TNF inhibition differentially impacts regional Abeta immunotherapy efficacy in the humanized hAb^SAA mouse model

Abstract

Background

The FDA approval of amyloid beta (Aβ) targeting immunotherapies offered the first opportunity to clinically modify Alzheimer’s disease progression and give patients back precious months in the course of disease progression. The administration of these therapeutics has uncovered a common side effect called Amyloid Related Imaging Abnormalities (ARIA). While often asymptomatic, these adverse events have potential to escalate, and their commonality has encouraged further studies to mitigate their incidence. Increases in immune activation are important in immunotherapy action, though heightened release of cytokines have also been shown in ARIA development. TNF has been shown in early studies to increase with immunotherapy administration and subsequent ARIA development and TNF itself has been implicated in blood brain barrier dysfunction which could contribute to hemorrhagic ARIA (ARIA-H) development.

Methods

This study assessed hemorrhage outcomes in an aged mouse model using MRI and Prussian blue histology, explored transcriptomic and protein changes with NanoString nCounter and Meso Scale Discovery, and we performed immunohistochemical stains for microglia/macrophages, astrocytes, and pericyte changes.

Results

In this study, we established ARIA-H pathology in a humanized mouse model (hAβ^SAA) via chronic Aβ immunotherapy administration. We saw significant hemorrhagic lesions in our model, despite low levels of cortical cerebral amyloid angiopathy (CAA) suggesting a role for CAA as a risk factor for ARIA but potentially not its driver. We also explored the role of TNF in ARIA progression by inhibiting soluble TNF alongside immunotherapy administration as a potential combination therapeutic to address ARIA. With TNF inhibition, we saw a reduction of hemorrhages in the hippocampus, but not in the cortex, suggesting a regional effect. Further, we saw additional regional differences between the cortex and hippocampus with TNF inhibition including macrophage activation markers and amyloid levels, suggesting a differential role for TNF signaling in these areas and a potential heterogeneity of cellular responses in each region.

Conclusion

We induced ARIA without significant cortical CAA pathology and identified regional differences with TNF inhibition including a potential benefit of treatment in the hippocampus and possible detriment to pathology in the cortex.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13195-026-02084-7.

Background

Alzheimer’s disease (AD) is a complex neurodegenerative disorder that is comprised of two distinct proteinopathies resulting in cognitive decline and behavioral changes in patients. These proteinopathies include aggregates of extracellular amyloid beta (Aβ) and intracellular neurofibrillary tangles (NFTs) that are toxic to cellular microenvironments, resulting in downstream degeneration. Extremely common in aging populations, AD has been estimated to impact 1 in 9 people over the age of 65 and presents a complicated financial and caretaking burden that is only projected to increase by 2050 [1]. To combat this disease, research efforts have been made for decades to develop therapeutics that could slow or stop the progression of AD. Antibodies targeting Aβ have been explored since 1999, justified by earlier studies showing that accumulation of Aβ begins prior to NFT development [2, 3]. Subsequent research efforts have culminated in groundbreaking FDA approval of passive immunotherapies for AD [4, 5]. As the first disease-modifying therapeutics in AD, these drugs directly target Aβ and have been shown to slow cognitive decline, giving patients back precious months [4, 5]. During early research and in recent clinical trials, severe side effects of these Aβ immunotherapies became apparent and are now named Amyloid Related Imaging Abnormalities (ARIA) and present a major hurdle to widespread clinical use for these therapies [6–9].

ARIA presents in MRI scans as either vasogenic edema (ARIA-E) or as hemorrhagic lesions (ARIA-H) that, while often asymptomatic, can result in serious health complications or mortality [10, 11]. This has made ARIA detection and prevention a clinical and research priority. Studies have shown a link between vascular amyloid (cerebral amyloid angiopathy; CAA) and risk of developing ARIA, with several proposed mechanisms behind this link [9, 12]. It is thought that clearance prompted by Aβ immunotherapy may cause an overload of amyloid to the vasculature, dramatically increasing CAA load while decreasing vessel integrity [13]. Separately, it has also been proposed that ARIA is an immune response to CAA itself, where Aβ antibodies may bind to existing CAA, exacerbating local inflammation and further compromising vascular stability [14, 15]. In either case, innate immune signaling is implicated via local and widespread inflammatory signaling that may exacerbate pathology. With these hypotheses in mind, there are still gaps in our understanding of ARIA mechanisms and key players that may be targetable in combination with Aβ immunotherapy to protect the efficacy of the antibody while eliminating the detrimental impacts of treatment.

Our laboratory has previously shown tumor necrosis factor (TNF) to be one of the earliest upregulated cytokines in the progression of CAA and other cerebrovascular disease models [16–18]. It has also been shown in a mouse model treated with Aβ immunotherapy that classical immune cytokines, including TNF, were increased [19]. While being an important master cytokine that can influence levels of other cytokine release, chronic TNF signaling may be detrimental and has been linked to necroptosis of hippocampal neurons in iPSC systems and human tissue [20]. Likewise, TNF has been implicated in blood brain barrier disruption by its ability to downregulate tight junction proteins and increase matrix remodeling protein expression via matrix metalloproteinases (MMPs) [21, 22]. In human psoriasis studies, use of TNF inhibitors lowered serum MMP9 and reduced activation markers on endothelial cells [23].

