Cortical circuit principles predict patterns of trauma induced tauopathy in humans

Abstract

In brains of individuals who had sustained repetitive head trauma, advanced pathologic tau protein in neurons and axons within temporal cortices followed patterns seen in homologous cortico-cortical connections in non-human primates. The relational Structural Model, which is based on the universal principle of the systematic variation of cortical laminar structure, has successfully predicted the relative laminar distribution of cortico-cortical connections based on the relative similarity/difference in laminar structure in pairs of linked areas. Here, the Structural Model successfully predicted the graded laminar distribution and density of pathologic tau in chronic traumatic encephalopathy (CTE) and was validated by a computational progression model. By contrast, early and sporadic tau pathology in the depths of sulci in CTE followed local columnar connectivity rules. These findings support applicability of a theoretical model to unravel the direction and progression of tau pathology in neurodegeneration via cortical connection mechanisms. Cortical pathways converging on temporal cortices may help explain the inexorable spread of pathologic tau to widespread cortical areas accompanied by decline in emotional and cognitive processes in humans with repetitive head trauma.

Introduction

Cortico-cortical connections are summarized succinctly by the rules of the Structural Model (reviewed in (Barbas 2015; Tucker and Luu 2023)), which uniquely predicts the laminar distribution and, by extension, the probable processing direction in the cortex (Aparicio-Rodriguez and Garcia-Cabezas 2023). The model is rooted in the classical theory of cortical systematic variation, traced from the phylogenetically ancient limbic areas characterized by the simplest lamination to areas with progressive elaboration of laminar structure seen through six-layer eulaminate cortices. The classical principle of cortical systematic variation was discovered independently by several investigators working on different continents and with different species (reviewed in (Pandya et al. 2015)). Recent studies using massive transcriptomic analyses have enriched the set of cortical markers (Hanisch et al. 2023; Hansen et al. 2023), but lack a theoretical framework to succinctly systematize the rich data to address cortical organization and disruption in disease (reviewed in (Garcia-Cabezas et al. 2019)).

Based on the classic theory of cortical systematic variation (reviewed in (Pandya et al. 2015)), the Structural Model predicts the laminar pattern of connections by the structural relationship of each pair of linked cortices. Connections that originate in a cortex with higher laminar structure originate in the upper layers and their axons terminate in the middle layers (deep 3–5a) of the target area, in a pattern commonly called feedforward. In the reverse direction, pathways originate in the deep layers 5 and 6 and innervate layer 1, in a pattern called feedback. The model is thus relational so that connections between areas that differ markedly in laminar structure involve fewer layers and are sparse. In contrast, connections between areas that have comparable laminar structure are balanced within layers and are dense. The key predictions of the relational model uniquely reveal the direction of pathways, while density of connections alone is uninformative for processing direction ((Barbas and Rempel-Clower 1997); reviewed in (Barbas 2015)).

Unlike animal studies, demonstration of the patterns of cortical connections in humans with postmortem use of tracers has encountered methodological limitations so that labeling mostly involved short-range pathways (e.g., (Tardif and Clarke 2001; Tardif et al. 2007)). Here, we used a different approach by study of the laminar pattern of phosphorylated tau (p-tau) pathology in a human neurodegenerative disease, chronic traumatic encephalopathy (CTE), in which the tau pathology may spread along connected networks via trans-synaptic transport between affected neurons. However, the circuit mechanism of spread in p-tau has remained elusive (Braak and Del Tredici 2011; Liu et al. 2012; Clavaguera et al. 2015; Vogel et al. 2020; Delpech et al. 2021).

We used brains from individuals with neuropathologically verified CTE, a neurodegenerative p-tau pathology associated with repetitive mild head trauma, including concussive and nonconcussive head impacts. CTE is defined pathologically by the perivascular accumulation of p-tau as neurofibrillary tangles and neurites preferentially at the sulcal depths of the cortex (McKee et al. 2009; McKee et al. 2013). The severity of CTE pathology can be divided into a 4-tiered staging scheme, based on the density and regional distribution of p-tau pathology (McKee et al. 2009; Alosco et al. 2020).

The goal of this study was to determine whether the laminar pattern of p-tau pathology in CTE, and thus direction of spread, is consistent with cortico-cortical connection rules, and to simultaneously address the broad question of whether laminar connections in humans can be predicted by a theoretical model. We found that the laminar patterns and density of p-tau pathology in the medial temporal lobe (MTL) in stage III CTE bear a striking resemblance to the graded laminar patterns of cortico-cortical connections of non-human primates. Further, we provide evidence that the initial p-tau pathology in the depths of sulci of CTE stages I-II follows a different pattern, based on connectivity within narrow columns. Collectively, these findings provide support for the use of a theoretical model to predict the direction of spread and progression of p-tau pathology in time, as well as patterns of connection in the human cortex.

Materials and methods

Human postmortem brain and tissue preparation

The study was approved by the Institutional Review Board of Boston University. Data of human postmortem brain tissue are summarized in Supplementary Table 1. CTE cases were from the BU CTE Brain Bank (N=12; Supplementary Table 1). Human tissue from the CTE Brain Bank was fixed in periodate–lysine–paraformaldehyde (PLP) and stored at 4°C until block preparation and sectioning. In addition, we examined the cytoarchitecture of temporal cortices in postmortem human brain tissue from neurotypical individuals (N = 3, one female) to determine the laminar structure and cortical types of temporal areas, using Nissl histological staining, as described in detail earlier (Garcia-Cabezas et al. 2020, 2022; Sancha-Velasco et al. 2023). Donated postmortem human brains from the National Disease Research Interchange (NDRI) were immerse-fixed in 10% formalin. We then sliced each cerebral hemisphere in coronal slabs of 1 cm thickness, photographed the anterior and posterior surfaces of each slab and post-fixed them in 10% formalin for 2–4 days. After post-fixation, we matched the slabs based on atlases of the human brain (von Economo and Koskinas 1925/2008; Mai et al. 2015). We cryoprotected tissue slabs in a series of sucrose solutions (10–30% in 0.01 M PBS). The temporal blocks then were frozen in −75°C isopentane (Thermo Fisher Scientific, Pittsburgh, PA, United States) for rapid and uniform freezing (Rosene et al. 1986). We then cut blocks on a freezing microtome in the coronal plane at 50 μm and 10 consecutive series of sections were collected. Sections were stored in antifreeze solution (30% ethylene glycol, 30% glycerol, 40% PB 0.05 M at pH 7.4 with 0.05% azide) at −20°C for future assays.

