In vivo effect of LATE‐NC on integrity of white matter connections to the hippocampus

Abstract

INTRODUCTION

TAR DNA‐binding protein 43 (TDP‐43) is a highly prevalent proteinopathy that is involved in neurodegenerative processes, including axonal damage. To date, no ante mortem biomarkers exist for TDP‐43, and few studies have directly assessed its impact on neuroimaging measures utilizing pathologic quantification.

METHODS

Ante mortem diffusion‐weighted images were obtained from community‐dwelling older adults. Regression models calculated the relationship between post mortem TDP‐43 burden and ante mortem fractional anisotropy (FA) within each voxel in connection with the hippocampus, controlling for coexisting Alzheimer's disease and demographics.

RESULTS

Results revealed a significant negative relationship (false discovery rate [FDR] corrected p < .05) between post mortem TDP‐43 and ante mortem FA in one cluster within the left medial temporal lobe connecting to the parahippocampal cortex, entorhinal cortex, and cingulate, aligning with the ventral subdivision of the cingulum. FA within this cluster was associated with cognition.

DISCUSSION

Greater TDP‐43 burden is associated with lower FA within the limbic system, which may contribute to impairment in learning and memory.

Highlights

• Post mortem TDP‐43 pathological burden is associated with reduced ante mortem fractional anisotropy.

• Reduced FA located in the parahippocampal portion of the cingulum.

• FA in this area was associated with reduced episodic and semantic memory.

• FA in this area was associated with increased inward hippocampal surface deformation.

1. BACKGROUND

Dementia and neurodegenerative diseases affected more than 55 million individuals worldwide in 2022 ¹ , ² and is very costly, resulting in US$604 billion in worldwide costs in 2010. ³ The underlying neural mechanisms of dementia syndromes are complex and poorly understood. Therefore, it is imperative to better understand the relationship between neurodegenerative disease and the brain in living patients. In particular, TAR DNA‐binding protein 43 (TDP‐43) ⁴ , ⁵ is a highly prevalent proteinopathy that has been implicated in multiple neurodegenerative diseases including Alzheimer's disease (AD). ⁶ A unique presentation of TDP‐43 pathology is defined as stereotypical TDP‐43 proteinopathy accumulation within limbic regions in older adults (typically aged 80 and older), with or without coexisting hippocampal sclerosis pathology, known as limbic‐predominant age‐related TDP‐43 encephalopathy neuropathologic change (LATE‐NC). ⁷ , ⁸ Without a distinct clinical or cognitive profile, and without ante mortem biomarkers, there is a great need to study the In vivo effects of LATE‐NC not only to inform the disease process but also to aid in differential diagnosis for improved clinical outcomes. While a large majority of current neuroimaging literature has focused on macrostructural aberrations within neurodegenerative disease, ⁹ , ¹⁰ , ¹¹ microstructural changes may be more sensitive to and better reflect the mechanistic drive of neurodegenerative diseases, ¹² particularly given the importance of non‐neuronal involvement in TDP‐43. ⁴

Diffusion tensor imaging (DTI) is a sensitive measure of white matter (WM) structural integrity that can be used to identify microstructural changes in aging and dementia cohorts. ¹³ , ¹⁴ , ¹⁵ Several studies have shown DTI to be a useful tool in measuring WM integrity in clinical populations. ¹⁶ , ¹⁷ Further, DTI metrics were shown to outperform gray matter atrophy in diagnostic classification accuracy ¹⁸ in neurodegenerative clinical samples. While these studies highlight the importance of WM integrity in dementia groups, research has been limited to clinical or gene mutation groups. Utilizing clinical samples as a proxy for underlying disease status does not allow for measuring the direct relationship between pathologic burden and brain structure. Clinical samples such as gene mutation carrier groups that are known to represent high aggregation of TDP‐43 allow for a specific exploration of disease; however, they are notably rare and may not represent the presentation of highly common and critical proteinopathy of LATE‐NC. Furthermore, mixed disease states are far more common than “pure” disease states, ¹⁹ enhancing the need for detailed pathologic quantification. Studies examining WM integrity with autopsy confirmed populations, while rare, have revealed that WM integrity significantly dissociates tau from TDP‐43 groups, ²⁰ and ex vivo magnetic resonance imaging (MRI) studies have shown a negative association of FA values within the temporal lobe with LATE‐NC. ²¹ Therefore, there is a great need to explore the relationship between post mortem neuropathologic disease burden and ante mortem neuroimaging to better understand the effect of disease on microstructural integrity. By utilizing data from the Rush Alzheimer's Disease Center (RADC), we have the unique opportunity to directly relate post mortem neuropathology and ante mortem neuroimaging metrics to measure In vivo signatures of disease.