In this study, our outcomes were two-fold; we aimed to induce ARIA-H in the humanized hAβ^SAA mouse model through chronic Aβ antibody administration to demonstrate efficacy of this strain as an ARIA model. We then directly compared immunotherapy-only treatment to a combination of immunotherapy with soluble TNF inhibitor to determine the role of TNF in ARIA related vascular dysfunction. XPro1595 is a soluble TNF targeting biologic that was chosen for this study due to its ability to cross the blood brain barrier and reach therapeutic levels in the CSF [24]. Preclinically, XPro1595 has shown promise in reducing plaque load in 5xFAD mice, decrease IBA1 and GFAP staining across various models, and improved functional outcomes in a model of cerebral ischemia, which encouraged testing its benefit in reducing ARIA-H, which may be driven by similar inflammatory and vascular mechanisms [24–28]. By specifically targeting soluble TNF, we hypothesized that we would see a reduction in MRI and Prussian blue measured ARIA-H by modulating neuroinflammation without disrupting Aβ clearance.

Methods

Animals

Female and male hAβ^SAA [B6(Cg)-APP^tm1.1Dnli/J] mice were used, which carry a humanized Aβ sequence within the murine App gene (N = 22). These animals carry the Swedish (KM670/671NL), Arctic (E693G), and Austrian (T714I) mutations. Mice were bred in house in 12 h light/dark cycle, provided food and water ad libitum and aged to 19 months before beginning treatments. This study was approved by the Indiana University Institutional Animal Care and Use Committee and conformed to the National Institutes of Health Guide of the Care and Use of Animals and Research and ARRIVE guidelines (Animal Research: Reporting of In Vivo Experiments).

Treatment

Mice were randomized and received an intraperitoneal injection of either mouse anti-Aβ immunotherapy (3D6 in saline; Eli Lilly, Indianapolis, IN) or IgG2a control (Eli Lilly, Indianapolis, IN) weekly at 10 mg/kg for 12 weeks. A single 3D6 dosing regimen was employed due to extensive literature indicating this dose reliably induced ARIA-like events [29–32]. Additionally, animals received either subcutaneous injection of XPro1595 (10 mg/kg in saline; INmune Bio. Boca Raton, FL) or saline control twice weekly for 12 weeks [25, 33]. Treatment groups will be referred to as IgG:S(aline), 3D6:S(aline), and 3D6:X(Pro1595) for the duration of the paper. Groups were arranged as follows: IgG:S (n = 6; 3 m, 3f), 3D6:S (n = 8; 5 m, 3f), and 3D6:X (n = 8; 3 m, 5f).

Anesthesia protocol for in vivo MRI

For MRI experiments, animals were initially anesthetized in an induction chamber under 3% isoflurane at 1 L per minute in 100% oxygen. The anesthetized mice were then transferred to an MR compatible cradle and positioned in an MRI compatible head holder to minimize head motion. Anesthesia was subsequently maintained at 1.5% isoflurane in 100% oxygen throughout imaging. Respiration rate was monitored using a pressure pad placed under the animal abdomen and animal body temperature was maintained by a warming pad (37 °C) placed under the animal. Respiration rate was maintained between 80 and 125 breaths per minute using manual adjustments to isoflurane vaporizer.

Magnetic resonance imaging

Baseline scanning was conducted one week prior to treatment initiation and endpoint data was collected ~ 1 week prior to euthanasia. Imaging staff were blinded to experimental groups. In vivo MRI was conducted at SNRI on a horizontal bore 9.4 Tesla Biospec pre-clinical MRI system (Bruker BioSpin MRI GmbH, Germany) equipped with shielded gradients (maximum gradient strength = 660 mT/m, rise time = 4570 T/m/s) and ¹H mouse cryogenic surface coil (Cryoprobe, Bruker, Biospin). Mice were anesthetized with 5% isoflurane (balance medical oxygen), placed on an MRI scanning platform, and maintained at 1–3% isoflurane during the entire scan period. Respiration was monitored and controlled continuously. For T2* weighted imaging, a 3D multi-echo gradient echo (ME-GRE) sequence with the following parameters was used: TE = 3.28 ms, ΔTE = 4.47 ms, 5 echoes with monopolar readout gradients, TR = 37 ms, flip angle = 12.5°, FoV = 16 × 16x9 mm, matrix size = 256 × 256x36, flow saturation = on, averages = 6.

MRI data processing

The root mean square (RMS) based echo-combined magnitude images (T2*W_TE-weighted) were created using the weighted combination of the multi-echo images. The echo dependent weighting factor ω_i is given by Eq. 1 [34]:

1

where i = 1–5, and S_i is the coil combined magnitude image for the i^th echo. The TE weighted magnitude T2*W_TEW image is given by Eq. 2:

2

Following image processing, microhemorrhages were manually counted per slice by locating areas of hypointensity on T2*W_TEw anatomical images with ROIs in the cortex and hippocampus specifically. Hypointense areas were noted as dark, often punctate points though could be larger depending on size of hemorrhage. If a hemorrhage was noted across several slices, it was still counted as a single bleed in our data unless clearly distinct from neighboring hemorrhage. Thin blood vessels and punctate points that were bilateral were not included, though puffiness surrounding vessels were counted as bleeds (Fig. 2A).