p-Tau visualization in human postmortem tissue sections

For AT8 and CP13 phospho-tau histopathology, tissue blocks were embedded in paraffin for sectioning. Serial 10 μm sections were cut and mounted for immunohistochemistry, as described (McKee et al. 2010; Armstrong et al. 2017; Butler et al. 2022). Briefly, mounted sections were incubated overnight at 4 °C in primary antibody AT8 (a mouse monoclonal antibody directed against phosphoserine 202 and phosphothreonine 205 of PHF-tau; Pierce Endogen, Rockford, IL; 1:2000), or CP13 (a monoclonal antibody directed against phosphoserine 202 of tau, considered to be the initial site of tau phosphorylation in neurofibrillary tangle formation; 1:200; courtesy of Peter Davies) followed by biotinylated anti mouse secondary antibody. P-tau was visualized with a 3-amino-9-ethylcarbazol HRP substrate kit (Vector Laboratories, Burlingame, CA). Sections were coverslipped with Permount medium (Thermo Scientific, Rockford, IL).

Neuropathological evaluation and stage classification of CTE

Diagnosis and stage classification were based on characterization of pathognomonic lesions, as described in detail in published criteria for the neuropathological diagnosis of CTE (McKee et al. 2016); (Bieniek et al. 2021). CTE is characterized by progressive buildup of hyperphosphorylated tau (p-tau) in neurofibrillary tangles and neurites that initially surrounds small blood vessels at sporadic sulcal depths of the cerebral cortex and later is seen in different cortical and subcortical areas. The density and regional deposition of p-tau determines four pathological stages of CTE, ranging from stage I (mild) to stage IV (severe), with gradually increasing and more widespread p-tau deposition, with fibrils that have distinctive molecular structural configuration unlike the configurations observed in aging, Alzheimer’s disease, or other p-tau pathology (Alosco et al. 2020).

Animals, surgery, tracer injections, and tissue preparation

We used available postmortem rhesus monkey (Macaca mulatta, ~2–3 years, n = 3, female = 1) brain tissue for architectonic and pathway analyses (Supplementary Table 2). Sources of all antibodies, reagents, and software used in this study are listed in Supplementary Table 3. Animal protocols were approved by the Institutional Animal Care and Use Committee at Boston University School of Medicine and Harvard Medical School, in accordance with the ILAR Guide for the Care and Use of Laboratory Animals (publication 80–22 revised, 1996). We designed experimental procedures to minimize animal suffering and to reduce the number of animals needed for research, by placing multiple distinct tracers in each case, which we used for this and other unrelated studies.

Tracer injection sites and quantification of projection neurons have been described in detail in previous studies and will be briefly described here (Bautista et al. 2023). First, we obtained high-resolution magnetic resonance imaging (MRI) scans of the brain in vivo for surgical planning of tracer injections. Animals were sedated with ketamine hydrochloride (10–15 mg/kg, i.m.), then anesthetized with propofol (loading dose 2.5–5 mg/kg, i.v.; continuous infusion rate 0.25–0.4 mg/kg/min), and placed in a stereotaxic apparatus (1430M; David Kopf Instruments) for the scan. We used MRI scans to calculate stereotaxic coordinates using the interaural line as reference for five injections of neural tracers in several temporal cortical areas (Table 2).

We performed surgery for tracer injections under general anesthesia (isoflurane, to a surgical level) with continuous monitoring of vital signs. We injected one or three tracers in each monkey, as shown in Supplementary Table 2. The dyes injected included fluorescent and other tracers: Fast Blue (FB, case AT, 3% dilution, 2 μl; Polysciences), Diamidino Yellow (DY, case AT, 3% dilution, 4 μl,), Fluoro-ruby (FR, case BC, (10% solution, mixture of 10 kDA and 3 kDA, 4 μl; Invitrogen), and Biotinylated Dextran Amine (BDA, cases AV, AT: equal parts 10% 10 kDA and 10% 3 kDA, 5 μl; Invitrogen). Dextran amines with molecular weight 10 kDA (BDA in AT and AV) are optimal for anterograde labeling of axonal projections and their terminations, while the 3 kDA variant is optimal for retrograde labeling of cell bodies and proximal dendrites (Veenman et al. 1992; Richmond et al. 1994; Reiner et al. 2000). Post-operatively we monitored animals and gave antibiotics and analgesics.

After a survival period of 18–20 days, the animals were anesthetized with a lethal dose of sodium pentobarbital (~50 mg/kg, i.v., to effect), and perfused transcardially with 4% paraformaldehyde in cacodylate buffer or in 0.1M PB at pH 7.4 (cases AT, AV), or 4% paraformaldehyde and 0.2% glutaraldehyde in 0.1M PB, pH 7.4 (case BC). The brains were removed, photographed, cryoprotected in ascending concentrations of sucrose solutions (10–25% sucrose in 0.1M PB, pH 7.4, with 0.05% sodium azide; Sigma Aldrich), and frozen in −80°C isopentane (Rosene et al. 1986). We used a freezing microtome (AO Scientific Instruments/Reichert Technologies) to section brains in 50 μm-thick coronal sections, for case BC, or 40 μm for cases AT, AV, producing ten matched series of sections that were stored in antifreeze (30% ethylene glycol, 30% glycerol, 0.05% sodium azide in 0.05M phosphate buffer, pH 7.4), until further processing.

