PET and MRI images were acquired simultaneously on a 3T Siemens Biograph mMR (software version VE11P, Siemens Healthcare, Erlangen, Germany). PET data were acquired from participants 90–110 minutes after intravenous injection of 230–333 MBq of [¹⁸F]florbetaben (FBB) radiotracer. [¹⁸F]-florbetaben was synthesized by Life Molecular, Inc., and delivered to the PET imaging facility at Stony Brook University for all research participants. Three-dimensional volumetric T1-weighted magnetization prepared rapid gradient echo (MPRAGE) images were collected to facilitate PET signal localization and to measure neurodegenerative burden using a vendor-provided 20-channel PET-compatible head/neck coil with TR/TE/TI = 1900/2.49/900 ms; Flip Angle = 9◦; acquisition matrix: 256×256; voxel resolution: 0.89 × 0.89 × 0.89 mm.

Diffusion tensor imaging was also performed with the following parameters: TE/TR = 87.6/4680 ms, b value = 1200, 64 diffusion directions, in-plane resolution = 2 mm, slice thickness = 2 mm, matrix size = 128 × 128, multiband factor = 2. After data collection, post-processing incorporated standard techniques for acquisition-based artifact elimination. Subject motion and eddy corrections were accounted for using the FMRIB Software Library (FSL) ³⁴.

T2-weighted anatomical scans used a turbo spin-echo pulse sequence (34 axial slices, TR=6170s, TE=96ms; Flip angle=150◦; acquisition matrix = 320×320; voxel size = 0.36 × 0.36 × 3mm) were acquired and read by a board-certified radiologist to determine incidental findings.

PET images were reconstructed offline using vendor-provided reconstruction software (e7tools, Siemens Healthcare, Erlangen, Germany). To improve the accuracy of attenuation correction with bone compartment, the “Boston method” was used to generate attenuation maps using the MPRAGE images ³⁵. PET reconstruction was performed using OSEM with point-spread function modeling and 100 iterations (21 subsets) to ensure convergence for optimal quantitative accuracy ³⁶, followed by 4 mm Full-width half maximum (FWHM) Gaussian post-filtering to control noise (effective spatial resolution = 5.1mm) determined using the Joshi method ³⁷.

Sinograms were binned covering 20 minutes starting 90 minutes post-injection of [¹⁸F]-florbetaben according to established protocols ³⁸. PET images were scaled to injected dose and body weight to produce standardized uptake value (SUV) images.

Cortical segmentation of the associated T1-MPRAGE images was performed using FreeSurfer ³⁹ [V.6.0.0] followed by regional parcellation according to the Desikan-Killiany brain atlas ⁴⁰. Using the PETSurfer tools within FreeSurfer we extracted the VOI data from the PET images which also allowed for partial volume correction (PVC) of the VOIs using the Symmetric Geometric Transfer Matrix method ⁴¹ using the 5.1 mm FWHM resolution determined above.

Diffusion images were visually inspected to check for major image artifacts or significant motion during the acquisition. All images passed inspection. No radiological abnormalities were identified in the images. The diffusion data were reconstructed in the Montreal Neurological Institute (MNI) space using q-space diffeomorphic reconstruction ⁴² to obtain the spin distribution function (SDF; diffusion sampling length ratio = 1.25) using DSI Studio (5/4/2019 build) ⁴³. Cortical mean diffusivity (MD; b=0 & 1000) is a highly sensitive measure of cortical inflammation ⁴⁴ and microstructural cortical changes ⁴⁵ that often reflects early changes in cerebral structure resulting from amyloidosis. We extracted for a Region of Interest (ROI) based analysis using Desikan-Killiany atlas ⁴⁰ and DSI studios’ FreeSurferSeg atlas. All 34 unilateral ROI pairs from the Desikan-Killiany atlas ⁴⁰ were studied along with left and right ROI pairs of the hippocampus and amygdala, and five subsections of the corpus callosum from the FreeSurferSeg atlas provided by DSI Studio.

