Iron-loading is a prominent feature of activated microglia in Alzheimer’s disease patients

Brain iron accumulation has been found to accelerate disease progression in Amyloid β-positive Alzheimer patients, though the mechanism is still unknown. Microglia have been identified as key-players in the disease pathogenesis, and are highly reactive cells responding to aberrations such as increased iron levels. Therefore, using histological methods, multispectral immunofluorescence and an automated in-house developed microglia segmentation and analysis pipeline, we studied the occurrence of iron-accumulating microglia and the effect on its activation state in human Alzheimer brains. We identified a subset of microglia with increased expression of the iron storage protein ferritin light chain (FTL), together with increased Iba1 expression, decreased TMEM119 and P2RY12 expression. This activated microglia subset represented iron-accumulating microglia and appeared morphologically dystrophic. Multispectral immunofluorescence allowed for spatial analysis of FTL+Iba1+-microglia, which were found to be the predominant Aβ-plaque infiltrating microglia. Finally, an increase of FTL+Iba1+-microglia was seen in patients with high Amyloid-β load and Tau load. These findings suggest iron to be taken up by microglia and to influence the functional phenotype of these cells, especially in conjunction with Aβ.


1
Alzheimer's disease is the most common cause of dementia, and is defined by the presence of amyloid-β (Aβ) 2 plaques and tau tangles. In addition, the brain's resident innate immune cells, microglia, have been found to be 3 at the centre-stage of the disease, as most identified risk genes are predominantly or even exclusively expressed 4 in microglia [1,2].

5
Not only can microglia modulate Alzheimer's disease, but many transcriptomic studies showed microglia to 6 undergo the most pronounced changes in response to pathology. In mice, a subset of responding microglia has 7 been found to lose their homeostatic molecular signature and transition into a so-called 'disease associated 8 microglial state' (DAM) [3]. In humans, a comparable yet disparate state coined the human Alzheimer microglia 9 (HAM) has been identified [4]. Upregulated genes in these subsets do not only indicate loss of homeostatic 10 function and increased pro-inflammatory activation, but also dysregulated iron-metabolism, manifested via 11 upregulation of the FTL-gene and downregulation of FTH1 and SLC2A11 [4,5].  Alzheimer's disease [6,7]. Though increased iron concentration likely plays a role, the exact link between the 15 two has not yet been established.

16
Iron accumulation, irrespective of microglial activation, on the other hand, has been reported in disease-affected 17 areas in Alzheimer's disease, using both in-vivo and post-mortem human MRI [8]. Several MRI and histology 18 studies found high correlations between iron accumulation and cortical Aβ and tau spreading [9][10][11]. Clinically,

19
increased iron concentrations were shown to accelerate cognitive decline in Aβ-positive Alzheimer patients, 20 indicative of a disease-modifying role for iron accumulation [12,13]. Again, how iron accelerates cognitive 21 deterioration is poorly understood.

22
Therefore, in this study we aimed to research the possible link between iron accumulation and functionally 23 activated microglia, and finally, its relation with Aβ-plaques. We performed a comprehensive investigation of 24 iron-accumulating microglia, and first identified that the iron-storage protein FTL, specifically reflected 25 increased iron accumulation in microglia. Secondly, by using multispectral immunofluorescence and an in-26 house automated cell-analysis pipeline, we found FTL + microglia to show significant activation, shown via both 27 downregulation of homeostatic markers TMEM119 and P2RY12 and dystrophic morphology, and to 28 4 predominantly infiltrate Aβ-plaques. This provides evidence for iron dysregulation as a prominent feature of 1 activated microglia in Alzheimer's disease in humans.