Tracer visualization in rhesus monkey tissue sections

In the cases with injection of fluorescent tracers, sections were mounted immediately after cutting, dried, coverslipped with Krystalon (Millipore), and labeled neurons were mapped directly, as described (Zikopoulos et al. 2018). Briefly, sections used to visualize neurons retrogradely labeled for tracers (FB, DY or FR) were viewed under a microscope with epifluorescence attachment (Nikon Optiphot) that includes an analog-digital microscope/computer interface (Nikon, Optiphot/Austin), for mapping and counting labeled neurons and axon terminals with custom-made software or using a commercial system (Neurolucida, v.10.40 Microbrightfield).

To visualize projection neurons labeled by BDA we incubated free-floating sections for 1 h at 4°C in 0.05M glycine (Sigma-Aldrich) and preblocked for 1 h at 4°C in 5% normal goat serum (NGS; Vector Laboratories), 5% bovine serum albumin (BSA; Sigma-Aldrich), and 0.2% Triton-X (Sigma-Aldrich) in 0.01M PBS. Sections were then incubated for 1 h at 25°C with avidin-biotin horseradish peroxidase (AB-HRP; Vectastain Elite ABC kit, Vector) at a 1:100 dilution in PBS, followed by three rinses in PBS (10 min each), and tracer was visualized with incubation for 1–3 min in diaminobenzidine (DAB substrate kit, Vector). Sections were mounted on gelatin-coated glass slides, dried, counterstained with Nissl (thionin stain, every other section) as previously described (García-Cabezas et al. 2016), and coverslipped with Entellan (Sigma-Aldrich).

Quantitative analysis of p-tau labeling in the human cortex in CTE

We quantified overall p-tau labeling density across layers and cortices in immunostained coronal sections from cases with early (I and II) and advanced (III) CTE stages. We estimated the laminar p-tau content from images of representative columns from each area in the human frontal and temporal cortices captured at the light microscope (Olympus BX 51), under brightfield illumination, with a CCD camera (Olympus DP70), connected to a workstation running imaging software (DP Controller, cellSens Standard 3.2, Olympus). We captured images using 4x, 10x, and 20x lenses with the same light exposure to minimize background variability, as described (Zikopoulos et al. 2018). We imported images into ImageJ (v2), for further analyses of the overall signal density, the number of labeled neurons, as a proxy for retrograde labeling, and the density of p-tau signal excluding neurons, as a proxy for anterograde labeling, since abnormal tau labels neurons and axons and their terminals (Liu et al. 2012; Braak and Del Tredici 2018; Vogels et al. 2020; Butler et al. 2022). To complement the approach of using representative columns, we performed the same type of analysis along the entire cortical ribbon, spanning all medial and lateral temporal regions with positive p-tau immunostaining. Moreover, to determine whether gyral, sulcal, and straight parts of the cortex show variability in the density of p-tau labeling, we also analyzed all gyral, sulcal, and straight cortical regions separately, which are squeezed or stretched at different rates, due to mechanical forces as the cortex folds (Van Essen 1997; Hilgetag and Barbas 2006). This analysis is needed because as the cortex folds p-tau appears denser at the depths of sulci where the cortex is compressed and less dense in parts where the cortex is stretched.

Overall and laminar signal density

To measure signal density, we converted images into 8-bit gray scale, and inverted them, so that signal intensity measurements ranged from 0–255 (Zikopoulos et al. 2018). We measured levels of background staining in gray matter regions with no specific immunostaining within each image and subtracted background pixel values from each image to eliminate staining inconsistencies, due to experimental variability. We obtained optical density (mean gray value) and surface area measurements from each layer within frontal and temporal cortical gray matter columns and used these to additionally estimate the integrated density of the signal (mean gray value / area). We additionally used image thresholding to estimate the area fraction of p-tau labeling for each image, as an independent measure of signal density.

Laminar distribution and density of labeled neurons (retrograde)

We computed the number of labeled neurons across layers in representative cortical columns within frontal (for early CTE stages I and II) and temporal (for CTE stage III) cortices. To describe the laminar distribution of labeled neurons within and across cases, we expressed laminar specificity as the ratio of labeled neurons in the upper layers (II-III), referred to as the supragranular index (SGI) for each site.

Laminar distribution and density of p-tau immunolabeling in axons and terminals (anterograde)

We estimated signal density excluding the signal from labeled neuron bodies, as a proxy for axonal p-tau signal. We segmented images using ImageJ to produce a binary image mask that facilitated extraction of only the pixels within cell bodies of interest and exclusion of other aspects of the image, as described (Joyce et al. 2020). The resulting mask contained an outline of all p-tau+ cells. Each manually segmented binary mask was then used to extract image pixels from the matched grayscale original image using MATLAB (2019b, MathWorks, Natick, MA). To perform the optical density analysis, we normalized each extracted image so that the maximum brightness was 1. For each image, the average p-tau intensity per pixel was computed for the neuropil and was used to estimate the density ratio of signal in the superficial, deep, and all layers to express the supragranular index (SGI) as labeling in superficial / all layers.

Analysis of anterograde and retrograde labeling in the rhesus monkey cortex

We mapped cortical pathways directed to medial and lateral temporal areas in a series of coronal sections through the medial and adjacent inferior temporal (visual) and superior temporal (auditory) association and multimodal areas in the depths of the superior temporal sulcus, using exhaustive plotting in 1 out of 20 coronal sections through the ipsilateral hemisphere, as described (Bautista et al. 2023). In each series of sections, we mapped all labeled cortical neurons in the temporal cortex in the superficial (II-III), and deep (V-VI) layers, at 200x with brightfield or epifluorescence microscopy (Olympus BX60) and a semi-automated commercial system (Neurolucida v.10.40, MicroBrightField). We normalized estimates by dividing the number of labeled neurons in a specific area by the total number of labeled neurons in the temporal lobe of each case, to account for likely variation across cases in the size of injection sites, tracer transport dynamics, and immunolabeling, even when identical procedures are used. We summed labeled neurons across sections for each cortical area in each case to describe the laminar distribution of labeled neurons within and across cases and expressed laminar specificity as the ratio of labeled neurons originating in the upper layers, the SGI. We estimated the mean and standard error of the mean of these ratios across cases.