Fig. 2.

3D6 administration results in significant hemorrhage pathology in hAβSAA mice. Hemorrhage assessment via magnetic resonance imaging and Prussian blue. Representative images of mouse brain at endpoint measure with cortical hemorrhages indicated by white dashed circlesred arrows (A). Hemorrhage counts from T2*W imaging at baseline and endpoint for each group [IgG:S (IgG + saline), 3D6:S (3D6 + saline), and 3D6:X (3D6 + XPro1595)]. Graphs are split by total hemorrhage count, cortical hemorrhages, and hippocampal hemorrhages. Manova values depicting the interaction of treatment and time are reflected on each graph (B). Representative image of Prussian blue histology with cortical and superficial bleeding denoted with arrows (C). Scale bar in C = 100μm. Cortical, superficial, and hippocampal Prussian blue positivity are shown by average number of hemorrhages, average size of the hemorrhages (μm2), or the average coverage of Prussian blue positivity (μm2) (D-F). Graphed correlation of MRI and Prussian hemorrhage data within a density ellipse (confidence interval 0.95) for cortex and hippocampus (G). Significant p-values from either all pairs Tukey HSD or Steel-Dwas all pairs were reported at *p < 0.05, **p < 0.01, ***p < 0.001. Error bars show mean ± standard deviation within each timepoint. All Prussian Blue images taken at 10 × objective. Male data points are shown in blue and female data points are shown in pink

During MR imaging, it was found that a majority of female hAβ^SAA mice developed apparent pituitary tumor pathology. Seven females began the study with apparent tumors, while nine ended the study with tumor pathology. Notably, 1 female had an apparent tumor in the left cortex, while 1 male was excluded from the study for presumed tumor pathology near the left hippocampus, resulting in a sizable hemorrhage.

Tissue collection and processing

Following lethal injection of Euthasol, blood was collected retro-orbitally into EDTA tubes, spun at 4 °C at 1000 g for 15 min. Plasma supernatant was removed and stored at −80 °C. Mice were perfused with 20 mL normal saline, brains were rapidly removed and bisected down the midsagittal plane. The right hemisphere was dissected into frontal and posterior cortex, striatum, hippocampus, thalamus, cerebellum, and rest of brain prior to flash freezing in liquid nitrogen and storage at −80 °C. The left hemisphere was fixed in 4% paraformaldehyde for 24 h. These were passed through 10%, 20%, and 30% sucrose gradients prior to sectioning on a sliding microtome, where 25 μm horizonal sections were collected and stored in 1xDPBS with sodium azide at 4 °C.

RNA and NanoString nCounter

RNA was extracted from the right frontal cortex (n = 22) using the Qiagen RNeasy Mini Kit according to the manufacturer’s instructions. On column DNase digestion was also performed during the protocol using the Qiagen RNase-free DNase Set according to manufacturer’s instructions. RNA quality and quantity were analyzed using the Agilent Bio-Tek Synergy Neo2 Hybrid Multimode Reader and the Take3 plate. 5μL of normalized samples (20 ng/μL) were run on the NanoString Technologies nCounter system housed within the Stark Neurosciences Research Institute Biomarker Core at Indiana University. Both the Mouse Neuroinflammation and Mouse Cardiovascular Disease panels were run on all samples (Bruker Spatial Biology, Inc).

Gene expression changes for all genes were investigated using two-way t-tests comparing each treatment group for males, females, and combined sexes. For each test, log2 fold change from their relevant control was assessed for each gene. FDR correction was performed on t-test p-values to account for the number of comparisons made. There were no statistically significant differences observed following corrections for multiple comparisons. Gene ontology terms (GO terms) were identified using the National Institutes of Health Database for Annotation, Visualization, and Integrated Discovery (DAVID). These terms were determined by combining the nCounter Neuroinflammation and Cardiovascular panels and using any gene with a p-value < 0.05.

Meso scale discovery (MSD)

Protein was extracted from right posterior cortex tissue (N = 22). Samples were homogenized in 500μL lysis buffer consisting of MPER and protease and phosphatase inhibitors. Samples were centrifuged for 15 min at 4 °C and 10,000 g. The supernatant was moved into a new tube and concentration of soluble protein was obtained using the Pierce BCA Protein Assay Kit (Thermo Scientific). Samples were then normalized to 1 mg/mL. The Meso Scale Discovery U-Plex Macrophage M1 combo 1 (mouse) assay plate was modified to include TNFRI and was used to quantify proteins of interest (U-Plex cat #K15408K). Samples were run in duplicate and plates were run according to the manufacturer instructions and read on the MESO QuickPlex SQ 120MM plate reader using the provided Methodical Mind software. Analysis was conducted using the Discovery Workbench analysis software, then exported to JMP for statistical comparisons. Proteins measured were IL-1β, IL-6, IL-12p70, IL-15, IL-23, IP-10, MCP-1, MIP-1α, TNF, and TNFRI. Any measurements that were found to be above or below standard curve were excluded. All samples had measurable quantities of each target protein, aside from IL-1β, which was not included in 6 samples due to measurements outside of our standard curve.

