Antiparallel structure in vascular aggregates

In this study, a total of 80 vascular amyloid deposits from one normal and four diseased mid-frontal brain tissue specimens corresponding to different stages of CAA severity were analyzed: 20 blood vessels from a normal, non-CAA patient, 20 blood vessels from a mild CAA patient, 10 vessels from a different mild CAA patient, 20 vessels from a moderate/severe CAA patient and 10 vessels from a different moderate/severe CAA patient. The details of the samples are provided in Table S1. Typically, CAA severity is graded according to the degree of amyloid deposition in the blood vessel walls and resulting pathologic features. In mild CAA, blood vessels show focal limited amyloid deposition without vascular damage, whereas moderate and severe CAA exhibit more pronounced circumferential amyloid presence with associated vascular damage, particularly for the latter [12]. Amyloid positive blood vessels were identified using immunohistochemical (IHC) staining and grouped into two categories: mild CAA and moderate/severe CAA based on the extent of amyloid deposition. Figure 1 shows a representative example of each IHC stained vascular aggregate type. Additional images are shown in the Supporting Information (Figures S1–S2). From a molecular perspective, it is well known that Aβ fibrils have an ordered parallel β-sheet structural motif, while prefibrillar aggregates can have a comparatively disordered structure. Both parallel and antiparallel β-sheets have unique vibrational signatures in amide I region of mid-infrared spectra, arising from excitonic coupling between the backbone amide moieties. Both exhibit a characteristic band at 1625–1635 cm−1, while antiparallel β structures exhibit an additional band at around 1690 cm−1[8, 9]. Figure 1e–h show O-PTIR images at 1628 cm−1 and 1692 cm−1 normalized to the spectral intensity at 1660 cm−1, reflective of overall β-sheet and antiparallel β-sheet populations respectively, for the blood vessels in Fig. 1a, b. An unstained, parallel tissue section was used for acquisition of IR data to avoid spectral interferences from the stain. This is a commonly used approach in IR imaging [4, 11] and we have previously used this strategy in our studies targeting Aβ plaques [17, 21, 33]. We observe increased O-PTIR intensity from the blood vessel walls for moderate/severe CAA, consistent with elevated presence of β-sheet rich amyloid aggregates. Interestingly, we also observe signatures for antiparallel β-structures from the blood vessel indicating a complex structural ensemble different from typical parallel cross-β fibrils. In comparison, the blood vessel corresponding to mild CAA does not exhibit significantly enhanced parallel or antiparallel β-sheet signatures. Of course, this does not indicate how generalizable this observation is and if such transient structures persist in other vascular aggregates as well. However, expanding the discrete frequency imaging approach above to a statistically relevant number of blood vessels is challenging for two reasons. First, while normalized IR intensities are reflective of relative populations of structural motifs such as β-sheets, they cannot be used to quantitatively assess the overall secondary structure of protein aggregates and changes therein. Furthermore, a discrete frequency approach essentially requires preempting the bands of interest, which is not always viable. Acquisition of the entire amide I spectra and subsequent spectral deconvolution represents a more quantitative alternative, which, however, significantly increases the experimental time. To mitigate this, we therefore employed a spatially resolved spectroscopic approach, wherein full spectra of specifically the blood vessel walls were acquired, as indicated in Fig. 2a (green dashed circle). This offers the same spectral insights into the composition of amyloid rich blood vessels. For each blood vessel, 10–30 spectra were acquired, depending on its size. Representative spectra from different spatial locations along the blood vessel are shown in Fig. 2b; the mean spectra of all blood vessels corresponding to mild and moderate/severe CAA thus acquired are shown in Fig. 2c. Mean spectra acquired from control blood vessels that are amyloid negative are also shown for comparison. Additional individual spectra from blood vessels and their corresponding spatial locations are shown in the Supporting Information (Figures S3-S7). An important aspect to note in this context is the spectral variation within individual blood vessels. Similar observations have been made for amyloid plaques in AD [17, 61]. In studies on vascular amyloids in CAA, data are typically grouped by blood vessels corresponding to a specific pathologic state/type; for example, in brain derived aggregation experiments, protein lysates from one or more blood vessels are used as seeds [37]. Data are rarely analyzed without regard to vessel identity or pathology. Vessel or histologic level grouping is necessary to preserve spatial and pathologic context, which is critical for interpretation. We have therefore used the mean spectra of blood vessels for further analysis in this work, which also allows for circumventing effects from their size variations. We note that the spectral heterogeneity within a single blood vessel can potentially offer additional insights into disease pathogenesis, which we aim to address in detail in future work. We observe that the mean spectra of vascular amyloids corresponding to moderate/severe CAA exhibit significantly enhanced intensity at ~ 1632 cm−1 compared to normal blood vessels, indicating an increase in β-sheet abundance. The spectra also show a distinct peak at ~ 1736 cm−1. This absorption is characteristic of carboxylic acids and ester functional groups, and in the context of cellular and tissue imaging, is attributed to lipid moieties [4, 46]. The spectra thus provide evidence supporting a concurrent increase in lipids along with β-sheet motifs, which is a unique finding and highlights the possibility of a lipid driven aggregation mechanism underlying the pathologic changes in CAA. In addition to the above, we also note an increase in the spectral intensity at ~ 1686 cm−1, which can be assigned to antiparallel β structures. In comparison, vascular aggregates from mild CAA exhibit a relatively smaller increase for both the overall and antiparallel β-sheet populations. An increase in the lipid band is also observed, compared to normal blood vessels. Taken together, the spectra indicate an overall increase in β-sheet population with increasing CAA severity, potentially correlated to lipids, and also raise the possibility of concurrent presence of antiparallel β-structure.

