Detection of molecular markers of ferroptosis in human Alzheimer's disease brains

Subjects and case selection

The investigations were performed on postmortem human brain tissue from the Institute of Neuropathology and from the brain bank of the Institute of Anatomy at Leipzig University. All donors signed a written declaration of consent for this purpose, whereby the use of human tissues for scientific purposes is approved by the Ethics Committee of the University of Leipzig (registration number: 129-21 ek). Twenty cases with a low NFT stage ≤ I and 20 cases with a high NFT stage ≥ IV were used (Supplemental Table 1). NFT staging was performed as described by Braak et al. ³³ The age range of the donors was from 46 to 93 years. We focused our research on the prefrontal cortex (PFC) because material of this brain region was sufficiently available from all cases. In three cases (41, 42, 43), we also analyzed TfR expression in the hippocampal region. In addition, we used hippocampal and prefrontal brain slices from three children, which were either stillborn or a few weeks old, to investigate expression of ferritin and TfR (Supplemental Table 1). Furthermore, we used liver sections from the Leipzig body donation bank as a positive control for TfR since this receptor is known to be expressed in hepatocytes. ³⁴ Exclusion criteria for case selection were a postmortem delay (PMD) of more than 48 h, cerebral hemorrhage, brain tumors, and ischemia as a cause of death because of the possible impairment of brain iron metabolism.

Sample preparation and fixation

After removal the human tissue material was fixed in 4% phosphate-buffered formaldehyde. Tissue samples were dehydrated, embedded in paraffin and cut into 10–12 μm thick sections. We also used fresh-frozen material, which after removal was embedded in Tissue-Tek (Tissue- Tek^® O.C.T.™ Compound), frozen on dry ice, stored at −80°C and cut on a cryostat (12 μm).

Immunohistochemistry (IHC)

Brain sections were deparaffinized and antigen retrieval was performed in citrate buffer at pH 6 for 20 min in the microwave at 95°C. After cooling and washing the sections in 0.02 M phosphate-buffered saline (PBS) pH7.4, endogenous peroxidase was blocked with H₂O₂ in PBS for 30 min at room temperature (RT). The sections were washed with PBS and a barrier created with Dako Pen (ScienceServices). After washing with PBS containing 0.3%Triton TX-100 (PBS-T), blocking was performed with normal goat serum (NGS, Jackson Cat# 005-000-121) diluted 1:20 in PBS-T for 2 h at RT. Sections were incubated overnight at 4°C with the respective primary antibodies diluted in PBS-T with 1% NGS. Anti-TfR1 3F3 FMA, CD71 3B8 2A, TfR H68.4 and anti-TfR, anti-ferritin (rb) (Supplemental Table 2). Negative controls were kept overnight without primary antibodies in blocking solution. The next day, after washing with PBS-T, sections were incubated with the respective biotinylated secondary antibodies for 2 h at RT using goat anti-mouse and goat anti-rabbit IgG antibodies diluted in PBS-T with 2% human serum (HS). After washing again in PBS-T, sections were incubated with ExtrAvidin peroxidase (Sigma, E2886) at a dilution of 1:200 in PBS-T for 1 h at RT. To visualize the antibody signals in brown/black, DAB H₂O₂ (Sigma, D-4418) or Vector substrate (Vector, Cat# SK-4700) was used. In single-stained preparations selected sections were counterstained with Mayer's hemalum for 30 s.

For double staining, incubation with the second primary antibodies was performed sequentially overnight at 4°C. Antibodies directed against Factor VIII, AT8, anti-β-amyloid, MBP, MOG and anti-ferritin (rb) were employed. Biotinylated secondary antibodies were applied and binding sites were visualized with Vector substrate or DAB H₂O₂. Selective sections were counterstained. Sections were dehydrated in ascending alcohols, placed in xylene and cover slipped with Histokitt (Roth).

Immunofluorescence (IF)

Sections were deparaffinized, antigen retrieval was performed by microwave treatment and sections encircled with the Dako-Pen. For staining of neurofilament proteins we used fresh frozen sections that were equilibrated to RT, washed in PBS and fixed with 100% EtOH for 20 s. After washing in PBS-T, all sections were blocked with NGS diluted 1:20 in PBS-T and subsequently incubated overnight at 4°C with the primary antibodies: anti-ferritin (rb), anti-ferritin (gt), anti-TfR, AT8, anti-GFAP, anti-Iba1, anti-MHCII, anti-NeuN, anti-MAP2, anti-neurofilament 68 kDa (NFL), anti-NFM, anti-NFH and anti-β-amyloid in PBS-T with 1% NGS. On day two, after washing in PBS-T, incubation with the appropriate secondary antibodies: goat anti-rabbit-AF568, goat anti-mouse-AF488, goat anti-guinea pig-AF488 diluted 1:200 in PBS-T with 2% HS for 2 h at RT was performed. For triple staining, the sections were first incubated with anti-TfR and AT8 with the secondary antibodies goat anti-rabbit-AF568 and goat anti-mouse-AF647. This was followed by incubation with anti-β-amyloid, biotin-goat anti-mouse IgG2a and FITC-Avidin. After several washing steps nuclear staining with DAPI for 12 min followed. Finally, auto-fluorescence was blocked with True Black (Biotium, Cat#23007) for 15 s. All sections were cover slipped with Dako Fluorescence Mounting Medium (Dako, Cat#C0563).

