Heme and hemoglobin suppress amyloid β–mediated inflammatory activation of mouse astrocytes

Glial immune activity is a key feature of Alzheimer's disease (AD). Given that the blood factors heme and hemoglobin (Hb) are both elevated in AD tissues and have immunomodulatory roles, here we sought to interrogate their roles in modulating β-amyloid (Aβ)-mediated inflammatory activation of astrocytes. We discovered that heme and Hb suppress immune activity of primary mouse astrocytes by reducing expression of several proinflammatory cytokines (e.g. RANTES (regulated on activation normal T cell expressed and secreted)) and the scavenger receptor CD36 and reducing internalization of Aβ(1–42) by astrocytes. Moreover, we found that certain soluble (>75-kDa) Aβ(1–42) oligomers are primarily responsible for astrocyte activation and that heme or Hb association with these oligomers reverses inflammation. We further found that heme up-regulates phosphoprotein signaling in the phosphoinositide 3-kinase (PI3K)/Akt pathway, which regulates a number of immune functions, including cytokine expression and phagocytosis. The findings in this work suggest that dysregulation of Hb and heme levels in AD brains may contribute to impaired amyloid clearance and that targeting heme homeostasis may reduce amyloid pathogenesis. Altogether, we propose heme as a critical molecular link between amyloid pathology and AD risk factors, such as aging, brain injury, and stroke, which increase Hb and heme levels in the brain.

effective therapy for AD and have much left to learn about the mechanisms involved in AD pathogenesis. Neuroinflammation, in particular, is increasingly recognized as an important aspect of AD pathology (2), but whether glial activity promotes pathogenesis (3) or is neuroprotective (4,5) remains an open question. Glial activity plays an important role in neuroprotection and maintaining tissue homeostasis by regulating metabolism (6), pruning neurites and synapses (7), and clearing pathogens such as the hallmark AD protein, ␤-amyloid (A␤) (8). However, glial inflammatory activity can also promote a neurotoxic microenvironment via overexpression of neurotoxic cytokines (9 -11) and reactive oxygen species (12), among other factors. Moreover, there is increasing evidence that glia efficiently clear A␤ early in AD but that they become dysfunctional with time (13)(14)(15), perhaps due to changes in environmental factors and immunomodulatory signaling during AD progression (11,13). Whereas the role of microglia in AD has long been acknowledged, astrocytes are increasingly emerging as important modulators of AD pathology. Specifically, astrocytes have been reported to migrate toward A␤ plaques and uptake and degrade A␤ (14, 16 -18), suggesting that deficits in astrocyte immune function may contribute to AD pathogenesis. Increased brain tissue levels of the immunomodulatory blood factors heme and hemoglobin (Hb) are characteristic of AD (19 -22) and are associated with a number of AD risk factors, including age, brain injury, and stroke (23, 24). Notably, analysis of post-mortem human AD tissue has shown increased heme in the temporal lobe (21) and increased Hb mRNA and protein in the inferior temporal gyrus and parietal gray and white matter (22), respectively. In fact, heme has been shown to colocalize with A␤ deposits in AD tissue (25), and Hb has been found within senile plaques and cerebral amyloid angiopathy (22). Furthermore, both heme and Hb have been reported to bind A␤ and alter aggregation state (19,21). Studies in macrophages and endothelial cells indicate that heme stimulates the immune response via Toll-like receptor 4 (TLR4) signaling (26). Moreover, Hb can promote inflammation independently of heme (27). Despite these prior observations pointing to potential roles for heme and Hb in modulating AD pathogenesis, the effects of heme and Hb on A␤-mediated inflammatory response and the physiologic consequences of heme and Hb interactions with A␤ remain unknown.
Herein, we sought to elucidate the effects of heme and Hb on astrocyte immune function and delineate how heme and Hb specifically affect astrocyte inflammatory response to A␤. Surprisingly, our data reveal that heme and Hb, which are proinflammatory in macrophages (26,28,29), largely suppress the A␤-mediated astrocyte expression of a broad collection of proinflammatory cytokines. Moreover, we determined that heme and Hb reduced expression of the scavenger receptor CD36 and internalization of A␤  and other substrates. Our results further demonstrate that heme or Hb alters A␤(1-42)-mediated inflammatory activation via dual mechanisms that involve both modulation of immune signaling and physical association with A␤ . With respect to signaling, heme and Hb modulate the phosphatidylinositol-4,5-bisphosphate 3-kinase (PI3K)/ Akt pathway. With respect to heme/Hb interactions with A␤(1-42), we find that a high-molecular weight (HMW) oligomeric soluble A␤ species is primarily responsible for inflammatory activation of astrocytes, and this species is no longer inflammatory when associated with heme or Hb. In total, our data reveal that heme and Hb are able to reduce astrocyte activation and phagocytic capacity by direct cell signaling and through interactions with A␤ . Given that Hb and heme levels are increased in AD brains (19,20), our findings represent a new paradigm for understanding astrocyte dysfunction and neuroinflammation in AD. They specifically suggest that targeting heme/Hb homeostatic machinery could represent a viable therapeutic strategy for AD. Last, the suite of cell biological and biophysical methods that we employed can be easily adapted to probe the effects of a multitude of molecules found to be dysregulated in the AD brain for their effects on neuroinflammation.

