TGFβ pathway deregulation and abnormal phospho‐SMAD2/3 staining in hereditary cerebral hemorrhage with amyloidosis‐Dutch type

Abstract Hereditary cerebral hemorrhage with amyloidosis‐Dutch type (HCHWA‐D) is an early onset hereditary form of cerebral amyloid angiopathy (CAA) pathology, caused by the E22Q mutation in the amyloid β (Aβ) peptide. Transforming growth factor β1 (TGFβ1) is a key player in vascular fibrosis and in the formation of angiopathic vessels in transgenic mice. Therefore, we investigated whether the TGFβ pathway is involved in HCHWA‐D pathogenesis in human postmortem brain tissue from frontal and occipital lobes. Components of the TGFβ pathway were analyzed with quantitative RT‐PCR. TGFβ1 and TGFβ Receptor 2 (TGFBR2) gene expression levels were significantly increased in HCHWA‐D in comparison to the controls, in both frontal and occipital lobes. TGFβ‐induced pro‐fibrotic target genes were also upregulated. We further assessed pathway activation by detecting phospho‐SMAD2/3 (pSMAD2/3), a direct TGFβ down‐stream signaling mediator, using immunohistochemistry. We found abnormal pSMAD2/3 granular deposits specifically on HCHWA‐D angiopathic frontal and occipital vessels. We graded pSMAD2/3 accumulation in angiopathic vessels and found a positive correlation with the CAA load independent of the brain area. We also observed pSMAD2/3 granules in a halo surrounding occipital vessels, which was specific for HCHWA‐D. The result of this study indicates an upregulation of TGFβ1 in HCHWA‐D, as was found previously in AD with CAA pathology. We discuss the possible origins and implications of the TGFβ pathway deregulation in the microvasculature in HCHWA‐D. These findings identify the TGFβ pathway as a potential biomarker of disease progression and a possible target of therapeutic intervention in HCHWA‐D.


INTRODUCTION
Sporadic cerebral amyloid angiopathy (sCAA) is a disease of the elderly due to amyloid b (Ab) deposition in cerebral leptomeningeal and cortical vessels, and is associated with intracerebral hemorrhages. CAA pathology is a common feature in Alzheimer's disease (AD) and is also a defining pathological feature in hereditary cerebral hemorrhage with amyloidosis-Dutch type [HCHWA-D; (22)]. HCHWA-D is caused by a Gln-to-Glu substitution at codon 693 of the amyloid precursor protein (APP) gene leading to the formation of the Ab E22Q peptide, a particularly aggregationprone and toxic variant of the Ab peptide (17). The Dutch mutation results in severe CAA pathology with loss of vascular smooth muscle cells and intracerebral hemorrhage typically between the ages of 40 and 65. Although the correlation between HCHWA-D carrier status, reduced cerebrovascular function and the clinical phenotype has been studied (40), the exact mechanisms underlying Ab accumulation in the vessel wall are still largely unknown.
Some earlier studies of HCHWA-D postmortem brain material have focused on Ab clearance and deposition in the vasculature by the induction and modification of extracellular matrix (ECM) proteins (15,39). Transforming growth factor b1 (TGFb1) has a key role in vascular fibrosis by inducing ECM production in vessels. In postmortem AD brain material, TGFb1 mRNA levels correlate positively with the extent of CAA pathology (47). Moreover, mouse models of TGFb1 overexpression in astrocytes or neurons demonstrated that high TGFb1 levels lead to vascular fibrosis (36,48). Whether these animal models actually accumulate murine Ab in the CAA pathology-like vascular plaques remains controversial, but many in vitro studies have shown a role for TGFb1 in promoting APP and Ab production by astrocytes (1,4,12,21). Interestingly, TGFb1 astrocytic overexpression in APP-overexpressing mice results in a CAA increase with a reduction in parenchymal Ab plaque load (47).
TGFb1, 2 and 3 isoforms are expressed in mammals and mediate their cellular effects through the TGFb type I (TGFBR1) and type II (TGFBR2) receptors. TGFb is present in an inactive form bound to the ECM and is activated by consecutive cleavage of the latent-associated-protein and pro-domain. Once activated, TGFb binding to TGFBR2 induces transphosphorylation of the TGFBR1 kinase, which subsequently recruits and phosphorylates the receptor regulated Smad (homolog of Drosophila mothers against decapentaplegic) signal transducing proteins, SMAD2 and/or SMAD3. Phospho-SMAD2/3 (pSMAD2/3) interacts with the common Smad, SMAD4, and translocate into the nucleus to regulate target gene expression. Genes related to ECM synthesis such as plasminogen activator inhibitor-1 (PAI-1), fibronectin (FN1) and collagen (Type I Col1A1 and Type III Col3A1 among others) are typical TGFb target genes (41), and evidence is mounting that an increase in ECM production is linked to CAA pathology (14,45).