We studied anterograde tracer termination patterns qualitatively in cases AT and AV (tracer injections in A36 and TE1/TS1) by analyzing the efferent projections from the injection sites to other temporal areas using ImageJ. We captured darkfield images using an Olympus optical microscope (BX53) with a CCD camera (Olympus DP74) connected to a commercial imaging system (DP Controller, Olympus). In ImageJ, we used the auto threshold feature and the moments method to analyze the images, as this best represented the signal of the tracer. We used the binary images to obtain the measurement for area fraction, which is the area of the image that was occupied by the signal. We acquired the area fraction measurement with the highest intensity within each area for the column (variable height among areas and width of 150μm). We then computed the proportion of area fraction of each area of the temporal lobe by dividing the measurement recorded per area over the total amount of area fraction measured in each case.

Statistical analysis

We used linear regression analysis to fit a line through each set of observations to examine the relationship between p-tau labeling and changes in cortical laminar structure. The fit can be used to predict the extent to which the laminar distribution and density of p-tau labeling may be affected by the gradual changes in cortical laminar architecture. We report significance and $F$ values for 95% confidence level.

Propagation model

We modeled the propagation of p-tau using a simple bounded integration model. The increment of the proportion of p-tau p in a given layer m of cortex area A at time step t given by

$$\mathrm{\Delta }{p}_m^A(t)=\sum _{n,B}{W}_{mn}^{AB}{p}_n^B$$

Where $W_{mn}^{AB}$ is the strength of a connection to layer $m$ of cortical area $A$ from the layer $n$ in a connected cortical area $B$. The sum is taken over all connected cortical areas ($B$) and layers ($n$). The values of proportion p are bound to be less than or equal to 1. For the simulation here, all cortices are assumed to be connected to each other. Note that local connectivity within a cortical area is omitted for simplicity.

We implemented the rules of the relational Structural Model by letting the difference in degree of lamination D regulate the laminar connection strength. The degree of lamination D was specified by a score varying from 0 to 1, with 0 corresponding to limbic cortex with the simplest laminar architecture, and 1 corresponding to eulaminate cortex with the most complex laminar structure. The connection to an infragranular layer (i) in cortex A from a supragranular layer (s) in a connected cortex B is given by:

$$W_{is}^{AB}=r(1-[D_A-D_B{]}^{+})$$

Similarly, the connection to a supragranular layer (s) in A from an infragranular layer (s) in B is given by:

$$W_{si}^{AB}=r(1-[D_B-D_A{]}^{+})$$

The expression $[x{]}^{+}=x$ when $x>0$ and is 0 otherwise. The constant $r$ determines the maximum possible strength of the connection.

If cortex A is maximally limbic ($D_A=0$) and cortex B is maximally eulaminate ($D_B=1$) then $W_{is}^{AB}=r$ and $W_{si}^{AB}=0$, corresponding to a pure feedforward projection from B to A. For the reverse situation, when A is maximally eulaminate ($D_A=1$) and cortex B is maximally limbic ($D_B=0$) then $W_{is}^{AB}=0$ and $W_{si}^{AB}=r$, corresponding to a pure feedback projection from B to A. If the two cortical areas have the same degree of lamination, then both feedback and feedforward projections have equal and maximal strength ($r$), corresponding to a ‘lateral’ pattern that involves more layers. Intermediate levels of difference in degree of lamination produce biased patterns that are either more feedforward or more feedback, depending on the sign of the difference.

The initial distribution of p-tau is determined by a Gaussian function as follows:

$$p_i^X=k\mathrm{ }\mathrm{e}\mathrm{x}\mathrm{p}\left(-\frac{{\left(D_X-D_o\right)}^2}{2{\mathrm{\sigma }}^2}\right)$$

The center of the Gaussian is given by $D_o$, and for the simulations shown here, it takes a value of zero. The constant $\mathrm{\sigma }$ determines the width of the Gaussian. This initial condition corresponds to a situation where the infragranular layers in the simplest limbic cortex are the densest with p-tau, and the density gradually falls off in the direction towards eulaminate cortex with the highest complexity in laminar structure. Constant $k$ specifies the peak level of p-tau in the initial distribution. Index $X$ runs over all the cortices (N = 10). Other initial conditions were also simulated. These results are shown in the Supplementary Materials.

To demonstrate the change in the SGI trend over time as CTE progresses, the slope of a linear fit of SGI versus degree of lamination was computed. These results could be contrasted with the statistical analysis of empirical data on early versus late stage III CTE.

Simulation of the above model was run for 30 timesteps, with $r=0.005$, $k=0.2$ and $\mathrm{\sigma }=0.25$. The outcomes for timesteps 1, 10, 20, and 30 are shown in the Results. Simulations were run using the Julia programming language.

Data availability

Data supporting the findings of this study are either included in the manuscript or are available from the corresponding author upon request. The code for the progression model is deposited at https://github.com/yohanjohn/CTE.

Results

The laminar structure of MTL areas varies systematically from a medial to lateral direction

As in other cortical regions, the MTL shows graded changes in lamination (Braak and Del Tredici 2015; Piguet et al. 2018; Sancha-Velasco et al. 2023). From a medial to lateral direction, agranular MTL A28 lacks layer 4, and is abutted laterally by dysgranular cortex with an incipient layer 4, and then by eulaminate areas which have six-layers and show elaboration of their layers in successive lateral sites. The architecture of MTL allowed straightforward analysis of p-tau pathology in neurons in single sections without the need to group areas into types (Barbas and Rempel-Clower 1997), in a region where areas are interconnected in macaque monkeys (Suzuki and Amaral 1994; Bautista et al. 2023).