Heme and hemoglobin modulate A␤(1-42)-induced astrocyte inflammatory cytokine expression
We used primary mouse astrocytes generated from CD-1 pups as a model system to delineate the effects of heme and Hb on A␤(1-42)-induced inflammatory response. Although neonatal-derived astrocyte cultures are widely used, microglial contamination is a concern in assessing inflammatory response (30). We did not, however, detect expression of the microgliaspecific marker, Iba-1, in A␤(1-42)-stimulated cultures (Fig.  S1), suggesting that our cultures were not significantly contaminated with microglia. Because physiologic concentrations of A␤  are reported to be in the pico-to nanomolar range (31)(32)(33)(34), we used 50 nM A␤  for all culture conditions. The concentrations of heme and Hb have not been defined in Alzheimer's disease, although they are thought to be on the order of 10 -100 M in hemolytic disorders (29,35). Thus, we tested our primary conditions at equimolar 50 nM heme or Hb and verified using high 25 M concentrations. Heme or Hb was added to astrocyte culture medium containing fetal bovine serum, which has ϳ60 nM heme and Ͻ10 nM Hb (Fig. S2). The heme present in serum is not bioavailable to A␤ , because it is almost all exclusively associated with high-affinity hemoproteins, Hb or hemopexin, which exhibit tight dissociation constants (K D ) for heme, K D Ͻ 100 fM (36).
Canonical markers of astrocyte activation, such as glial fibrillary acidic protein (GFAP), are not reliable activation markers in culture due to high baseline expression (37,38). Therefore, we used a multiplexed immunoassay to robustly quantify astrocyte activation in terms of protein expression of 32 cytokines into the culture medium (EMD Millipore, Billerica, MA). As expected, astrocytes conditioned for 24 h with A␤(1-42) increased expression of numerous pro-inflammatory cytokines, including IL-1␤ (39), RANTES (40), and GM-CSF (41) (Fig. 1a). Because we measured a total of 32 cytokines, we aimed to create a cytokine profile that could be used to discern differences between groups. To do so, we utilized a discriminant partial least squares regression (D-PLSR) analysis (42). We have previously used this approach to identify a cytokine profile distinguishing post-mortem human control and AD brain tissues (9). Applying this analysis here identified an axis called a latent variable (LV1) that distinguished A␤  conditioned wells from all other conditions (Fig. 1b). The LV1 axis consisted of a profile of cytokines that were most different between groups (Fig. 1c), whereas LV2 defined a second axis of cytokines that were most different in the heme ϩ A␤(1-42) condition (Fig.  1d). By plotting each sample in terms of its score on LV1 (representing a composite indicator of cytokine expression), we found that A␤(1-42)-induced cytokine expression was significantly increased compared with controls or A␤(1-42) wells that were co-treated with heme or Hb (Fig. 1e). Plotting of selected individual cytokines revealed a trend where certain pro-inflammatory cytokines, such as RANTES and GM-CSF (43), were down-regulated by heme and Hb, whereas others, such as KC (44) and MCP-1 (45), were not substantially modulated (Fig. 1f), reflecting our multivariate analysis ( Fig. 1, b-e).
We additionally conditioned astrocytes with a high concentration (25 M) of heme associated with hemolytic disorders (29,35), which promotes cytokine expression in the RAW264.7 macrophage cell line (Fig. S3a). Analysis of the same panel of 32 cytokines in astrocytes demonstrated that 25 M heme produced no change in cytokine expression compared with vehicle ( Fig. S3b). Moreover, when applied together with A␤(1-42), 25 M heme further reduced cytokine expression compared with treatment with 50 nM heme ϩ A␤(1-42) (Fig. S3b). Applying the same concentrations of heme to the SIM-A9 cell line (46), we found that neither low nor high concentrations of heme alone were highly pro-inflammatory (Fig. S3c). Moreover, 50 nM heme or Hb suppressed cytokine expression compared with treatment with A␤(1-42) alone. In contrast, 25 M heme applied together with A␤(1-42) amplified cytokine expression compared with treatment with A␤(1-42) alone (Fig. S3c).

Heme and hemoglobin reduce astrocyte uptake of A␤(1-42) and phagocytic capacity
Because heme and Hb reduced A␤(1-42)-induced expression of multiple inflammatory cytokines (Fig. 1), we next investigated their effects on astrocytic capacity to scavenge A␤  and other substrates. First, we conditioned astrocytes with 50 nM A␤(1-42), either alone or together with heme or Hb. We then used immunocytochemistry to stain for A␤(1-42) using the 6E10 antibody (BioLegend, San Diego, CA). Astrocytes treated with A␤(1-42) alone showed A␤(1-42) aggregates within the plane of the cell (Fig. 2a, arrows). In contrast, astrocytes co-conditioned with heme, and to a lesser extent Hb, showed little A␤(1-42) within the plane of the cell and substan-tial labeling on the cell surface, suggesting that heme and Hb suppress A␤(1-42) internalization (Fig. 2a, arrows).
To determine whether heme or Hb affected scavenger activity for other substrates, we next interrogated their effects on astrocyte internalization of killed Escherichia coli particles. We treated astrocyte cultures with killed E. coli microparticles that were labeled with a trypan-quenched fluorescein (Thermo Fisher Scientific). Fluorescence intensity ( ex ϭ 480 nm; em ϭ 520 nm) was quantified on a microplate reader and revealed that Hb and, to a lesser extent, heme reduced microparticle internalization compared with control (Fig. 2b). To determine whether these effects also modulated phagocytosis, we incubated astrocytes with pH-sensitive pHrodo Zymosan particles, which fluoresce in phagosomes. Quantification of phagocytic cells using fluorescence microscopy (Fig. S4) ). b, primary astrocytes, preincubated with 50 nM heme or 50 nM Hb, were incubated with trypan-quenched, fluorescein-labeled, killed E. coli particles. Particle internalization, measured by fluorescence intensity using a microplate reader, significantly decreased upon incubation with 50 nM Hb (n ϭ 28 wells, p ϭ 0.0129; vehicle versus Hb). c, primary astrocytes, preincubated with 50 nM heme or 50 nM Hb, were incubated with pH-sensitive pHrodo beads to assess phagocytic capacity. The percentage of total cells uptaking beads, quantified by fluorescence microscopy (Fig. S4

Heme and Hb suppress astrocyte inflammation
reduction of phagocytosis in Hb-treated astrocytes and a nonsignificant reduction by heme (Fig. 2c), mirroring our observations with killed E. coli particles. Because heme and Hb appeared to have similar effects, although to differing degrees, we next wanted to determine whether these effects were associated with changes in astrocyte phagocytic receptor expression. A␤ is highly promiscuous (47), although key astrocyte scavenger receptors include CD36, RAGE, and CD47 (48). Of these, we chose to quantify CD36 expression because it mediates both phagocytosis and inflammatory signaling in astrocytes (49). Indeed, Western blot analysis revealed decreased CD36 expression in the presence of heme or Hb (Fig. 2, d and e). This finding is consistent with heme-and Hb-mediated modulation of A␤-induced inflammation ( Fig. 1), because CD36 is required for astrocyte activation (49) and mediates A␤-induced inflammatory signaling (50).