12
Formalin fixed paraffin embedded (FFPE) tissue was serially cut into ten 5-μm-thick and four 10-um-thick 13 sections. Consecutive 10-μm-thick sections were used for histological detection of iron using an enhanced Perl's 14 stain and IHC detection of Ferritin Light Chain (FTL). 5-μm-thick sections were used for staining of the 15 microglia multispectral immunofluorescence (mic-mIF) panel (Supplementary Table 2) to verify expression of 16 FTL in microglia/macrophages (Iba1), look at the activation state of these cells (P2RY12/TMEM119) and study 17 the interaction with Amyloid β-plaques (Aβ). Finally, of three subjects, 20-μm-thick sections were obtained for 18 3D confocal imaging.
Step-by-step histological and IHC optimization protocols, together with the imaging 19 parameters, are reported in the Supplementary Methods. A step-by-step mIF protocol and further analysis of the 20 described histological, IHC and mIF staining will be described in the following sections.

23
One 5-μm-thick section of each subject was stained with the mic mIF panel with the following protocol.

24
Sections were deparaffinized with 3× 5 min xylene, rinsed twice in 100% alcohol and subsequently washed with 25 100% ethanol for 5 min. Endogenous peroxidases were blocked for 20 min in 0.3% H2O2/methanol, after which 26 the slides were rinsed with 70% and 50% alcohol. Heat induced antigen-retrieval was performed by cooking the 27 slides for 10 min in pre-heated citrate (10 mM, pH=6.0) buffer for 10 minutes. After cooking, excess buffer was 28 removed and slides were cooled for 60 min. Non-specific antibody binding sites were blocked with blocking 29 buffer (0.1% BSA/PBS + 0.05% Tween) for 30 min. Firstly, slides were incubated with anti-TMEM119 (1:250,

11
for 5 min, after which they are mounted with 30 uL Prolong diamond (ThermoFisher).

14
MRI data and T2*-w severity scores were obtained from a previous study by Bulk et al. [10], on the same 15 tissue-blocks. In this study, tissue blocks were put in proton-free fluid (Fomblin LC08, Solvay), and scanned at 16 room temperature on a 7T horizontal-bore Bruker MRI system equipped with a 23 mm receiver coil and  Solution platform (Philips, the Netherlands) and imported into ImageJ. RGB images were converted into 8-bit 25 greyscale images. Subsequently, while blinded for diagnosis, for each subject an optimal threshold was set to 26 include DAB-positive intracellular iron depositions, but exclude extracellular background signal. The cortex of 27 the MTG was delineated and the number of positive cells was determined using the ImageJ particle analyser, 28 with a size threshold of 4-100 pixels. Subject AD5 was excluded from this analysis, as iron-accumulating cells 29 could not be distinguished due to high extracellular iron load.  Supplementary Fig. 2a). Hence, accurate segmentation of the whole microglia cell area is of 3 paramount importance for our method. Solutions currently available for microglia cell segmentation 4 (Abdolhoseini et al., 2019 [14], Inform, PerkinElmer) typically fall short of capturing the whole microglia area 5 ( Supplementary Fig. 2b). These are focused on either capturing the skeleton of the cells, without properly 6 identifying the cell boundaries ( Supplementary Fig. 2c), or segmenting the microglia's soma excluding their 7 processes, which in the acquired 2D images are typically detached from the soma ( Supplementary Fig. 2d). As a 8 result, a novel segmentation algorithm for this type of data was developed.

19
(TMEM, PRY12, FTL, Iba) component images was utilized as input and level-set parameters were set to ν=2 20 and μ=3. In both cases, level-sets were initialized with regions obtained using the Otsu thresholding method [19] 21 which is robust to intensity variation between images originating from the white and grey matter. Additionally,

22
somas and nuclei with a total area smaller than 50 and 30 pixels, respectively, were removed.

24
For the extension of the obtained segmentation to the whole cytoplasmic area, the approach previously 25 described for soma was repeated with less strong regularization (v=2, μ=2). The result of this step was a finer 26 segmentation capturing microglia areas that are less bright than the soma. Connected components overlapping 27 with the previously identified somas were regarded as microglia cells, whereas not overlapping components 28 were considered as possible detached processes. At this step, in case a blood vessel was identified in an image, 29 the Li thresholding method [20] was chosen over Otsu for the initialization of the level-sets algorithm, as it is 8 less sensitive to the high intensity pixels representing the vessel. Vessels were defined as components larger 1 than 4000 pixels, after Otsu thresholding of the autofluorescent component image.