Fluorescence in situ hybridization (FISH)

FISH was performed with the Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostic (ACD), Berlin Germany) using hippocampal brain slices of one case with NFT stage II and one with NFT stage IV from the brain bank of the Anatomy Institute Leipzig, which were fixed in 4% buffered formaldehyde for 24 h only and were paraffin embedded. Here, we combined the probe of the PTGS2 gene/TFRC gene with RBFOX3 and performed immunofluorescence staining with AT8.

Sections were first baked at 60°C for 1 h, followed by deparaffinization and antigen retrieval. Following the kit instructions, a sample was prepared with 1:50 diluted TFRC (ACD, Cat# 437351)/PTGS2 (ACD, Cat# 406801) and RBFOX3 (ACD, Cat# 415591-C3) and the sections were incubated for 2 h at 40°C in a manual HybEZ™ II assay hybridization system (ACD). After 3 amplification steps, the combination with the corresponding opals at dilution 1:750 followed. PTGS2/TFRC was coupled with Opal 570 (ACD, Cat# OP-001003) and RBFOX3 with Opal 690 (ACD, Cat# OP-001006). Between each step the slices were washed with wash buffer. This was followed by immunofluorescence staining in which AT8 was coupled with goat anti-mouse-AF488 as described above.

Image acquisition and analysis

Images of the IHC stainings were taken with the Olympus photomicroscope using OM Capture software version 3.0. Images of IF stainings were taken with Olympus BX 40 epifluorescence and Cell Sens standard 1.18 software. LSM 700 confocal laser scanning microscopy was used to acquire FISH images.

For analysis whole slide brightfield images of IHC staining were acquired using a digital slide scanner (Pannoramic Scan II, 3D HISTECH Ltd, Budapest, Hungary) at 20x magnification. Data sets were examined and manually cropped to relevant image regions using CaseViewer (Version 2.3, 3D HISTECH Ldt., Budapest, Hungary). Resulting images were imported into QuPath (Version 0.4.3) ³⁵ and analyzed with the StarDist plugin (Version 0.4). ³⁶ Tissue detection of anti-ferritin (rb) and anti-TfR stained images was performed by applying a simple thresholder at full resolution (2500 µm² minimum hole area).

Within the tissue area of the ferritin-stained images a pixel classifier trained on ferritin-positive structures was applied. Classification results underwent manual inspection and regions with insufficient segmentation quality due to staining artifacts, folded or overlapping tissue and out-of-focus tissue areas were omitted from further analysis. Finally, the percentage of ferritin-positive area within the remaining tissue area was calculated.

Within the tissue area of the anti-TfR stained images a two-step process for the detection of dot-like TfR-positive structures was employed to exclude false positive findings from further analysis. In a first run, endothelia stained by the anti-FVIII antibody in brown were detected using a simple thresholder for brown intensities (DAB channel in QuPath) and performing a cell detection with StarDist for other structures (mainly endothelium) with modified parameters (threshold = 0.0005; pixel size = 2.0). Resulting detection areas were merged and dilated by 1 µm. Subsequently, TfR-positive structures were detected with StarDist and another set of parameters (preprocessing with a 2pixel wide median filter; threshold = 0.6; pixel size = 0.085). Due to the variability of staining quality (TfR-signal color ranged from blue/purple to brown) and different tissue characteristics (light refraction, visual “granularity”) depending on the fixation time an appropriate detection correction had to be implemented. Objects were excluded based on the following criteria: (1) area size > 7 µm², (2) blue intensity median < 0.165, (3) blue intensity deviation > brown intensity deviation and (4) blue intensity max < brown intensity max. Finally, the large structure detection areas from the previous step were subtracted from the remaining signal detections and the percentage of final TfR-positive area within the tissue area was calculated.

Using CaseViewer, the number of cells stained by anti-TfR and the TfR-positive accumulations was counted manually and respective densities were calculated based on the relevant tissue area.