Heme and hemoglobin modulate A␤(1-42) aggregation state and morphology
The ability of heme and Hb to interact with and alter the aggregation state of A␤(1-42) (21, 22) may be responsible for attenuating A␤(1-42)-mediated inflammatory activation of astrocytes and clearance of A␤ . To better understand heme and Hb interactions with A␤(1-42) and the effects of these interactions on A␤ secondary structure, aggregation state, and morphology, we used a combination of UV-visible spectroscopy, CD spectroscopy, thioflavin T (ThT) fluorescence assays, size-exclusion chromatography (SEC), and transmission EM (TEM). First, using UV-visible spectroscopy, we found that titration of A␤(1-42) into aqueous solutions of heme and Hb resulted in changes in heme absorbance spectra consistent with high-affinity interactions. In the case of heme, we found that A␤(1-42) interacts with heme in a 2:1 stoichiometry, with apparent dissociation constants of K D1 Յ 100 nM and K D2 ϭ 3 M (Fig. 3, a-c), consistent with prior studies (51,52). In the case of Hb, we found that each monomer of A␤(1-42) interacts with each monomer of tetrameric Hb (Fig. 3, e and f). Second, using CD spectroscopy, we found that the ␤-sheet secondary structure of A␤(1-42) remains intact upon heme binding (Fig. 3d). Further, CD spectroscopy revealed that a mixture of Hb and A␤(1-42) yields a CD-derived secondary structure that is distinct from the sum of the individual Hb and A␤(1-42) spectra (Fig. 3g), confirming biophysical association between Hb and A␤(1-42). Third, using ThT fluorescence ( ex ϭ 450 nm; em ϭ 482 nm) as a probe for A␤(1-42) fibrillization (53), we found that heme and Hb could prevent the formation of ThT-positive A␤(1-42) aggregates (Fig. S5a), and heme de-aggregated preformed ThT-positive A␤(1-42) fibrils ( Fig. S5b), consistent with prior studies demonstrating that heme and Hb suppress A␤(1-42) fibrillization (19,54).
We next analyzed the effects of heme and Hb on the morphology and size of insoluble and soluble A␤(1-42) species using SEC and TEM. TEM analysis of insoluble A␤  revealed the presence of four morphologically distinct species that shift in distribution upon the presence of heme or Hb: extended long fibrils (Type I), tangled fibrils (Type II), amorphous aggregates (Type III), and short fibrils (Type IV) (Fig. 4a).  the presence of Type I and Type II fibrils. We therefore do not consider Type IV species in our analysis of heme and Hb effects on A␤(1-42) aggregation and fibrillization. Incubation with heme results in the conversion of the Type I species to the amorphous Type III species in the insoluble fraction, with little effect on the Type II species. In the case of Hb, whereas there is greater variation between two independent trials, it is clear that Hb also has a profound effect on fibril morphology, shifting the species distribution from Type I extended fibrils to Type II tangled fibrils (Fig. 4b). Further, it is worth noting that the insoluble species generated from the application of heme and Hb are not enriched with heme ( Fig. S6), suggesting that heme remains largely associated with soluble A␤(1-42) species. Altogether, our studies with heme and Hb are consistent with prior work demonstrating that it reverses and/or suppresses A␤(1-42) fibril growth (19,21). However, our work shows that heme and Hb also have the capacity to increase the formation of certain HMW soluble oligomers of A␤(1-42) and, with respect to Hb, can alter fibril morphology from extended (Type I) to tangled fibrils (Type II).
To analyze the effects of heme and Hb on soluble A␤(1-42) aggregation state, size, and morphology, we subjected the soluble A␤(1-42) fraction, after pelleting out insoluble A␤ , to SEC analysis. SEC analysis of soluble A␤(1-42) revealed the presence of HMW aggregates, Ͼ75 kDa, that elute in the void volume, 8 ml, as well as low-molecular weight (LMW) species, Ͻ6.5 kDa, that elute at ϳ19 ml (Fig. 4c). Both heme and Hb resulted in the loss of the LMW species eluting at 19 ml in favor of the HMW species eluting at 8 ml. Further, UV-visible spectroscopy indicates that the 8-ml species is associated with heme (Fig. 4, c and d), and immunoblot analysis of the 8-ml peak with Hb further demonstrated the co-elution of both Hb and A␤(1-42) (Fig. S7a). By comparison, free Hb eluted at 12 ml (Fig. S7b). The HMW oligomers that elute in the 8-ml fraction are exclusively Type III amorphous A␤(1-42) aggregates both in the absence and presence of heme or Hb, as indicated by TEM analysis (Fig. 4e). The LMW A␤(1-42) species eluting at 19 ml appears by TEM to be a smaller amorphous aggregate than the Type III HMW species that elutes at 8 ml (Fig. S8).
To determine whether the Type III species are on-pathway to form amyloid fibrils, we incubated a 1 M concentration of the SEC isolated Type III species, with or without heme/Hb, as well as the unseparated stock mixture of 2 M A␤(1-42) for 48 h at 37°C and analyzed them by TEM. Interestingly, unlike the unseparated stock mixture of A␤(1-42), which readily forms fibrils, the Type III species did not readily form fibrils (Fig. S9).
Altogether, these data demonstrate that heme and Hb do not simply de-aggregate or prevent fibril formation, but also act to promote the formation of distinct heme-and Hb-associated soluble HMW oligomers. The isolated A␤(1-42) species and their inflammatory response are summarized in Table 1.