Fig. 1

a, b Brightfield optical images IHC stained vascular amyloid deposits from a severe and a mild CAA case. c, d DC image of vascular amyloid deposits from (a) and (b). e, f IR ratio images (I1628/I1660) highlighting the presence of overall β-sheet content. g, h IR ratio images (I1692/I1660) reflective of antiparallel β-sheet populations in the vascular deposits

Fig. 2

a Schematic representation of the spatially resolved spectroscopic approach used for vascular amyloid characterization. The dashed circle represents the locations along the blood vessel wall from which spectra are acquired. b Representative IR spectra from different spatial locations on the blood vessel shown in (a). The locations are indicated by circles that are color coded to respective spectra. c Mean IR spectra of vascular aggregates from control, mild and moderate/severe CAA

Antiparallel β-structure in CAA is correlated to lipids

The amide I band in proteins is a convolution of multiple underlying components, and as a result, the spectral intensities are not quantitative reflections of the relative populations of different structural motifs. It is thus difficult to unambiguously establish the above conclusions without further analysis. Therefore, to gain further insights into the secondary structure of the vascular amyloids and their evolution with disease severity, we performed spectral deconvolution using Multivariate Curve Resolution- Alternating Least Squares (MCR-ALS). The gold standard in IR spectral deconvolution is band fitting; however, it requires a priori knowledge of the spectral components, such as frequency and linewidth. MCR-ALS is a matrix factorization approach that enables blind deconvolution of spectral data without the need for information of the pure spectra [19, 20, 62]. By iteratively applying alternating least squares optimization with constraints such as non-negativity and unimodality, MCR-ALS decomposes the data matrix into pure spectral profiles and their corresponding concentration profiles, similar to global fitting of spectra. This allows for quantitative identification of individual constituents and tracking their evolution across datasets, making it an ideal tool for assessing the evolution of secondary structure in CAA, as presented above. The results from MCR-ALS are presented in Fig. 3a. The spectra were grouped by blood vessel for this analysis to mitigate any potential effects from size variations. We find that the spectra can be adequately described as a weighted sum of four bands centered at 1628 cm−1, 1656 cm−1, 1690 cm−1 and 1732 cm−1, attributable to β-sheets, random coils/disordered structures, antiparallel β-sheets and lipids, respectively. The second derivatives of the mean spectra of the blood vessels (shown in Fig. 2a) further validate these components indeed constitute the spectra from different CAA stages. The second derivative spectra are shown in Fig. S8. The weights of each component from the different blood vessel groups are shown in Fig. 3b. The trends, as determined from MCR-ALS, for each of the spectral components, across the vascular aggregates from different CAA stages are shown in Fig. 3c–f. We observe that the overall β-sheet increases in mild and moderate/severe CAA compared to control. The exact opposite trend is observed for the random coil/disordered population. Both antiparallel β-sheets and lipids also exhibit a trend similar to overall β-structure, increasing steadily from control to mild and moderate/severe CAA. We note the relative changes in secondary structure populations are small, which is somewhat expected in complex tissue environments, where other proteins can also contribute to the overall spectrum. To verify if our observations are statistically significant, we performed Welch’s ANOVA followed by a Games-Howell post-hoc test for each structural component. The p values, listed in the Supporting Information (Table S2), indicate that there exists statistically significant difference in overall and antiparallel β-sheets and lipids between normal blood vessels and those from moderate/severe CAA. The overall populations for mild CAA are not always significantly different from those observed for the control and moderate/severe case, which likely arises due to a relatively low abundance of Aβ aggregates. We address the implications of potential contribution from non-Aβ protein components to the spectra later in the manuscript. Taken together, the above trends suggest a direct relationship between abundance of β-sheet structures, parallel and antiparallel, and lipids, which in turn implies a correlation between the respective weights or populations of these moieties. The correlation of both overall and antiparallel β-sheet weights with those of lipids, as obtained from MCR-ALS are shown in Fig. 4a–b. We find that lipids correlate with presence of antiparallel β-structures, but not with the overall β-sheet population. This indicates that a lipid mediated aggregation pathway specifically contributes to formation of transient, antiparallel structures. The overall β-sheet abundance is not strongly correlated to lipids likely because of presence of both parallel and antiparallel species, where only the latter is connected to lipids. To ensure that these trends are not dominated by individual spectra from a few vascular aggregates, we extended the MCR-ALS analysis to all spectra acquired from the 80 blood vessels studied. The results, shown in Fig. S9, are consistent with the findings from the analysis of mean spectra. We find that individual spectra can also be deconvoluted into similar basis constituents, and the weights/populations of the overall and antiparallel β-sheet components exhibit similar correlations with the lipid populations. This provides further validation that the conclusions drawn herein are not misguided by a handful of spectral outliers and represent a consistent trend in the composition of vascular amyloids.