The hippocampus has been identified as a central structure within neurodegenerative diseases and is a highly interconnected region within the limbic system. ²² , ²³ The hippocampus plays a key role in complex cognitive functions and has been identified as a site of early neuropathological proliferation. While current literature has focused on gray matter changes in the hippocampus and other limbic structures, few have explored the WM connections to the hippocampus. WM tracts that connect the hippocampus form both short‐ and long‐range connections. A major short‐range connection to the hippocampus is the perforant pathway involving the entorhinal cortex and parahippocampal cortex. Long‐range connections to distal regions of the brain include the fornix and cingulum bundles. In this study, we examined the effect of post mortem neuropathologic burden on the integrity of ante mortem WM connections to the hippocampus utilizing a voxel‐based approach. We hypothesized that FA will be negatively associated with pathological burden with the strongest relationships between the hippocampus and neighboring limbic structures.

2. METHODS

2.1. Study participants

This study consisted of 69 individuals, 66% of whom were women, with a mean age of death of 91.2 years (standard deviation [SD] 6.1 years). Clinical diagnoses upon last study visit included 30 individuals with no cognitive impairment (NCI; 42.3%), 16 with mild cognitive impairment (MCI; 22.5%), and 25 with probable AD (35.2%). Sample characteristics including age, sex, education, MRI and autopsy interval, and clinical diagnosis are described in Table 1. Data were drawn from two longitudinal cohort studies at the RADC: (1) the Religious Orders Study, which consisted of older Catholic nuns, priests, or brothers from across the United States, and (2) the Rush Memory and Aging Project, which consisted of older adults from across northeastern Illinois. ²⁴ , ²⁵ The studies started in 1994 and 1997, respectfully. The sole inclusion criteria were being without known dementia and agreeing to annual clinical evaluation and organ donation; optional, biennial MRI began in 2009. Both studies were approved by an Institutional Review Board of Rush University Medical Center. All signed an informed consent, an Anatomical Gift Act for brain donation, and a repository consent to permit data sharing. Participants underwent annual detailed clinical evaluations, including medical history, neurologic examination, cognitive testing summarized by a neuropsychologist, and clinician diagnosis. ²⁶ At the time of death, a neurologist specialized in dementia was blinded to post mortem and imaging findings, reviewed select clinical data, and provided a clinical diagnostic opinion regarding the most likely clinical diagnosis at the time of death. In cases requiring adjudication, case conferences including one or more neurologists and a neuropsychologist were conducted to finalize a diagnosis. ²⁶ , ²⁷ Subjects were selected for the current cross‐sectional study based on availability of neuroimaging data as described below.

TABLE 1.

Sample demographic characteristics.

Characteristics	Value
N	69
Male, n (%)	23 (33%)
Age at last visit, M (SD); median [range]	88.1 (6.2); 88.2 [68.4,100.9]
Age at death, M (SD); median [range]	91.2 (6.1); 91.8 [71.7,103.8]
Years between last MRI and death, M (SD); median [range]	3.0 (1.4); 3.0 [0.5, 6.0]
Years of education, M (SD); median [range]	15.6 (3.2); 16 5 , 22
Clinical diagnosis at last study visit
NCI	30 (42.3%)
MCI	16 (22.5%)
Probable AD	25 (35.2%)

Abbreviations: AD, Alzheimer's disease; M, mean; MCI, mild cognitive impairment; NCI, no cognitive impairment; SD, standard deviation.