Inflammatory activation of astrocytes by soluble A␤(1-42) aggregates is reversed by association with heme or hemoglobin
A combination of centrifugation and SEC identified a number of soluble and insoluble A␤(1-42) species (Table 1), including ones that associate with and/or are produced as a conse-

Heme and Hb suppress astrocyte inflammation
quence of heme and Hb. Recent studies of isolated A␤ from post-mortem human tissues have revealed that different species have distinct cytotoxicities and ligand affinities (56,57). We therefore sought to identify which A␤(1-42) species were pro-inflammatory and assess the effect that heme or Hb association with these species had on inflammatory activation of astrocytes. Toward this end, we conditioned astrocytes with 50 nM preparations of each soluble and insoluble A␤(1-42) species for 24 h and assayed their ability to stimulate the expression of inflammatory cytokines relative to a 50 nM concentration of the unseparated A␤(1-42) stock mixture and vehicle control. Most interestingly, we found that application of the soluble Ͼ75-kDa HMW Type III oligomer is highly inflammatory and comparable with that of the application of the unseparated A␤(1-42) stock mixture (Fig. 5a). In marked contrast, preparations of the A␤(1-42) pellet, which primarily consists of the Type I extended fibrils, or the soluble LMW species yielded minimal cytokine expression in astrocytes compared with the unseparated A␤(1-42) stock mixture (Fig. S10a).
We next determined what role heme and Hb association play in mediating A␤(1-42) inflammatory activation. Strikingly, heme or Hb association with the highly inflammatory soluble Ͼ75-kDa HMW Type III oligomer completely reverses its inflammatory activation of astrocytes (Fig. 5). On the other hand, the Hb-associated insoluble pellet, which has a greater fraction of Type II "tangled fibrils" relative to the noninflammatory A␤(1-42)-only pellet that is primarily composed of Type I "extended fibrils" (Table 1), produced a high inflammatory response in astrocytes (Fig. S10b). It is worth noting that the heme-associated insoluble A␤(1-42) pellet could not be tested for inflammatory activation of astrocytes due to the very low amount of A␤  in the insoluble fraction. This is probably because the insoluble fraction does not contain fibrils and only consists of the Type III amorphous aggregate, which is present in both soluble and insoluble fractions (Fig. 4). However, given that the heme-associated soluble Type III A␤(1-42) species are noninflammatory (Fig. 5a), we would predict that the insoluble Type III A␤(1-42) species is likewise noninflammatory.
Overall, these data paint a complex and nuanced picture of the roles of heme and Hb in modulating the inflammatory activity of A␤(1-42) on astrocytes. 1) A␤(1-42) alone forms a soluble HMW Ͼ75-kDa species that is highly inflammatory. It also forms insoluble extended fibrils (Type I) and smaller oligomeric species that have minimal inflammatory properties. 2) Direct association of heme or Hb with the soluble HMW Ͼ75-kDa A␤(1-42) species reduces astrocyte cytokine expression.
3) The presence of heme and Hb changes the A␤(1-42) species found in the insoluble fraction. 4) The Type II "tangled fibrils" in the insoluble fraction produced by Hb are highly inflammatory.

Heme and hemoglobin modulate the PI3K/Akt pathway
We have established that heme and Hb exert their modulatory effects on astrocyte immune activity both by physically associating with A␤(1-42) (Figs. 3 and 4) and by a second mechanism independent of A␤(1-42) (Fig. 2). The latter suggests that heme and Hb have the capacity to impact A␤(1-42) clearance through its effects on immune signaling. To gain insight into how heme or Hb modulates immune signaling, we quantified phosphorylation of 11 phosphoproteins in the PI3K/ Akt signaling pathway. The PI3K/Akt pathway is of particular interest because of its known role in modulating autophagic clearance of A␤ in neurons (58) and because of its known ability to inhibit M1 macrophage polarization (59), which is linked to expression of scavengers receptors, including CD36 (60). Further, this pathway is important to astrocyte immune function, as it is involved in regulating astrocyte viability, migration, autophagy, and production of cytokines and inflammatory mediators (61)(62)(63)(64)(65).
Because phosphoprotein signaling occurs on a much faster time scale (on the order of minutes) than other phenotypic responses (66), we analyzed phosphoproteins from astrocytes conditioned with combinations of A␤(1-42), 50 nM heme, and 50 nM Hb at 5-and 15-min time points (Fig. S11, a and b). To simultaneously account for data from both time points, we concatenated the time point data and used D-PLSR analysis to identify signaling differences between conditions. Analyzing the effects of heme alone identified two axes of interest with respect to heme and A␤(1-42) (Fig. 6a). First, LV1 separated heme ϩ A␤  to the right with all other conditions to the left. Among other signals in the pathway, LV1 consisted of phospho-Akt, phospho-PTEN, and phospho-TCS2 at the 5-min time point as top correlates with the heme ϩ A␤(1-42) condition (Fig. 6b). The D-PLSR analysis also determined that both heme and heme ϩ A␤(1-42) were increased along LV2 (Fig. 6a), which consisted of phospho-mTOR at 15 min, and phospho-IRS1 at both 5-and 15-min time points as top correlates with heme or heme ϩ A␤(1-42) (Fig. 6c). Plotting all condition groups along LV1 revealed that heme ϩ A␤(1-42) was significantly different from heme alone (Fig. 6d), whereas plotting along LV2 revealed that treatment either with heme alone or with heme ϩ A␤(1-42) was significantly different from vehicle controls (Fig. 6e). Thus, these data indicate that heme can significantly shift signaling within the PI3K/Akt signaling pathway, which is modulated by A␤(1-42). Applying the same analysis to Hb revealed that Hb did not significantly mod- Table 1 A␤ species isolated and assayed for immunomodulation of astrocytes Shown is a summary of A␤(1-42) species isolated from centrifugation and SEC. Indicated are the isolation method and immunomodulatory effects of each isolated species on astrocytes as determined by Luminex-based cytokine profiling in Fig. 5 and Fig. S10. NA, not applicable. Heme and Hb suppress astrocyte inflammation ulate signaling within the pathway compared with either control or A␤(1-42) alone (Fig. S11, c-e).

Method of isolation
Because heme more substantially modulated PI3K/Akt signaling (Fig. 6) and CD36 expression (Fig. 2, d and e) than Hb, we next hypothesized that inhibition of the pathway would restore astrocyte scavenger activity. To test this, we used rapamycin, which inhibits signaling through mTOR, a central node within the PI3K/Akt pathway (Fig. 6g). Indeed, co-treatment of heme ϩ A␤(1-42) with 10 nM rapamycin yielded partial recovery of CD36 expression ( Fig. 6f and Fig. S12). Importantly, these signaling data reveal that 1) the PI3K/Akt pathway is stimulated by heme and 2) the PI3K/Akt pathway regulates astrocyte scavenger activity.

