sequencing dataset from the PFC of humans and macaques. Details of RNA sequencing data used for these analyses are shown in Figure 2. We selected genes with log-based 10 expressions in reads per kilobase per million (RPKM) > 1 averaged across samples per species, but we did not filter RNA sequencing datasets by hemisphere or sex. We fit a glmnet model (cross-validation: n = 10; repeat = 5; tune length = 5) to log10(RPKM + 1) in humans. We then used this model to predict ages from normalized gene expression in macaques. We considered this model accurate because ages predicted from these and other time points accounted for 95% of the variance when nonlinear regressions were fit to these data (Fig. 1). There may be variation in extrapolating ages across methods so that time points from the glmnet model were included with other time points to find corresponding ages across humans and macaques.
Transcriptional definition of cortical long-range projecting neurons.
We used gene expression data to test for modifications to FC circuits (Figs. 2, 3). We considered SE genes because of differences in expression patterns between humans and mice, and because they are expressed in layers III–V where neuronal somas that project over long distances are located (Gilbert and Kelly, 1975). It is, however, not clear whether other genes are better suited to study LRP neurons. For example, supragranular-enriched genes were identified from a list of only ∼1000 genes so that many genes are here excluded from consideration (Zeng et al., 2012). Moreover, evolutionary modifications in the expression of supragranular-enriched genes may reflect modifications to large layer III neurons as well as other cell populations. We aligned temporal variation in gene expression and structure to identify genes that are expressed by LRP neurons.
We identified LRP markers by testing for associations between transcription and maturation over the course of human development. We considered myelin water fraction (MWF) as an index of white matter maturation across lobes (Deoni et al., 2012), and we used multiple RNA sequencing datasets (Liu et al., 2016; Zhu et al., 2018; Fig. 2). LRP markers were defined as genes (1) that were expressed by layer II–III neurons, but not non-neuronal cells; and (2) that have expression patterns significantly associated with MWF. We added a value of 1 before logging the expression of each sample to consider genes that may not be expressed at a specific age. We fit smooth splines (df = 4) through log10(RPKM) values versus age to extrapolate data at matching ages [n = 10; from 405 d after conception (DAC) to 6 years of age]. Only expression profiles that were significantly associated with MWF across all tested areas were considered LRP markers. We used single-nucleus transcriptomes from the human primary motor cortex (n = 2; age, 18–68 years; n = 76,535 nuclei; Bakken et al., 2020) to filter these genes by cell types. An expressed gene was defined by a count > 0. We also used an in situ hybridization (ISH) Allen Brain Atlas dataset to evaluate the spatial expression of LRP markers and SE genes, but we did not filter these data by hemisphere or sex. We considered RNA sequencing data from macaques (n = 26) and humans (n = 36) and across cortical areas (n = 11; Zhu et al., 2018). We translated age in macaques to that of humans and fit smooth splines (df = 4) to compare normalized gene expression across these two species (Fig. 1).
Structural MR scans to test for variation in FC white matter maturation.
The white matter houses long-range projections. We compared white matter growth to test for variation in the timeline of FC circuits across species. Our definitions of FC and PFC follow those used previously (Hendy et al., 2020; Charvet, 2021). Structural MR scans of macaque brains were obtained from the UNC-Wisconsin Rhesus Macaque Neurodevelopment database (n = 32). We used Fiji to measure the PFC white matter, which was defined as white matter anterior to the corpus callosum, consistent with previous definitions (Fig. 4). For this, we used data generated by this and other studies (Sakai et al., 2011; Young et al., 2017; Charvet, 2021). We measured the PFC white matter area across sections (in at least every other section) using an approach similar to that used previously (Charvet, 2021). We reconstructed volumes by multiplying the area, section thickness, and section spacing. Nonlinear regressions (library easynls; model = 3) were used to detect the age of growth cessation (Fig. 4) with the caveat that some growth may persist beyond identified time points.
Diffusion MR imaging protocols and tractography.
We used diffusion MRI datasets of 14 individual subjects (humans, n = 5; mice, n = 4; macaques, n = 4; and Sykes monkey, n = 1) and combined some of these with tract-tracers (Figs. 5–10). Some of these datasets were previously collected (e.g., Japanese Monkey Brain Center; Calabrese et al., 2015; Ding et al., 2016; Sakai et al., 2018). We used diffusion MR datasets of human brains (n = 4) scanned on a 3 T scanner with a 32-channel head coil (Tim Trio, Siemens) at the Massachusetts General Hospital Athinoula A. Martinos Center for Biomedical Imaging. The resolution of the human MR scans was 0.75 mm isotropic (diffusion MRI data acquisition duration, ∼31 h). Diffusion-weighted data were acquired with a 3D steady-state free precession sequence. Diffusion weighting was applied along 90 directions distributed over the unit sphere (effective b value = 4080 s/mm2; 12 b0 values; TR, ∼28.87 ms; TE, ∼24.44 ms). We also used one diffusion MR scan of a 34-year-old human brain made available by the Allen Brain Institute (Ding et al., 2016). Diffusion-weighted data were acquired with a 3D steady-state free precession sequence (TR = 29.9 ms; α = 60°; TE = 24.96 ms; 900 μm and 1.2 mm; two averages). Diffusion weighting was applied along 44 directions distributed over the unit sphere (effective b value = 3686 s/mm2; 8 b = 0 images; Ding et al., 2016). An additional lower-resolution scan was also acquired in the same individual. We also collected diffusion MR scans of mouse brains (n = 8) at postnatal day 21 (P21) and P60 using a 9.4 T Bruker scanner at the University of Delaware (Figs. 7–9). A 3D diffusion-weighted spin-echo echoplanar imaging sequence (TR, ∼500 ms; TE, ∼40 ms; resolution, 100 μm isotropic) was used to image the mouse brains. Sixty diffusion-weighted measurements (b = 4000 s/mm2) and non-diffusion-weighted measurements (5 b = 0 s) were acquired.
Diffusion MRI data were processed with Diffusion Toolkit (www.trackvis.org; threshold angle, 45° angles for 13 of 14 brains). In some panels, fibers were skipped for visualization but not for the analyses. With the exception of the human brain scans from the Allen Brain Atlas, we used high-angular resolution diffusion MR imaging (HARDI) tractography to generate whole-brain tractography. The orientation distribution functions (ODFs) were normalized according to the maximum ODF length within each voxel. Fractional anisotropy (FA) was calculated from orientation vectors by fitting the data to the tensor model (Takahashi et al., 2012). We used fiber assessment by continuous tracking with HARDI. No fractional anisotropy threshold was applied in reconstructing tracts, which is consistent with previous work (Takahashi et al., 2012). We used TrackVis (http://trackvis.org) to visualize and quantify pathways. Details on the individuals scanned, including their age, and the spatial resolution are in the legend of Figure 5.