Fig. 3

a Spectral deconvolution of the amide I IR spectra showing its individual components. Overall β-sheet (purple), disordered structures (green), antiparallel β-sheet (orange) and lipids (red). b Contributions of each spectral component determined from deconvolution across blood vessels from control, mild and moderate/severe CAA. c–f The relative change in the populations of each spectral component, shown in (b). The values have been normalized to that of the control, for clarity

Fig. 4

The Correlations of lipid populations with β-sheets, as determined from spectral deconvolution. a Scatter plot of spectral weights of overall β-sheets versus lipid content. b Scatter plot of spectral weights of antiparallel β-sheets versus lipid, exhibiting a more linear correlation, which implies that presence of antiparallel β-sheet structures depends on lipids. The colors correspond to spectral data from different patients with mild (blue, green) and moderate/severe CAA (orange, red). The dashed lines represent fits to a linear regression model. The corresponding correlation coefficients are indicated on the plots

The severity of CAA is linked to increase in fibrillar amyloid aggregates in cerebral vasculature, leading to fragmentation of the blood vessel walls and hemorrhage [12, 28]. Our results are consistent with this view: we find gradual increase in the overall β-sheet population, characteristic of amyloid fibrils, going from normal blood vessels to those corresponding to mild and moderate/severe CAA. The presence of antiparallel structures, however, is unexpected. It is well known that Aβ fibrils have a parallel cross β arrangement, whereas oligomers can have an antiparallel structure. The antiparallel structure, however, has also been demonstrated to exist in Aβ mutants, where transient fibrillar intermediates can have this structural motif, while mature fibrils are parallel [7, 57]. More recently, the presence of antiparallel structure in wild-type Aβ40, the isoform most relevant to CAA, has also been evidenced [5]. The increase in antiparallel population in moderate/severe CAA can thus be attributed to increased presence of either antiparallel fibrils or oligomers, or both. It should be noted that our findings do not point to presence of exclusively antiparallel fibrils in severe/moderate CAA, and it likely that a mix of parallel and antiparallel fibrils exist in these vascular deposits. Furthermore, the presence of only antiparallel structures would lead to an identical evolution for overall β-sheet population between the specimens; we observe that the trends of overall and antiparallel β-sheets, while similar, are not identical. However, presence of antiparallel structures is inconsistent with recent findings from cryo-EM, which have unequivocally demonstrated that the fibrillar polymorphs associated with CAA have a parallel cross-β motif [43, 75]. It is important to note in this text that cryo-EM studies of brain derived fibrils usually do not focus on fibrillar aggregates from individual blood vessels or plaques. Furthermore, it is not fully understood how different polymorphs isolated from diseased brains propagate relative to each other in seeded growth. Therefore, it is possible that the overall conformational ensemble is largely dominated by parallel fibrils, with the antiparallel aggregates representing a smaller subset of structures. Recent studies on Aβ mutants from mouse models reveal a similar finding, where a small fraction of fibrillar polymorphs exhibited an antiparallel arrangement [76]. Our results underscore the critical significance of obtaining structural information about amyloid aggregates directly from vascular aggregates to complement seeded growth from brain-derived fibrils. Recent studies have revealed similar findings for Aβ plaques, where presence of antiparallel structures in a subset of plaque cores has been demonstrated [33]. Interestingly, in recent work, Irizarry et al. has used IR and Nuclear