To determine whether potential WTC amyloidosis was similar in cortical topology to that found in AD, we performed a centiloid score analysis of [¹⁸F]-florbetaben data ⁴⁶, in which we used PMOD software (PMOD Technologies, Zurich, Switzerland). We strictly followed the protocol for this tracer to calculate a centiloid value for each responder, where centiloid values nominally fall on a 0–100 scale such that a centiloid score exceeding 25 can be used as a cutoff for AD positivity ⁴⁷. The centiloid method provides a universal scaling metric targeted at identifying AD amyloidosis.

Since [¹⁸F]-florbetaben distribution in the left and right hemispheres was strongly correlated (r = 0.99), and directions were similar across subregions, regional analyses focused on bilateral abnormalities. We defined three subregions of interest in this study. First, we defined a meta-region consisting of regions near the olfactory bulb and tract including the pars orbitalis, medial orbitofrontal, and lateral orbitofrontal regions. Next, we examined a meta-ROI made up of smaller ROIs that were adjacent to the olfactory bulb and cortex including the accumbens, rostral anterior cingulate, superior frontal, caudal middle frontal, and precentral regions. Finally, a meta-ROI made up of WTC-affected regions identified during a neuroimaging study of WTC responders with early-onset dementia^{11, 48, 49} consisted of the following regions: hippocampus, entorhinal cortex, parahippocampal gyrus, inferior parietal cortex, inferior/middle/superior temporal gyri (including banks of the superior temporal sulcus), supramarginal gyrus, inferior frontal gyrus, posterior cingulate, isthmus of the cingulate gyrus, precuneus, amygdala, putamen, and ventral diencephalon. All regional SUVs were normalized by the uptake of the whole cerebellum for SUV ratios (SUVR).

Each [¹⁸F]-florbetaben image was qualitatively read by a trained analyst (SC) and a board-certified fellowship-trained neuroradiologist with ten years of clinical and research experience in PET/MRI (AF) without knowledge of prior study results or clinical status. The global level of amyloid positivity (positive/negative) and any regional findings were reported.

WTC exposure duration was assessed using an interview-administered exposure assessment questionnaire that was completed by all responders upon enrollment in the parent monitoring program used to document exposures at the WTC response sites ³¹. All responders were dust-exposed during the WTC response efforts, so we defined exposure duration as the total duration (in months) of response efforts. None of the imaged responders reported regularly using personalized protective equipment, including masks, while working at the WTC sites.

Cognitive Throughput is a corrected response rate measure giving the number of successes per unit of discretionary time and was measured using the CogState^™ computerized battery. The battery consisted of five validated game-like tasks (detection, identification, one-card learning, continuous paired associate learning, and Groton maze learning and recall) and was administered to evaluate domains of cognitive dysfunction following a standard operating protocol ⁵⁰. Across these five tasks, we assessed episodic memory (accuracy measured by the paired associates learning test), visuospatial functioning learning and recall (total errors), visual working memory (accuracy), visuospatial memory, throughput (accurate responses per second), intra-individual item-response variability (measured in standard deviations), processing and response speed (responses per millisecond), and attention (accuracy).

Cognitive status was determined using the Montreal Cognitive Assessment ⁵¹; a cutoff of <22 was used to identify individuals with CI, whereas CU individuals had a score ≥26. Prior work in this population has found that apolipopreotein-ε4 (APOE4) status is not a central predictor of cognitive performance, but because this study examined amyloidosis ⁵², we were able to examine apolipoprotein-ε4 (APOE4) allele possession among 28/35 participants who consented to genotype analysis. To accomplish this, deoxyribonucleic acid was extracted from blood samples of WTC responders and genotyped using the Infinium Global Screening Array (Illumina, San Diego, CA, USA), and APOE4 status was recorded. In this small study, none of the participants in this small study were homozygous carriers so we dichotomized the genotype into possessing one allele (actually, only the e3/e4 allele combination was identified) versus non-possession.