We quantified the relative number of pathway types coursing through the frontal cortex white matter. We randomly selected a region of interest (ROI) consisting of a voxel in the frontal cortex white matter. Voxels ranged from 0.1 × 0.1 × 0.1 mm to 1.2 × 1.2 × 1.2 mm in humans. ROIs were placed at equidistant sites across the anterior-to-posterior axis of the frontal cortex. We used coronal planes to randomly select sites along the dorsal-to-ventral and medial-to-lateral axis of the frontal cortex white matter. We then classified pathway types coursing through these ROIs to test for species differences in pathway types.
Varying sampling and imaging procedures of diffusion MR scans.
We tested how sampling, resolution, and imaging protocols impact the percentage of pathway types, as it is unclear how these parameters impact results from tractography. We randomly subsampled the number of sites and extracted the average proportion of pathway types in the FC for each individual (Fig. 8). Variation in the percentage of pathway types was minor regardless of the number of randomly selected voxels, imaging procedures, and resolution (Fig. 8). Our analyses suggest that the differences in the pathway types or a lack thereof are robust to sampling size, resolution, and imaging protocols.
Tractography quantification and comparison with tract-tracers.
The accuracy of diffusion MR tractography has remained elusive because of a lack of alternative tools to map the pathways in the human brain. We compared diffusion MR tractography with EGFP injections to assess which metrics should be extracted from tractography in mice (Fig. 7). Given that the accuracy of tractography appeared compromised at the gray matter–white matter boundary (Figs. 7a–h, 9j–s), we randomly selected voxels across spaced planes along the anterior-to-posterior axis of the FC white matter, with planes varying across individuals. We then classified pathways based on orientation and direction within the white matter. Fibers were classified as those belonging to the corpus callosum, cingulate bundle, other corticocortical, or subcortical–cortical pathways (Fig. 10). If a pathway was observed coursing through the dorsal midline, it was considered callosal regardless of its terminations. Pathways connecting the cortical and lateral limbic structures were considered cortico–subcortical pathways. U fibers or long-range pathways connecting cortical areas (e.g., arcuate fasciculus) were considered corticocortical. We focused on corticocortical pathways because of the expansion of layer II–III in primates relative to rodents. We tested how sampling, resolution, and imaging protocols impacted pathway types (Fig. 8). We observed that HARDI and diffusion tensor imaging (DTI) reconstruction yielded comparable results in the mouse FC at P60 (y = 0.82× + 4.52; R2 = 0.89; p < 0.01). We also tested for temporal variation in the relative percentage of pathway types of mouse brains at P21 versus P60. An ANOVA with age and pathway types as factors showed trends but no signific effect for age (p < 0.05) on the pathway types (F = 13.16, p < 0.01; n = 32). Frontal cortex white matter growth in mice ceases before P60 (Hendy et al., 2020); therefore, P60 should represent adult proportions in pathway types.
We compared diffusion MR tractography with tract-tracers in mice to assess which metrics should be extracted from tractography. We compared diffusion MR tractography in P60 mouse brains with viral tracer experiments, which involved EGFP injections into selected regions of the mouse brain. These data were made available by the Allen Mouse Brain Connectivity Atlas (https://connectivity.brain-map.org/; Fig. 7). We identified injection sites and set ROIs to compare projections identified from the tract-tracers with those from tractography. We found strong concordance in the location and the orientation of pathways coursing through the white matter. We observed that tracts did not necessarily penetrate the gray matter in the same location as the tract-tracers. Axons make an abrupt turn at the junction of the gray matter and the white matter (Figs. 7, 9). These sharp turns challenge the accuracy of tractography, as evident from the qualitative observations of tract-tracers and tractography (Figs. 7, 9).
We compared tract-tracers with diffusion MR tractography to guide the development of quantitative approaches to study FC pathways. As there is no evidence of a relationship between fiber numbers and circuits (e.g., axons), we did not quantify fibers. Instead, we selected voxels through the FC white matter, classified pathway types, and quantified the proportion of pathway types across the FC and the PFC. In relatively rare cases, we refrained from classifying the pathway types if the tractography was not clearly classifiable.
Comparative analyses of supragranular-enriched gene expression in the FC.
We leveraged RNA sequencing datasets from bulk samples (n = 24; Hawrylycz et al., 2012; Bozek et al., 2014; Miller et al., 2014; Li et al., 2018), single cells, and ISH to detect cross-species variation in SE and LRP gene expression across the mouse, macaque, and right FC of humans (Figs. 2, 11). We performed principal component analyses (PCAs) on these data, and measured gene expression intensity across layers as we had done previously (Charvet et al., 2017; Charvet, 2021).
We measured the intensity of gene expression across layers in a manner similar to that done previously (Charvet et al., 2019; Hendy et al., 2020; Charvet, 2021). We downloaded ISH images of the FCs from humans, macaques, and mice from the Allen Brain Atlas. We analyzed the expression of select SE genes within layers II–IV and V–VI. The boundary across layers IV and V was based on the cytoarchitecture from Nissl staining and RORB mRNA expression. Layer IV was defined as a cell-dense zone characterized by preferential expression of RORB. Layer VI was bound by the white matter. We measured the expression intensity from ISH images of CRYM in humans, macaques, and mice (Fig. 10m). We used ImageJ software to randomly select areas within the FC, then we placed a rectangular grid to capture gene expression intensity across layers II–IV and V–VI. The grid was perpendicular to the cortical surface. The height varied with cortical thickness. Frame widths were 1000 µm in mice, macaques, and humans. We binarized the images and measured the intensity of expression in layers II–IV and V–VI. We computed the ratio of these values to compare the relative expression of genes of interest in the upper and the lower layers. A value >1 indicates that the gene is preferentially expressed in layers II–IV.
Data availability.
Data and scripts are available on DRYAD (https://datadryad.org/stash/share/gcCyHjb2tbV2Irb5D2H5U3gIt5Apq1v1dKCF83C-nA8) and as Extended Data 1.
Results
We used structural and transcriptional variation to find corresponding ages across species. We then integrated these data to test for cross-species variation in FC neural circuits.