Immunohistochemistry and histology

Eight 25 μm sections spaced 600 μm apart were selected for free floating immunohistochemistry (IHC; N = 22) for total Aβ using the 4G8 antibody (1:3000 dilution, Biolegend, Cat#800,704), IBA1 (Rabbit monoclonal, 1:1000 dilution, Wako/Fuji, Cat#019–19741), CD206 (Rat monoclonal, 1:1000 dilution, BioRad, Cat#MCA2235), GFAP (Rabbit monoclonal, 1:10,000 dilution, Dako, Cat#Z0334), DP71 (Rabbit monoclonal, 1:1000 dilution, Abcam, Cat#ab15277), CD13 (Rat monoclonal, 1:1000 dilution, MBL International, Cat#M101-3). IHC was performed as previously described [35]. In short, sections were stained, mounted, allowed to dry overnight, then dehydrated and cover slipped in DPX (Electron Microscopy Sciences, Hatfield, PA). For histology, eight sections spaced 600 μm apart were selected and mounted on slides for Congo red and Prussian Blue staining as previously described (N = 23) [6, 36]. Slides were scanned on an Olympus Slideview VS200 at 10x. Imagers were blinded to experimental groups. Regions of interest for the cortex and hippocampus were drawn and staining was analyzed using NIS Elements General Analysis 3 module (Nikon Instruments, Melville, NY) as described previously [37]. Briefly, in our IHC we assessed percent area of stain, number of positively stained cells/vessels, and their average size. For Congo red histology, we assessed total Aβ, dense core plaques, and CAA, as well as Aβ density and size of plaques. For Prussian blue histology, we identified number of Prussian positive hemosiderin deposits and their respective areas for each region.

Immunofluorescence and confocal imaging

Eight 25 μm sections spaced 600 μm apart were mounted for on-slide immunofluorescence staining (IF). For microglial/plaque staining (N = 12) we used 4G8 (Mouse monoclonal, Alexa 488 primarily conjugated. Biolegend. Cat#800,714), IBA1 (Rabbit monoclonal, 1:1000 dilution, Wako/Fuji, Cat#019–19741. Secondary: Goat anti-Rabbit, Alexa 594. Invitrogen. Cat#A21207), and CD68 (Mouse monoclonal, Alexa 647 primarily conjugated. Bio-Rad. Cat#MCA1957A647). Tissue sections were mounted on slides and allowed to dry overnight. The following day, slides were blocked for an hour, then incubated in IBA1 primary antibody overnight. On day 2, slides were incubated in anti-rabbit Alexa 594 secondary antibody for 2 h, then incubated in CD68 and 4G8 primarily conjugated antibodies overnight. On the third day, slides were washed and cover slipped using ProLong Diamond mounting media (Invitrogen. Cat#P36966) and allowed to dry prior to imaging.

Imagers were blinded to experimental groups. Slides were scanned on an Olympus Slideview VS200 at 10x. Regions of interest for the cortex and hippocampus were drawn and staining analyzed using NIS Elements General Analysis 3 module (Nikon Instruments, Melville, NY) [38]. Briefly, we assessed percent area of each stain and average number of microglia/mm² to look at differences across the entirety of each region of interest.

Slides were then imaged on a Zeiss LSM 900 confocal on an Axio Observer 7 motorized inverted microscope. Three diode lasers were used for excitation, 488, 561, and 640 nm. Using a Plan-Apochromat 20x/0.8 objective, four images were taken from a single central slice per animal. Image location was determined by positive presence of 4G8 and IBA1. Three images were taken in the cortex, one image per region (frontal, parietal, posterior cortex) and then combined together for comprehensive cortical data. One image was taken within the hippocampus. Images were 319.45 by 319.45 μm in size. Z-stack was collected across 11 slices covering 20 μm. Data was exported and analyzed using NIS Elements General Analysis 3 module (Nikon Instruments, Melville, NY) [38]. In brief, we performed a Scholl’s concentric circle analysis wherein we identified plaques and created a ring around the plaques to assess microglia (IBA1+ cells) within 10 μm of the plaque. Using these microglia, we identified which were CD68 and 4G8 positive, then graphed the distribution of these markers and average volume of the cells.

Statistical analysis

All statistical analyses were conducted utilizing JMP Pro 17 (JMP Statistical Discovery, LLC). Outliers were detected via repeated Grubbs testing (Supp. Table 1). Data was graphed using GraphPad Prism (GraphPad Software). A Manova for our repeated measures MRI data was used and run with treatment and sex as covariates of interest. All individual comparisons were made using multiple comparison, all pairs Tukey HSD or Steel-Dwas all pairs depending on distribution. All data are expressed as mean ± standard deviation from the mean with pink and blue dots representing male and female mice, respectively. P < 0.05 was considered statistically significant and p-values less than 0.1 were reported with their value.

Results

Aβ staining was conducted to assess total amyloid clearance, and the amount of dense core plaques vs diffuse in hAβ^SAA mice in both the cortex and hippocampus (Fig. 1A-D). In each group, we saw that a majority of the total percent Aβ deposition was from diffuse deposition compared to dense core deposits (Fig. 1C). In animals treated with 3D6, we saw a significant reduction in both diffuse Aβ percent area and the average size of the diffuse Aβ in the cortex (Fig. 1C). Only diffuse Aβ average size was significantly decreased in the hippocampus, and only in animals treated with 3D6 alone (Fig. 1D). We did not see any significant changes in dense core Aβ plaque deposition.