Discussion
Neuroinflammation is recognized as a key component of AD pathology (2). However, it is unclear whether glial activity promotes neurotoxicity (3) or is neuroprotective (4,5). In reality, the consequences of neuroinflammation occupy a continuum between neuroprotective (3) and neurodegenerative (4,5). In terms of protection against AD, glial activation is essential for clearance of cytotoxic A␤ species. In terms of aggravating AD pathogenesis, excessive neuroinflammation contributes to a number of detrimental effects, including "fatigued" glia that are unable to clear A␤, generation of toxic reactive oxygen species, and hyperactivated microglia that indiscriminately phagocytize neurons (11)(12)(13)(14)(15). Further complicating matters is the reality that different species of A␤ may affect neuroinflammation in very different ways. Moreover, the effects of each species may be modulated by other aspects of tissue pathology. Herein, we have identified heme and Hb as key AD-relevant immunomodulators and have probed mechanisms that mediate the inflammatory activation of astrocytes. Moreover, we have identified key A␤(1-42) species that are responsible for astrocyte activation and the effects of heme and Hb on the inflammatory potential of these species. Overall, our data indicate that heme and Hb suppress the A␤(1-42)-mediated inflammatory activation of astrocytes, suggesting that these factors contribute to AD pathogenesis by impairing A␤ clearance mechanisms.
A pathological hallmark of AD is a weakened blood-brain barrier (BBB) and a concomitant increase in blood and serum factors in the brain, including infiltrating red blood cells, heme, Hb, and haptoglobin (19,20,67). Because both heme and Hb have previously been shown to alter A␤ oligomerization (21,22) and macrophage immune activity (28,68), they have the potential to modulate AD pathogenesis. Although heme is known to have neurotoxic properties at high concentrations associated with hemorrhage (69), ours is the first study to evaluate the effects of heme and Hb on astrocyte immune activity at concentrations that are physiologically relevant outside of hemorrhage. Astrocytes are one of three glial cell types that take on varying immune and neuronal support functions (70). In addition to being essential regulators of neuronal metabolism, astrocytes possess vital immune functions and play essential roles in clearance of A␤ and regulation of microglial activity