2.2. Post mortem neuropathologic quantification

Post mortem neuropathologic quantification procedures have been described elsewhere. ¹⁹ , ²⁸ , ²⁹ , ³⁰ , ³¹ , ³² , ³³ , ³⁴ Here we provide a brief summary. At autopsy, left and right brain hemispheres were removed, and one hemisphere was immersion fixed in 4% phosphate‐buffered paraformaldehyde solution and subsequently cut into 1‐cm‐thick coronal slabs. Hemispheric selection was procedurally determined based on the integrity of tissue and pathologic burden. These slabs underwent tissue dissection and pathological diagnosis. ²⁷ , ³² , ³⁵ TDP‐43 was assessed in five brain regions, including (1) amygdala, (2) hippocampal CA1, (3) hippocampal dentate gyrus, (4) entorhinal cortex, and (5) midfrontal cortex. Slabs were stained with monoclonal antibodies to phosphorylated TDP‐43 (pS409/410; 1:100). A semiquantitative rating of TDP‐43 cytoplasmic inclusions (both neuronal and glial) in a 0.25‐mm² area of greatest density within each region was assigned as follows: 0 = none, 1 = sparse [1 to 2 inclusions], 2 = sparse to moderate [3 to 5 inclusions], 3 = moderate [6 to 12 inclusions], 4 = moderate to severe [13 to 19 inclusions], and 5 = severe [20 or more inclusions]. An overall burden measure was created by summing the severity score in each region and dividing it by 5. Coexisting disease was measured according to previously published methods ²⁹ , ³² , ³⁶ , ³⁷ , ³⁸ for use as covariates, including amyloid beta (Aβ), paired helical filament (PHF)‐tau, hippocampal sclerosis (HS), Lewy bodies, ³⁶ gross infarcts, ²⁹ atherosclerosis, arteriolosclerosis, and cerebral amyloid angiopathy (CAA). ³⁸ Individuals presented at autopsy with varying levels of pathology with the vast majority (91.3%) containing more than one key pathology of interest (Aβ, PHF‐tau, or TDP‐43) and 42.0% presenting with all three. PHF‐tau pathology was most common, with 98.5% of individuals presenting with tangles (68 of 69 individuals had tangles in the hippocampus). Aβ was second most common, with 86.9% of individuals presenting with plaques (48 individuals had plaques within the hippocampus). There were 35 individuals without presence of TDP‐43, 15 with subscribed TDP‐43 within the amygdala, 10 with amygdala and hippocampal inclusions, and nine with inclusions additionally in middle temporal gyrus. Information on the neuropathological characteristics of the sample can be seen in Table 2. 

TABLE 2.

Sample neuropathological characteristics.

Characteristics	Value
PHF‐tau tangle presence	68 (98.5%)
M (SD)	8.6 (8.9)
Hippocampal presence	68 (98.5%)
Amyloid‐β presence	60 (87.0%)
M (SD)	3.9 (3.2)
Hippocampal presence	48 (69.6%)
TDP‐43 presence	33 (47.8%)
M (SD)	0.6 (1.0)
Hippocampal presence	17 (24.6%)
Hippocampal sclerosis	3 (4.3%)
Lewy body disease	19 (27.5%)
Cerebral infarctions (micro/gross)	39 (56.5%)
Atherosclerosis	16 (23.2%)
Arteriolosclerosis	13 (18.8%)
CAA	26 (37.7%)

Abbreviations: CAA, cerebral amyloid angiopathy; PHF, paried helical filament; SD, standard deviation; TDP‐43, TAR DNA‐binding protein 43.

RESEARCH IN CONTEXT

1. Systematic review: In reviewing the literature, the authors identified a current gap in the study of TAR DNA‐binding protein 43 (TDP‐43), as it relates to white matter structural integrity, measured In vivo, given the lack of studies that utilize post mortem quantification of pathological burden.

2. Interpretation: Our review of the literature led to an experiment to test the hypothesis that increased TDP‐43 burden would be associated with decreased white matter integrity in regions closely connected to the limbic system. Our findings indicated that white matter structural integrity is a sensitive measure of TDP‐43 burden and that TDP‐43 targets specific interconnections of the hippocampus to affect learning and memory.

3. Future directions: The mechanism of TDP‐43 targeting the limbic system remains an unmet challenge. Research exploring the underlying biological processes would enhance this research to understand the vulnerability of this region and the role of TDP‐43 in its degradation.