Fig. 1.

3D6 immunotherapy treatment reduces diffuse amyloid in hAβSAA animals, but not dense core plaque load. Amyloid quantification for total Aβ in the cortex and hippocampus (A-D). Representative images of total Aβ for each group [IgG:S (IgG + saline), 3D6:S (3D6 + saline), and 3D6:X (3D6 + XPro1595)] in the cortex (A), and hippocampus (B). Cortical and hippocampal quantification of total, dense and diffuse Aβ shown as percent positive stain, amyloid density (Aβ/μm2), and average Aβ area (μm2) (C-D). Representative images of Congo red staining for each group in the cortex (E) and hippocampus (F). Cortical and hippocampal quantification of total, dense plaque, and CAA shown as percent positive stain, amyloid density (Aβ/mm2), and average congophillic plaque area (μm2) (G-H). Significant p-values from either all pairs Tukey HSD or Steel-Dwas all pairs were reported at *p < 0.05, **p < 0.01, ***p < 0.001. Error bars show mean ± standard deviation. All images taken at 10 × objective. Scale bar in all images = 250μm. Male data points are shown in blue and female data points are shown in pink

Congo red staining was used to evaluate the distribution and severity of dense core Congophilic plaques and CAA in the hAβ^SAA mice across treatment groups in the cortex and hippocampus (Fig. 1E-H). Percent area of Congo staining showed a majority of Congophilic plaque deposition was from dense core plaques, with very little Congophilic CAA presence noted (Fig. 1G, H). There was, however, a significant reduction in Congophilic CAA density with administration of 3D6. There were no other differences in percent area, average Congophilic plaque area, or plaque density across treatment groups.

Magnetic resonance T2*W imaging was conducted at baseline and endpoint to assess for presence and development of hemorrhagic lesions due to treatment administration (Fig. 2A-B). There were very few hemorrhages in any group at baseline, however, there was a time by treatment effect of 3D6 administration shown in our Manova (p < 0.0001) which caused significant increase in total cortical hemorrhage number irrespective of secondary treatment (saline or XPro1595) (Fig. 2B). We did not see this treatment effect in the hippocampus (p < 0.5252). When we included sex as a covariate in our Manova, we identified a significant effect of sex on number of bleeds over time in the cortex (p < 0.001) but not the hippocampus (p = 0.6039) (Supp. Figure 1). Using Prussian blue to further explore observed hemorrhagic events (Fig. 2C-F) it was found that 3D6 administration caused significant superficial and cortical bleeding in hAβ^SAA mice compared to IgG controls, with no statistical effect of XPro1595 dosing. In the hippocampus, 3D6 significantly increased the number of hemorrhages, however, this was attenuated by XPro1595 treatment. Correlating the MRI and Prussian blue data, there was a significant positive correlation between manual MRI hemorrhage counts and Prussian blue hemorrhage counts (p = 0.0001, r² = 0.5756), though no correlation between counts from the hippocampus (p = 0.6089, r² = 0.0167) (Fig. 2G).

To elucidate transcriptomic changes across treatment group and sex, we used NanoString nCounter panels to identify targeted neuroinflammatory and cardiovascular changes in the frontal cortex. Comparing 3D6:S treated animals to IgG controls, we saw an increase in expression of genes including Mmp12, C3, Fbn1, and Col1a1 across our two panels (Fig. 3A). From our p-value significant genes, we identified GO terms related to apoptosis, innate immunity, and positive regulation of NFκB (Fig. 3B). Comparing 3D6:X treated animals to 3D6:S treated controls, we saw upregulation of genes including Bik, C6, Cd69, Cd163, and Il31ra (Fig. 3C), and GO terms were related to transcription regulation, chromatin remodeling, and autophagy (Fig. 3D). When comparing male IgG treated hAβ^SAA animals to their female counterparts, we did not observe overt transcriptomic differences (Fig. 3E). To further assess inflammatory changes, we ran Meso Scale Discovery assays using posterior cortex tissue to quantify protein levels of several inflammatory proteins (Fig. 3F). We saw no changes in IL-6, TNFR1, IL-1β, or IL-15. There was an impact of 3D6 administration, which lowered levels of MIP-1a and IL-23. Interestingly, 3D6 increased levels of MCP-1, which was attenuated by XPro1595 administration.

Fig. 3.

Distinct transcriptomic and MSD signatures following Aβ-immunotherapy and TNF-targeting biologic administration. Transcriptomic analysis via NanoString nCounter using targeted neuroinflammation and cardiovascular panels. Volcano plots showing 3D6:S differentially expressed genes (DEGs) compared to IgG:S control for both neuroinflammation and cardiovascular nCounter panels (A). Gene ontology terms from significant genes in 3D6:S/IgG:S comparison (B). Volcano plots showing 3D6:X DEGs compared to 3D6:S control for both neuroinflammation and cardiovascular nCounter panels (C). Gene ontology terms from significant genes from 3D6:X/3D6:S comparison (D). Volcano plots showing male IgG:S DEGs compared to female IgG:S animals (E). Significant (p < 0.05) downregulated genes are colored blue (left side of graph, above dotted lines denoting significance) and upregulated genes are colored red (right side of graph, above dotted lines denoting significance). No genes passed correction for false discovery rate. Proteomic assessment of the posterior cortex using MSD for inflammatory proteins (F). Significant p-values were reported at *p < 0.05, **p < 0.01, ***p < 0.001. MSD was assessed with either all pairs Tukey HSD or Steel-Dwas all pairs depending on distribution of data. Error bars show mean ± standard deviation. Male data points are shown in blue and female data points are shown in pink