Accordingly, the aim of this study was to investigate in postmortem material whether the TGFb pathway is involved in the pathogenesis of HCHWA-D, based on gene expression levels and histological observations. We specifically investigated in HCHWA-D if there is a correlation between the deregulation of the TGFb pathway and the extent of CAA pathology.

Experimental design
HCHWA-D, healthy controls and sCAA brain material was used in the study as summarized in Table 1. Both frontal and occipital cortex were used in all studies, based on the assumption that the CAA pathology seems more severe in the occipital lobes in HCHWA-D (23) which therefore is expected to represent a more advanced disease stage compared with the frontal lobe. sCAA individuals were included to investigate whether the Dutch mutation in HCHWA-D results in a different effect on TGFb signaling compared to CAA pathology in general. As TGFb signaling increases with age (9), the control group was age-matched to the HCHWA-D patients. sCAA patients were significantly older (see Table 1).
We evaluated TGFb pathway activation by immunohistochemistry in HCHWA-D brain tissue, staining for the dually phosphorylated pSMAD2 and/or pSMAD3 indicative of active TGFb receptor signaling. These studies were performed in 11 HCHWA-D (age 60.5 years 6 10.7 years), 11 control (age 69.2 years 6 14.9 years) and 10 sCAA (age 74.8 years 6 8.0 years) cases. We also analyzed gene expression levels for several pathway components (RT-PCR, see Table 2) in a sub-set of these patients of whom frozen brain was available: 7 HCHWA-D patients (age 56.4 years 6 7.7 years) and 7 age-matched controls individuals (age 59.1 years 6 0.1 years). Frozen sCAA material was not available for this measurement.

Brain tissue
Frontal and occipital human postmortem brain tissue was obtained from the Netherlands Brain Bank and from our hospital (LUMC). Written informed consent was obtained for each donor and all material and data were handled in a coded fashion maintaining patient anonymity according to Dutch national ethical guidelines (Code for Proper Secondary Use of Human Tissue, Dutch Federation of Medical Scientific Societies). The study was approved by the local Ethics Committee. Quantitative RT-PCR Frozen brain tissue was cut with a sliding microtome (Leica SM2010 R), homogenized with ceramic MagNA lyser beads (Roche) and grinded using a Bullet Blender (Next Advance). RNA was extracted immediately with Aurum Total RNA Mini Kit (Biorad), including removal of remaining genomic DNA by an oncolumn DNaseI treatment for 25 minutes. Total RNA was eluted in 60 lL of provided buffer and the RNA content was measured with Nanodrop at 260 nm. All RNA extractions were performed in duplicate and cDNA was synthesized directly after extraction with the Transcriptor First Strand cDNA Synthesis Kit (Roche) using Random Hexamer primers at 658C. The cDNA was then adjusted and aliquoted at 20 ng/lL. Evaluation of RNA Integrity was performed with on-chip electrophoresis using an RNA 6000 Nano kit and a Bio-Analyzer 2100 (Agilent Technologies). Intron-spanning primers targeting TGFb pathway components and target genes (indicated in Table 2) were designed for qPCR using Primer3 Plus software (38). Primer pairs were first spotted into the wells (2.5 pmol of each in 2 mL). The qPCR was performed in a 384 wells plate using 6 ng of cDNA in a PCR master mix (Roche; 1 time PCR buffer with MgCl 2 , 0.2 mM dNTPs, 0.28 U FastStart Taq DNA Polymerase) containing 1 time EvaGreen-qPCR dye (Biotum) and PCR grade water to a final volume of 8 mL per well. All samples were run in duplicate on the same plate along with three reference genes: Hydroxymethylbilane Synthase (HMBS), Ribosomal Protein L22 (RPL22) and TATA-Box Binding Protein (TBP). The amplification was performed on a LightCycler 480 (Roche) with an initial denaturation of 10 minutes at 958C, followed by 45 cycles of 10 s denaturation at 958C, 30 s annealing at 608C and 20 s elongation at 728C. Relative expression of the transcript levels was calculated using LinRegPCR v11.1 (32) with the raw fluorescence values as input. Transcript levels were calculated with the Geomean of the biological and technical repeat (four points) normalized with two of the reference genes (HMBS and RLP22). The third reference gene (TBP) was used to check the normalization efficiency and the inter-plate variance. Changes in relative transcript levels were analyzed in GraphPad Prism version 6.00 using an unpaired two-sided Student's t test. Differences between groups were considered significant when P < 0.05.