2.3. Neuroimaging

High‐resolution T1‐weighted anatomical data were obtained using a 3D magnetization‐prepared rapid acquisition gradient‐echo (MPRAGE) sequence (echo‐time [TE]  =  2.98 ms, repetition time [TR]  =  2.3 s, preparation time  =  900 ms, flip angle  =  9°, field of view [FOV]  =  25.6  × 25.6 cm, 176 sagittal slices, slice thickness  =  1 mm, no gap, 256 × 256 acquisition matrix, parallel imaging acceleration factor [AF]  =  2 along phase encoding direction), for a total imaging time of 5 min and 30 s. Spin‐echo echo‐planar DTI was conducted: TE  =  85 ms, TR  =  8.1 s, FOV  =  22.4 × 22.4 cm, 65 axial slices, slice thickness  =  2 mm, no gap, 112 × 112 acquisition matrix with 6/8 partial Fourier acceleration, b  =  1000 s/mm² for 40 diffusion directions uniformly distributed in 3D space, 6 b  =  0 s/mm² volumes, for a total scan time of 6 min and 37 s.  Scans at the time point closest to death were used for the current project. Preprocessing of DTI data was performed with TORTOISE ³⁹ , ⁴⁰ to correct for distortions in the diffusion‐weighted volumes caused by eddy currents and magnetic field non‐uniformities, bulk‐motion correction, B‐matrix reorientation, and calculation of anisotropy. Preprocessed DTI images were aligned to the IIT Human Brain Atlas version 5.0 ⁴¹ , ⁴² , ⁴³ using Advanced Normalization Tools (ANTs). ⁴⁴ We generated a hippocampal connection mask by utilizing a “connectome approach” from the IIT probabilistic WM atlas. This approach consists of utilizing track density imaging (TDI), which defines the connectivity between all possible pairs of gray matter regions of interest of the atlas. This approach makes it possible to account for multiple fibers that may pass through the same voxel to be identified and included in the mask. A sum of all the TDI maps that connected to either the left or right hippocampus was generated to create the hippocampal connection mask. The TDI maps were thresholded at 5% of the maximum number of streamlines per track density image. A visual representation of the mask created using BrainNet Viewer ⁴⁵ can be seen in Figure 1. We determined which known WM bundles were captured within the mask by overlaying the IIT Human Brain Atlas WM bundles onto our hippocampal connection mask and selected the overlapping voxels. Voxels within the hippocampal connection mask coincided with voxels from the cingulum, corpus callosum, and the fornix.

FIGURE 1.

Hippocampal probabilistic mask. Hippocampal connection mask generated utilizing a “connectome approach” from the IIT probabilistic WM atlas. Track density image maps of connections to the left and right hippocampus were thresholded at 5% of the maximum number of streamlines for each connection. After thresholding the mask totaled to 147,007 voxels.

2.4. Statistical procedures

We utilized regression models to relate overall TDP‐43 burden to ante mortem FA within each voxel of the hippocampal mask. Specifically, a linear model was constructed for each voxel within the hippocampal mask, with FA as the dependent variable and overall TDP‐43 pathology as an independent predictor. Coexisting diseases including a composite measure of AD pathology, cerebrovascular pathology along with demographic variables were used as covariates. Models are described below. Model 1 represents univariate prediction of FA using only overall TDP‐43 burden. Model 2 includes the variables from Model 1 and adds demographic variables as covariates to rule out the influence of these factors. Model 3 includes all the variables of Model 2 and additionally adds coexisting cerebrovascular disease and AD burden as described earlier. Specifically, these covariates included number of chronic infarcts, cerebral amyloid angiopathy, Lewy bodies, arteriolosclerosis, and atherosclerosis. Hippocampal sclerosis was not included as a covariate. ⁷ Results were corrected using random field theory family wise error rate (RFT FWER) ⁴⁶ , ⁴⁷ adj‐p < .05 to control for multiple comparisons. Significant beta coefficients were visualized as color maps within the hippocampal mask. An additional experiment was included to better explore the effect of LATE‐NC given the proportion of subjects without the presence of TDP‐43. To do this, we included two variables to model the variation of TDP‐43, including (1) presence or absence and (2) observed variance from the group mean of non‐zero TDP‐43. Results from this experiment are included in the supplemental information.