Whole slide fluorescence images were acquired using a digital slide scanner (Zeiss Axioscan 7, Carl Zeiss Microscopy GmbH, Jena, Germany) at 20x magnification (red channel: anti-ferritin or anti-TfR, respectively; green channel: AT8; blue channel: DAPI). Data sets were examined and center positions of AT8-positive DD were manually marked using the microscope software ZEN (Version 3.7, Carl Zeiss Microscopy GmbH, Jena, Germany).

Marker coordinates were imported into Mathematica (Version 12.2, Wolfram Research Inc., Champaign, IL, USA) and 401 × 401 pixel (about 140 × 140 µm) wide image sections centered around the marker coordinates were exported using CZIcmd. ³⁷ These images were imported into Mathematica, split into the corresponding color channels and submitted to segmentation.

AT8-positive DD (green channel), TfR-positive structures (red channel) and nuclei (blue channel) were detected by local adaptive segmentation (25-pixel window width).

Local DD boundaries were defined by multiplication and subsequent thresholding (Otsu's method) ³⁸ of the detected DD with a 75-pixel wide Gaussian filter. Within these boundaries the number of DD with and without TfR-positive signals was counted.

Single nucleus RNA sequencing (snRNAseq) analyses

For the snRNAseq analyses, we chose a total of three publicly available datasets that applied snRNAseq with the 10x Genomics platform to healthy and AD-affected human postmortem brain samples.

Respective to our own analyses using either FISH or IHC/IF, we chose to include data from the entorhinal and the prefrontal cortex. We additionally included a dataset of the prefrontal cortex that consists of aged-dependent data. The raw datasets were downloaded from the National Center for Biotechnology Information Gene Expression Omnibus (data accessible at NCBI GEO database, accession GSE138852, ³⁹ GSE174367, ⁴⁰ GSE129308, ⁴¹ for the age-dependent dataset accession GSE168408 ⁴² ) and imported into the Python runtime using the Scanpy package version 0.1.0. We applied preprocessing to all datasets and manually assigned cell-type annotations to UMAP partitions using canonical marker genes. We classified oligodendrocytes with MOBP and MOG, astrocytes with AQP4, GFAP, and SLC1A2, oligodendrocyte precursor cells with VCAN, PDGFRA, and OLIG1, microglia with APBB1IP and ITGAM, inhibitory neurons with SYT1 and GAD1, excitatory neurons with SYT1 and SLC17A7, pericytes with PDGFRB, epithelial cells with HTR2C, endothelial cells with FLT1 and PECAM1, mesenchymal cells with COL1A1, lymphocytes with CD96 and monocytes/macrophages with CD163. We chose this method instead of taking over the authors assignments to ensure a better similarity of cell-type specific groups across the datasets. All plots were created using either seaborn 0.13.1 and Matplotlib 3.8.2 or Bokeh 2.4.3.

Age dependent data

For the Herring et al. ⁴² dataset, we mapped the normalized TFRC expression values according to their age for each cell type on a line plot and measured the linear correlation of the TFRC expression over time by calculating Pearson's correlation coefficients with SciPy 1.12.0.

Gene set enrichment analysis (GSEA)

Differential gene expression analyses were carried out with Scanpy's rank_genes_groups function using the Wilcoxon test between either conditional groups (AD versus control) or groups that were created based on TFRC gene expression (TFRC Z-score < 0 versus TFRC Z-score > 1). The genes were considered to be differentially expressed if they had an absolute log fold change greater than 0.25 and a p value less than 0.05.

GSEApy 1.1.1 was used to perform gene set and gene set enrichment analysis. For gene set analysis, we carried out the hypergeometric enrichment test using GSEApy's enrichr function and the GO Biological Process 2018 dataset. For gene set enrichment analysis, we first filtered for genes expressed in at least 30 cells and then used GSEApy's prerank function and the KEGG_2021_HUMAN gene set to determine relevant enriched KEGG pathways (Supplemental Figure 1).

Gene set variation analysis (GSVA)

For evaluating ferroptosis driver/suppressor/marker enrichment scores in individual pseudobulk groups, we downloaded each dataset from the FerrDB V2 website ³⁰ and then filtered the candidate genes by only including genes of a certain confidence level (marked as “validated”) and genes that were tested in humans. We were left with 207 ferroptosis driver, 203 ferroptosis suppressor and 3 ferroptosis marker genes and removed TFRC as this was our gene of interest.