Immunohistochemical staining and quantification of pSMAD2/3 in blood vessels
Formalin-fixed, paraffin-embedded blocks of brain tissue were cut into serial 5 mm thick sections and mounted on coated glass slides (SuperFrost V R Plus, VWR). Deparaffinization in xylene and rehydration through a series of ethanol concentrations were followed by antigen retrieval by cooking for 40 minutes at 0.76 bar steam pressure (Steba DD 1 ECO) in an acidic pH 6 solution (H-3300, Vector labs). Sections were then blocked for endogenous peroxidase with 3% H 2 0 2 in dH 2 0 for 10 minutes and for unspecific epitopes binding with blocking buffer [1% BSA suspension in washing buffer (0.1% Tween Intron-spanning (exon 6-7) Intron-spanning (exon 10-11) Intron-spanning _ reference gene After, the sections were incubated with rabbit anti-pSMAD2/3 antibody (#3101S, Cell Signaling, 1:500 dilution) overnight at 48C in the blocking buffer. Incubation with secondary anti-Rabbit HRP was followed by a DAB reaction kit (SK-4100, Vector lab) and mounting with Entellan V R New (107961, Merck). Sections were scanned (Philips Ultra Fast Scanner 1.6 RA) for grading. Collagen IV and laminin staining followed identical procedure (details in Supporting Information Table S1). The grading of pSMAD2/3 staining was reproduced on scanned sections by two independent researchers (LvdG; JMdJ) blinded to the clinical diagnosis of each case. Six fields throughout gray matter areas of the slides were randomly selected at 1003 magnification (2.016 mm 2 per field view). Per area, radially crosscut parenchymal angiopathic arterioles were counted. CAA load is defined here as the average number of angiopathic arterioles, identified by a typical thickened vessel wall, per mm 2 . Presence of Ab in angiopathic arterioles was checked on a consecutive slide (data not shown, not used for the grading). pSMAD2/3 deposits in the tunica media of angiopathic arterioles was determined for each vessel at 4003 magnification. Difference in CAA load between frontal and occipital cortex was assessed with GraphPad Prism version 6.00 using a paired twosided Student's t test. Differences between groups were considered significant when P < 0.05.
Immunohistofluorescent double staining and pSMAD2/3 specificity Formalin-fixed, paraffin-embedded frontal and occipital cortex 5 mm sections were used from patient material as specified in Table 1. and occipital (Occ) cortex compared to age-related controls. Transcript expression levels in postmortem brain cortex were normalized with two reference genes and represented in a dot plot with mean 6 SD of seven samples; *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001 as determined by a two-tailed unpaired Student's t test.