The laminar distribution of neurons with p-tau pathology in MTL

Systematic p-tau pathology in the cortex emerges first in entorhinal A28 of MTL in CTE stage III (McKee et al. 2013; McKee 2020) (Fig. 1A). In the cortex, A28 also has the densest p-tau pathology (Fig. 1B-K), and thus is akin to the origin or seed site (Fig. 1A, box B). As in cortico-cortical connections, p-tau transmission is bidirectional, proceeding from neurons to axons and vice versa. Accordingly, just as all areas that receive connections from A28 (Fig. 1B) reciprocate, we found p-tau beyond A28 in other MTL areas. Neurons labeled with p-tau in Fig. 1 (magnified panels below the cross section in A) parallel the origin of reciprocal pathways from different MTL sites that project to A28, as seen in animal connection studies. According to model circuit rules, the relative laminar distribution of neurons that reciprocate projections varies by the difference/similarity in laminar structure between the linked areas (areas with label and the seed site). The smaller the difference in structure, the more evenly neurons are distributed in the upper layers (2–3) and in the deep layers (5–6), as seen in Fig. 1, panels C and D. Similar to connections in non-human primates, in more lateral MTL sites (panels E-K), the laminar distribution of neurons with p-tau pathology shifts gradually to the upper layers. The graded laminar distribution of neurons with p-tau pathology reflects the increasing difference in laminar structure between A28 and sites from E to K, where the few labeled neurons are found increasingly in the upper layers.

Figure 1. P-tau pathology in temporal cortices in late stage III CTE.

(A) Cross section through the medial temporal lobe (case K128) at the level of the amygdala immunostained against hyperphosphorylated tau (box B shows the densest p-tau pathology in A28). (B-K) At high magnification, regions (outlined in a) show a graded pattern in the laminar distribution of neurons with p-tau pathology that parallels the laminar connectivity between the agranular A28 (B) through adjacent dysgranular and progressively with eulaminate (six-layer) lateral temporal areas. (L) Quantitative analysis shows a gradual increase of neurons with p-tau pathology in the upper layers (SGI) from medial to lateral parts of MTL for three cases with late stage III CTE. (M) Quantitative analysis of the density of neurons with p-tau pathology in the supragranular layers of three cases with late stage III CTE, gradually decreasing from A28 (B) to progressively adjacent lateral sites within MTL.

The pattern of p-tau pathology was corroborated in two additional cases of stage III CTE. Quantitative analysis showed monotonic increase of neurons with p-tau pathology in the upper layers from medial to lateral parts of MTL (p = 0.0002, F(1,8) = 41.07, df = 9. Fig. 1L (for individual maps and combined analyses in these and additional cases see Supplementary Figures 1-4). The number of p-tau labeled neurons was normalized for each of the ten sites and expressed as the supragranular index (SGI), which is the ratio of p-tau neurons in the upper layers over all neurons with pathology at each site. There is about an equal number of neurons with p-tau pathology in the upper and deep layers in medial sites (C-D), with an increasing ratio of neurons in the upper layers in progressively adjacent lateral MTL sites. The trend in the density of neurons with p-tau pathology is seen with quantitative analysis by using the SGI in Figure 1M for the three late-stage III cases (and Supplementary Figure 4B for combined density: p = 0.011684, F(1,8) = 10.57, df = 9).

In another case, we saw the same trend of label with p-tau. Remarkably, this case, which was classified by refined staging as early stage III, revealed a crucial laminar pattern: p-tau pathology in A28 (the seed site) was found predominantly in the deep layers (Fig. 2A, B), unlike advanced stage III, where all layers were affected (Fig. 1B). This finding revealed that p-tau pathology spreads from A28 to lateral MTL sites via feedback pathways. As in the cases above, the laminar distribution pattern and density of label were graded from medial to lateral MTL sites (Fig. 2B-H), and by quantitative analyses (Fig. 2L, M: (p = 0.000029, F(1,8) = 71.39, df = 9).

Figure 2. P-tau pathology in MTL cortices in early stage III CTE.

(A) Overview of cross section through the medial temporal lobe at the level of the amygdala (case AT8-CV-SLI-67) immunostained against hyperphosphorylated tau. (B-H) Sites shown at high magnification (boxes outlined in A) show a graded pattern in the laminar distribution of neurons with p-tau pathology that parallels patterns of laminar connectivity between A28 and areas with increasing laminar differentiation in the MTL. (L) Quantitative analysis shows monotonic increase of neurons with p-tau pathology in the upper layers (SGI) from medial to lateral parts of MTL for a case with early stage III CTE. (M) Quantitative analysis of the density of neurons with p-tau pathology in the supragranular layers of a case with early stage III CTE, shows gradually decreasing density from A28 to progressively adjacent lateral sites within MTL.

Quantitative analysis of p-tau labeling in the entire cortical ribbon spanning all medial and lateral temporal regions with signal showed the same pattern of gradual increase in SGI as the analysis of representative columns (Supplementary Figure S3A). Moreover, the same pattern of labeling was evident in straight, gyral, and sulcal parts of the cortex, when these were analyzed separately (Supplementary Figure S3B-C).

The termination of pathways from seed A28 to other MTL sites

From the seed site in A28, feedback-like pathways innervate other MTL areas. Antibodies for p-tau pathology label both neurons and axons and their terminals, which allowed us to address whether p-tau pathology in A28 reflects the pattern of termination of pathways known for MTL areas in non-human primates. This analysis included the density of the fine ‘background’ label in each panel, which is made up of axon fibers and terminations but excluded labeled neurons (see Methods). We normalized the findings from each site as the SGI, as above. As seen in Figure 3, there is a significant correlation in the SGI within the MTL sites, which is low at sites closest to the seed A28 and increasingly higher in lateral panels (Fig. 3A: p = 0.0028, F(1,8) = 18.04, df = 9; Fig. 3B: p = 0.012, F(1,8) = 10.64, df = 9). This relationship is consistent with the predominant ‘feedback’ pathways that A28 issues, which terminate increasingly in the upper layers through sequential lateral MTL sites, in areas that differ more and more in laminar structure from A28.