3
For correct identification of microglia cells in the proximity of Aβ-plaques, the watershed segmentation was 4 applied specifically to those cells whose cytoplasmic area is shared among multiple microglia somas [21]. Aβ-5 plaque identification was performed employing a semi-supervised approach using Ilastik [22].

7
Finally, branches identified within a 10 pixel radius from the region corresponding to each identified microglia 8 soma were identified as detached processes and assigned to the microglia cell.    The median expression value of each marker for each phenotype was illustrated with a heatmap. The similarities 1 among the identified phenotypes were observed from a t-SNE [26] embedding using the same input as in 2 Phenograph and the default parameters. The t-SNE embedding was coloured according to the cluster of each 3 cell, its cohort or its individual marker expression values [27].

5
To explore the differences between the Alzheimer patients and controls regarding their phenotypes and their 6 spatial relationship with the Aβ-plaques, an interactive, data-driven pipeline described by [28] was utilized.    with a small soma and many thin processes (Fig. 1a) and quantification indicated a significant increase of iron-5 positive cells in Alzheimer patients compared to controls (P = 0.0024; Fig. 1b). Additionally, iron-positive cells 6 appeared to cluster in groups, something that was not observed in control patients (Fig. 1a). All MTG tissue 7 blocks have also previously been scanned using T2*-w MRI, sensitive for paramagnetic substances such as iron.

8
MRI images were scored based on alterations in signal intensity reflecting overall parenchymal iron 9 accumulation and focal iron depositions, and were published by Bulk et al. [10]. An increase of iron-positive 10 microglia appeared to be only present in cases with the highest MRI severity score, indicating a significant 11 increase of iron-positive microglia only to occur in subjects with a pronounced macroscopic iron-phenotype 12 (Fig. 1c). Subsequently we studied the correspondence of iron accumulation with altered expression of the main 13 iron-storage protein ferritin light chain (FTL), as FTL is known to be expressed in microglia and 14 oligodendrocytes, whereas heavy chain ferritin is primarily expressed by neurons in Alzheimer tissue [30]. The

15
Perl's staining and the FTL staining showed a highly similar staining pattern, with focal clusters of cells 16 representing microglia morphology (Fig. 1d). Thus, increased expression of the main iron-storage protein FTL 17 appears to reflect iron accumulation in microglial cells.

Quantitative analysis enables microglia phenotyping 1
To confirm the microglial origin of FTL + cells, study their activation state and potential interaction with Aβ, we 2 designed the microglia multispectral immunofluorescence (mic-mIF) panel that can simultaneously detect 6 3 different markers (Supplementary Table 2). The MTG of 12 Alzheimer patients, both of early-and late onset, 4 and 9 control subjects (Supplementary Table 1) was stained and imaged. After image acquisition and 5 multispectral unmixing of the data, images were exported for automated segmentation, phenotyping and spatial 6 analysis (Fig. 2). In total, 3149 images (110-236 per subject) were obtained. Multispectral unmixing allowed 7 for simultaneous detection of FTL with the nuclear marker DAPI, TMEM119, P2RY12, Iba1 and Aβ at 0.5x0.5 8 μm resolution (Fig. 3a). TMEM119 and P2RY12 are generally considered homeostatic microglia-specific  Alzheimer patients. Images were segmented using a targeted in-house segmentation pipeline allowing 13 segmentation of cells with processes (like microglia) in 2D images ( Fig. 3b; Supplementary Fig. 2). After 14 segmentation, unsupervised clustering using Phenograph assigned single segmented cells to 20 separate clusters.