Deparaffinization, antigen retrieval and blocking steps were identical to the immunohistochemical staining. Rabbit anti-pSMAD2/3 antibody (#3101S, Cell Signaling; 1:500 dilution) was incubated overnight at 48C with mouse antibodies (references and dilutions in Supporting Information Table S1). The antibodies were visualized with anti-rabbit Alexa Fluor V R 488 and anti-mouse Alexa Fluor V R 594, respectively (1 h at RT). Alternatively, Tyramide Signal Amplification (TSA V R biotin detection kit, NEL700A001KT, Perkin Elmer) followed by streptavidin, Alexa Fluor V R 488 conjugate was used when specified in the text (Supporting Information Table S1). Nuclei were stained with DAPI (1 lg/mL) during the secondary antibody incubation step. After each incubation, the slides were extensively washed in washing buffer. Sections were mounted in Pro Long Diamond (Life technologies). Images of the fluorescent staining were acquired using a confocal laser-scanning microscope (Leica SP8, Leica Microsystems). The specificity of pSMAD2/3 staining was assessed either with calf intestine alkaline phosphatase treatment removing the phosphor-epitopes [adapted from Ref. (37), details in Supporting Information Figure S1] and by using different phospho-antibodies for the same epitope (Supporting Information Figure S1 and Table  S1). No cross-reactivity were observed with anti-pSMAD1/5/9 (data not shown). A granular perivascular pSMAD2/3 staining was also detected by Dr. Ueberham.  To assess TGFb pathway implication we measured  gene expression levels of TGFb1, TGFb2, TGFBR1, TGFBR2, SMAD2, SMAD3, SMAD4 and SMAD7 by qRT-PCR. The relative gene expression levels of TGFb1 and TGFBR2 were significantly higher in the frontal and occipital lobes of HCHWA-D patients compared to age-related controls ( Figure  1). TGFBR1 followed a similar trend, but did not reach statistical significance. TGFb2 levels were significantly higher in the frontal lobe of HCHWA-D samples, especially in the two eldest patients from the gene expression study (H6 and H7); ( Figure  1). Interestingly, these two samples also present the highest level of the inhibitory Smad, SMAD7, suggesting an enhanced TGFb pathway activation compared to the other HCHWA-D patients (Supporting Information Figure S2). Other signaling effectors of the canonical SMAD pathway (SMAD2, SMAD3, SMAD4 and SMAD7) were further not significantly different (Supporting Information Figure S2). Plasminogen activator inhibitor-1 (PAI-1), fibronectin1 (FN1), Col1A1 and Col3A1 are known SMAD-dependent downstream targets of TGFb. PAI-1 and FN1 gene expression levels were both significantly higher in frontal and occipital cortex of HCHWA-D patients. Col3A1 and Col1A1 were upregulated in both brain area, but reach statistical significance in the frontal cortex only (Figure 2). pSMAD2/3 is accumulating in HCHWA-D, but not in sCAA

RESULTS
To evaluate TGFb pathway activation by immunohistochemistry in HCHWA-D brain tissue, we stained for the TGFb1 down-stream signaling effector dually phosphorylated pSMAD2/3. Frontal and occipital cortex of controls, sCAA and HCHWA-D cases were assessed for pSMAD2/3 staining. In all the observed cases, pSMAD2/3 labeling was predominantly located in nuclei of neurons, with little or no cytoplasmic staining detected in both brain area.
Both in frontal and occipital cortex, pSMAD2/3 staining in amyloid-laden vessels were found uniquely in HCHWA-D cases. The staining pattern showed distinct granules in the tunica media as well as diffuse staining covering the entire vessel wall. Examples of these granular deposits are given in Figure  3A. pSMAD2/3-positive and negative vessels were not morphologically different. Notably, pSMAD2/3 deposits were mainly found in parenchymal arterioles with Ab covering the entire vessel circumference (double staining with anti-Ab antibodies, Figure 3B), corresponding to an advanced CAA grade [grade 2 or 3 based on the Ab content as defined by Greenberg and Vonsattel (13)]. Nevertheless, these granular deposits in the tunica media were not colocalized with the smooth muscle actin (SMA) staining, and we could observe in some angiopathic vessels an accumulation of pSMAD2/3 granules in vacuoles  Figure S3).

pSMAD2/3 accumulation is correlated with CAA load, but less with age
Entire tissue sections were graded to assess pSMAD2/3 staining in HCHWA-D samples. For the same sections, the CAA load was measured as the number of angiopathic arterioles in a defined area. We found no significant differences in CAA load between frontal and occipital cortex ( Figure 4A). However, a significant positive correlation between pSMAD2/3-positive angiopathic vessels and CAA load was found independently of the brain area studied (Figure 4B). This confirmed that pSMAD2/3 deposits are only present once the CAA pathology becomes more severe. In areas with a high CAA load, most angiopathic vessels are pSMAD2/3-positive.
Although there is a strong correlation between pSMAD2/3 and CAA pathology severity, CAA load in itself is only moderately dependent on the age of the patients ( Figure 3C,D). In particular, the 81-years-old patient, who reached an unusual age for HCHWA-D, was much less affected than expected.

Perivascular pSMAD2/3 granules are present in occipital cortex of HCHWA-D
Apart from granules localized on angiopathic vessels in HCHWA-D, pSMAD2/3 granules were also found in the parenchyma. These were observed as perivascular parenchymal rings (see Figure 5A,B) in the occipital cortex in about 50% of the cases (Supporting Information Table S2), but never in the frontal cortex of HCHWA-D brains or in sCAA and control cases. Perivascular rings were found both around angiopathic and non-angiopathic vessels, as well as around capillaries, and were found in clusters with a predilection for the first and last cortical layer of the gray matter. SMAD4 binding to pSMAD2/3 is a prerequisite for SMAD2/3 to enter the nucleus and initiate gene transcription. However, in our samples, SMAD4 was neither co-localized with granular vascular, nor with perivascular halo  Figure S4); questioning the participation of the granules in the active signaling.