Figure 3. Laminar distribution of p-tau pathology in processes (excluding neuron bodies) as a proxy for anterograde pathways in stage III CTE.

Early stage III CTE (blue circles), case AT8-CV-SLI-67; (corresponding retrograde signal is shown in Figure 2). Late stage III CTE (red squares), case K128, (corresponding retrograde signal shown in Figure 1).

The graded laminar pattern of p-tau pathology in axons in the anterograde direction appeared to be more subtle than the neurons labeled with p-tau pathology or connection patterns in animals, likely due to removal of degenerating axons by glia, a process that starts early after axon damage and has a variable duration (Zhang et al. 2021); the timing of this process is not known for CTE, or the time interval between repetitive head impact and death. Similar trends in the spread of anterograde p-tau pathology were seen in the other cases (Supplementary Figures 1 and 2).

Comparison with findings in macaque studies

The above findings of p-tau pathology in MTL in humans mirror cortico-cortical connections seen in animal studies, as seen here by a direct comparison with interconnections within the homologous MTL region of rhesus monkeys. The data were computed from cases with the use of gold-standard neural tracers injected in different parts of MTL and adjacent inferior temporal visual and superior temporal auditory association areas. Figure 4A shows a cross section with connections in MTL after injection of a bidirectional neural tracer in MTL A36 in a rhesus monkey. As seen in the magnified panels below, the predominant termination in A28 is in the middle-deep layers (4C), reflecting a connection from an area with more elaborate laminar structure than the destination in A28. Fig. 4D shows the distribution of terminations throughout the layers at a site close to the seed area, reflecting a lateral pattern of connections, while panels E and F show predominant termination in the upper layers, consistent with a pathway from an area with less elaborate to an area with more elaborate structure (feedback). The bidirectional tracer labels neurons as well, which are mapped in Fig. 4G, and are mingled in a background of anterograde label (white label in Fig. 4C-F). In Fig. 4G, projection neurons directed to area 36 thus are seen in the deep layers of A28 and 35 which have fewer layers, through a shift to both the upper and deep layers in an adjacent site of A36 (lateral pattern) and seen mostly in the upper layers in eulaminate visual area TE1 (red dots), as predicted by the Structural Model. The density of label gradually decreased in a direction from areas with the lowest, to areas with the highest, laminar complexity.

Figure 4. Interconnections of MTL areas in rhesus monkeys.

(A) Overview of representative coronal section through the temporal lobe at the level of the amygdala shows pathway terminations and projection neurons (white label) connected with medial temporal A36 (neural tracer was biotinylated dextran amine (BDA cocktail of 3kDA for retrograde and 10kDA for anterograde transport) injected in area 36. Magnified boxes (from A) (C-F) Medial A28 is connected with A36 via the deep layers, D, shows connections in all layers and progressively in the upper layers of more lateral MTL sites. (B) Quantitative analysis shows gradual increase of projection neurons in the upper layers (SGI) from medial to lateral parts of MTL. Connection data were from 5 cases with retrograde tracer injections in areas TE1, TS1/TE1, TS2, TPro, and A36, and were expressed as a function of the difference in laminar type with respect to the injection site in each case. (G) Projection neurons in the supragranular layers (red dots) and infragranular layers (brown dots) mapped in a cross section through MTL show a shift of projection neurons from the deep layers in A28, to both superficial and deep layers in adjacent MTL A36, and mostly in the upper layers of visual association area TE. The dotted line shows the upper part of layer V. ls, lateral sulcus; rs, rhinal sulcus. (A-F, darkfield photomicrographs).

Figure 4B shows connection individual data from 5 cases with injections of tracers in several temporal cortical areas. Data were expressed as a function of the difference in laminar structure between the injection site and areas with label in each case. The relational Structural Model makes it possible to combine data from different cases with injection of tracers (seed sites) in diverse temporal areas and to express the findings by quantitative analysis of labeled neurons in each area based on the structural relationship for pairs of linked areas in each case. Accordingly, in Fig. 4B on the x axis, 0 represents projection neurons from areas that have comparable laminar organization as the injection (seed) site, as in a lateral pattern (e.g., (Hilgetag et al. 2016)). Sites in increasingly lateral sites within MTL have progressively higher laminar differentiation than the injection site (seed site), as shown on the x axis, with progressive positive values in increasingly lateral sites, as in feedforward pathways. Negative values on the x axis in Fig. 4B show the SGI from areas that have a less differentiated laminar structure than the injection site (seed area), as in feedback pathways.

The graded patterns depend on the relative laminar structural similarity/difference between pairs of linked areas, regardless of their geographic distance, as discussed in previous studies (Hilgetag et al. 2016), and our observations here that p-tau pathology in the dysgranular insula, which is situated at a considerable distance from the seed site of A28, revealed a similar bilaminar pattern as the MTL site situated next to A28, which is also dysgranular (Fig. 1C; and Supplementary Figure 5). In summary, the human CTE pathology in neurons and axons mimics the laminar patterns and density of cortico-cortical connections in monkeys.

Propagation model

To illustrate how the Structural Model can account for the spread of p-tau pathology in CTE, we developed a propagation model to simulate key qualitative patterns visible in stage III CTE and beyond. Details of the model for propagation of p-tau pathology are in Methods.

Figure 5 shows the relative distribution of p-tau pathology in the upper layers (L2/3) and in the deep layers (L5/6) for the progression of the disease. We included ten types of cortices that differ by degree of lamination. The first affected area is the cortex with the simplest laminar structure, which lacks layer 4 (type 1, agranular, akin to A28). All model cortices are connected to each other. The difference in degree of lamination determines the laminar pattern of connectivity: the greater the difference in laminar structure for pairs of linked cortical areas the more asymmetrical the connectivity pattern, reflected in laminar distribution and density of connections. The more similar two connected areas are, the more symmetrical is the connectivity pattern. The difference in laminar architecture thus determines the extent to which the connectivity reflects feedforward/feedback versus lateral connections.