Demographics included age in years, sex (male versus female), race/ethnicity (White, Black, Hispanic, and other), educational attainment (high school or less, some college/technical school, and college or more), and occupation on September 11, 2001 (law enforcement vs. other) were included as covariates. Post-traumatic stress disorder was diagnosed using the Structured Clinical Interview for the DSM-IV (SCID-IV) ⁵³, a semi-structured interview administered by trained clinical interviewers.

The sample was described using means and standard deviations, or frequencies and percentages comparing diagnostic categories using non-parametric trend tests and describing qualitative findings. Next, Welch’s t-test was used to compare mean tracer uptake in regions previously identified as vulnerable in WTC responders using the WTC regional SUVR and, secondarily, to consider regions commonly examined in AD comparing the Centiloid.

The regional distribution of β-amyloid used Statistical Parametric Mapping (SPM12) to complete voxel-wise analyses of Gray Matter SUVR completed. All models were adjusted for age, education, and race/ethnicity. Models adjusted for the centiloid score to account for differences in [¹⁸F]-florbetaben burden predominantly indicative of aging-related sporadic AD. To reduce the potential identification of false positives, p-values were estimated for clusters exceeding 50 voxels in size. FDR-adjusted p-values were estimated for cluster peaks. Results were then mapped onto the cerebral surface using SPM12’s Computational Anatomy Toolbox.

We compared correlations between WTC exposure duration and tracer uptake in the olfactory and surrounding regions using scatter plots and linear regression. Spearman’s ρ was reported for these analyses to account for the potential for nonlinear associations between amyloidosis and potentially correlated features including markers of neurodegeneration and cognitive function. We observed correlations between localized and regional [¹⁸F]-florbetaben burden, and between [¹⁸F]-florbetaben burden and cortical atrophy. All models were adjusted for age, education, and race/ethnicity. We also adjusted for the centiloid score to account for differences in [¹⁸F]-florbetaben burden predominantly indicative of aging-related sporadic AD.

A two-tailed α = 0.05 was used to determine statistical significance; results from repeated testing analyses were adjusted for the false discovery rate (FDR = 0.05) ⁵⁴. Statistical analyses were completed using Stata 17/MP [StataCorp].

This study was reviewed by the institutional review board. All participants provided informed written consent.

In total, 36 participants were deemed ineligible at screening. The 50 remaining individuals were scheduled for neuroimaging, but five withdrew and 9 were unable to complete the neuroimaging visit due to delays or imaging protocol failures (i.e., insufficient radioactive dose, delayed injection past the anticipated injection window, scanner failure). At one appointment, ligand production failed, and the participant refused to try again. These outcomes resulted in a final sample of 35 responders who underwent neuroimaging.

We begin by showing the demographic and clinical characteristics of the WTC responder group (Table 1). On average, neuroimaging participants were 56 years old, had at least some college, and were trained as first responders. When stratifying the sample into high versus low exposure, we found no statistically significant differences in demographics or other covariates between the two groups.

Voxel-wise analyses mapped to the cerebral surface identified elevated amyloid that was generally bilaterally symmetric, presenting with cluster centroids in the medial orbitofrontal, superior frontal, and precentral regions (Figure 2). In this figure, you can see evidence of cortical uptake both in the frontal lobe and in the cingulate, as well as in the left temporal pole and across regions in the parietal lobe. These differences were supplemented by regional analyses showing evidence of [¹⁸F]-florbetaben retention in WTC-related ROIs (P=0.025), such as the superior temporal and parahippocampal cortices, isthmus cingulate, and hippocampus (Supplemental Figure 1).

Qualitative clinical reads assessed one responder as FBB-positive consistent with ADRD (heterozygous apolipoprotein-ε3/ε4). Three other responders had at least one visual reader indicating the possibility of [¹⁸F]-florbetaben positivity in one region of interest. Centiloid analysis (mean = −9.31, SD = 10.25) identified one responder who met centiloid criteria for the presence of cerebral amyloidosis. Apolipoprotein-ε4 allele possession was not associated with amyloidosis in any analysis, though individuals carrying one apolipoprotein-ε4 allele had evidence of higher SUVRs in WTC-related regions that might suggest that larger studies would find a difference (APOE- = 1.11, APOE+ = 1.70, difference = 0.60, P=0.16).