15
Following manual evaluation of the unsupervised clusters, 6 clusters were excluded based on non-microglial 16 morphology and/or sub-threshold expression of all microglial markers (TMEM119/P2RY12/Iba1). In addition, 17 three times two clusters were merged based on similarity in protein expression levels and their visual 18 appearance ( Supplementary Fig. 3). Exclusion of the non-microglial cells resulted in identification of 69227 19 cells, with no significant differences in the number of microglia per mm 2 between control and Alzheimer 20 patients in either grey matter (GM) or white matter (WM) (Fig. 3c). The remaining 11 clusters (C1-C11) were 21 identified as major microglia phenotype clusters (Fig. 3d). Though the 11 different phenotypes clustered on the 22 t-SNE plot, the low degree of separation suggests a rather continuous spectrum of expression of the microglia 23 markers (Fig. 3e). The control and Alzheimer patients did cluster together, and the marker-based t-SNE plots 24 already revealed more cells with high TMEM119 and P2RY12 expression in controls, but increased FTL 25 expression in Alzheimer patients (Fig. 3f). With regard to anatomical region, only C1 and C2 appeared to be 26 more present in the grey matter (GM), whereas C5 and C6 appeared to be proportionally more present in the 27 white matter (WM) (Fig 3g). Four FTL + clusters (C1-C3, C5) were identified, with differing expression levels 28 and co-expression levels of P2RY12, TMEM119 and Iba1 (Fig. 3d). Cluster C1 (FTL + Iba1 + ) appeared 29 significantly more present in Alzheimer patients (P = 0.0264), while C2 (P2RY12 + TMEM119 + FTL + Iba1 + ) was more present in controls (P = 0.0055; Fig. 3h). FTL + Iba1 + clusters lacking either P2RY12 (C3) or TMEM119 1 (C5) did not differ significantly in prevalence between control and Alzheimer patients. Cluster C4 showed 2 solely Iba1 expression, meaning that this cluster likely also consists of non-resident infiltrating macrophages.

3
Additionally, three P2RY12 + clusters (C6-C8) were identified, with the highest expressing cluster (C8) being 4 more present in controls. The same applied for the TMEM119 + clusters (C9-C11), with C10 and C11 having  image as a 'glyph' (Fig 4a)[29], with the different colours corresponding to the respective cluster of the 14 infiltrating microglia, to analyse which clusters predominantly infiltrated Aβ-plaques. Subsequently, all 15 individual cells represented as cluster-colored dots or the cluster-colored glyphs were plotted back onto the 16 original whole slide image (Fig 4a), to assess differences in cluster composition of microglial Aβ infiltration on 17 a whole-section scale. As expected, quantification showed significantly more identified Aβ-plaques in 18 Alzheimer patients, although some were found in controls as well (P = 0.0002; Fig 4b). Furthermore, a higher 19 percentage of the plaques showed microglia infiltration in Alzheimer patients (P = 0.013; Fig 4c). Looking at 20 the whole slide distribution, Aβ-plaques were found to be more present in the coronal sulcus rather than the 21 gyrus. This also appeared to be associated with the regional microglia phenotype, as can be seen for the 22 predominantly purple (C6-C8) microglia populating the Aβ-plaque deplete regions (Fig. 4d). To quantify the 23 influence of Aβ-plaques on microglia phenotype, we compared all phenotyped microglia (all-mic) with the 24 subset of microglia infiltrating Aβ-plaques (Aβ-mic). Controls showed a slight percental increase of C1 and C5 25 in Aβ-mic compared to all-mic, and less Aβ-plaque infiltration of TMEM119 + clusters C9-C11 (Fig. 4e), though 26 this was based on a limited total number of Aβ-plaques. Alzheimer patients on the other hand, showed a large 27 percental increase of FTL + -clusters C1 and C3 in the Aβ-mic population (Fig. 4E), which was also statistically 28 significant when looking at subject-specific proportional increases (C1: P < 0.0001, C3: P = 0.0004; Figs. 4f,g).