We then subdivided excitatory and inhibitory neurons in each dataset and performed pseudobulking (at least 10 cells, at least 100 counts) of these clusters for individual samples and condition groups (depending on TFRC expression). For evaluating the possibility of TFRC being a specific marker gene for any of the examined ferroptosis gene sets, we used different TFRC expression groups: TFRC baseline expression, TFRC Z-Score > 1, TFRC Z-Score > 2, TFRC Z-Score > 4. These bulk groups were then mapped with the ferroptosis gene sets and GSVA was performed for each gene set, condition group and individual (respectively, every bulk group). The calculated scores were imported to GraphPad Prism and compared according to the relevant comparing algorithm (TFRC high versus low expression, AD/NFT versus control) (Supplemental Figure 1).

Statistical analyses

Statistical analysis of parameters from image processing was performed using Mathematica. Descriptive statistics were calculated and bar charts, boxplots as well as scatter plots were generated. Data were tested for normal distribution using Shapiro-Wilk test and Kolmogorov-Smirnov test, respectively. Group comparisons were tested using the Mann-Whitney-U test. Correlation analysis was performed using Spearman rank correlation. Level of significance for all tests was set at p < 0.05.

For the snRNAseq analyses we either performed two-way ANOVA between multiple condition groups or Mann-Whitney U test to compare between two condition groups with a significance level of p < 0.05.

TfR expression on neuronal structures displays typical features of DD

To detect TfR as a potential molecular marker of ferroptosis in the AD brain, we first tested monoclonal mouse antibodies of different clones against TfR1/CD71, i.e., [3F3 FMA], [3B8 2A1], and [H68.4] as well as a polyclonal rabbit antibody, anti-TfR, which all should bind TfR1. In our hands, the mouse monoclonal antibodies showed no signals on fresh frozen, fixed cryostat and paraffin-embedded human brain tissues except for faint endothelial staining, as has been reported. ⁴³ The rabbit polyclonal anti-TfR showed membrane-bound signals not only at the endothelium, but also revealed beaded neurites in grey and white matter in IF (Figure 1) and IHC (Figure 2). Hepatocytes which are known to express TfR ³⁴ served as positive controls (Figure 2(h)).

OPEN IN VIEWER

Figure 1.

Fluorescence staining with anti-TfR. Fresh frozen sections of human prefrontal cortex stained with anti-TfR in red and cell nuclei with DAPI in blue. The endothelium (*) and a beaded neuritic process (arrow) are stained (a). Beaded neuritic processes surrounding a cell nucleus can be seen (b, arrowhead). Scale bar = 20 μm.

OPEN IN VIEWER

Figure 2.

Comparison of AT8 and TfR immunostaining. (a, b) Paffin sections of human hippocampus show AT8-positive DD. (c-g) Paraffin-sections of human prefrontal cortex stained with anti-TfR in black. Beaded TfR-positive neurites resemble DD in (c, d). Cell bodies connected to beaded processes are visible in (e, f). Beaded neuritic processes of different lengths are shown in (g). (h) Paraffin-section of human liver stained with anti-TfR serves as a positive control. TfR signals can be seen in hepatocytes and on their membrane. (h*) omission control. Scale bar = 20 μm.

We detected a striking morphological similarity between droplet degeneration spheres positive for AT8 (Figure 2(a), (b)) and TfR staining patterns (TfR accumulations) occasionally observed in PFC and hippocampus (Figure 2(c), (d)). In addition, TfR-positive cells with stained soma and fragmented processes suggestive of degeneration (Figure 2(e)-(g)) were present. To identify these TfR-positive cell types, we performed IHC and IF double staining with antibodies against neuronal markers such as NeuN, MAP2, Neurofilament 68 kDa (NFL), NFM and NFH, astrocyte marker GFAP, oligodendrocyte markers such as MOG and MBP and microglia markers such as Iba1 and MHCII. Since the antibodies against different neurofilaments and MAP2 were not suitable for paraffin-embedded tissues, fresh-frozen material was used.

We found co-localization of TfR-positive neuronal spheres with neurofilament markers NFM (Figure 3(a), (b)) and NFL (Figure 3(c)). TfR signals surrounded NFL-stained neurofilaments in cross section (Figure 3(c)), suggesting that the signals are membrane-bound and located on the surface of neurites. This location would be expected from TfR as a transmembrane glycoprotein.^44,45 However, it must be noted that the resolution of this staining makes it difficult to distinguish between internal labeling and the cell surface. TfR signals were not co-localized with GFAP, Iba1, MAP2, MBP, MOG, and MHC II and no direct overlap of TfR-positive cells with NeuN was seen (Supplemental Figure 2). We conclude that TfR is not only present in endothelial cells ⁴⁶ but also in neurites and a population of degenerating neurons as observed in AT8-labeled sections.

OPEN IN VIEWER

Figure 3.