$$\mathrm{Model}1=\mathrm{FA}\sim \mathrm{TDP}-43$$ (1)

$$\begin{array}{ccc}\hfill \mathrm{Model}2& =& \mathrm{Model}1+\mathrm{age}+\mathrm{sex}+\mathrm{years}\mathrm{between}\mathrm{scan}\mathrm{and}\mathrm{death}\hfill \\ & & +\mathrm{education}\left(\mathrm{years}\right)\hfill \end{array}$$ (2)

$$\mathrm{Model}3=\mathrm{Model}2+{\mathrm{Alzheimer}}^{\prime }\mathrm{s}\mathrm{disease}$$ (3)

$$\mathrm{Model}4=\mathrm{Model}3+{\mathrm{Alzheimer}}^{\prime }\mathrm{s}\mathrm{disease}+\mathrm{cerebrovascular}\mathrm{disease}$$ (4)

2.5. Association of WM integrity and cognition

Interested in exploring how reductions in FA due to LATE‐NC may impact cognition, we measured the relationship between neuropsychological test score performance and FA within the identified cluster. The following measures of cognition were used as previously described ⁴⁸ : episodic memory, perceptual orientation, processing speed, semantic memory, and working memory. Individuals’ raw scores for each neuropsychological test were converted to z‐scores using the mean and SD of the entire cohort at their baseline visit for each test and averaged to generate a domain level z‐score for each of the five domains. Global cognition was calculated for each participant by averaging the z‐scores from all 17 cognitive tests. To measure the relationship between the effects of reduced FA due to LATE‐NC disease burden on cognition, we first identified the voxels significantly related to LATE‐NC from Model 3, described earlier, and calculated an average FA score across those voxels for each subject (FA score at each voxel/total number of voxels). Voxels were not bound by bundle boundaries but instead represented all significant voxels that were identified from regression models. Finally, we calculated Pearson correlation coefficients between average FA within the cluster and the cognitive z‐scores. Results can be seen in Table 3.

TABLE 3.

Correlation of average cluster FA to cognition.

Cognitive variable	Pearson correlation coefficient
Episodic memory	0.25 *
Perceptual orientation	n.s.
Processing speed	n.s.
Semantic memory	0.24 *
Working memory	n.s.
Global memory	0.25 *

Abbreviation: n.s., not significant.

^* p < .05.

3. RESULTS

3.1. Regression analysis

Results from univariate analysis Model 1 indicated one cluster in the left temporal lobe containing 2114 voxels with a significant negative relationship between FA and LATE‐NC. The top 10 most probable connections included the following: 8.67% left parahippocampal gyrus and left hippocampus, 6.90% left entorhinal cortex and left hippocampus, 5.72% left superior‐temporal gyrus and left temporal pole, 5.22% right hippocampus and right entorhinal cortex, 4.69% left fusiform gyrus and left inferior‐temporal gyrus, 4.38% left entorhinal cortex and left temporal pole, 4.02% left temporal pole and left amygdala, 3.99% left fusiform gyrus and left temporal pole, 3.84% left entorhinal cortex and left parahippocampal gryus, 3.44% left temporal pole and left insula. Results from Model 2 revealed largely similar findings with one cluster containing 1044 voxels with a significant negative relationship between FA and LATE‐NC. Significant voxels contained fibers, which were exclusively contained in the left hemisphere. The top 10 most probable connections included the following: 20.07% left parahippocampal gyrus and left hippocampus, 13.36% left entorhinal cortex and left hippocampus, 7.56% left entorhinal and left parahippocampal gyrus, 7.48% left isthmus of cingulate and left hippocampus, 5.80% left fusiform gyrus and left inferior‐temporal gyrus, 4.73% left isthmus of cingulate and left parahippocampal gyrus, 3.16% left fusiform gyrus and left temporal pole, 2.74% left fusiform gyrus and left hippocampus, 2.49% left fusiform gyrus and left amygdala, 2.13% left precuneus and left hippocampus. Results from Model 1 and 2 can be seen in the supplemental material.

Model 3 additionally controlled for coexisting AD pathology. One significant cluster (Figure 2A) contained 997 voxels, which contained the following probable connections: 20.07% left parahippocampal gyrus and left hippocampus, 13.36% left entorhinal cortex and left hippocampus, 7.56% left entorhinal and left parahippocampal gyrus, 7.48% left isthmus of cingulate and left hippocampus, 5.80% left fusiform and left inferior‐temporal gyrus, 4.73% left isthmus of cingulate and left parahippocampal gyrus, 3.16% left fusiform gyrus and left temporal pole, 2.74% left fusiform gyrus and left hippocampus, 2.49% left fusiform gyrus and left amygdala, 2.13% left praecuneus and left hippocampus. Connecting regions are overlaid surrounding the cluster in Figure 2B. Results from Model 3 with overlapping WM bundle of the cingulum can be seen in Figure 2C. Results from Model 4, which includes cerebrovascular disease, did not survive FWER correction.