29
While C1 and C3 microglia together make up less than 20% of all-mic, they constitute almost 50% of the Aβ-30 13 mic population (Fig. 4e). P2RY12 + clusters C6-C8, on the other hand, showed a small contribution to Aβ-mic 1 compared to all-mic (Fig. 4e). Finally, not only did C1 and C3 make up the majority of Aβ-mic, but also when 2 examining the proportions of these individual clusters that directly infiltrated Aβ-plaques, they showed much 3 higher proportion of infiltration than all the other clusters (Fig 4h). A visual example of the C1 and C3 microglia 4 infiltrating an Aβ-plaque on the original mic-mIF images can be found in Fig. 4i. All in all, these results suggest 5 Aβ-plaques to be predominantly infiltrated by a specific subset of microglia, characterized by increased FTL 6 and Iba1 expression and loss of expression of homeostatic markers P2RY12 or TMEM119 and P2RY12.  Figs. 4a,d)). This suggests that it is especially the marked increase of FTL expression found in C1-microglia that 14 reflects substantial iron loading, while moderate FTL expression is also found in non-iron accumulating cells in 15 controls. Although we already found C1-microglia to significantly infiltrate Aβ-plaques, we also checked for its 16 correlation with overall Aβ and Tau load, as assessed by a neuropathologist using Thal stage and Braak stage, 17 respectively. A marked increase of the number of C1-microglia was solely found in high-pathology load 18 subjects with Thal phase V, and Braak stage V/VI (Fig. 5b,c), though not all high-pathology load subjects show 19 increase of C1-microglia. C2-microglia were primarily found in controls with low Braak stage I/II and Thal I-II, 20 whereas C3-microglia were present in both controls and Alzheimer patients with varying pathological burdens 21 (supplementary Fig. 4b,c,e,f). This is in line with the finding that iron-positive microglia were particularly   Figs. 4g-q). In addition, we looked at 28 differences between APOE3 and APOE4 carriers, as the latter have been found to have elevated ferritin levels in 29 the CSF [37]. As expected, APOE4 carriers had more Aβ-plaques (Fig. 5d), but did not show overall increased microglia infiltration (Fig. 5e). Though sample sizes for both groups were small (n = 4-6), a trend indicating 1 higher prevalence of C1-microglia in the GM could be observed (P = 0.0667; Fig. 5f), which was not the case 2 for C2 and C3-microglia ( Supplementary Fig. 4r,s) However, no difference was observed when looking at the 3 proportion of Aβ-plaques infiltrated by C1 microglia (Aβ-mic) (P = 0.5096; Fig. 5g). This suggests that even 4 though a higher percentage of C1-microglia infiltrate Aβ-plaques (P = 0.0381; Fig. 5h), this is likely due to the 5 increased number of Aβ-plaques present in the APOE4 carriers.

11
phagocytic and perivascular macrophages (Fig. 6a), based on previously described morphological phenotypes 12 [38]. The parenchyma of controls was predominantly populated by C6-C11-microglia, which consistently 13 expressed TMEM119 and/or P2RY12. These cells presented with homeostatic morphology, showing small 14 circular or oval cell bodies, with thin highly ramified processes and extensive branches (Fig. 6b). Morphological 15 appearance therefore appeared to be in line with the homeostatic protein phenotype. Occasionally activated 16 microglia were identified, which have larger cell bodies and noticeably fewer branches and ramifications 17 (especially second degree) (Fig. 6a). Activated cells generally showed higher Iba1 and FTL expression and were 18 often phenotyped as C2-microglia (Fig. 6a). Microglia in Alzheimer patients, on the other hand, had a much 19 more heterogeneous appearance; homeostatic, activated, dystrophic and phagocytic microglia could all be 20 observed within the coronal sulcus of a single patient (Fig. 6b). Though almost all phenotype clusters and 21 morphological clusters could be observed, we focussed on the C1-microglia, as they reflected iron + -microglia.

22
We found the striking majority of C1-microglia to have a dystrophic morphological appearance. The dystrophic 23 cells show a very distinct phenotype, often with a cloudy or cytorrhexic (fragmentation of the cytoplasm) 24 appearance which results in ill-defined processes (Fig. 6a). There is often deramification and the remaining 25 branches show spheroids and fragmentation. Especially microglia (both C1 and C3) infiltrating Aβ plaques 26 showed highly dystrophic morphological characteristics, indicative of an advanced activated/neurodegenerative 27 state (Fig. 6c). The dystrophic morphology was also verified using 3D confocal microscopy, which also showed 28 the same cytorrhexic appearance of microglia surrounding the Aβ-plaques (Fig. 6d). All in all, the finding of a 29 dystrophic phenotype in C1-microglia was in line with the increased Iba1 and decreased TMEM119 and