Comparison of TfR and neurofilament immunostaining. Fresh frozen human prefrontal cortex stained with anti-TfR in red, anti-NFL or NFM proteins in green, and nuclei with DAPI in blue. Co-localization of TfR-positive structures with NFM (a, b) and NFL (c) can be seen. In (c) cross sectioned neurofilaments (positive for NFL) co-localize with TfR-positive beads (arrows).

mRNA of signature genes of ferroptosis are detected in pretangles and p-tau negative neurons, but less in DD

To investigate whether typical marker genes of ferroptosis are detectable in p-tau + structures, we performed FISH using probes of TFRC and PTGS2 as ferroptosis marker genes and RBFOX3 to identify neurons. FISH was followed by IF staining using AT8 antibody to visualize AD pathologies. Here we examined one case with NFT stage IV and one with NFT stage II and analyzed the FISH quantitatively. Signals from both probes clearly occurred in the endothelium, consistent with analyses of the human protein atlas. ⁴⁶ Outside the endothelium, PTGS2 and TFRC were present in cells including neurons in or around cell nuclei. More signals of the TFRC probe were detectable compared to PTGS2 in both cases. We could not detect a clear quantitative difference of TFRC and PTGS2 between both NFT stages but were able to detect PTGS2 and TFRC in pretangles, in p-tau negative neurons located in the vicinity of p-tau + structures and less in DD (Figure 4).

OPEN IN VIEWER

Figure 4.

FISH with probes for ferroptosis related genes. Paraffin sections of human hippocampus were hybridized with PTGS2 (a-c)/TFRC (d-e) in red and RBFOX3 in white, stained with AT8 in green and DAPI in blue for nuclear staining. Signals of the mRNA of PTGS2 (circle in b, c) and TFRC (circles in d) can be seen in pretangles, but less in DD (a, e). There is also mRNA of the PTGS2 and TFRC gene in neurons around p-tau + structures (arrows in a, d, e). Scale bar = 20 μm.

GSEA reveals downregulation of pathways involved in fatty acid metabolism in AD

To explore a tendency to ferroptosis in individuals affected and not affected by AD, we compared the gene expression of common ferroptosis markers, drivers and suppressors by visualizing the cumulative gene expression scores for each individual cell cluster. This comparison revealed only minor differences in the mean expression values of the genes of interest, with slightly higher mean expression of all genes in the low NFT stages. It is noteworthy that specific ferroptosis markers like GPX4, PTGS2 and TFRC showed an enrichment in both excitatory and inhibitory neurons, but gene expression did not differ between the AD-affected and AD-unaffected individuals (Figure 5(a)).

OPEN IN VIEWER

Figure 5.

Differential gene expression and gene set enrichment analysis for TFRC expressing neurons. To explore a tendency to ferroptosis in individuals affected and not affected by AD, we compared the gene expression of common ferroptosis markers, drivers and suppressors by visualizing the cumulative gene expression scores for each individual cell cluster (a). Differential gene expression analysis was performed with two snRNAseq datasets between TFRC expressing neurons (TFRC expression higher than 1 standard deviation of mean TFRC expression) and TFRC non-expressing neurons (TFRC expression lower than mean expression) in the AD and control group (b, c, f, g). Venn diagrams of the DEGs and ferroptosis related genes showed partial overlaps, indicating the presence of ferroptosis related genes in the DEG groups (e, h). Further GSEA of the DEG list showed certain ferroptosis associated pathways to be either up- or downregulated in the TFRC expressing cells in both the AD and the control group. The major pathways identified were terpenoid backbone synthesis, O-glycan biosynthesis and butyrate metabolism. Specific GSEA of the ferroptosis pathway in the DEG list of TFRC expressing neurons of the high NFT stages showed overrepresentation of ferroptosis related genes (d). It is to note that this GSEA did not reach significancy (p = 0.2857). (ODC: oligodendrocytes; MG: microglia; PER.END: pericytes, endothelial cells; OPC: oligodendrocytes precursor cells; ASC: astrocytes; EX: excitatory neurons; INH: inhibitory neurons).

We then investigated differentially expressed genes (DEGs) in TFRC expressing neurons (TFRC expression higher than 1 standard deviation of mean TFRC expression) and TFRC non-expressing neurons (TFRC expression lower than mean expression). Differential gene expression analysis revealed a total of 329 to 2223 genes to be differentially expressed (p-value < 0.05), depending on the conditional group and used dataset.

By performing TFRC DEG analysis, we found many genes of both the AD and control groups to be ferroptosis related (Figure 5(e), (h)). According to the data of Grubman et al. ³⁹ there were less genes overlapping, because of the lower number of the AD DEGs (Figure 5(h)).