Perivascular pSMAD2/3 granules do not colocalize with studied cell types or neuropathological features
Since perivascular pSMAD2/3 granules were arranged in a linear alignment, reminiscent of cytoskeletal filaments ( Figure 5), we investigated whether the granules co-localized with a particular cell type using neuronal and astrocytic cytoskeleton markers. MAP-2 staining revealed a presence of the granules along perivascular dendrites ( Figure 5C) without co-localization within the neuronal processes. pSMAD2/3 granules never aligned with GFAP processes ( Figure 5D) but were topographically restricted to GFAP-positive perivascular areas, as depicted in Figure 6. Further, parenchymal cytoplasmic pSMAD2/3 granules were incidentally identified in neurons in some HCHWA-D individuals (Supporting Information Figure S5).
Previous HCHWA-D studies reported perivascular ubiquitinated and phosphorylated neurites associated with preamyloid and amyloid deposits around angiopathic vessels (24,33). Therefore, we examined in HCHWA-D occipital samples whether the pSMAD2/ 3 perivascular granules were co-localized with any of these previously reported deposits. No co-localization of pSMAD2/3 granules was found in the perivascular rings with Ab, hyperphosphorylated Tau (pTau, AT-8 antibody) and ubiquitin. Likewise, we stained for perivascular coarse deposits of ECM as described previously for HCHWA-D using collagen IV and laminin antibodies (39). Although we found these coarse deposits, they did not co-localize with pSMAD2/3 granules ( Figure 7A,B). Furthermore, an extracellular pSMAD2/3 deposition with amyloid deposits and neuritic plaques was described in AD (6,37). Even though a similar colocalization with diffuse parenchymal Ab plaques in HCHWA-D was occasionally detected, these extracellular pSMAD2/3 deposits had a fibrous-like and diffuse staining that is different from the bright round-shaped dots composing the pSMAD2/3 granular rings (Supporting Information Figure S6).

DISCUSSION
Our findings suggest that TGFb is implicated in the pathogenesis of HCHWA-D. We found an upregulation in gene expression of several components of the TGFb pathway and its direct downstream signaling targets as well as a strong correlation of pSMAD2/3 deposits with CAA load. Based on previous studies in HCHWA-D describing an increased CAA pathology in the occipital cortex compared to the frontal cortex (23), we hypothesized that the CAA load and the associated vascular pathology in the occipital lobe would represent a more advanced disease stage. However, in our HCHWA-D samples, individual perforating arterioles presented a consistent moderate to severe CAA grade (grade 2 to 3), irrespective of the brain area studied. Furthermore, the CAA load, based on the number of angiopathic vessels per mm 2 , was not significantly different in the two lobes ( Figure 4A). A possible explanation for this finding is that the higher CAA load in the occipital cortex compared to the frontal cortex in previous studies is influenced by the presence of angiopathic capillaries (25). Capillary CAA is typically a feature of aged patients, while our cohort was relatively young. Additionally, the high occipital CAA load in the occipital cortex was in most cases confined to the end of the occipital horn (28), whereas our samples were obtained from diverse occipital area.
Although we did not find a difference in CAA severity between occipital and frontal lobes, we found a high correlation between the CAA load and the presence of pSMAD2/3 granules. Strikingly, these granules were only found in HCHWA-D patients, but not in sCAA cases, even in similarly affected vessels. The association of CAA pathology with pSMAD2/3, which is a direct TGFb downstream signaling effector, suggests an involvement of TGFb on the vessel wall pathology. TGFb is thought to be a key mediator of vascular remodeling (31) that is found in sporadic small-vessel diseases (34). TGFb deregulation in the vessel wall is a central mechanism common to several hereditary brain microvasculopathies, like cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy [CADASIL; (18)] or cerebral autosomal recessive arteriopathy with subcortical infarcts and leukoencephalopathy [CARASIL, (3,49)]. TGF-b signaling was recently proposed as a common denominator of several forms of cerebral small-vessel diseases (26). Common microvasculopathies include microaneurysms, fibrinoid necrosis, obliterative intimal changes, and hyaline thickening. Interestingly, these secondary structural microvasculopathies, are found more frequently in HCHWA-D in comparison to sCAA (42) and are correlated to CAA load (28). This aggravated remodeling may be due to a fast progression of the disease in HCHWA-D or to a direct effect of Dutch-type Ab binding to TGFBR2, directly activating the signaling pathway (16). Accumulation of pSMAD2/3 granules in the vessel wall in HCHWA-D supports the hypothesis that TGFb deregulation contributes to secondary microvascular remodeling.