We next used immunohistochemistry (IHC) to assess key cells known to modulate blood brain barrier permeability. GFAP was targeted to capture astrocyte activation, where we saw no changes in percent area of stain, astrocyte density, or average astrocyte area across any groups in the cortex or hippocampus (Fig. 4A-C). Dystrophin 71 (DP71) is structurally important for astrocytes, acting as an anchor between the astrocytic endfoot and the cerebrovasculature [39]. There were no observed differences in DP71 percent area or in vessel density or vessel area across any groups or brain region (Fig. 4D-F). Pericytes were targeted due to their proximity to small vessels and regulatory role in cerebral blood flow [40]. CD13 staining was conducted to identify pericytes along the vasculature (Fig. 4G), though we did not see any significant differences in pericyte CD13 expression, pericyte density, nor pericyte average area across our treatment groups (Fig. 4G-I).

Fig. 4.

No major changes in neurovascular unit glial regulators. Immunohistochemical assessment of astrocytes and pericytes. Representative image of cortical astrocyte GFAP staining (A). Quantification of percent area GFAP, astrocyte density, and average astrocyte area for the cortex (B) and hippocampus (C). Representative image of DP71 staining in cortex (D). Quantification of percent area DP71, DP71 positive vessel density, and average vessel area in cortex (E) and hippocampus (F). Representative image of cortical pericyte CD13 staining (G). Quantification of percent area CD13, pericyte density, and average pericyte area in the cortex (H) and hippocampus (I). Significant p-values from either all pairs Tukey HSD or Steel-Dwas all pairs were reported at *p < 0.05, **p < 0.01, ***p < 0.001. Error bars show mean ± standard deviation. All images taken at 10 × objective (scale bars = 250μm). Male data points are shown in blue and female data points are shown in pink

In response to the minimal changes in astrocyte or pericyte engagement, we explored microglial and macrophage involvement. IBA1 immunohistochemistry was performed to identify changes in brain macrophage (primarily microglial) activation (Fig. 5A-D). In the cortex, we saw a decrease in average microglial area with 3D6 administration, which was also decreased with the addition of XPro1595 treatment (Fig. 5B). Likewise, we saw a significant reduction in microglial branch endings in 3D6:X treatment compared to IgG:S, indicative of heightened microglial activation (Fig. 5B). In the hippocampus, there were fewer changes, though there was still a significant decrease in average microglial area in 3D6:X treated animals compared to IgG:S (Fig. 5D). We next explored perivascular macrophages (PVMs) by performing CD206 staining, which distinguishes these cells from microglia (Fig. 5E-H) [41]. While there were no significant differences in our CD206 staining, we did see a mild decrease in average PVM area (p = 0.0746) and soma area (p = 0.0617) in the cortex (Fig. 5F). Interestingly, in the hippocampus we saw a slight increase in CD206 percent area (p = 0.0776) and an increase in PVM density (p = 0.0512) (Fig. 5H).

Fig. 5.

Microglial and macrophages show exacerbated activation when administered both Aβ immunotherapy and TNF targeting biologic. Immunohistochemical assessment of microglia and perivascular macrophages (PVMs). Representative images of IBA1 staining in cortex (A) and hippocampus (C). Quantification of percent area IBA1, microglial density, microglial area, and microglial branch endings in cortex (B) and hippocampus (D). Representative images of CD206 staining for PVMs in cortex (E) and hippocampus (G). Quantification of percent CD206, PVM density, average PVM area, and PVM soma area in cortex (F) and hippocampus (H). Significant p-values from either all pairs Tukey HSD or Steel-Dwas all pairs were reported at *p < 0.05, **p < 0.01, ***p < 0.001. Error bars show mean ± standard deviation. All images taken at 20 × objective. Scale bar in all images = 100μm. Male data points are shown in blue and female data points are shown in pink

Following our observation of reduced microglial area in IBA1 IHC, we performed immunofluorescence (IF) staining to identify phagocytic microglia near amyloid plaques. We utilized 4G8 to show amyloid, IBA1 to label activated microglia, and CD68 to identify phagocytic microglia (Fig. 6). In full tissue images, we saw more microglia that were IBA1 and CD68 positive compared to IBA1 alone or IBA1/CD68 positive and colocalized with 4G8 in every group. We did not observe differences in the distribution of microglial markers across treatment groups, nor in the density of microglia populations in either the cortex or hippocampus (Fig. 6B-C). In confocal imaging, we looked specifically at the makeup of microglia within 10 μm of a plaque (Fig. 6D-E). In the cortex, we saw no changes across treatment groups in average plaque volume, nor in the distribution of microglial phagocytic markers (Fig. 6D). We also did not see changes in the average microglia volume across groups. Interestingly, when we looked in the hippocampus, we saw a decrease in average plaque volume in the 3D6:X treated group compared to IgG:S. There was a higher percentage of IBA1 + CD68- microglia in 3D6:X compared to IgG:S, with significantly lower percentages of cells positive for both IBA1 and CD68, or with 4G8, indicative of reduced phagocytic microglia (Fig. 6E). There were no changes in the average microglial size across groups.