FIGURE 2.

Voxel‐wise regression model results. Results from regression analysis Model 3 visualized onto descriptive sagittal, coronal, and axial views. (A) Relationship between LATE‐NC pathological burden and ante mortem FA within each voxel of the hippocampal mask. Colors indicate strength of significant relationships after correction for multiple comparisons. Darker colors indicate a more negative relationship. (B) Connecting regions to the hippocampus are overlaid surrounding the cluster. Regions include left hippocampus (yellow), left parahippocampal gyrus (green), left entorhinal cortex (blue), and left isthmus of the cingulate (not pictured). (C) The IIT Human Brain Atlas cingulum bundle is displayed underneath the cluster (cyan).

3.2. Post hoc analyses

3.2.1. Correlation to cognition

To better understand the relationship between structural integrity in WM connections to the hippocampus and cognition, we measured the correlation between cognition scores across several domains and the average FA within the significant cluster identified from regression analysis from Model 3. Table 3 reports the correlation coefficients for each cognitive domain assessed. The results show episodic memory and semantic memory as significantly related to FA within this cluster. Cognitive domains of perceptual orientation, processing speed, and working memory were not significantly correlated with FA within the previously identified significant cluster.

3.2.2. Correlation to hippocampal shape

In a previous study, we measured the relationship between post mortem LATE‐NC pathology and ante mortem hippocampal surface deformation in a superset (N = 99) of the same sample. ⁴⁹ That study utilized the same imaging and pathological quantification methods and found that increased LATE‐NC burden was associated with inward hippocampal surface deformation measures predominately in the left lateral and anterior regions (approximating CA1, subiculum). Here we related this hippocampal surface deformation profile to the WM cluster identified in the current study via correlation. We assessed the correlation between average hippocampal surface deformation and voxels that were previously identified as associated with LATE‐NC and average FA in the voxels of the WM cluster described earlier. This correlation revealed a significant positive relationship (Pearson's rho = 0.49, p value < .001) between WM and hippocampal shape profiles unique to TDP‐43, as seen in Figure 3.

FIGURE 3.

Correlation between average cluster FA and hippocampal deformation due to TDP‐43. Correlation between hippocampal deformation and WM hippocampal connection integrity due to LATE‐NC. Average hippocampal deformation due to LATE‐NC was calculated by taking the average deformation score across all voxels in a cluster that was uniquely associated with LATE‐NC. Similarly, average WM integrity affected by LATE‐NC was calculated by taking the average FA score across all voxels within the cluster identified by Model 3, as seen in Figure 2A.

4. DISCUSSION

Here, we assessed the relationship between ante mortem FA and post mortem neuropathological burden to identify In vivo markers of LATE‐NC on WM connections to the hippocampus. Findings revealed a significant negative association of FA and LATE‐NC in one WM cluster connected to the hippocampus. That is, higher LATE‐NC pathological burden was associated with lower FA. The location of the cluster was within the left medial temporal lobe and connected the hippocampus to highly proximal limbic structures including parahippocampal gyrus, entorhinal cortex, and the isthmus of the cingulate gyrus. These connections include the perforant pathway, which is a major input path to the hippocampus. Using the bundles of the IIT Human Brain Atlas, this cluster was identified to be part of the left cingulum bundle, specifically the ventral or parahippocampal segmentation of the cingulum. ⁵⁰ These results indicate WM microstructural changes within focal regions of the limbic system, specifically circumscribed regions connecting the hippocampus, may be implicated by LATE‐NC.

The cingulum bundle is a major connecting fiber pathway within the limbic system. ⁵¹ , ⁵² This large C‐like structure connects both prefrontal and temporal regions to major limbic structures. Because the structure is known to be composed of several long‐ and short‐range fibers, different segmentations of the cingulum have been proposed. The significant cluster identified in the current study aligns with the ventral or parahippocampal segment of the cingulum. In one study using deconvolution of fiber tract densities, ⁵⁰ , ⁵³ the authors noted that this portion of the cingulum predominately projects to the medial temporal lobe and the cingulate. Findings from the current study indicate LATE‐NC is associated predominately with connections between the hippocampus and parahippocampal gyrus via this portion of the cingulum bundle, along with a smaller connection to the entorhinal cortex. The parahippocampal gyrus, including the entorhinal cortex, is a key area involved in encoding and retrieval of memory and has been identified as one of the earliest areas of neuropathologic proteinopathy within AD and related diseases. ⁵⁴ This finding is supported by the post hoc correlation analysis with cognition. FA within this cluster was significantly associated with memory, whereas other cognitive functions, such as processing speed and perceptual orientation, were not associated with FA within this region.