1
In this manuscript, we confirmed that increased FTL expression reflects an increase in iron accumulation in 2 microglia in the cortex of Alzheimer patients. Microglia with increased FTL expression also showed higher Iba1 3 expression, but loss of homeostatic markers TMEM119 and P2RY12, indicative of an activated phenotype. On 4 further investigation this FTL + Iba1 + phenotype appeared to be increasingly present in Alzheimer patients and 5 the predominant Aβ-plaque infiltrating microglia phenotype. Morphologically they appeared to be in a 6 dystrophic activation stage.

7
Firstly, in this study we confirmed that previously identified iron-positive cells in Alzheimer patients [39,40] are 8 of microglial rather than astrocytic origin, and show high FTL expression. Subsequently, using multispectral 9 fluorescence and unsupervised clustering, we identified an several FTL + -clusters, which were variably present in 10 controls and Alzheimer disease stages. C2-microglia, which displayed positivity for all included microglia 11 markers were almost exclusively present in control patients. Conversely, C1-microglia (FTL + Iba1 + ) were 12 significantly more present in AD patients, and C3-microglia (TMEM119 + FTL + Iba1 + ) were marginally present in 13 either group. Interestingly, both C1 and C3-microglia showed a strong tendency to infiltrate Aβ-plaques. C1-14 microglia were almost exclusively present in advanced stage Alzheimer patients, whereas C2-microglia were 15 primarily detected in controls (with low Thal/Braak stages), and C3-microglia were variably present across 16 controls and Alzheimer patients of all stages. Regarding the temporal dynamics of these clusters, one could 17 therefore hypothesize that in Alzheimer's disease microglia surround Aβ-plaques and lose P2RY12 expression, 18 as has been observed previously by others (transition from C2 to C3) [35,41]. As of yet we do not know what 19 the relevance is of the preserved TMEM119-expression. Over time, these microglia take up iron, causing a 20 pronounced increase of FTL expression and loss of TMEM119. This corresponds to the fact that only C1-21 microglia appeared to correlate with iron-accumulating microglia. However, our study population is not ideal to 22 dissect the temporal dynamics of these clusters, since the majority of Alzheimer patients showed advanced 23 disease (Braak V/VI) and only two patients showed mild to moderate (Braak III/IV). Future work studying these 24 phenotypes in a larger cohort with a larger range of disease stages would be highly relevant to accurately 25 determine at what stage of the disease C2 microglia prevalence decreases and C1 and C3 microglia prevalence 26 increases.

28
Several qualitative studies had previously identified increased presence of dystrophic ferritin + microglia in brain 1 tissue of Alzheimer patients [6,39,42,43]. The dystrophic morphological appearance was also confirmed in this 2 study, though the functional insights of these morphologically defined states remains debatable. Our spatial 3 analysis revealed a strong tendency of FTL + Iba1 + to infiltrate Aβ-plaques; significantly more than can be 4 expected based on prevalence of the cluster itself, and more than any other identified microglia cluster.

5
Although some other studies had already looked into the association of dystrophic ferritin + microglia with Aβ-6 plaques [6,7,30,39,44], results were inconsistent, as none of these studies so far looked into the relative 7 proportion of these microglia in the total population. The importance of this is also stressed in a recent study by

19
To date, the reason for the observed increase of FTL-expression remains disputed. With FTL being the long-20 term storage component of ferritin, its expression is likely to be increased in response to increased intracellular 21 labile iron concentrations. Yet, ferritin is also widely recognized as an acute phase reactant and it has also been 22 suggested that microglia upregulate ferritin as a response to exhaustion, caused by the attempting to 23 phagocytose aggregated Aβ [44]. However, our findings show that the identified FTL + Iba1 + -microglia closely 24 reflected microglia with high levels of the metal iron, and therefore suggest that the observed increased FTL-25 expression at least does not merely reflect inflammatory activation or exhaustion, but also increased iron levels.