Heme and Hb suppress astrocyte inflammation
during AD pathogenesis (71,72). Our integrated analysis of cytokines, signaling, scavenger activity, and biophysical analysis of heme/Hb association with A␤(1-42) reveals for the first time that Hb and heme strongly modulate astrocyte immune activity. Moreover, heme/Hb effects are imparted both by directly modulating astrocyte function and by physically binding to A␤(1-42). Simultaneous analysis of the relative expression of 32 cytokines provided us with a detailed view of astrocyte inflammatory response to A␤(1-42), heme, Hb, and heme/Hb-bound species of A␤(1-42). We began our study by applying A␤(1-42) and either heme or Hb to astrocyte cultures (Fig. 1a). Our multivariate analysis (Fig. 1, b and c) identified a composite cytokine variable that integrated all readings from each condition applied. Scoring each sample on this composite variable demonstrated that both heme and Hb reduced cytokine expression (Fig.  1e). From this analysis, A␤(1-42) conditioning strongly correlates with RANTES, GM-CSF, and IL-1␤, which were all down-regulated in cultures co-treated with heme or Hb (Fig. 1f). Of these, RANTES is a pro-inflammatory chemokine and involved in microglial recruitment (73,74), GM-CSF promotes microglial proliferation (75), and IL-1␤ is a highly pro-inflammatory cytokine up-regulated early in AD (76) that has been shown to promote A␤ clearance in a mouse model (77). Interestingly, application of heme or Hb at low concentration (50 nM) did not strongly suppress expression of other cytokines, including IP-10, KC, and MCP-1, which are all involved in immune cell recruitment (44,45,78). However, application of a high dose of heme (25 M) reduced these   or heme-only conditions. LV1 identifies upstream elements of the pathway, including p-PTEN, p-Akt, and p-TSC2, at 5 min as top correlates with the heme ϩ A␤(1-42) condition. c, LV2 depicts a linear combination of phosphoproteins at the 5-and 15-min time points that correlate with the heme and heme ϩ A␤(1-42) conditions or the vehicle control. LV2 identifies p-mTOR at 15 min and p-IRS at 15 and 5 min as top correlates with both heme conditions. d, plotting LV1 scores for each group shows that the LV1 profile significantly segregates the heme ϩ A␤(1-42) signaling effects from heme-only signaling effects (n ϭ 3, p ϭ 0.0022; heme versus heme ϩ A␤ ). e, plotting LV2 scores for each group shows that the LV2 profile significantly segregates all heme conditions from the vehicle control (n ϭ 3, p ϭ 0.0125; vehicle versus heme, p ϭ 0.0275; vehicle versus heme ϩ A␤(1-42)). f, CD36 expression in the presence of A␤(1-42) and heme, quantified by Western blotting (Fig. S12) cytokines as well (Fig. S3, a and b), suggesting that multiple mechanisms are associated with heme/Hb suppression of inflammatory response. We note that whereas all of the cytokines modulated by heme and Hb are well-established to modulate immune activity, they have not generally been found to be neurotoxic.
Dual mechanisms of heme and Hb immunomodulatory activity (effects on astrocyte immune signaling and effects produced by binding to A␤(1-42)) represent one possible explanation for why low doses of heme/Hb suppress certain cytokines and not others, whereas high doses of heme suppress cytokines globally. Evidence for heme and Hb down-regulating astrocyte inflammatory activity through physically modifying A␤(1-42) stems from our finding that heme and Hb bind to a particularly inflammatory species of A␤ . Indeed, our SEC analysis revealed that a soluble HMW oligomeric species of A␤(1-42) (Ͼ75 kDa) produced the principal inflammatory response compared with other fractions (Figs. 4 and 5 and Fig. S10). Moreover, the A␤(1-42) HMW soluble oligomer was the only species that was verified to be associated with heme or Hb and strongly suppressed cytokine expression compared with the unbound oligomer. Further, this result is particularly important because it provides a physiological context to previously described A␤heme and A␤-Hb interactions, as soluble A␤ oligomers are abundant in human AD brain tissue and have high binding affinities and neurotoxic properties (57, 79).
On the other hand, whereas fibrillar A␤(1-42) is noninflammatory, the effect of Hb on altering A␤(1-42) fibril morphology contributes to the formation of a highly inflammatory "tangled" fibrillar species (Fig. 4a and Fig. S10). Thus, whereas Hb renders a soluble oligomer noninflammatory, it acts to increase the inflammatory potential of an insoluble fibril. Altogether, the isolation of distinct A␤(1-42) species and the effect of heme or Hb on their distribution and inflammatory activation of astrocytes highlight the complex and nuanced nature of A␤-mediated immune signaling. Indeed, the observation of changes in neuroinflammation over the course of AD may reflect the competing effects of multiple A␤, A␤-heme, and/or A␤-Hb species that have differing immunomodulatory activities. Additionally, the peroxidase activity of heme-A␤ complexes may further act to modulate the inflammatory response (54).
A second mechanism of heme/Hb control of inflammatory activity is via direct action on astrocyte signaling, which is supported by our observations that heme and Hb suppressed microparticle internalization, phagocytosis, and CD36 expression (Fig. 2) in experiments free of A␤ . Our results in astrocytes generally point toward heme as having anti-inflammatory effects at both low and high heme concentrations. Moreover, we found that scavenger activity of astrocytes was inhibited by low (50 nM) heme concentrations and that both low and high (25 M) heme concentrations inhibited scavenger activity of SIM-A9 microglia ( Fig. 2 and Fig. S13). Nevertheless, the canonical role of heme in immune signaling, primarily delineated in macrophages and endothelial cells, is that it stimulates inflammation via TLR4 (26). In contrast, our astrocyte data indicate that heme or Hb reduce the inflammatory response and stimulate phosphorylation of multiple signaling molecules in the PI3K/Akt pathway. In addition to regulating metabolism, this pathway regulates autophagy and is a known modulator of inflammation and phagocytosis (60,80,81). Additionally, the mTOR inhibitor rapamycin partially recovered expression of CD36 ( Fig. 6f and Fig. S12), defining a novel and causal role for the PI3K/Akt pathway in heme signaling and suggesting that inhibition of this pathway has potential therapeutic efficacy for promoting A␤ clearance.
How are heme signals integrated to control immune activity? Extracellular "free" heme can be internalized by heme transporter HRG-1, and Hb may be internalized by the Hb receptor CD163, although the latter is primarily expressed in the brain by microglia (82)(83)(84). Once in the cell, free heme or Hb-derived heme can be catabolized into CO, biliverdin, and bilirubin, which all possess anti-inflammatory properties (85)(86)(87)(88). Importantly, CO is known to activate the Akt pathway, providing a plausible explanation for heme-mediated immune signaling (89). An alternative mode of heme signaling may involve heme binding to a number of heme-regulated transcription factors, including p53, Bach1, and Rev-erb-␣/␤, which control genes important for immune function (90 -92). Given that it is increasingly recognized that heme is a dynamic and mobile molecule important for a number of signaling pathways (36,(93)(94)(95), future work will involve elucidating the targets of heme signaling during inflammation.
What is the physiologically relevant concentration range and source of bioavailable heme and Hb in human control and AD brains? These unresolved questions would dictate the extent to which heme or Hb association with A␤ or heme-mediated immune signaling would occur in vivo. In the extreme case of hemolysis and hemorrhage, heme and Hb have been estimated to be in the 10 -100 M regime (35). Given that estimates of brain [A␤] span the pico-to nanomolar range (96,97) and the relatively tight heme-A␤ and heme-Hb interactions, K D Ͻ 100 nM and K D ϭ 350 nM (Fig. 3c), it is likely that a significant fraction of A␤ is associated with heme and Hb in the AD microenvironment. An obvious source of heme and Hb during AD pathogenesis is from the vasculature because a weakened BBB is associated with AD. However, given the recent discovery of heme exporters (98,99), an intriguing alternative possibility is that astrocytes and/or other brain cells modulate neuroinflammation via control of extracellular heme export. The recent development of fluorescence-based (36, 100, 101) and activity-based (95) heme sensors will be critical for elucidating brain heme homeostasis and the absolute concentrations of heme within cells and in the extracellular space in Alzheimer's disease.
We close by noting that changes in heme and Hb represent only one aspect of a complex mosaic of factors in brain pathophysiology that affects A␤ aggregation and brain immune activity during Alzheimer's disease. Indeed, many factors with the potential to modulate both A␤ aggregation and glial immune activity have been identified within A␤ plaques, including proteoglycans, cytokines, metals, apolipoprotein E, and proteases, among others (102). Nevertheless, there has been limited characterization of the effects of these molecules on A␤ speciation and on astrocyte and microglial immune activity. In this work, we have established an integrated methodology to elucidate the individual effects of particular AD-relevant molecules on astrocyte immune activity via 1) direct effects on astrocyte immune Heme and Hb suppress astrocyte inflammation signaling and 2) effects due to physical association with A␤ or modulation of A␤ speciation. Our approach is readily generalizable and may yield broad new insights into the mechanisms promoting immune dysfunction in AD.
Altogether, our data indicate that Hb and heme are potent modulators of astrocyte immune activity by dual mechanisms. The first is by direct signaling to astrocytes, mediated at least in part by the PI3K/Akt pathway. The second is by physical association with a highly inflammatory A␤(1-42) oligomer, whereby heme or Hb suppresses this inflammatory behavior. Given reports of increased Hb in late-stage AD and in transgenic mouse models, Hb and heme signaling and physical activity represent possible mechanisms responsible for astrocyte fatigue in AD tissues, thereby permitting amyloid pathogenesis. Additionally, recent findings of BBB leakage early in AD suggest that Hb concentration may be locally increased at the vascular wall (103). By extension, Hb and heme activity may be responsible for a high prevalence (ϳ90%) of cerebral amyloid angiopathy in AD patients (103). The recent report of heme-specific single-domain antibodies may represent a new therapeutic strategy to limit heme availability to A␤ (35). Further, our observation that rapamycin was able to partially restore astrocyte immune activity suggests that intervening in Hb/heme signaling represents a promising therapeutic strategy for AD. More broadly, our approach establishes a rigorous methodology to interrogate the immunomodulatory effects of diverse proteins and other molecules that co-localize or associate with A␤.