We did not find ferroptosis to be a significantly regulated pathway in either group or dataset. Nevertheless, for the high NFT stages of the prefrontal cortex neurons, we identified certain genes of the KEGG pathways that are part of the fatty acid metabolism (butanoate metabolism ↓, mucin type O-glycan biosynthesis ↑, terpenoid backbone synthesis ↓) and vesicular transport (↓). Interestingly, especially genes of pathways that are relevant for glycolysis were highly enriched (pentose phosphate pathway, fructose and mannose metabolism) (Figure 5(b)).

For the low NFT stages of the prefrontal cortex neurons, we only found negatively enriched genes of pathways, that are mainly part of anabolic cellular processes (histidine, arginine, proline metabolism, oxidative phosphorylation) (Figure 5(c)).

We additionally evaluated the ferroptosis pathway to point out a general overrepresentation of ferroptosis related genes. The enrichment plot shows a clear enrichment of ferroptosis in the present DEG list of the high NFT stages with a distinct peak in the beginning, indicating besides not being a significant pathway that TFRC expression might be accompanied by the expression of ferroptosis-related genes (Figure 5(d)).

We similarly carried out these analyses in the Grubman et al. ³⁹ dataset (entorhinal cortex cells). GSEA results showed a similar picture in comparison to the pathways described above. In the AD group, pathways for vesicular transport were downregulated and pathways of fatty acid metabolism (mannose type O-glycan biosynthesis, terpenoid backbone synthesis, steroid biosynthesis) were downregulated as well (Figure 5(f)).

In the control group, we found comparable results with the exception of a high enrichment of O-glycan biosynthesis (Figure 5(g)).

GSVA indicated TFRC as an early marker of ferroptosis

The second part of our snRNAseq analyses focused on evaluating GSVA scores for ferroptosis markers, drivers and suppressors in different conditions, namely comparing TFRC high expressing neurons versus low expressing neurons and AD versus control/neurons affected by neurofibrillary tangle pathology versus control neurons.

In the dataset of Otero-Garcia et al. ⁴¹ consisting of neurons affected and unaffected by neurofibrillary tangle pathology, we found a significant difference in the GSVA scores of both ferroptosis markers (p = 0.0005) and suppressors (p = 0.007) between the AT8 and control group (Figure 6(a), (b)). Also, there was a significant difference in the ferroptosis marker GSVA scores between AT8 and MAP2 cells (p = 0.0012; Figure 6(a)). While there were no significant differences between the high and low groups when setting the cut-off value for TFRC high expression at Z-score > 2 (more than two standard deviations), we found a significant difference for ferroptosis drivers between the high and low groups for a Z-score > 4 (AT8: p = 0.0093; MAP2: p = 0.0217; Figure 6(c)). This indicates that only cells with very high TFRC expression are currently undergoing a shift towards ferroptosis, as ferroptosis driver genes are then enriched in these groups.

OPEN IN VIEWER

Figure 6.

Gene set variation analysis for ferroptosis gene sets. We evaluated GSVA scores for ferroptosis markers, drivers and suppressors in different conditions, namely comparing TFRC high expressing neurons versus low expressing neurons, and AD versus control / neurons affected by neurofibrillary tangle pathology versus control neurons. In the Otero-Garcia et al. dataset significant differences in the GSVA scores of both ferroptosis markers (p = 0.0005) and suppressors (p = 0.007) between the AT8 and control group (a, b) and in the ferroptosis marker GSVA scores between AT8 and MAP2 cells (p = 0.0012) (a) were present. For ferroptosis drivers there were no significant differences between the high and low groups. When setting the cutoff value for TFRC high expression at Z-score > 4 (expression four standard deviations higher than mean expression), we found a significant difference between the high and low groups (AT8: p = 0.0093; MAP2: p = 0.0217) (c). The other two datasets showed similar results that did not reach significancy (d, e), except for the control group of the Grubman et al. dataset that showed a significant difference (p = 0.033) in the GSVA scores of ferroptosis drivers between the high and low group (e).

The above observations were relatively similar in the other two datasets. Although only reaching significance (p = 0.033) for the control group in the Grubman et al. ³⁹ dataset, in the TFRC high expression groups, we were able to observe a difference in the GSVA scores of ferroptosis drivers between the high and low groups, which goes in line with the above findings (Figure 6(d), (e)). This indicates that TFRC might function as an early marker of ferroptosis that is that cells might co-express ferroptosis driver genes but have not yet reached the ongoing process of ferroptosis.