Magnetic Resonance (NMR) spectroscopy to show that CAA-derived fibrils can have both parallel and antiparallel architecture [37]. This agrees with our findings and supports the view that antiparallel fibrils can be concurrently present along with the expected parallel structure in vascular amyloids. Another explanation of the discordance between our results and cryoEM is that the antiparallel structures observed here arise from oligomers and are hence not observed in brain derived fibrils investigated in cryoEM. However, there are no reports that have identified significant accumulation of oligomeric species in blood vessels with CAA progression. Hence, we disregard this possibility and consider the alternative: the antiparallel structure arises from fibrillar aggregates. This leads to the question: what leads to antiparallel fibrils? The other intriguing insight from this work is the potential role of lipids in modulating the structural distribution of aggregates. We find that the relative abundance of lipids in blood vessels correlates specifically to antiparallel β-sheets and not to the overall β-sheet population, which indicates that a. vascular amyloids have a mixed structural ensemble comprised of both parallel and antiparallel aggregates, and b. the formation of the latter may be connected to or mediated by interaction with lipids. The role of lipids in accelerating and modulating the amyloid aggregation pathway is well known [52, 63]. Charged lipids surfaces can aid in the nucleation of monomers through electrostatic interactions, thus accelerating the formation of Aβ aggregates and fibrils. Lipid interactions can also influence the structure of Aβ aggregates. While often promoting the formation of the typical parallel β-sheet fibrils, lipids can stabilize distinct oligomeric intermediates, some potentially possessing antiparallel character, which are strongly implicated in membrane disruption and cellular toxicity [44, 79]. Amyloid plaques are known to contain lipids colocalized with Aβ aggregates, which further highlights their potential role in altering amyloid aggregation pathways in disease [23, 46]. Recent studies using cryo-EM have demonstrated that Aβ40 forms predominantly parallel fibrils in-vitro in presence of 1,2-dimyristoyl-sn-glycero-3-phosphoglycerol (DMPG) vesicles [25]. Interestingly, the fibrillar arrangement identified in this work closely resembles that reported by Ghosh and coworkers, who identified an additional outer cross-β layers with antiparallel arrangements in brain-derived fibrils [27]. The surface charge and headgroup chemistry of lipids can differentially affect protein–lipid interactions, which in turn can modulate the aggregation pathway and the secondary structure distribution in the resulting aggregates. Zhaliazka and coworkers have provided a more nuanced view of such potential interactions of different lipids with Aβ and have shown that while most lipids promote higher parallel β-sheet populations, some lipid compositions can lead to an increase in the antiparallel character [80, 82]. In particular, the β-sheet character can be distinctly different when Aβ interacts with charged lipids such as cardiolipin and phosphatidylcholine and neutral lipids such as cholesterol, with the former generally favoring both parallel and antiparallel structures. This suggests that Aβ, when interacting with a heterogeneous mixture of lipids that likely prevail in the brain, could form a mixture of parallel or antiparallel fibrils, or fibrils with mixed β-sheet character. It is critical to note here that the ex-vivo tissue spectra reported do not demonstrate causality but rather indicate a correlation between the lipid abundance and antiparallel β-character in vascular amyloids. Identifying the precise origins of this correlation necessitates additional studies that can accurately pinpoint the molecular identities of the lipids and proteins in different blood vessels. We hope to address this in future work.