Corresponding time points during postnatal development
We used time points to find corresponding ages between humans and macaques (n = 96). These data included time points from a glmnet model applied to normalized gene expression in RPKM from human and macaque PFCs of different ages (n = 38 humans, n = 31 macaques; Figs. 1, 2; Liu et al., 2016). We selected genes with minimum expression averaged across samples [i.e., log10(RPKM) > 1; n = 8014]. We did not predict ages beyond 10.6 years in macaques because samples at late stages are sparse (Fig. 2i,j). We trained a glmnet model to predict ages from normalized gene expression in humans (cross-validation = 10; n = 38). This model has high predictive accuracy because sampling 70% of the data to train the model (R2 = 0.99) resulted in strong correlations between log-10 transformed predicted and observed ages (R2 = 0.98). The same approach applied to normalized gene expression in macaques predicted age of macaques (R2 = 0.99). We then used a glmnet model (R2 = 0.97; λ = 0.053) trained from human samples (n = 38) to predict ages from normalized gene expression in macaques. This approach yielded 23 corresponding time points in humans and macaques. We fit a quadratic model on the log-transformed time points in humans and macaques (y = −1.96 + 2.74× – 0.31 × 2; R2 = 0.95, n = 96) to find corresponding ages across these two species. Early in development, ages are approximately similar in the two species, but time points occur much later in humans than in macaques by adulthood (Fig. 1). We next tested whether FC development at the transcriptional and the structural level (Figs. 1, 2, 4) is unusually extended in humans.
a, We identified corresponding ages between humans and macaques from abrupt and gradual changes in structural and transcriptional variation. We extracted corresponding time points by training a glmnet model to predict age from RNA sequencing data in humans and subsequently predicted ages from normalized gene expression in macaques. Ages are expressed in days after conception (DAC), unless otherwise noted. b, c, Models trained to predict age in each species have a high accuracy as evidenced by the strong correlation between predicted and observed values in humans (b) as in macaques (c). d, With these data, we identified corresponding ages across fetal and postnatal time points across humans and macaques. For example, a macaque at gestational week 12 (GW12) is equivalent to a human at GW21, and a 10-year-old macaque is equivalent to a 22-year-old human. We then addressed whether the development of FC circuits is protracted in humans relative to macaques after controlling for variation in developmental schedules. Smooth surfaces of MR scans are from multiple sources (Shi et al., 2011; Rohlfing et al., 2012; Miller et al., 2014; Bakker et al., 2015; Ding et al., 2016; Reveley et al., 2017; Liu et al., 2020; Extended Data Tables 1-1, 1-2). The ages in d do not correspond exactly to the labeled age.
Table 1-1
Corresponding time points across humans and macaques. Download Table 1-1, XLSX file.
Table 1-2
List of RNA sequencing datasets and ages of samples used in the present study. Download Table 1-2, XLSX file.
a–c, We tested whether trajectories in SE gene expression are extended in humans relative to macaques after cross-species age alignment. We mapped age in macaques to that of humans, and we tested whether the temporal trajectories of SE genes in the FC are significantly different relative to other cortical areas in humans and macaque. a, Correlation coefficients from the FC are not significantly different from other cortical areas. b, c, SE gene expression of NEFH and VAMP1 in the dorsolateral frontal cortex of humans and macaques overlap extensively in the two species. d–h, We also used LRP markers to test for variation in the development of FC circuitry relative to other cortical areas (d–h) in humans and macaques. f, We fit a smooth spline through normalized gene expression versus age in humans and macaques as exemplified in the primary motor cortex (M1C). We then extracted normalized gene expression from smooth splines at corresponding ages across humans and macaques (n = 10). Two example genes (NRGN, RGS4) are shown (f). g, h, We correlated the expression of LRP markers across humans and macaques for each cortical area and tested for differences in correlation coefficients (g) and significance tests (h). No significant differences in correlation coefficients are observed between the FC and other cortical areas. We find conservation in the developmental time course of LRP marker expression in the human FC relative to macaques (Extended Data Table 2-1). i, k, We also evaluated PCA, read numbers, age ranges, and replicates across RNA sequencing datasets Liu et al. (2016; i), Zhu et al. (2018; j), and the BrainSpan Atlas of the Developing Human Brain (Li et al., 2018; k). PCAs on log10-transformed RPKM samples cluster primarily by species with no obvious outliers. i, j, Age ranges and number of reads overlap across datasets. Scatter plots of log10-transformed RPKM values from biological replicates show that variance in expression decreases with increasing expression. A1, Primary auditory cortex; DFC, dorsolateral frontal cortex; IPC, inferior parietal cortex; ITC, inferior temporal cortex; MFC, medial frontal cortex; OFC, orbitofrontal cortex; S1C, primary somatosensory cortex; STC, superior temporal cortex; ITC, inferior temporal cortex; VFC, ventral frontal cortex; V1, primary visual cortex.
Table 2-1
Post hoc Tukey's tests between correlation coefficients of LRP markers across different regions and corticocortical pathway types across humans, macaques, and mice. Download Table 2-1, XLSX file.
Transcriptionally defining neural circuits
We considered the expression of SE genes (n = 16) and other sets of markers to test for cross-species variation in FC circuit development (Figs. 1–3, 10–15). We mapped age in macaques onto humans, and we extrapolated normalized gene expression at corresponding time points in both species. We then correlated SE gene expression over the course of development in humans and macaques. We extracted correlation coefficients to test whether the pattern of gene expression over the course of development is unusual in the frontal cortex of humans versus macaques (Figs. 2, 11). We found that these correlation coefficients are not significantly different between frontal and other cortical areas (ANOVA: F = 0.145; p = 0.99; n = 176). Although these findings suggest conservation in FC circuit maturation, it is not clear whether genes other than SE genes are best suited to track FC circuit maturation. We therefore systematically tested for genes that could be used as markers of neurons that project through the white matter.
Temporal variations in SE gene expression in humans and macaques are very similar in the two species once cross-species age alignment, Data are from Zhu et al. (2018). a, We extracted normalized gene expression at corresponding time points across cortical areas and correlated normalized gene expression across species. Here, age in macaques is translated to humans according to Figure 1. b, Significance tests overlap extensively across areas, which indicate that temporal profiles in SE genes are very similar in humans and macaques. c, Indeed, temporal profiles in SE gene expression (e.g., NEFH, VAMP1, CRYM) from the orbitofrontal cortex (OFC), ventral frontal cortex (VFC), and primary visual cortex (V1) appear highly similar in humans and macaques. Together, these data show a lack of evidence for protracted development in FC circuits in humans relative to macaques. d, In situ hybridization images show the expression of select genes, including myelin basic protein (MBP), NEFH, and VAMP1, increases postnatally in humans. MBP, NEFH, and VAMP1 expression is weak shortly after birth, but increases postnatally. Arrows point to large pyramidal neurons. These qualitative observations align with our computational analyses, which also show that the expression of these genes increases postnatally. In situ hybridization images are from the Allen Brain Institute. Images of NEFH and MBP expression are from the frontal cortex. VAMP1 expression is from the visual cortex. Abbreviations are the same as in Figure 2.