Primary mouse astrocyte cultures
Astrocyte cultures were derived from postnatal day 0 -1 CD1 mice (Charles River Laboratories) under a protocol approved by the Georgia Institute of Technology Institutional Animal Care and Use Committee. Cortices were isolated following an existing protocol (104) and triturated in plating medium with a 1-ml sterile pipette tip. Plating medium consisted of minimum essential medium (Thermo Fisher Scientific) with 10% horse serum (Sigma), 1% antibiotic/antimycotic solution (Sigma), and 0.3% glucose solution (Sigma). Cells were left to attach overnight to T-75 flasks coated in 0.1 mg/ml poly-D-lysine (Sigma). After 24 h, flasks were knocked to remove debris and rinsed with PBS, and plating medium was replaced with astrocyte medium (ScienCell) with 2% fetal bovine serum (ScienCell), 1% penicillin/streptomycin solution (ScienCell), and 1% astrocyte growth serum (ScienCell), in which cultures were maintained for up to four passages for conditioning. Cultures were maintained in a 37°C, 5% CO 2 humidified incubator.

Cell conditioning and lysis
For cytokine expression, phosphoprotein signaling, and Western blot analyses, primary astrocytes were plated in 6-well plates and conditioned with combinations of hemin chloride (50 nM; EMD Millipore), human hemoglobin (50 nM; Sigma), rapamycin (10 nM; Selleck Chemicals, Houston, TX), and A␤(1-42) (50 nM; rPeptide) in 1% (w/v) NH 4 OH. For CD36 analysis, cells co-conditioned with rapamycin were first preconditioned with rapamycin for 1 h. Conditions were applied at 75% confluence for 24 h, after which conditioning medium was collected for cytokine analysis, and cell lysates were collected for phosphoprotein signaling and Western blotting analyses using the Bio-Plex cell lysis kit (Bio-Rad), with the addition of one cOmplete mini protease inhibitor tablet (Roche, Basel, Switzerland) and 20 l of phenylmethylsulfonyl fluoride (Sigma) per 5 ml of lysis buffer. Lysates were placed in microcentrifuge tubes and inverted at 4°C for 10 min. Lysates and medium were centrifuged at 4°C for 10 min at 13,200 rpm, and supernatant was collected and stored at Ϫ80°C until analysis.

Multiplexed phosphoprotein and cytokine signaling analysis
For phosphoprotein signaling analysis, cell lysates were thawed on ice and centrifuged at 4°C for 10 min at 13,200 rpm. Protein concentrations were determined using a Pierce BCA protein assay (Thermo Fisher Scientific) and normalized with Milliplex MAP assay buffer (EMD Millipore) to 2 g of protein/25 l for Akt/mTor pathway analysis or 1 g of protein/25 l for mitogen-activated protein kinase pathway analysis. These protein concentrations were selected because they fell within the linear range of bead fluorescence intensity versus protein concentration for detectable analytes. Multiplexed phosphoprotein analysis was conducted for the Akt/mTOR pathway by adapting the protocols provided for the Milliplex MAP Akt/ mTOR 11-Plex (p-Akt, p-GSK3␣/␤, p-IGF1R, p-IR, p-IRS1, p-mTOR, p-p70S6K, p-PTEN, p-RPS6, and p-TSC2) and phosphoprotein magnetic bead kits (EMD Millipore).

Western blot analysis
Cell lysates, obtained as described above, were thawed on ice and then centrifuged for 10 min at 10,000,000 rpm and 4°C. Protein concentration was determined using a Pierce BCA protein assay and equal amounts of protein were dissolved in reducing sample buffer, boiled, and loaded onto SDS-polyacrylamide gels. Following separation by electrophoresis, proteins were transferred to a Hybond P 0.45-m polyvinylidene fluoride membrane (GE Healthcare). Membranes were blocked at room temperature (RT) for 1 h with 5% milk in TBS containing 0.01% Tween 20. Membranes were probed at 4°C overnight with rabbit anti-CD36 (1:500; Novus Biologicals, Littleton, CO) and mouse anti-␣-tubulin (1:2000; Sigma). Membranes were then incubated with Alexa Fluor-conjugated secondary antibodies (1:2000; Thermo Fisher Scientific) for 2 h at RT. Imaging of blots was performed using an Odyssey CLx imager (LI-COR Biosciences, Lincoln, NE). Protein quantification was performed using Image Studio Lite version 5.2 (LI-COR Biosciences).
Confocal microscopy was performed on a Zeiss LSM 700 laser-scanning inverted microscope to obtain 15-30 optical sections with 1-m interval thickness. Orthogonal projections were rendered using Zen version 2.3 software (Zeiss, Oberkochen, Germany).

E. coli particle internalization assay
Primary astrocytes or SIM-A9 microglia were plated on 96-well plates at a density of 10,000 cells/well and left to adhere overnight in a 37°C, 5% CO 2 humidified incubator. Cells were treated with either control, 50 nM hemin chloride, or 50 nM human hemoglobin conditions for 4 h. Conditioning medium was aspirated, and cells were incubated with the E. coli fluorescent BioParticle suspension from the Vybrant TM phagocytosis assay kit (Thermo Fisher Scientific) for 1 h. Extracellular fluorescence was quenched with trypan blue. Fluorescence was read on a SpectraMax M3 microplate reader ( ex ϭ 480 nm; em ϭ 520 nm) (Molecular Devices, Sunnyvale, CA).

Phagocytosis assay
Primary astrocytes or SIM-A9 microglia were plated in 0.1 mg/ml poly-D-lysine-treated half-area 96-well, glass-bottom plates at a density of 10,000 cells/well and left to adhere overnight in a 37°C, 5% CO 2 humidified incubator. Cells were treated with either control, 50 nM hemin chloride, or 50 nM human hemoglobin conditions for 24 h. Conditioning medium was aspirated, and cells were incubated with a fluorescent pHrodo TM Red Zymosan BioParticle suspension (Thermo Fisher Scientific) diluted in astrocyte medium for 2 h. After removing BioParticle suspension, cells were fixed with 4% paraformaldehyde, permeabilized for 10 min at RT with 0.1% Triton X-100, and blocked with a 5% BSA, 3% goat serum (Sigma) solution for 1 h. Primary antibody incubation was performed overnight at 4°C, with rabbit anti-GFAP (1:1000; Novus Biologicals). After washing with wash buffer, fixed cells were incubated with Alexa Fluor 488 goat anti-mouse secondary antibody (1:200; Thermo Fisher Scientific) for 2 h at RT. Cells were co-stained with DAPI (1 ng/ml; Thermo Fisher Scientific) for nuclei. Fluorescence microscopy was performed on a Zeiss Axio Observer Z.1 inverted microscope and quantified using ImageJ.