Fig. 6.

TNF inhibition differentially impacts phagocytosis markers in the hippocampus compared to cortex. Immunofluorescent (IF) assessment of microglia and amyloid. Representative confocal images of plaque (4G8; Green) and microglia (IBA1; Red, CD68; White) in the cortex and hippocampus (A). Quantification of IF shown as microglial distribution and density in cortex (B) and hippocampus (C). Microglial distribution is shown as percentage of marker combinations (IBA1 + CD68-, IBA1 + CD68 +, and IBA1 + CD68 + 4G8 +) from total IBA1 positive staining. Microglial density showed as number of microglia exhibiting each marker combination. Quantification of confocal imaging assessing microglia within 10 μm of plaque in cortex (D) and hippocampus (E). Data shown as 4G8 volume, microglial distribution as percentage of total IBA1 + staining, and average microglial volume for each marker combination. Significant p-values from either all pairs Tukey HSD or Steel-Dwas all pairs were reported at *p < 0.05, **p < 0.01, ***p < 0.001. Error bars show mean ± standard deviation. All images taken at 20 × objective. Scale bar in all images = 100μm. Male data points are shown in blue and female data points are shown in pink

Discussion

The approval of amyloid targeting immunotherapies was groundbreaking and offered the first real opportunity to modify Alzheimer’s disease progression and give patients back a sense of agency in their fight against dementia. While giving back precious months to patients, Aβ targeting immunotherapies pose the risk of ARIA development. These side effects are extremely common and, in rare cases, can prove fatal, encouraging studies to explore the mechanisms of ARIA and potential avenues for mitigating ARIA risk. In this study, we demonstrated ARIA induction in mice and assessed the role of TNF in the development and progression of ARIA pathology. In this paradigm, we demonstrated regional differences in the impact of soluble TNF inhibition on drug efficacy and in potential microglial/macrophage phenotype following chronic soluble TNF inhibition.

Using the hAβ^SAA mouse model, we were able to induce significant hemorrhaging following chronic treatment with mouse monoclonal anti-Aβ antibody 3D6, indicative of ARIA-H. A majority of hemorrhages were within the cortex, and most were found near the surface of the brain as possible superficial siderosis, a common ARIA-H presentation [8]. This was particularly interesting due to the lack of cortical Congophilic cerebral amyloid angiopathy (CAA) found in these animals. Our finding suggests that while CAA may be a strong risk factor for ARIA development during immunotherapy treatment, cortical CAA may not be necessary to drive ARIA-H as is commonly hypothesized and requires further studies to elucidate the nuances of this hemorrhage mechanism. Additionally, these findings suggest the potential of the hAβ^SAA animals as a model of superficial siderosis in the context of immunotherapy administration without significant cortical Congophilic CAA presence.

In hypotheses of ARIA mechanisms, the immune system is thought to be central to the development of edema and hemorrhagic events linked to immunotherapy administration. Cytokine release encourages cellular responses to amyloid and further immune recruitment which may be detrimental to vascular integrity [13–15]. When assessing cytokine levels in an ARIA model, TNF and others were found to be upregulated with immunotherapy treatment [19]. TNF is a master cytokine capable of regulating downstream inflammation and has been linked to neurodegeneration, vascular disease, and autoimmune diseases [42]. Studies have suggested TNF inhibition may have a protective effect on dementia development, further implicating it as a therapeutic target in both dementia and ARIA [43, 44].

Targeting TNF in the brain has historically been challenging, as most approaches do not cross the blood brain barrier and global inhibitors may lead to immunocompromise by reducing transmembrane signaling [45–47]. XPro1595 is a soluble TNF targeting biologic that was chosen for this study due to its ability to cross the blood brain barrier and reach therapeutic levels in the CSF [24]. By specifically targeting soluble TNF, we hypothesized that we would see a reduction in ARIA by modulating neuroinflammation without disrupting Aβ clearance.

Inhibition of soluble TNF in this concurrent-dosing paradigm was associated with a modest attenuation of 3D6-mediated amyloid reductions in select regions, as shown by slightly diminished amyloid clearance following immunotherapy treatment. In addition to potential reduced efficacy, XPro1595 treatment did not reduce 3D6 induced hemorrhages in the cortex but did in the hippocampus, though it should be pointed out that there were very few hemorrhagic events in the hippocampus in any group.

Interestingly, XPro1595 treatment was associated with significant changes in microglia and subtle changes in PVM area, suggesting increased microglia activation in the cortex and hippocampus. Notably when we investigated potential functional changes, soluble TNF inhibition was associated with less microglial CD68 expression and less colocalization near plaques, suggesting a phenotypic shift from phagocytic to a more general inflammatory state specifically in the hippocampus, but not in the cortex. There are several possibilities for this, including regional differences in TNF receptor distribution, regional differences in drug availability, or phenotypic heterogeneity in microglia between the hippocampus and cortex. While we found no overt IHC protein difference between groups in the cortex, we did see some changes in the posterior cortex while using MSD. We also saw changes in RNA related to transcription regulation and inflammation in the 3D6/XPro1595 treated group, which supports the idea of differential microglial responses rather than drug availability or receptor distribution.