We identified LRP markers by testing for associations between gene expression and myelination used as an index for long-range projection maturation (Fig. 11; Deoni et al., 2012; Zhu et al., 2018) across cortical areas. Genes were filtered to have minimum expression averaged across regions and ages [log10(RPKM) > 1]. We extrapolated MWF values from previous work (Fig. 11d; Deoni et al., 2012). We fit a smooth spline through log-based 10(RPKM) values versus age (i.e., log-10 d after conception) to extrapolate gene expression and MWF at matching ages (n = 10) and across cortical areas (n = 10). We iteratively tested for associations between gene expression and MWF across cortical areas. We selected significant and positive associations (slope > 0) after correcting for multiple testing [Benjamini–Yekutieli (BY) test, p < 05]. We then used single-cell RNA sequencing (Bakken et al., 2020) from the motor cortex to exclude genes expressed by non-neuronal cells but include those expressed by layer II–III neurons. These filtering steps resulted in 250 LRP markers. These genes are expressed by large pyramidal neurons in the cortical gray matter and increase in expression postnatally, and a subset of these are SE genes (e.g., NEFH, VAMP1; Figs. 2, 3, 10–12, 15). These observations confirm the validity in using these genes as LRP markers.
We tested whether temporal profiles in the expression of these LRP markers are different between humans and macaques. We fit a smooth spline through the log2(RPKM) values versus log-10 ages expressed in days after conception, and we extrapolate gene expression at equivalent ages in macaques and humans (n = 10). We correlated the expression of LRP markers (n = 234) across cortical areas in humans and macaques (n = 11) over the course of development. An ANOVA on the correlation coefficients pointed to significant differences across cortical areas (ANOVA: F = 3.12, p < 0.01; n = 2574), but post hoc Tukey's HSD tests showed that correlation coefficients from the FC areas were not significantly different (significance threshold set to p < 0.01) from those of other cortical areas (Fig. 2). These data showcase strong similarity in FC development between humans and macaques.
Structural variation in FC circuit development
We compared white matter growth across species to test whether LRP neuron maturation is unusually extended in humans relative to macaques (Fig. 4). Over the course of development, the white matter grows and ceases postnatally in both humans and macaques. We fit nonlinear regressions with age in days after conception as the predictor variable, and the FC and log10-based PFC white matter volume as the dependent variable to compare the time course of white matter growth across species (Fig. 4). We captured the ages at which the FC and PFC white matter reach percentages of adult volumes (e.g., 80%, 90%, 100%; Fig. 3d–h). We found that these time points overlap with other time points (Fig. 4c). For example, the PFC white matter ceases to grow at 1.74 years of age in humans (R2 = 0.76, n = 28, p < 0.001; Fig. 4f) and at ∼1 year of age in macaques (R2 = 0.66, n = 32, p < 0.001; Fig. 4g). The addition of time points from these growth trajectories did not account for a significant percentage (p = 0.24) of the variance (ANOVA: F = 1355, p < 0.05; n = 102). The age of growth cessation was largely invariant of sample size, demonstrating that these results were not driven by outliers (Fig. 4i–n). These data show no evidence for protracted FC white matter maturation in humans once cross-species variation in developmental schedules are accounted for.
We compared the timeline in frontal and prefrontal cortex white matter growth across humans and macaques. a, b, The PFC white matter was defined as the white matter anterior to the corpus callosum as highlighted on a macaque template (Bakker et al., 2015; Calabrese et al., 2015). The red horizontal line illustrates the posterior boundary of the PFC white matter. c, We extracted epochs from growth trajectories in the PFC and FC white matter in humans and macaques and found that these time points align with other data points. d–g, We fit nonlinear regressions to capture the age at which the growth of the FC (d, e) and PFC (f, g) white matter ceases and when these volumes reach percentages of adult volumes. For example, PFC growth cessation occurs at ∼1.7 years of age in humans and 1 year of age in macaques. The choice of epochs was based on the age ranges from which volumetric data were available. h, We applied the same nonlinear regression on other data (Knickmeyer et al., 2010) to assess the range of variation in the timetable of PFCw growth across datasets. We first fit a smooth spline with the prefrontal cortex white matter (PFCw) volume versus age for males and another one for females. We subsequently fit another smooth spline through these data to average values across males and females. i–n, To test how sampling impacts the age of growth cessation, we subsampled the number of individuals and applied a nonlinear regression on the log-transformed PFCw volumes versus age expressed in days after conception in macaques and humans. We extracted the age in which PFCw growth cessation (i, l), the percentage of variance accounted for by the model (j, m), and significance tests (i.e., p values; k, n). e, The 95% confidence intervals show that the macaque PFCw ceases to grow between ∼0.6 and ∼1.3 years of age. j, k, m, n, These nonlinear regressions systematically account for a high percentage of the variance (approximately >70%; j, m) and are statistically significant (p < 0.05; k, n). These analyses show a lack of evidence for protracted FC development in humans (Extended Data Table 4-1, Table 4-2).
Table 4-1
Nonlinear regressions used to compare growth of the PFC and FC white matter. Download Table 4-1, XLSX file.
Table 4-2
PFC measurements collected at various time points in macaques. Download Table 4-2, XLSX file.
Cross-species variation in tractography of adult FC circuits
We used diffusion MR tractography to test for modifications to FC connectivity across humans, macaques, and mice (Figs. 5–10). We compared the orientation and location of fibers from tractography, tract-tracers, immunohistochemical markers, and myelin stains to assess sites of potential inaccuracies in the tractography. These observations were used to develop a quantitative procedure to compare FC pathway types across species (Figs. 7, 9). In mice, the direction and location of tracts within the white matter were concordant across methods, but tracking accuracy appeared to be limited at the gray matter–white matter boundary. We therefore classified pathways based on the orientation and direction of pathways within the white matter (Fig. 9a–i). We placed voxels in spaced sections through the FC white matter, and we classified pathways according to their orientation (Figs. 7–10). The percentage of corticocortical pathways was significantly greater in the primate FC relative to mice. No significant differences were observed between humans and macaques (ANOVA: F = 14.82, p < 0.01; n = 14). Overall, we found no significant differences in the relative proportion of FC and PFC pathway types across primates (t tests; Fig. 10g–i). Subcortical pathways were significantly expanded in the PFC of humans versus macaques (t = 5.1, p < 0.008). Relative FC pathway types were relatively invariant with respect to sampling, tested directions, pathway reconstruction, and resolution (Figs. 8–10). Although the cross-species variation in pathway types or lack thereof were robust to variations in imaging parameters, the limitations inherent to tractography led us to generate evidence from multiple scales to ensure the accuracy of these findings.