Our findings align with and supplement current literature surrounding WM relationships with neurodegenerative diseases. While typical aging has been shown to demonstrate an anterior‐posterior gradient of WM degeneration, AD has been associated with WM degeneration particularly within posterior regions ¹⁶ , ⁵⁵ and regions that connect critical memory structures. ⁵⁶ Utilizing neurite orientation dispersion and density imaging (NODDI), ⁵⁷ a highly sensitive measure of brain microstructure, previous work found an [¹⁸F]Fluorodeoxyglucose positron emission tomography (FDG‐PET) marker of TDP‐43 was associated with reduced microstructural integrity within the parahippocampal cingulum, a location similar to that presented in the current analysis. A meta‐analysis of voxel‐based morphometry studies comparing AD patients and healthy controls revealed a similar pattern of findings. ⁵⁸ This meta‐analysis revealed an association between WM integrity and cognitive status within left‐parahippocampal WM and within the corpus callosum. The direct link between TDP‐43 neuropathology burden and ante mortem WM microstructure integrity as revealed by our study lends credence to these findings from living patients.

Findings in this study indicated that LATE‐NC had a unique effect on WM structural integrity, after controlling for coexisting AD pathologies and cerebrovascular diseases. The imaging patterns revealed in this study have the potential to be used as a biomarker for the underlying neuropathological process in clinical populations. For example, Zhang and colleagues ¹⁷ examined commonalities and differences in frontotemporal dementia (FTD) and AD clinical groups and found FTD to have greater reductions in FA in frontal brain regions compared to AD. Further, compared to controls, FTD individuals showed reduced FA in the anterior corpus callosum, bilateral anterior, descending cingulum tracts, and uncinate tracts, while AD individuals showed reduced FA in the left anterior and posterior cingulum, bilateral descending cingulum, and left uncinate tracts. These findings highlight that distinct clinical phenotypes exhibit unique WM alterations, which may indicate unique effects of pathological processes, as corroborated by our study.

Studies that explore biomarker signatures of disease allow for a closer understanding of the effects of disease on the brain, rather than exploring clinical groups alone. The present work aimed to enhanced this literature with direct comparisons of post mortem disease accumulation and WM structural integrity for an In vivo signature of disease. A previous study of the associations of WM integrity within the fornix using amyloid‐PET imaging ⁵⁹ revealed significant longitudinal effects of amyloid accumulation and reduced radial diffusivity in this region. Conversely, when exploring the relationship between WM integrity and tau pathologic burden, Tian ⁶⁰ found associations with the uncinate fasciculus, cingulum adjoining hippocampus, and inferior fronto‐occipital fasciculus. Further highlighting regional differences in WM degeneration due to disease, in an ex vivo analysis within a superset of subjects from the current analysis, Tazwar ²¹ found reduced WM integrity associated with LATE‐NC in numerous limbic connecting regions, providing supporting evidence for the results in the current analysis. This further supports the congruence of ex vivo and in vivo imaging signatures, highlighting the potential utility of in vivo imaging as a biomarker for disease. Additionally, this group found LATE‐NC was associated with lower R₂ values within WM providing connecting temporal lobes to frontal lobes, temporal lobes, and basal ganglia. ⁶¹ Our post hoc analysis indicating a modest correlation between hippocampal surface deformation and WM structure integrity supports the close relationship of DTI and atrophy measures. However, while sensitivity comparisons of atrophy and DTI were beyond the scope of this analysis, future research should consider comparing the sensitivity and specificity of these two modalities to better inform biomarker development. Together, this differential degeneration of WM and subcortical regions suggests these temporolimbic WM structures may present with unique sensitivity to pathologic processes, which could be utilized in diagnosis.