Direct co-localization between TfR- and p-tau + structures is a rare event

To investigate whether p-tau + structures express ferroptosis markers, we performed double staining with anti-TfR/anti-ferritin (rb) and AT8. We could not detect TfR (Figure 7(a)) or ferritin at pretangles. Direct co-localizations of TfR-positive structures and AT8-positive DD were rarely noted (Figure 7(b)). In a few cases, TfR signals were detected in the DD spheres of debris (Figure 8(a)). More often, an adjacency of the TfR signals to the DDs was observed (Figure 7(c)-(e); Figure 8(b)-(e)). In most cases, the DDs remained negative for TfR both within and in the vicinity (Figure 8(f)). To describe this quantitatively, we evaluated 44 DDs from four cases (case numbers 22, 24, 26 and 27), of which 17 (38.6%) were positive for TfR. A DD was considered positive for TfR, if TfR signals were detectable directly within the DD spheres of debris or directly outside in a defined distance from the DD center as described above. TfR-positive structures resembling DD, did not co-localize with p-tau in most, but not all cases. Neurons stained with the anti-TfR were also not p-tau + . In summary, TfR-positive structures often exist in the vicinity of AT8-positive DD.

OPEN IN VIEWER

Figure 7.

Immunohistochemical double staining with anti-TfR and AT8. Paraffin sections of human prefrontal cortex double-stained with anti-TfR (black) and AT8 (brown). There is no TfR signal in pretangles (a). A rare direct overlap of TfR-positive structures and p-tau + DD can be seen in (b). More often there is adjacency of DD and TfR-positive structures (c-e). Scale bar = 20 μm.

OPEN IN VIEWER

Figure 8.

Fluorescence staining with anti-TfR and AT8. Paraffin sections of human prefrontal cortex stained with anti-TfR in red, AT8 in green and cell nuclei with DAPI in blue. Beaded neuritic processes in DD can be seen (a), but more often there is adjacency of DD and TfR-positive structures (b-e). In most cases, DD spheres occur without TfR-positive structures (f). Scale bar = 10 μm.

Ferritin-positive microglia were usually present at sites of DD (Figure 9(a)-(c)), as described previously. ⁴ However, occasionally they were undetectable (Figure 9(d)).

OPEN IN VIEWER

Figure 9.

Fluorescence staining with anti-ferritin and AT8. Paraffin sections of human prefrontal cortex stained with anti-ferritin in red and AT8 in green and cell nuclei with DAPI in blue. Ferritin-positive microglia in or around DD spheres can be seen (a-c). DD spheres occur rarely without ferritin-positive microglia (d). Scale bar = 10 μm.

IHC double staining with anti-TfR and anti-ß-amyloid antibodies showed that most amyloid deposits were negative for TfR signals. In some cases, beaded processes were seen in amyloid deposits (Figure 10(a)). To identify neuritic plaques, we performed IF triple staining with anti-ß-amyloid, AT8 and anti-TfR. Again, neuritic plaques remained negative for TfR in most instances. In Figure 10(b) a rare case with TfR signals in a neuritic plaque is shown.

OPEN IN VIEWER

Figure 10.

Spatial correlation between Aß and TfR. (a) Paraffin sections of human prefrontal cortex stained with anti-Aß (black) and anti-TfR (brown). In some cases, beaded neuritic processes are seen in Aß plaques (circle). Scale bar = 20 μm (b) Paraffin sections of human prefrontal cortex stained with anti-TfR in red, AT8 in green, anti-Aβ in light blue and cell nuclei with DAPI in blue showing a neuritic plaque. Scale bar = 20 μm.

We investigated the spatial relationship between TfR and ferritin signals using IHC double staining with anti-TfR and anti-ferritin (rb). We found that these two antibodies stain different structures and a spatial proximity was not observed. Only rarely processes of ferritin-positive microglia contacted TfR-positive structures (Figure 11(a)-(c)). Furthermore, we investigated whether the expression of TfR and ferritin correlate in their levels and found only a modest correlation (r = 0.44, p = 0.005, Figure 11(d)).

OPEN IN VIEWER

Figure 11.

Spatial and quantitative relationship between ferritin and TfR. (a-c) Paraffin sections of human prefrontal cortex stained with anti-TfR (brown) and anti-ferritin (black) showing the same image section in three different focal planes. Ferritin-positive microglia are in contact with TfR-positive structures. Scale bar = 20 μm (d) Correlation analysis of the amount of ferritin expression (Ferritin area ratio in %) and TfR expression (TfR area ratio in %) shows a modest correlation (r = 0.44, p = 0.005).

TfR and ferritin expression increases significantly with NFT stages

During our microscopic examinations we observed an increased TfR and ferritin staining pattern in high NFT stages. Consequently, we compared TfR and ferritin expression quantitatively in one group with NFT stage ≤ I (n = 20) and another with NFT stage ≥ IV (n = 20). To compare TfR expression levels, signals stained by anti-TfR were detected as a total TfR-positive area in relation to the total area evaluated (TfR area ratio in %). Because anti-TfR labels endothelium and neuronal structures, double staining was performed with anti-TfR and anti-FVIII antibody, an endothelial marker. Finally, signals of the endothelium labeled by anti-TfR and anti-FVIII were subtracted from the total anti-TfR signals in order to evaluate only TfR expression on neuronal structures.