Fluorescence assay for heme quantification
Heme concentration in pellet fractions was determined by a fluorescence-based assay (105). Briefly, samples were boiled in the presence of 1 M oxalic acid to remove iron from the protoporphyrin ring of heme. Fluorescence of protoporphyrin IX was measured with excitation at 400 nm and emission at 662 nm on a Tecan Infinite 200 Pro plate reader. Heme concentration was determined by comparison with serial dilutions of heme standards quantified by UV-visible spectroscopy using the extinction coefficient of aqueous heme at 612 nm of 4431 cm Ϫ1 M Ϫ1 (36) on a Cary 60 UV-visible spectrophotometer.

UV-visible spectroscopy
For heme-A␤(1-42) binding, freshly dissolved A␤  was diluted to 100 M in PBS and titrated into 500 nM heme in PBS. The UV-visible spectral changes were monitored on a Cary 60 UV-visible spectrophotometer. The heme Soret band at 395 nm was plotted versus A␤(1-42) concentration and fit with the two-site binding model described in the supporting material to determine the apparent dissociation constant. For Hb-A␤(1-42) binding, freshly dissolved A␤(1-42) was diluted to 200 M in PBS and titrated into 500 nM Hb. The heme Soret band of Hb at 412 nm was plotted versus A␤(1-42) concentration and fit with the one-site binding model described in the supporting material to determine apparent dissociation constant. Before each scan, the samples were allowed to equilibrate Heme and Hb suppress astrocyte inflammation for 5 min, consistent with other reports indicating that heme binding to A␤(1-42) occurs within 300 s (52,54).

CD spectroscopy
A␤(1-42) was freshly dissolved from frozen stocks at 20 M in the presence or absence of 20 M heme or 5 M Hb in 1ϫ PBS. Spectra were measured on a Jasco J-815 CD spectropolarimeter. Spectra were prepared from an average of 70 scans from 300 to 190 nm with 1-nm steps at 200 nm/min.

Size-exclusion chromatography
For SEC analysis, A␤ was diluted to 100 M in 1ϫ PBS with or without 200 M heme or 25 M Hb. The A␤ solutions were incubated at 37°C for 16 -18 h without agitation. To separate soluble and insoluble species, A␤ samples were pelleted at 21,100 ϫ g for 5 min to remove insoluble fibrillar material. The pellet was resuspended in 100 l of 1ϫ PBS for subsequent TEM analysis. The supernatant was collected and chromatographed over a Superdex 10/300 GL size-exclusion column on an Agilent 1260 Infinity HPLC with an in-line photodiode array detector. Elution of A␤ was monitored by reading the absorbance at 220 and 280 nm, whereas elution of heme associated species was monitored by reading absorbance at 400 nm. The ratio of heme to A␤ in SEC fractions was determined by a fluorescence-based assay for heme detection (105). Quantification of A␤ in SEC or pellet fractions was accomplished by UV-visible spectroscopy using the peptide aromatic absorbance at 220 nm relative to serial dilutions of an A␤ stock solution. The ratio of A␤ and Hb in mixed A␤/Hb SEC fractions was determined by immunoblotting for A␤ or Hb using 6E10 (1:5000; BioLegend) and H4890 (1:5000; Sigma) antibodies, respectively. Absorbance at 220 and 408 nm was then used to determine A␤ and Hb concentrations, respectively.

Transmission electron microscopy
TEM images were taken on a Jeol 100 CXII transmission electron microscope operating at 100 kV. A␤ SEC and pellet samples of 25 or 100 M A␤(1-42) alone or with 2ϫ heme and 100 M A␤(1-42) with 25 M Hb were stained using 1% uranyl acetate on 400-mesh continuous Formvar-coated grids (Ted Pella, Redding, CA). Briefly, 2 l of sample was placed on the grid for 1 min, and excess liquid was blotted away with filter paper. Next, uranyl acetate was added for 45 s, and excess liquid was again removed by blotting with filter paper before the grid was allowed to air-dry. Grids were then stored in a desiccator. Images were collected from across each grid. The distribution in morphology of various A␤ species was scored by counting species types in ϳ150 -200-m 2 areas for the pellet and ϳ100 m 2 for the SEC fractions in two independent experimental trials.

Thioflavin T assays
A␤ was freshly dissolved from frozen 1 mg/ml 1% NH 4 OHtreated aliquots to 110 M stock solution in 1ϫ PBS for ThT assays. A final concentration of 20 M ThT and 6 M A␤ was used to test for aggregation in the presence of heme or Hb. ThT fluorescence was monitored for 24 h at 37°C on a Tecan Infinite 200 Pro plate reader with excitation at 450 nm and 9-nm bandwidth and emission at 482 nm with a 20-nm bandwidth. For the fibril de-aggregation assay, A␤ at 110 M in 1ϫ PBS was allowed to aggregate into fibrils overnight at room temperature without agitation. At each time point, fibrils were then diluted to 6 M in the presence or absence of heme and placed at 37°C. 10 l of each sample was added to 200 l of 20 M ThT. ThT fluorescence was monitored at various time points on a Biotek Synergy Mx with 5-s shaking before measuring fluorescence at 435-nm excitation and 9-nm bandwidth and 486-nm emission with 9-nm bandwidth, as was reported previously for an A␤ and heme ThT assay (54).

Partial least-squares regression
D-PLSR analysis was performed in MATLAB using the partial least squares algorithm by Cleiton Nunes available on the Mathworks File Exchange. All data were z-scored before inputting into the algorithm. For all analyses, an orthogonal rotation in the LV1-LV2 plane was performed to identify LVs that best separated conditions.

Statistics
All statistical analyses were performed using GraphPad Prism version 7 (GraphPad Software, La Jolla, CA). Values are presented as mean Ϯ S.E. Statistical significance was determined, as appropriate, using Student's t test, ordinary one-way analysis of variance (ANOVA) followed by Dunnett's or Sidak's post hoc test, or Kruskal-Wallis ANOVA followed by Dunn's post hoc test. Normality of data was tested using the Shapiro-Wilk test of normality. Levels of significance were set as follows: *, p Ͻ 0.05; **, p Ͻ 0.01; ****, p Ͻ 0.0001.