Some examples of the diffusion MR tractography of human brains in lateral, dorsal, and ventral views. The scans used in the study show consistency across individuals and imaging protocols. a–g, Whole brains (a–d) and a left hemisphere (e–g), some of which are also housed by the Allen Brain Institute, were scanned on a 3 T Siemens Tim Trio scanner at the Massachusetts General Hospital Athinoula A. Martinos Center for Biomedical Imaging (c, d). See Extended Data Table 5-1 for scan details. A 1-mm-thick horizontal slice filter set through the human brains show fibers coursing across the brain such as the cingulate bundle and the arcuate fasciculus. These data show consistent high resolution in fiber tracking, and consistency across individuals and scanning protocols. c, d, Two of the brains used from the Allen Brain Atlas. These scans vary in their resolution (isotropic: 1.2 and 0.9 mm, respectively). e–g, A left hemisphere shown at different minimum length thresholds (e, 0 mm; f, 20 mm; and g, 50 mm). g, We visualized pathways at different thresholds to reveal fibers of various lengths coursing through the white matter such as the arcuate fasciculus. The color coding of tractography was based on a standard red-green-blue (RGB) code based on average fiber direction. A, Anterior; P, posterior; R, rostral; C, caudal; M, medial; L, lateral; D, dorsal; V, ventral.
Table 5-1
Scanning parameters used for diffusion MR tractography of humans, macaques, and mice. Download Table 5-1, XLSX file.
Brain tractography shows pathways coursing through the brain of Old World monkeys used in the present study. a–g, Those include several macaques such as a rhesus macaque (a–c, scanned with 120 directions; d, scanned with 60 directions), a crab-eating macaque (e), a toque macaque (f), and a Sykes monkey (g). The tractography is consistent across individuals. Slices set through the prefrontal cortex (c) and larger areas through the FC (c, h–k) capture fibers emerging from and terminating within the FC. These settings show that the FC is composed by a preponderance of pathways emerging and terminating within the FC. A minimum-length threshold (15 mm) was set for visualizations purposes only. Brain masks are overlaid on these pathways. The color coding of tractography is based on a standard red-green-blue (RGB) code based on fiber direction. A, Anterior; P, posterior; R, rostral; C, caudal; M, medial; L, lateral; D, dorsal; V, ventral.
a–i, k–n, We compared diffusion MR tractography of mouse brains (a) with tract-tracers (b–h), at different ages (P21 and P60; i), and histologic data (k–n) to evaluate the accuracy of the tractography. ROIs were used to capture well known pathways coursing through the white matter (b–d, primary somatosensory pathway; e, hippocampal commissure; f, corpus callosum; g, anterior commissure; h, thalamocortical fibers from the lateral geniculate nucleus to the primary visual cortex; white arrowheads). Tract-tracers injected in the primary somatosensory area (upper limb) show fibers radially aligned within the gray matter (a, b, white arrows). Fiber orientation determined from tract-tracers and tractography are concordant across methods within the white matter, but sharp turns at the gray matter–white matter boundary are associated with reduced accuracy of diffusion MR tractography (e, f). For example, an ROI set to capture the corpus callosum accurately tracks fibers coursing contralaterally but does not accurately trace fibers penetrating the gray matter (f, white star). We refrained from using end points within the gray matter as a basis for classification because of the potentially reduced accuracy of the tractography at the gray matter–white matter boundary. i, j, The tractography (i) and relative proportion of pathway types (j) are very similar across mice at P21 and P60 (j). k, l, Gallyas stains show myelinated axons course radially within the gray matter but course preferentially along the medial axis to the lateral axis in the white matter (red boxes, close-up views). m, n, Antibodies against neurofilament medium polypeptide (NEFM) and neurofilament heavy polypeptide (NEFH), which label neurons of small (m) and large (n) calibers, respectively, support these observations. NEFH+ and NEFM+ neurons are oriented radially within the gray matter but course across the medial to lateral axis in the white matter; white boxes highlight the close-up views shown in m and n of NeuN- and Pval-labeled neurons. These observations suggest that axons of either large- or small-caliber axons make a sharp turn at the gray matter–white matter boundary. Sharp turns at the gray matter–white matter boundary are associated with reduced accuracy of diffusion MR tractography. Tract-tracer data are from the Allen Brain Institute Mouse Connectivity Database (Extended Data Table 7-1).
Table 7-1
List of tract-tracer experiments used to compare diffusion MR tractography with tract-tracers. Download Table 7-1, XLSX file.
a–k, We tested how tractography reconstruction, image resolution, and sampling impacts the tractography across mice (a-d), humans (e–k), and macaques (g–k). We considered how DTI and HARDI (a, b) impact pathway types in mice. We randomly selected voxels through the white matter, and we quantified the percentage of pathway types reconstructed with DTI and HARDI. a, b, Dorsal views show tracts reconstructed with HARDI (a) and DTI are similar (b). c, The relative percentages of pathway types reconstructed from DTI and HARDI from P60 mice strongly correlate (R2 = 0.89, p < 0.01, n = 16). d, We randomly subsampled voxels and computed the relative percentages of pathways coursing through the mouse FC and found that the relative number of pathway types is relatively invariant with respect to sample size. We considered how image resolution (900 µm vs 1.2 mm) impacts the pathway proportions in humans (e, f) and how direction number (60 vs 120) impacts pathway types in macaques (g, h). Pathway proportions vary little with sample size in humans (f) and macaques (h). The percentages of corticocortical (i), cortico–subcortical (j), and collosal pathways (k) are similar in humans and macaques regardless of sampled voxels used to classify pathways. Horizontal dashed lines and associated values show the relative mean pathway types per scanning parameters with varying sample size in humans and macaques. The percentages of pathway types are robust to variation in resolution and sampling. A minimum length threshold was set to 7.2 mm for humans, 5.8 mm for macaques, and 2 mm in mice for visualization purposes. This human brain is made available by the Allen Brain Institute (Ding et al., 2016). Ninety percent of the fibers are skipped to better visualize pathways in macaques. A, Anterior; P, posterior; M, medial; L, lateral; D, dorsal; V, ventral.