While none of our genes passed FDR corrections, we plotted genes with significant p-values to identify potential avenues of exploration in future studies. In 3D6 treatment alone compared to IgG, transcriptomic changes included upregulation of complement component 3 (C3), which supports ongoing hypotheses of complement having a role in AD and ARIA development [48]. We also saw changes in ECM and angiogenic modulators, including Mmp12 and Cd36, which may be contributing to vessel breakdown/dysfunction contributing to the hemorrhagic events that we observed within this study. With the addition of XPro1595, we further saw changes in complement (C6) which is a key component of the membrane attack complex (MAC) which allows for cell lysis as well as Bik (a pro-apoptotic factor) which may promote further damage via cell death signaling in already overwhelmed vasculature. Inhibition of TNF was also associated with upregulation of IL33, a common alarmin that has been found in brain capillaries and has been proposed to decrease CAA formation [49]. When looking at proteomics, we saw that 3D6 administration decreased inflammatory markers MIP-1α and IL-23, potentially in an anti-inflammatory fashion. Interestingly, 3D6 increased MCP-1, a macrophage chemoattractant, which was attenuated with XPro1595 treatment further suggesting XPro1595 modulating the immune response in our study.

Limitations

There are some limitations to our study to discuss. The presence of tumors in female mice was unexpected. We compared males and females together and saw no clear differences driven by sex or tumor presence, though this does not mean that the tumors had no physiological impact, only that they did not significantly impact areas we explored. We hypothesize that the aforementioned tumors are of pituitary origin, and likely benign, though further histopathology would need to be performed to determine exact origins and makeup of the tumors and any functional impact they may pose. In addition, we were not powered for a deeper investigation of sex differences in this study and future exploration of minute differences is warranted.

Separately, we only assessed these mice at their endpoint, following 3 months of treatment. A future study should include a time course of ARIA development throughout treatment to determine exactly when pathology began and look at cellular responses across timepoints. It is possible that the changes we saw in our hippocampal immunofluorescence were due to successful clearance and microglia slowly returning to homeostasis, rather than purely being from TNF inhibition. This is separate from our original hypothesis, but still worth further exploration.

This study also focused primarily on ARIA-H due to ARIA-E being difficult to assess in mice and we were attempting to limit time under anesthesia during MRI. We anticipated ARIA-H to manifest cortically and thus limited our focus to cortical and hippocampal regions of interest. However, future studies mapping lesions to more precise anatomical regions will help to identify specific areas of change. Additionally, moving forward our group is assessing ARIA-E and vascular leakage using two photon microscopy, though that was not included in this study design. In our assessment of ARIA-H, we explored cortical and hippocampal CAA burden using Congo Red. In future studies, inclusion of Aβ_42/40 staining would be incredibly valuable to assess any non-fibrillar vascular amyloid to include in CAA quantification. It should be mentioned that our method of brain removal may result in loss of leptomeninges, thus our CAA quantification is limited to cortical and hippocampal vessels, and our data encourages pointed inclusion of leptomeninges in future studies.

We also targeted TNF at the same time as initiating immunotherapy. If the relevant protective mechanism of TNF modulation involves vascular stabilization, blood–brain barrier repair, or normalization of perivascular inflammatory tone, such effects may require a period of pre-treatment to manifest. Accordingly, this experimental design does not address whether TNF inhibition administered prior to antibody-mediated amyloid mobilization could differentially impact vascular outcomes. There is a chance that this combination would be more efficacious if neuroinflammation was reduced earlier, prior to immunotherapy treatment beginning rather than trying to reduce inflammation while simultaneously stimulating an immune response.

Conclusion

In this study, we targeted TNF in combination with Aβ immunotherapy to mitigate ARIA-H development. We identified a regional difference in outcomes following TNF inhibition, including a reduction in hemorrhage counts in the hippocampus but a significant increase in the cortex, supporting future studies in regional heterogeneity of cellular responses. We also showed ARIA-H presentation despite low cortical CAA burden. This provides new insight into ARIA mechanisms without cortical CAA in a causal role and encourages further studies looking at these as two distinct yet synergistic pathologies.

Authors’ contributions

KEK: Design/methodology, data collection, analysis, writing. JLC: Data collection, analysis. JTL: Data collection. HW: Data collection. CAF: Data collection, analysis. SSS: Design/methodology, analysis. EMW: Design/methodology, analysis, project administration, supervision, review and editing. DMW: Design/methodology, funding acquisition, supervision, review and editing. All authors read and approved the final manuscript.

Funding

Research reported in this publication was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under Award Number 1RF1NS130834 (DMW). NIH provided 100% of total project costs ($100,000). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary files. NanoString data will be made available upon request.

Declarations

Ethics approval and consent to participate

This study was approved by the Indiana University Institutional Animal Care and Use Committee and conformed to the National Institutes of Health Guide of the Care and Use of Animals and Research and ARRIVE guidelines (Animal Research: Reporting of In Vivo Experiments).

Consent for publication

Not applicable.

Competing interests

The authors declare the following financial interests/personal relationships which may be considered potential competing interests: DMW reports relationships with the following: Longeveron LLC (consulting or advisory), Vigil Neuroscience Inc. (consulting or advisory), SynapsDx (consulting or advisory), Alzheimer’s & Dementia (board membership).