The mechanism of neuropathology resulting in reduced WM structural integrity is important to consider for an understanding of the relationship between these processes. In gray matter, neuropathology is associated with pathological protein accumulation resulting in cell death, while in WM, mechanistic relationships focus on reduction in myelin integrity due to gliosis. ⁶² , ⁶³ In a study exploring the effect of TDP‐43 pathology in a frontotemporal lobal degeneration sample, degeneration of WM was characterized by increased vacuolation and glial inclusions. In previous work from our group, TDP‐43, Aβ, and PHF tau tangles ⁴⁹ , ⁶⁴ were associated with unique hippocampal surface deformation patterns. Thus, neuropathologic disease may differentially target unique subfields of the hippocampus, which in turn have unique WM connections. This may represent unique vulnerabilities within the hippocampus, and therefore, exploring the structural connections of the hippocampus may better inform the impact of disease on WM and further on cognition and clinical presentation. The regions identified in the current analysis were circumscribed to the medial temporal lobe, rather than distal regions of frontal or parietal cortex. Further, the affected connections were ipsilateral, indicating cross‐hemisphere communication may be less affected by disease accumulation. Additional research in an amytrophic lateral sclerosis (ALS) population revealed TDP‐43 to follow a stereotyped staging process, ⁶⁵ as also seen within LATE‐NC. ²¹ , ³⁷ Though not explored in the current analysis, staging patterns of WM degeneration could additionally inform disease progression. It is important to consider the way in which TDP‐43 inclusions may result in reduced WM integrity to better understand the mechanism of action to inform potential therapeutic targets.

While the research presented here offers several innovative findings, it is not without limitations. First, the relatively small sample consists of individuals with advanced age (mean age at MRI = 88 years) who were highly educated (median years of education = 16) and predominately white. Thus, due to both the sample size and the unique characteristics of the subjects, the generalizability of our findings may be limited. Along with the small sample size, the distribution of pathology represents a large portion of subjects without TDP‐43 burden and of those with TDP‐43 burden, a high amount of disease confined to the amygdala. Because this study utilizes a community‐based cohort, high rates of less common pathologies are more infrequent compared to memory clinic or specialty populations. The present distribution of TDP‐43 within the amygdala aligns with the greater community studies performed at the RADC ⁶⁶ and with community samples from other centers, ⁶⁷ particularly for subjects without memory impairment, as a large portion of subjects in the present sample were cognitively normal before death. Additionally, while half of the subjects present with TDP‐43 confined to amygdala, the other half have TDP‐43 extending to other medial temporal lobe structure and the cortex, representing a broad spread of pathology along the continuum of the disease. While the TDP‐43 distribution may be a limitation, the present findings may support early disease changes of WM structural integrity. Second, the lack of clear hemisphere selection information limits our ability to interpret the laterality effects seen here. Findings from the regression analyses were circumscribed to the left hemisphere, revealing unexpected laterality effects, though supported by previous research using this sample. ⁴⁹ Future studies can broaden this sample by including more diverse range of participants and in greater number.

In conclusion, we measured the relationship between ante mortem WM structural integrity and post mortem neuropathological burden utilizing a voxel‐based approach. We found a significant negative relationship between FA and LATE‐NC within the left temporal lobe, particularly the ventral portion of the cingulum bundle, while controlling for the effects of demographics and coexisting pathologic burden. This study allowed for an innovative comparison of post mortem quantification of pathological burden and ante mortem WM integrity, which offers a more direct measure of the relationship between pathology and brain structure than utilizing clinical groups alone. Results from this study highlight the significant effect of disease on WM tracts, which connect the hippocampus to widespread brain regions. These relationships can help explain the contribution of WM in dementia syndromes.

CONFLICT OF INTEREST STATEMENT

The authors report no competing interests. Author disclosures are available in the supporting information.

CONSENT STATEMENT

Both studies were approved by an Institutional Review Board of Rush University Medical Center. All participants signed an informed consent, an Anatomical Gift Act for brain donation, and a repository consent to permit data sharing.

Supporting information

ICMJE Disclosure Form

Heywood A, Stocks J, Schneider JA, et al. In vivo effect of LATE‐NC on integrity of white matter connections to the hippocampus. Alzheimer's Dement. 2024;20:4401–4410. 10.1002/alz.13808

DATA AVAILABILITY STATEMENT

The data used in this work can be made available upon request at www.radc.rush.edu