a–i, Tractography of macaque (a–c), human (d–f), and mouse brains (g–i). Pathways identified from randomly selected voxels include collosal fibers (b, e, h), U fibers (f), and long-range cortically projecting pathways (c, i). Fibers preferentially terminate within gyri in macaques and humans. j–s, We investigated how brain structure associates with biases in the termination of fibers in macaques. We evaluated fiber terminations on FA images of macaque brain scans. j–m, Dashed lines at the intersection between the gray matter (GM) and white matter (WM) of FA images show that FA is particularly low (dark) at the gray matter–white matter boundary. FA varies across the cortex, with high FA within gyri (i.e, high likelihood of fiber crossing) and low FA toward sulci (low likelihood of fiber crossing). Spheres capturing fibers through the white matter show that fibers preferably penetrate the cortex within gyri. n–s, Close-up views through select cortical areas show that fibers preferentially penetrate the gray matter at areas of high FA (o–s, blue arrowheads) rather than low FA (red arrowheads). These observations suggest that fibers penetrating gyri may be overrepresented at the expense of those closer to sulci because of increased uncertainty in tracking fibers penetrating sulci. It is because of these kinds of uncertainties in the tractography that the classification of pathways is not constrained by their precise termination of fibers within the gray matter but is focused instead on their orientation within the white matter. FA images and tractography are from the study by Calabrese et al. (2015). a, b, Brain pathways for macaques (a) and humans (b) are partially transparent.
a–f, j–s, Analysis of pathways with tractography (a–f), and cytoarchitecture (j–s) reveal modifications in adult FC circuits across humans, macaques, and mice. a–f, The diffusion MR scans identify pathways coursing through the brain. c, Brain tractography in humans with variable minimum length thresholds (0, 20, and 40 mm) reveal different pathways across the cortex. In particular, there appears to be a preponderance of short fibers within the FC, as evidenced by a lack of tracts within the anterior cortex (white arrowheads) when 20 or 40 mm minimum length thresholds are set. d, Horizontal slices (1 mm wide) show pathways coursing across the white matter, some of which span the FC (e.g., cingulate cortex, arcuate fasciculus; arrowheads). e, f, Diffusion MR scans of macaque (e) and mouse (f) brains were also used to classify pathway types. g–j, Percentage of pathway types in the FC white matter of humans, macaques, and mice show that the relative percentage of corticocortical pathways is significantly greater in humans and macaques compared with mice (*p < 0.05). h, i, No significant differences in pathway type are detected between humans and macaques whether we consider the PFC (h), or the rest of the FC (i). j–k, The relative thickness of layer II–III, as defined by CALB1 expression and Nissl stains, is significantly greater in the primate FC compared with mice but not between humans and macaques. This is true whether we consider the anterior cingulate cortex (AC) or the superior frontal gyrus (SFG). Details of these data are in Extended Data Tables 10-1, 10-2, 10-3, and 10-4. l–o, PCAs were used to test whether transcriptional profiles of LRP or SE genes differ between humans and macaques. A PCA of orthologously expressed genes (m) shows that samples cluster by species but PCAs of expressed SE genes (n) and LRP markers (o) show that macaque samples cluster with those of humans and chimpanzees. These observations support the notion that transcriptional profiles of LRP neurons are highly similar between humans and macaques. p–s, CRYM (an SE gene) expression is higher in supragranular layers of humans and macaque SFG compared with mice (arrowheads). s, The expression of CRYM is significantly higher in primate supragranular layers compared with the FC of mice, but no significant differences were observed between humans and macaques. This is true whether we compare the mouse FC with the anterior cingulate cortex (AC), the SFG, or the precentral gyrus (PG) of macaques and humans. Humans and macaques share a constellation of conserved features in FC neural circuits.
Table 10-1
Statistics on pathway types through the PFC and remaining FC in humans and macaques. Download Table 10-1, XLSX file.
Table 10-2
Proportion of pathway types across the FC of humans, macaques, and mice. Download Table 10-2, XLSX file.
Table 10-3
Relative gene expression and thickness measurements across FC layers in humans, macaques, and mice. Download Table 10-3, XLSX file.
Table 10-4
Statistics results from Tukey's HSD tests to compare the relative thickness of FC layer II–III in humans, macaques, and mice. Download Table 10-4, XLSX file.
Cross-species variation in transcription of adult FC circuits
Given that FC corticocortical pathways are expanded in primates, we focused on layer II–III neurons, many of which form corticocortical projections. First, we considered the cytoarchitecture to test for modifications in FC long-range projecting fibers across mice, macaques, and humans (Fig. 10). We observed that layer II–III in the FC, as defined transcriptionally (i.e., CALB1 expression) and cytoarchitecturally, is relatively expanded in humans and macaques compared with mice. This was true whether we considered the anterior cingulate (CALB1: ANOVA: F = 22.72, p < 0.01; n = 8; Nissl: F = 20.17, p < 0.01) or the superior frontal gyrus (CALB1: ANOVA: F = 15.62, p < 0.01; n = 8; Nissl: F = 15.16; p < 0.01; Fig. 10j). Post hoc Tukey's HSD tests showed that the relative thickness of FC layer II–III is significantly different between primates and mice but not between humans and macaques (Fig. 10k). Moreover, the expansion of layer II–III in macaques is concomitant with an expansion of SE gene expression in the layer (Figs. 10–15). There are major differences in layer II–III between primates and mice, but strong similarity between humans and macaques.
We next considered transcriptional variation of SE and LRP markers to investigate cross-species variation in adult FC circuits (Figs. 10–15). In a PCA applied to log-transformed expressed orthologous genes, genes clustered according to the phylogenetic relationship of species [Fig. 10m; first three principal components (PCs): 71.77% of the variance; n = 10 682]. In PCAs on the log-transformed expressed SE genes (Fig. 10n; first three PCs: 73.9% of the variance; n = 17) and LRP markers (Fig. 10o; first three PCs: 77.07%; n = 235), primates clustered together, demonstrating the strong similarity in expression of SE and LRP genes between humans and macaques. We further evaluated the spatial pattern of expression of an SE gene named CRYM to confirm similarities in SE gene expression in the FC of humans and macaques (Figs. 10m–p, 14). We quantified the relative expression of CRYM across supragranular and infragranular layers in different cortical areas of humans, macaques, and mice. We found that expression patterns varied across species whether we compared the mouse FC with the superior frontal gyrus (ANOVA: F = 8.94, p < 0.01; n = 11), the anterior cingulate (F = 4.86, p < 0.01; n = 11), or the precentral gyrus (F = 22.04, p < 0.01; n = 8) of macaques and humans (Fig. 10m). Tukey's HSD tests showed no significant differences in the expression profile of CRYM between humans and macaques. We found significant differences between primates and mice. Together, these data showcase the strong similarities in FC neural circuitry between humans and macaques.