Aβ aggregates in parenchyma are structurally different

The common role of Aβ in both CAA and AD has led to the hypothesis that the pathologies of the two diseases can be interlinked. A well-known pathologic signature of CAA is the prevalence of dyshoric changes: the spread of Aβ aggregates from blood vessel walls into the surrounding tissue [12, 60, 66]. Aβ accumulation in blood vessels can hinder its clearance from the tissue, leading to more deposits in the adjoining parenchyma. This can potentially contribute to and affect the pathogenesis of AD. However, it is not known if the same aggregation pathways are prevalent for Aβ in blood vessels and the tissue microenvironment. To understand the structural correlation between vascular and parenchymal Aβ aggregates, we therefore acquired spectra from the Aβ deposits in the tissue adjoining a total of 25 blood vessels, as identified by IHC (Fig. 1, also see Figure S10), from two diseased patients corresponding to mild and moderate/severe CAA. The spectra were grouped by blood vessel and projected on to the same basis as those from vascular aggregates to determine the relative contributions of the different secondary structure components. The results of the deconvolution analysis are shown in Fig. 5. We observe that the relative population of overall β-sheets increases from mild to moderate/severe CAA, as expected. In contrast, both the antiparallel β-sheet and lipid populations decrease. Furthermore, only the antiparallel population exhibits correlation with lipids (Figure S11). This persistent correlation between the antiparallel population and lipids suggests that the mechanistic pathways adopted by Aβ depends strongly on the chemical environment. As a result, the resulting secondary structure of aggregates can vary significantly between blood vessels and parenchyma, leading to distinctly different disease-specific polymorphs. These findings also provide a rationale for why antiparallel structures have been elusive in cryo-EM studies of brain-derived fibrils: if fibrillar seeds are isolated from the entire parenchyma and not specifically from blood vessels, the dominant structure can be parallel β-sheets, as indicated by our results. In fact, the only study that has reported antiparallel structures in brain-derived fibrils also involved isolating Aβ specifically from vascular aggregates, which provides further credence to this hypothesis and underscores the potential heterogeneity in structure of Aβ fibrils in brain lesions and the need for targeted isolation of seeds for characterization of different pathologies.

Fig. 5

a Mean spectra from amyloid deposits in tissue microenvironment adjoining vascular aggregates, as identified by IHC. b Contributions of each spectral component determined from deconvolution, representing overall β-sheets (purple), random coil (green), antiparallel β-sheets (orange) and lipids (red). c–f The relative change in secondary structure populations, as shown in (b). The values for each component have been normalized to that of the mild CAA specimens

Nanoscale IR spectroscopy validates lipid-induced antiparallel character in Aβ40 aggregates

However, a key caveat needs to be considered here. The results shown above demonstrate the presence of antiparallel β-structure in vascular amyloids; however, these structural fingerprints cannot be uniquely attributed to Aβ or any specific protein from the IR spectra. It is expected from the immunostaining that the spectral features are primarily from Aβ aggregates, but the possibility of other proteins such as ApoE and tau that are known to associate with Aβ contributing to the antiparallel spectral markers observed here cannot be ruled out. All major ApoE isoforms can bind to Aβ peptides, influencing their aggregation and deposition in blood vessel walls [16, 38]. While tau aggregates are mainly associated with Neurofibrillary tangles (NFTs) in AD [13, 35], it has been shown that the deposition of Aβ induces accumulation of the Tau protein in the blood vessels [32, 77], potentially playing an important role in neurotoxicity. However, mass spectrometry studies have conclusively shown that Aβ is the primary constituent of vascular amyloids, particularly the Aβ40 isoform [34, 78]. Therefore, spectral contributions from other proteins toward the antiparallel β-character, if any, is expected to be small. Secondly, native tau and ApoE aggregates have not been reported to contain significant antiparallel character; rather tau fibrils are known to contain parallel, in-register β-sheets [35]. On the other hand, the possibility of formation of antiparallel transient intermediates have been demonstrated for Aβ [7, 57]; furthermore, antiparallel structures have also been identified in amyloid plaques [33] and in brain-derived vascular Aβ fibrils [37]. Therefore, we attribute the presence of antiparallel β-sheets predominantly to the deposition of Aβ. Of course, this does not conclusively preclude the possibility that other associated proteins can also exhibit antiparallel structure, which would require necessitate experiments and analysis beyond the scope of this work. We hope to address this in future efforts. To further verify if these antiparallel structures can indeed arise from Aβ and whether they result from lipid interactions, we used nanoscale IR spectroscopy to probe the structure of individual Aβ-40 aggregates. Integrating IR with modalities that offer nanoscale spatial resolution, such as Atomic Force Microscopy, provides a means to characterize the spectra of individual fibrils and the effect of lipids on their structural arrangement. AFM-IR leverages the same photothermal signal as O-PTIR but uses the modulation of the cantilever oscillation for detection, which provides the desired spatial resolution on the scale of individual aggregates [18,