TGFb1 was upregulated in our HCHWA-D postmortem brain tissue, similar to a previous study where a correlation was found in the cerebrovascular pathology of AD (46). Although part of the observed result is possibly influenced by differences in the cellular composition of HCHWA-D cortex, normalization per cell type cannot be achieved, nor is commonly done in RT-qPCR. We corrected for the total amount of transcript per sample with stably expressed references gene. Another potential confounder in our study is that all patients suffered one or more hemorrhagic strokes before death, which influences TGFb1 expression. It is known that the expression of TGFb1 increases rapidly after brain injury to restrict brain damage and as part of the healing process (9,11). Still, we found consistent upregulation in all patients in our cohort, despite very different survival times after hemorrhage (from days to several years), variable hemorrhage sites and even unrelated causes of death. This suggests that the TGFb upregulation we found cannot solely be explained by an acute response after stroke, but is likely linked to the CAA pathology itself, as evidenced by the histological spatial correlation described above. We also found TGFBR2, the ligand-binding receptor, upregulated in HCHWA-D occipital and frontal cortex. Previous studies have also found high levels of TGFBR2 in a mouse model of traumatic cerebral injury and stroke in the chronic phase (10,30). These findings indicate that TGFBR2 upregulation is not an acute response. Last, the upregulation of TGFb-induced pro-fibrotic target genes such as PAI-1, FN1, Col1A1 and Col3A1 (41) indicate that the TGFb signaling pathway is likely activated.
TGFb upregulation is a double-edged sword with both protective and deleterious consequences. The vascular remodeling and fibrosis induced by TGFb might have a protective effect in CAA pathology, due to the cross-linked increased ECM and basement membrane which might prevent the weakening of the vessel wall  (8,50). Despite this protective effect in terms of stroke survival and prevention of hemorrhage, persistent TGFb upregulation is also thought to have a major downside. The resultant ECM synthesis which modifies the composition of basement membrane impairs perivascular drainage, thereby triggering further amyloid deposition and aggravating the CAA (5,43). This was demonstrated in mouse models of both inducible neuronal-and astrocytic-TGFb overexpression, where perivascular astrogliosis is preceding and promoting the vascular angiopathy (36,48). In our study, we found a perivascular ring of pSMAD2/3 granules around vessels in the occipital lobe of HCHWA-D samples. Comparable pSMAD2/3 granules were found in other neurodegenerative disorders, such as AD, Pick's syndrome, progressive supranuclear palsy and corticobasal degeneration (2,6,7). Typically, in AD the granules were found intracellularly and associated with neuronal aggregates of pTau or granulovacuolar inclusion of ubiquitin (2,20,27,37). This aberrant cytoplasmic dislocation of pSMAD2/3 could impair the normal TGFb signaling pathway by sequestration of this transcription factor (2,6,27,37). In our study, we did not find pTau, ubiquitin or Ab colocalization with the perivascular ring of pSMAD2/3 granules, which is not surprising considering the general lack of neuronal degeneration and pTau involvement in HCHWA-D. Nevertheless, the granular sequestration of pSMAD2/3 may point to a similar deregulation of the TGFb pathway.
In the current study, we often found that the perivascular granules were positioned following a linear pattern, reminiscent of astrocytic processes. In a recent study, a decrease in cerebrovascular reactivity in the occipital cortex of HCHWA-D was described as an early biomarker of the disease (40) and astrocytes are key mediators in this process (19). In transgenic mice with cerebral angiopathy due to overexpression of TGFb or APP, the cerebrovascular reactivity was impaired, due to neurovascular decoupling (29,35). This decoupling is the result of retraction of astrocyte end feet from the vessel wall. Similar underlying mechanisms likely occur in HCHWA-D patients. Such perivascular astrocytic remodeling has been linked in AD mouse models with CAA pathology severity and astrocytic phenotypic switch, defined by a loss of GFAPpositivity (44). The perivascular granules could be the remnants of astrocytic cytoskeletal remodeling.