We identified 250 LRP markers by systematically testing for variation in transcription and myelination across cortical areas. a, We considered gene expression across multiple areas and datasets as highlighted on a schematic of a human brain (Liu et al., 2016; Zhu et al., 2018). b, c, We used a single-cell RNA sequencing dataset where clusters of cell populations are defined with tSNE (t-distributed stochastic neighbor embedding) distributed stochastic neighbor embedding (Bakken et al., 2020) to filter candidate genes by cell type. c, LRP markers include genes that are expressed by layer II–III neurons but not those expressed by non-neuronal cells. d, We extracted MWF values from equations of different lobes. e–g, We fit a smooth spline through the log10(RPKM) values versus age expressed in days after conception (DAC) to extrapolate normalized gene expression. We tested for associations between myelination and gene expression across areas and datasets. Two examples of association are shown in e (for the PFC) and f (for the MFC) using the data (Liu et al., 2016; Li et al., 2018). We selected significant and positive (slope, >0) associations after correcting for multiple testing (BY test, p < 0.05). g, LRP markers (e.g., NEFH, NRGN, RGS4, CAMK2A) are expressed by pyramidal neurons (arrows) in the frontal cortex and other cortical areas. Boxes indicate regions shown in higher-power views. AC, Anterior cingulate, T, temporal; A1, primary auditory cortex; DFC, dorsolateral frontal cortex; IPC, inferior parietal cortex; ITC, inferior temporal cortex; MFC, medial frontal cortex; M1C, primary motor cortex; OFC, orbitofrontal cortex; S1C, primary somatosensory cortex STC, superior temporal cortex; VFC, ventral frontal cortex; V1, primary visual cortex. Details of these data and results from these analyses are in Extended Data Tables 11-1, 11-2, and 11-3.
Table 11-1
Cell clusters from the human primary motor cortex used to define layer II–III neurons and non-neuronal cells. Download Table 11-1, XLSX file.
Table 11-2
Equations used to capture temporal trajectories in MWF. Download Table 11-2, XLSX file.
Table 11-3
List of LRP markers identified from associations between normalized gene expression and MWF. Download Table 11-3, XLSX file.
Expression of the SE genes Vamp1 and Scn4b in the FC of mice, macaques, and humans. In situ hybridization and corresponding Nissl stains are from the Allen Institute for Brain Science. a–f, Vamp1 (a–c) and Scn4b (d–f) expressions in coronal sections through the FC of a mouse at P56 (a, d), a macaque at 48 m (male; b, e), and an adult human (46-year-old male control; c, f). Vamp1 (a) and Scn4b (d) expressions in the mouse FC cortex region are shown, along with enlarged views and corresponding Nissl images. Both genes appear to be widely expressed throughout layers, including layers II–III. In macaques, the ISH signal was variable (n = 3). However, staining was observed in a region anterior to and within the motor cortex. Sections through these regions in the macaque, along with enlarged detail views, show VAMP1 (b) and SCN4B (e) expression in layers II–IV of the superior frontal gyrus (SFG) and precentral gyrus (PrG). The ISH signal was also variable in humans in the dorsolateral PFC (DL-PFC). A section through the middle frontal gyrus (MFG) in the DL-PFC shows that (c) VAMP1 and (f) SCN4B are expressed in layers II–III, similar to what is observed in macaques. In macaques and humans, VAMP1 and SCN4B appear preferentially expressed in layers II–III. In contrast, these genes are more widely expressed across layers in mice. All in situ hybridization and Nissl images were obtained from the Allen Institute for Brain Science. Scale bars: a, 1 mm and 500 µm; b, 5 mm and 500 µm; c, 5 mm and 500 µm.
Expression of CALB1 in cortical layers II–III of the FC across mice, macaques, and humans. a, Coronal sections through a P56 mouse cortex show that Calb1 is restricted to layers II–III. The enlarged detailed views show a region of the primary and secondary motor cortex. The anterior forceps of the corpus callosum (fa) is also shown. b, c, Calb1 expression in the FC of macaques (n = 3; b) and humans (n = 15; c) is similar and shows the expansion of cortical layers II–III in the two species. b, Sections through two regions of the FC (blue lines, schematic) of a macaque (48-m, male) show Calb1 expression and enlarged detailed views, with corresponding Nissl images, of several gyri: a, c, anterior cingulate gyrus (ACG); b, middle frontal gyrus (MFG); d, lateral orbital gyrus (LOrG). c, Sections and enlarged detailed views (with corresponding Nissl images) of three regions of the FC in a human (28-year-old male control; blue boxes, schematic) showing Calb1 expression in the following gyri: e, superior frontal gyrus (SFG); f, MFG; g, cingulate gyrus (CgG); and h, inferior frontal gyrus (IFG). All in situ hybridization and Nissl images were obtained from the Allen Institute for Brain Science. Schematics modified from the National Institutes of Health blueprint nonhuman primate (NHP) Atlas and the Allen Human Brain Atlas. Scale bars: a, 1 mm and 500 µm; b, 1 mm and 500 µm; c, 5 mm and 500 µm.
Variation in CRYM expression in the supragranular layers across the mouse, human, and macaque FC. CRYM in situ hybridization and Nissl data and modified atlas schematics are from the Allen Institute for Brain Science. a, In the mouse (P56, male), Crym is not expressed in the supragranular layers (II–III) of the cortex. Enlarged detailed views show the primary and secondary motor areas of the cortex. For these views, a corresponding Nissl image is also shown. b, By contrast, strong CRYM expression is observed in these cortical layers in humans (n = 4), as observed in sections of several gyri at different rostral–caudal levels of the FC (31-year-old male control). Enlarged detailed views show CRYM expression and corresponding Nissl staining in the superior frontal gyrus (SFG) and middle frontal gyrus (MFG), and the SFG and PrG (precentral gyrus) in a rostral and caudal region of the FC, respectively. Overall, CRYM expression is similar across the four human specimens analyzed. c, CRYM is also expressed in the supragranular layers in macaques (n = 3), similar to humans, although its expression is variable across cortical regions. In the specimen shown (48-m, male), for example, CRYM expression in the SFG and MFG in a rostral region of the FC appears to be weaker than in similar gyri in more caudal regions. This variation is consistent across the three specimens analyzed. CgG, Cingulate gyrus; fro, frontal operculum; fa, anterior forceps of the corpus callosum; MFG (blue in schematic); PrG (dark green); SFG (light green); STG, superior temporal gyrus. For others, we used legends available at the Allen Brain Map (https://portal.brain-map.org/). The following anatomic reference atlases were used: the Allen Mouse Brain Atlas, the National Institutes of Health Blueprint Non-Human Primate (NHP) Atlas, and the Allen Human Brain Atlas. Scale bars: a, 1 mm and 500 µm; b, 5 mm and 500 µm; c, 5 mm and 500 µm.