Figure 5. Simulation of propagation of p-tau across cortical types.

Ten cortical types were simulated, with the simplest limbic cortex (degree of lamination = 0, agranular) being the seed area from which p-tau initially spreads. The disease advanced in stages, moving from left to right. Subplots A depict the relative uptake of p-tau in supragranular layers (L2/3) and infragranular layers (L5/6). Uptake is quantified as a proportion varying between 0 and 1: (A) value of 1 means that all the neurons in the corresponding layer have p-tau. (B) Subplots depict the corresponding supragranular gray index (SGI), which captures the extent to which the supragranular layers are preferentially stained with p-tau. SGI of 0.5 corresponds to equal staining in supragranular and infragranular layers. Red dashed lines indicate the linear regression used to calculate the slope of the gradual change in SGI across cortical types and time (C) Subplots depict the total amount of p-tau in infragranular layers (blue plots), supragranular layers (orange plots), and the sum of the two (green plots) for each area. The total takes a maximum value of 2 when both supragranular and infragranular layers are fully stained (a.u.: arbitrary units). As the disease progresses, p-tau expression tends towards more SGI uniformity across L2/3 and L5/6 within each cortical area, and also across cortices (subplot B-4).

In Figure 5, moving from left to right, the tauopathy progresses from an early to a late stage. As time elapses, the upper layers of all the model cortices gradually become affected (subplots A-1 to A-3). This leads to SGI that increases with degree of lamination (subplots B-1 to B-3) from values close to 0 to values closer to 1. The total density of p-tau decreases with degree of lamination (subplots C-1 to C-4, green plots), but the density in upper layers shows an upward trend (subplots C-1 to C-4, orange plots). Over time, more and more cortical areas serve as seeds for spreading p-tau (subplots A-2 to A-4), resulting in a progressive reduction of the slope of the SGI pattern (subplots B-2 to B-4). This pattern of gradual flattening of the SGI pattern can be seen in the empirical data by comparing Figure 1L to Figure 2L. Alternative connectivity patterns that ignore the laminar pattern of connectivity were also tested using the same underlying propagation model. These did not reflect the key qualitative pattern (Supplementary Results). Briefly, patterns that assigned either a uniform or a random connectivity between supragranular and infragranular layers of connected cortices, with strength determined by similarity of degree of lamination, did not display a clear monotonically increasing pattern of SGI (Supplementary Figures 7-15).

Differential pattern of label in the depths of sulci in early CTE (stages I, II): columnar connections

The pattern of cortical p-tau pathology in MTL in CTE stage III follows cortico-cortical circuit rules, in which the only layer that does not include labeled neurons is layer 4 (Figs. 1 and 2; and Supplementary Figures 1, 2). By contrast, within the focal label in the depths of sulci in early CTE (stages I and II), ~10% of p-tau labeled neurons were found in layer 4, and included labeled neurons in the other cellular layers as well (Fig. 6). This pattern resembles a different type of circuitry, namely connections within a cortical column, which includes participation of layer 4 neurons (reviewed in (Douglas and Martin 2004; Yoshimura et al. 2005)). Layer 4 neurons have short axons that do not leave the cortical mantle, as noted first by Ramon-y-Cajal (DeFelipe and Jones 1988), but participate in connections within a column and across the adjacent 1–2 columns. In both types of circuits, the termination of axons involves all layers.

Figure 6. Expression of p-tau pathology in dorsal and lateral cortices in early CTE stages.

(A) Cross section through the temporal lobe at the level of the anterior hippocampus immunostained against p-tau in stage I CTE. Outlined region shows initial area of injury in the depths of a sulcus, magnified below to highlight labeled neurons in all layers, including layer 4. (A1, A2) Outlined regions are shown at higher magnification on the right side of panel A. In the depths of the sulcus in A1 there are labeled neurons in all layers, a pattern that resembles a different type of circuitry, namely connections within a cortical column. Columns in nearby areas (A2) do not include neurons in layer 4, a pattern that resembles cortico-cortical connections. (B) Overview of cross section through the temporal lobe at the level of the temporal pole immunostained against hyperphosphorylated tau in stage II CTE. Outlined region (box) shows early p-tau pathology at higher magnification with labeled neurons in all layers, including layer 4. (B1) Outlined region is shown at higher magnification on the right side of panel B.

Discussion

Our findings revealed a striking parallel in the pattern of p-tau pathology in human MTL in CTE and cortico-cortical connections in monkeys. Significantly, the p-tau pathology pattern conforms to the predictions of the theoretical Structural Model that uniquely reveals the direction of connections, and in this instance p-tau pathology and its distribution in successive MTL sites by laminar distribution of p-tau in neurons and axons. The fundamental predictions of the Structural Model have been substantiated in the connections in diverse animal species and systems (reviewed in (Barbas 2015; Tucker and Luu 2023)) and specifically within the MTL here. Another prediction was the absence of neurons with p-tau pathology in layer 4, which do not participate in cortico-cortical connections but interconnect neurons up and down their particular narrow column (Douglas and Martin 2004; Yoshimura et al. 2005), as seen in the earliest stages of CTE in sporadic sites within the depths of sulci. These predictions were consistently substantiated for the first time in the human cortex through quantitative analyses in single cases and sections attesting to the robustness of the findings.

According to the reciprocity rule of cortico-cortical connections, the primate A28 projects to several MTL areas and receives projections from these areas in non-human primates (Suzuki and Amaral 1994) (Fig. 4). P-tau pathology was seen in neurons and axons, consistent with reports that misfolded tau complexes are internalized in dendrites and axons and travel in both directions (e.g., (Wu et al. 2013)). Accordingly, neurons with p-tau pathology in CTE stage III showed graded differences in laminar distribution and density in MTL. In areas that abut A28, neurons with p-tau pathology were densely distributed in more layers (layers 2–3 and 5–6), reminiscent of lateral connections in monkeys (see (Hilgetag et al. 2016)). In sequential lateral MTL areas, sparser neurons with p-tau pathology were found progressively in the upper layers, as seen with feedforward projection patterns in animal connectional studies and here, and consistent with model predictions.