Since WTC responders do not match the topographical [¹⁸F]-florbetaben burden pattern seen in AD, but rather in ROIs outside centiloid regions, we next focused on WTC susceptible ROIs, such as the olfactory bulb/tract and inferior frontal lobe. Next, we examined associations between WTC exposure duration and [¹⁸F]-florbetaben retention (Figure 3) to find a positive association between [¹⁸F]-florbetaben SUVR and longer WTC exposure duration in the olfactory regions (Spearman’s ρ =0.43, P=0.011, Panel A). We also identified a significant positive association between olfactory [¹⁸F]-florbetaben uptake and [¹⁸F]-florbetaben uptake in neighboring ROIs (ρ =0.93, P<0.001, Panel B), and a significant positive association between [¹⁸F]-florbetaben uptake in olfactory neighboring ROIs and previously identified WTC ROIs (ρ =0.41, P=0.016, Panel C).

Examinations of ROI-specific associations between tracer uptake and increased cortical mean diffusivity or decreased cortical thickness (Table 2) revealed statistically significant increases in cortical mean diffusivity across the entire parietal lobe, and throughout the occipital, temporal, and frontal lobes.

Next, we evaluated associations between cognitive throughput and region-specific [¹⁸F]-florbetaben burden (Table 3). These analyses revealed that Amyloidosis in olfactory-adjacent regions, and the centiloid scores were not associated with measures of cognitive functioning. However, we did find that the amyloid burden measured across the WTC meta-ROI showed statistically significant correlations with cognitive throughput (ρ=−0.59), response speed (ρ=−0.72), and visuospatial functioning (ρ=−0.49) that remained after accounting for multiple comparisons.

This study has several limitations including, for example, its small sample size (n=35) and the relatively unique exposures this population endured. The small sample size can cause several problems including a reduced ability to thoroughly map the full clusters of cerebral amyloidosis, and a reduced ability to identify correlations between amyloid and cognitive performance levels. Exposures at the WTC are like air pollution exposures, but the chemical mix is potentially unique and may therefore be generalizable to other occupational studies on disaster responses, where buildings are destroyed or burned unexpectedly or because of war conflicts, or also to the impact of climate-induced wildfires. Prior results from this population are consistent with mouse models and other in vivo studies of brain structure in people exposed to high levels of air pollution globally ^{3–5, 57}. We cannot determine whether these results align, since WTC dust contained a potentially unique mixture of neurotoxins, or whether the presence of any PM2.5 at the WTC was amenable to cerebral infiltration was sufficient. While our study included predominantly male subjects, this distribution matches our population as a whole and the makeup of WTC first responders in general. Finally, while the Alzheimer’s disease pathological cascade is believed to cause cerebral amyloidosis before the distribution of cerebral tauopathy and neurodegeneration, in this study we reported that cerebral amyloidosis is associated with both cognitive dysfunction and indicators of cortical atrophy. Future research is needed to assess the extent to which cerebral tauopathy is evident in this potentially novel condition.

WTC exposures included severe inhaled exposure to fine particulate matter as well as air pollutants like polycyclic aromatic hydrocarbons and dioxins. In this study, we found that responders who were exposed to the WTC worksites for months and who did not wear personalized protective equipment had a heightened burden of amyloidosis in the olfactory cortex as compared to those individuals with reduced lengths of exposure. Cerebral amyloidosis is concerning because it implies the presence of aging-related amyloidosis at younger ages. Such findings imply that exposure to air pollution may be a cause of a novel form of neuropathology in severely and chronically exposed individuals. If true, then efforts to remediate workplace exposures and protect workers might reduce the long-term burden of dementia in the population.