NEFH expression is greater in cortical layers II–III in the macaque compared with the mouse. Histological information from macaques (a) and mice (b–d) shows NEFH expression is greater in cortical layers II–III in the macaque (a) compared with mice (d–e). The expression of NEFH is of particular interest because it is expressed by large neurons thought to project over long distances. Spatial variation in NEFH expression can be used to make inferences about modifications to connectivity patterns. Somas of large layers II–III neurons are expected to project cross-cortically, whereas somas of neurons located in layers V–VI are expected to project subcortically. Accordingly, an increase in NEFH expression in layers II–III suggests an amplification of long-range cross-cortically projecting neurons. a, In macaques, NEFH expression is relatively high in layers II–III and V–VI across the frontal cortex. This is in contrast to mice, which have relatively thin layers II–III. d, e, We used Calb1 as a marker for layers II–III; the expression of Calb1 spans layers II–III in mice. NEFH mRNA (d) and protein expression (e) is high in layers V–VI but extremely low in layers II–III in mice. d–g, Close-up views through the mouse FC in d and e also show that NEFH mRNA (f) and protein (g) expression in layers II–III are particularly low in mice. In contrast, macaques show relatively stronger expression of NEFH protein in layers II–III across multiple frontal cortical areas (h–k) except for the cingulate gyrus (j). These qualitative observations demonstrate major modifications in frontal cortex circuits between macaques and mice. The increased NEFH expression in layers II–III of macaques relative to mice suggests an expansion in cross-cortically projecting FC neurons may have emerged early in primate evolution.
Discussion
The integration of transcription with neuroimaging is an effective approach to identify conservation and variation in biological programs linked to modifications of circuits. This integrative approach creates novel opportunities to study circuits in human health and disease, and across species.
Corresponding ages from transcriptional and structural variation
We identified corresponding postnatal ages in humans and macaques. This work builds on a previous line of work called the Translating Time Project, which relied on abrupt changes that unfold during development to find corresponding ages across model organisms and humans (Finlay and Darlington, 1995; Clancy et al., 2001; Zhu et al., 2018; Charvet, 2021). We collected 354 transformations across 19 mammalian model organisms to find corresponding time points during prenatal development. Extracting time points from gradual changes in transcription and structure, in addition to abrupt transformations, reveals corresponding postnatal ages across species (Charvet et al., 2017; Zhu et al., 2018). Each metric has uncertainties, but the use of multiple metrics ensures the robust determination of age alignment across species. The findings serve to extrapolate information from model systems to humans and to identify which developmental programs occur for an unusually long or short time in humans.
Limitations and opportunities for diffusion MR imaging
Diffusion MR tractography reveals a three-dimensional perspective of pathways (Wedeen et al., 2012; Edlow et al., 2019; Charvet et al., 2020). The sole use of tractography to trace neural circuits is problematic because of its limited ability to resolve crossing fibers or locate tract termination sites (Figs. 7, 9). Considering these caveats, we developed a novel approach to quantify FC pathway types. This approach was guided by a host of techniques to label axons, including tract-tracers, immunohistochemistry, and myelin stains. The orientation and location of fibers from diffusion MR tractography aligned with observations from histology and tract-tracers in the white matter (Fig. 7). Our analyses, which were designed to overcome limitations of diffusion MR tractography, withstood variations in sampling, direction, and resolution (Fig. 8). Nevertheless, we still lack methods to ensure the accuracy of diffusion MR tractography (McNab et al., 2009; Thomas et al., 2014). Given the lack of alternative methods to map connections, we integrated our findings with transcriptomic and structural information to more accurately trace pathways across species.
Enhanced methods to study FC circuits in primates reveal conservation
Although RNA sequencing from bulk and single cells offers an unprecedented perspective to track developmental programs, these metrics lack key information about the structural composition of circuits. There is often a lack of one-to-one correspondence between projection patterns and gene expression (Molyneaux et al., 2009). We, therefore, identified LRP markers by aligning structural and transcriptomic variation during human development. This approach was instrumental in identifying genes that can be used to track modifications to long-range circuits in different species without specifically tracking the location of pathway end points. Modification in the absolute or temporal profiles in the expression of these genes may be used to detect modifications in connectivity profiles of long-range projection patterns.
Past work considered white matter volumes or transcriptional information to assess whether FC circuits are unusual in humans. However, these studies have traditionally used these methods in isolation, and did not reach a consensus on whether the size or the developmental timeline of the frontal cortex or prefrontal cortex is unusually expanded in humans versus other primates (Semendeferi et al., 2002; Carlén, 2017; Donahue et al., 2018). The lack of tools to compare brain development (e.g., age alignments) and the small samples available at a single scale of study have precluded our ability to rigorously test the notion that the human frontal cortex timeline is extended relative to other species. We have worked to expand the repertoire of integrative tools to study pathways across the mammalian brain. We integrated diffusion MR tractography with histologic information (Charvet et al., 2017) and, subsequently, with one or a handful of genes (Charvet et al., 2019; Charvet, 2021). The present study stands out relative to past studies in harnessing a multiomics approach to trace the evolution of frontal cortex pathways.
We drew from multiple lines of evidence to test for modifications in FC circuitry across species over the course of development and in adulthood. We compared trajectories in transcriptional profiles of LRP neurons, white matter maturation, pathway types, and transcription across layers. Testing for differences across scales highlighted many conserved features in FC circuitry across humans and macaques (Sakai et al., 2011; Zhu et al., 2018), though there were some intriguing differences across species. Specifically, we found that the human PFC white matter possesses increased cortico–subcortical pathways relative to macaques. More work is needed to transcriptionally define pathway types across the primate frontal cortex to explore these species differences. More generally, the study shows that tracing human FC circuits from tractography, transcriptomic, and temporal dimensions is an exciting approach, one that will provide a more complete understanding of the evolution of circuits in the human and nonhuman primate brain.
Footnotes
This work was supported by Eunice Kennedy Shriver National Institute of Child Health and Human Development Grants 1R21-HD-101964-01A1 and 7R21-HD101964-02 (to C.J.C.); Institutional Development Award (IDeA) Networks of Biomedical Research Excellence (INBRE) Pilot Grant P20-GM-103446 from the National Institute of General Medical Sciences (NIGMS; to C.J.C.); NIGMS Core Center Access Award P20-GM-103446 (to C.J.C.); startup funds from Auburn University (to C.J.C.); Centers of Biomedical Research Excellence Grant 5P20-GM-103653 for research at Delaware State University; National Institute of Neurological Disorders and Stroke Grant R21-NS-109627 (to B.L.E.); and a James S. McDonnell Foundation grant (to B.L.E.). Opinions are not necessarily those of the National Institutes of Health. We thank Drs. Harrington, Whitaker, and Halley for help, and the Japan Monkey Center for access to brain scans.
The authors declare no competing financial interests.
- Correspondence should be addressed to Christine J. Charvet at charvetcj{at}gmail.com