26
This is in line with a previous study, which found ferritin levels in the CSF to not be associated with an 27 inflammatory response in Alzheimer patients and hypothesized ferritin levels to rather reflect changes in iron 28 associated with tangle and plaque pathology [46]. Why iron increases with age and even more profoundly in neurodegenerative diseases is still largely unknown 1 [8,47]. It is hypothesized to be caused by several factors including increased blood-brain barrier permeability 2 and disorganization of the iron-dense myelin sheaths [48][49][50]. Alongside a general increase of iron in the 3 parenchyma, iron was also shown to accumulate inside Aβ-plaques [50,51]. Therefore, a possible hypothesis 4 for why iron is sequestered in microglia surrounding Aβ-plaques, could be that the iron is taken up as byproduct 5 while attempting to phagocytose the Aβ aggregates. Conversely, considering we only found approximately 25% 6 of iron-accumulating C1-microglia to infiltrate Aβ-plaques, iron is more likely sequestered using either DMT1 7 or Transferrin-receptors and stored inside FTL, in an attempt to mitigate the potentially toxic effects of free iron,  activation, by showing that in human brain tissue of Alzheimer patients, microglia are exposed to a combination 19 iron and Aβ. Finally, these findings are also in line with recent clinical studies, in which iron was found to act as 20 a potential disease modifier by accelerating deterioration in Alzheimer patients with high Aβ load [12,13].

21
Thanks to the possibility to visualize up to six protein markers on the same section using mIF, we could better 22 study the great heterogeneity in microglia phenotype and its spatial relationship with pathology. A limitation of 23 mIF compared to other high-dimensional techniques such as single-cell or imaging mass cytometry is the 24 limited number of markers available to characterize the complex microglial activation states. However, single-25 cell mass cytometry lacks the spatial component, which is essential when studying the relation with Aβ. Imaging 26 mass cytometry, on the other hand, does capture the spatial distribution, however to date does not enable high-27 throughput analysis and offers limited resolution. Since microglia have a very complicated and variable 28 morphology, solely evaluating protein expression directly surrounding the nucleus is insufficient, and high-29 resolution images are required for proper segmentation and phenotyping. Secondly, as we are studying relatively rare activated microglia subtypes that will not be present in every ROI or even subject, we required high-1 throughput quantitative analysis methods. The mIF-mic panel, together with our optimized microglia 2 segmentation pipeline for 2D-images, enabled accurate segmentation and analysis of > 60000 cells to carefully 3 identify the FTL + -microglia in an unbiased fashion.

4
In this study we adopted an unsupervised learning approach to generate distinct clusters in our dataset, and 5 avoid bias in the identification in clusters, as can be present in more classical IHC studies. However, as already 6 indicated in the results section, even though distinct clusters were identified, the low degree of separation on the 7 t-SNE mapping and similarity on the associated heatmap, suggest these clusters may be more of a continuum 8 rather that distinct subsets. This is in line with other transcriptomic and proteomic studies, in which they also 9 showed the microglia clusters to be more of a continuum, even when studying substantially more genes or

23
In summary, we showed that our multispectral immunofluorescence pipeline allowed for accurate identification 24 of specific microglia clusters, and more importantly for the spatial analysis with respect to pathological 25 hallmarks. In this specific study we identified dystrophic FTL + Iba1 + TMEM119 -P2RY12 --microglia to be 26 significantly more present in Alzheimer's disease patient, and to be the predominant Aβ-plaque infiltrating 27 microglia cluster. Finally, in correspondence with the increase of FTL-expression, FTL + Iba1 + -microglia showed 28 massive iron-loading.

1
All material has been collected with written consent from the donors and the procedures have been approved by 2 the Medical Ethical committee of the LUMC and the Amsterdam UMC.

17
The authors have no conflicts of interest to declare. All co-authors have seen and agree with the contents of the 18 manuscript and there is no financial interest to report.

20
The data that support the findings of this study are available from the corresponding author upon reasonable