A significantly (p = 0.007) higher TfR area ratio was found in the NFT stage ≥ IV-group compared to the NFT stage ≤ I-group (mean = 0.071 versus mean = 0.030; Figure 12(a), (b)). Since the anti-TfR antibody labels various structures such as beaded processes, as well as entire neurons (Figure 2(e), (f)), we counted these neurons in all cases (cell density). No clear difference in the cell density was found between the two groups (p = 0.797, Figure 12c, d). TfR-positive structures resembling DD (Figure 2(c), (d)) were also counted in all cases (accumulation density). They were significantly more frequently detected in high NFT stages (p < 0.001, Figure 12(e), (f)), which had a higher mean age than the low NFT stages (Figure 12(g)). To rule out an age dependency of the TfR area ratio, we performed a correlation study, resulting only in a modest (according to Akoglu, 2018) ⁴⁷ correlation of r = 0.35 (p = 0.026, Figure 12(h)).

OPEN IN VIEWER

Figure 12.

Measurements of TfR expression. The area of TfR-positive signals stained by the anti-TfR antibody is set in relation to the total area evaluated (TfR area ratio in %). In (a), TfR area ratios of all selected cases are shown in the histogram. In (b), comparison of TfR area ratio between one group with NFT stage ≤ I (light grey) and one with NFT stage ≥ IV (dark grey) is shown. There is a significant (p = 0.007) higher TfR area ratio in NFT stage ≥ IV, shown in a boxplot. In (c), a histogram of all cases with the total number of TfR-positive neurons (cell density) is shown. The boxplot in (d) shows no significant (p = 0797) difference of cell density in both groups. In (e), a histogram of all cases with the total number of TfR-positive accumulations (accumulation density) is shown. In (f), a significantly (p < 0.001) higher accumulation density in NFT stage ≥ IV is shown in a boxplot. In (g), the age difference between the two groups is shown in a boxplot. In (h), the correlation analysis between the age of the cases and the TfR area ratio is shown, with a modest correlation (r = 0.35, p = 0.026).

For comparison of ferritin expression in the two groups, we performed IHC single staining with a polyclonal anti-ferritin antibody which we have used in previous studies on microglia in AD.^4,48,49 We detected the area of all ferritin-positive microglia relative to the total area of the evaluated section (ferritin area ratio in %). A significant higher (p = 0.001) ferritin expression was detected in the NFT stage ≥ IV-group compared to the NFT stage ≤ I-group (mean = 0.1033 versus mean = 0.22365, Figure 13(a), (b)). Again, the difference in age between the two groups had to be considered; therefore, we investigated whether there was an age dependency of ferritin expression with a correlation study. Here we could also find a modest correlation (r = 0.50, p = 0.001, Figure 13(c), (d)).

OPEN IN VIEWER

Figure 13.

Measurements of ferritin expression. The area of microglia stained by the anti-ferritin antibody is set in relation to the total area evaluated (ferritin area ratio in %). In (a), ferritin area ratios of all selected cases are shown in the histogram. There is a significant (p = 0.001) higher ferritin expression in NFT stage ≥ IV than in NFT stage ≤ I, which can be seen in the boxplot in (b). In (c), the age difference between the two groups is shown in a boxplot. In (d), the correlation analysis between the age of the cases and the ferritin area ratio is shown, with a modest correlation (r = 0.50, p = 0.001).

We also investigated the presence of TfR and ferritin signals in the hippocampal and prefrontal brain slices of newborns. Similar to elderly brains, stained structures were detected by both the anti-ferritin and anti-TfR antibodies. The beaded processes and the completely TfR-positive neurons (Figure 2(e), (f)) appeared to be more numerous than in adult brains. TfR-accumulations resembling DD (Figure 2(c), (d)) occurred only rarely. Altogether, TfR area ratio was higher than in adult brains. Ferritin-positive microglia were detected in reduced numbers (Supplemental Tables 3 and 5). Interestingly, by visualizing TFRC expression in a snRNAseq dataset, that included age-depended tissue samples of the prefrontal cortex ranging from age 0 to 40, we were able to depict a comparable picture of TFRC expression. TFRC expression was highest at a young age and remained relatively stable after the age of two. We found a positive correlation of TFRC expression for both inhibitory (r = 0.69, p = 0.000) and excitatory (r = 0.65, p = 0.001) neurons over the aging process (Supplemental Figure 3).