The medial part of A28 showed the earliest cortical and densest p-tau pathology in CTE stage III, operationally establishing it as the cortical seed site. In the example in Figure 2 of early stage III CTE, p-tau pathology was overwhelmingly seen in the deep layers. From A28, a graded pattern of p-tau pathology could be traced in the anterograde direction. By circuit rules, the deep layers of A28 project to more lateral parts of MTL via feedback-like pathways that target all layers of the nearby dysgranular sites, and progressively the upper layers in more lateral areas that have increasingly differentiated laminar structure. By circuit rules and application of a computational model here, these findings provide concrete evidence of the direction of p-tau pathology in time, with the implication that pathology affects all areas connected with A28 (the seed site) in a graded manner. In time, as areas close to the seed site become increasingly affected with p-tau pathology, they turn into seed areas themselves, affecting all areas they are connected with, in a chain reaction leading to the inexorable progression of the disease with widespread distribution of pathology in cortical layers by CTE stage IV (e.g., (McKee et al. 2013), Fig. 5, Supplementary Figure 6).

Progression of pathology via connections has been suspected in animal models of traumatic brain injury (e.g., (Zanier et al. 2018)) and in neurodegenerative diseases (e.g., (Hof et al. 1992; Goedert et al. 2010; Braak and Del Tredici 2015; Clavaguera et al. 2015; Kriegel et al. 2018)) such as Alzheimer’s disease (AD) (Vogel et al. 2020; Delpech et al. 2021). In the latter, the seed cortical area is the ‘transentorhinal’ region with pathology seen more medially in entorhinal cortex in progressive stages of Alzheimer’s disease (Braak and Del Tredici 2015), and seen in the hippocampus later. By contrast, in CTE III the seed cortical area is the medial part of A28, with progression of p-tau pathology to more lateral sites within MTL, revealing that in AD and CTE, p-tau pathology proceeds in opposite directions.

Hypotheses have also been advanced that pathology in neurodegenerative diseases may spread by a non-circuit basis through microglia or exosomes (Asai et al. 2015; Leyns and Holtzman 2017; Liddelow et al. 2017). Our findings cannot rule out some contribution by such mechanisms. However, the ordered laminar distribution of p-tau pathology in both neurons and axon terminals strongly points to circuit mechanisms as the primary mode of spread, since any substantial non-circuit spread would have obscured the consistent laminar distribution of p-tau pathology found here.

Previous studies have also distinguished the early focal, perivascular p-tau pathology in the depths of sulci in frontal or temporal lobes in CTE stages I - II from the broad p-tau pathology that occurs in the MTL in CTE stage III (e.g., (McKee et al. 2013; Cherry et al. 2020)). The early focal perivascular p-tau pathology likely affects locally the sites of greatest physical deformation during head impact injury (reviewed in (Kriegel et al. 2018)). Perivascular damage likely affects the glymphatic system that helps remove pathological tau after injury (reviewed in (Rasmussen et al. 2018; Hussain et al. 2023). On the other hand, the progressive pathology found in the MTL in stage III likely represents a spreading phenomenon (Kaufman et al. 2021). In the latter, tauopathy in neurons is seen in layers that participate in cortico-cortical connections, but not in layer 4 neurons, which participate in focal connections within one or two cortical columns. A key distinction noted for the first time here, is the presence of neurons with p-tau pathology in layer 4 in the depths of sulci in early CTE, consistent with the distinct circuitry within cortical columns (reviewed in (Douglas and Martin 2004; Yoshimura et al. 2005)) but not for cortico-cortical connections.

Functional implications

The systematic variation of the cortex can be traced to development and has implications for cortical evolution (Garcia-Cabezas et al. 2019; Garcia-Cabezas et al. 2022). The linkage of cortico-cortical connections to the systematic variation of the cortex has provided a set of rules to predict connections and probe direction and progression of pathology in neurodegenerative diseases, as applied here for CTE. The Structural Model also predicts the cortical plasticity-stability continuum (Garcia-Cabezas et al. 2017) and areas of vulnerability to disease (Barbas 2015; Garcia-Cabezas et al. 2019). By this measure, the agranular (limbic) A28 and adjoining dysgranular areas are the most plastic areas in MTL, consistent with their involvement in learning, memory and emotions. The plasticity of limbic areas also renders them vulnerable to neurodegenerative diseases, as seen here for CTE, which affect cellular metabolic, bioenergetic and repair processes (e.g., (Cerasuolo et al. 2023; Ceccarelli Ceccarelli and Solerte 2025; D’Alessandro et al. 2025)).

The involvement of A28 in the systematic p-tau pathology in CTE has important implications for function. Entorhinal A28 receives pathways from high-order sensory and other association cortices (Insausti et al. 1987), projects to hippocampus, receives the hippocampal output in its deep layers, and projects to high-order association cortices (e.g.,(Rosene and Van Hoesen 1987)). These pathways endow A28 with information to process memories in context (Suzuki and Amaral 2004; Eichenbaum 2017), engage in complex spatial tasks, and form relational associations for abstract processes within a cognitive and mnemonic domain (Alvarado et al. 2016; Constantinescu et al. 2016; Wilming et al. 2018). The pathology in MTL likely disrupts pathways from hippocampus and cortical areas. One such pathway, which originates in medial prefrontal A25 and innervates heavily the deep layers of A28, is associated with autobiographical memory and emotions (Joyce and Barbas 2018). These processes are severely affected in psychiatric diseases, including PTSD and depression, which are also experienced by young people later diagnosed with CTE (McKee et al. 2023). The strong parallels in p-tau pathology with cortico-cortical connections points to the potential that repetitive head impact may damage early diverse cortical pathways that converge on A28 and collectively contribute to the systematic p-tau pathology seen later in the MTL in CTE stage III.

Supplementary Material

Supplementary materials are available at Cerebral Cortex online.