Amyloid plaques are among the most recognizable features of Alzheimer’s disease – but their chemical surroundings are far less understood. While imaging studies have mapped where plaques form, much less is known about how the lipid landscape shifts within and around them, and whether those changes differ across brain regions.
In a recent Nature Communications study, a research team led by Jonathan Sweedler and Fan Lam developed a multimodal mass spectrometry imaging framework to examine plaque-associated lipids at high spatial resolution. Their work reveals striking regional heterogeneity and links specific lipid signatures to underlying metabolic pathways.
Here, study first author Timothy Trinklein discusses the rationale behind the multimodal design, how the platform works in practice, and what the findings suggest about lipid dysregulation in Alzheimer’s disease.
What initially motivated your team to take a multimodal approach to plaque-associated lipid analysis?
Amyloid plaques are extracellular deposits of the peptide amyloid. Curiously, the occurrence of these plaques seems to disrupt the local lipid environment. We wanted to find out if this process differed across brain regions, and what biochemical mechanisms were responsible for this process.
Our approach was to image the lipids with MALDI-2 mass spectrometry imaging (MSI), before staining and imaging the tissue with fluorescence microscopy to determine where the plaques were located post hoc. This way, we could computationally "mask" the plaque-containing regions and compare lipid compositions between regions. Therefore, our approach is multimodal, combining both MALDI and fluorescence microscopy. Because the same tissue section is used for both modalities, it is straightforward to coregister (i.e., spatially align) the data.
There is a lot of excellent work using imaging techniques like MSI to map plaques. However, a lot of the existing work has focused on examining a few plaques in a small, focused brain region. We wanted to expand the scale to hundreds or thousands of plaques across the brain, and among many animals, so that we could uncover, if present, heterogeneous “populations” of plaques.
Could you briefly walk us through how your workflow functions?
We leveraged a number of advanced analytical tools to enable this analysis. MALDI-2 – a post-ionization approach – was used to boost lipid signals, along with trapped ion mobility to resolve lipid isomers. After MSI, we used fluorescence to localize plaques in the data.
This enabled, in my opinion, the most exciting and unique aspect of the work – our computational approach. After discovering altered lipids by comparing plaque against non-plaque pixels, we segmented pixels according to brain region. We then used t-SNE and DBSCAN clustering to see if there were interesting populations in the data. Notably, the resulting pixels grouped broadly by brain region, before clustering into finer subpopulations. For those familiar with single-cell RNA-seq, the pattern resembles well-defined cell types visualized by t-SNE or UMAP, although here each point represents the MSI signal from a plaque.
We observed this heterogeneity even after zooming into even smaller brain regions. By training machine learning models, we identified the lipids that were responsible for each population and were able to predict, with high accuracy, the specific brain region a plaque was from, based purely on its lipids.
Finally, we searched the genes responsible for synthesizing or degrading those lipids in spatial transcriptomic databases to try and understand the underlying biochemical pathways responsible for these alterations.
What were the greatest challenges your team faced – and how did you overcome them?
One major challenge in "large-scale" MALDI imaging experiments involving multiple animals is batch effects. For instance, we often observe major differences in signal intensity between different days of analysis and between the glass slides on which samples are mounted – how does one correct for these signal alterations without removing meaningful biological variation? This is a well-known problem in MALDI imaging, yet at present, there’s no widely agreed-upon solution.
Our first line of defense against these problems is sound experimental design. For instance, we simultaneously mounted brains from many animals on the chuck of a cryostat, sectioned them all at once, and then mounted them on the same slide. Choosing an appropriate method for normalization is also critical. TIC normalization is common in MALDI, but only accounts for variation within an image, and when applied inappropriately, it can alter apparent ion distributions.
An internal standard can be sprayed on the tissue sections – but this, too, is suboptimal. Because of ion suppression, the internal standard signal can differ between tissue regions despite having the same concentration, such as between white and gray matter. The internal standard, or standards, also have to be chosen with some understanding of the analyte’s classes, though this commonly isn’t known beforehand in non-targeted analyses. This problem has been better studied in transcriptomic circles, where approaches such as “combat” emerged.
In our work, we combined careful experimental design (e.g. many animals per slide) with a “transcriptomics” batch correction algorithm. However, this problem always requires careful consideration.
Which results most surprised you?
In wild type mice, we and others have observed that lipids are surprisingly region-specific. Just based on the lipid profile of a mass spectrum, we can predict the brain region it originated from with incredible accuracy. Here, we found the same principle is true of plaques.
We also found that the lipids accounting for broader regional variation were different from those that classified plaques. That immediately raises the question of what drives these differences. Is it cell type? In principle, plaques in both the hippocampus and cortex are associated with disease-activated microglia and astrocytes, among others. It may be that the relative proportions of these cell types differ between brain regions.
How do your findings change how we think about lipid dysregulation in Alzheimer’s disease?
When we think about Alzheimer's disease, we mostly think about amyloid plaques and neurofibrillary tangles. But in fact, Alois Alzheimer's landmark 1906 report described not two, but three distinct pathologies – the third being fatty lipid accumulations. So, even though lipid dysregulation in Alzheimer's has been known for over 100 years, it remains one of the most poorly understood aspects of the disease.
We and others consistently observed that a class of lipids called gangliosides strikingly accumulate in plaques. However, when we observe lipid accumulation in Alzheimer's disease, it doesn't necessarily mean that a given lipid is being synthesized more. It could, in fact, be that a larger, more complex lipid is being broken down, and we’re observing one or more of these degradation products. Fueled by our collective curiosity, we searched for alterations in synthesis and degradation genes in publicly available spatial transcriptomics data. Here, we found that lysosomal degradation genes were most affected, suggesting that this could be the primary mechanism. However, the synthetic genes were upregulated as well, albeit by a much, much smaller amount.
It's possible there is some "compensation" effect. However, I think it’s important to mention that many lipids besides just gangliosides are known to be dysregulated in Alzheimer's disease, and different metabolic pathways are altered. Beyond that, there are thousands of known lipids in the mammalian brain, though we only detect a couple hundred of them in MALDI imaging experiments. I can confidently say that our understanding of lipid alterations is still far from complete.
Looking ahead, how do you see this multimodal framework being applied next?
There are many exciting areas for future development. Multimodal approaches to mass spectrometry imaging are increasingly popular, and for good reason, too. To obtain a deep understanding of the biology of the brain (or any complex organ/model), one first needs to map the distribution of many classes of molecules – and across different scales – from tissue to single cells. This usually necessitates multimodal imaging.
The biggest challenge moving forward, however, is not deciding what modalities to combine, but how to meaningfully fuse data from distinct technologies with different spatial resolutions and statistical assumptions underlying the data. When this problem is solved, we will really see multimodal approaches take over spatial biology.
Expanding this approach to other disease models is a natural next step. There is still considerable scope to apply it in Alzheimer’s disease, particularly to ask whether specific lipid alterations are associated with the formation of neurofibrillary tangles, another hallmark pathology.
Human tissue is also of great interest for these and related investigations, although it brings its own set of challenges. For obvious reasons, one does not have the same level of control over the sample: with a mouse model, we can perfuse the brain with ice-cold saline to freeze metabolic processes, remove interferences from blood, and immediately freeze or section the tissue. With a human, there may be a significant post-mortem interval where lipids have degraded. With these considerations, human tissues allow us to confirm what we have seen in animal models and observe alterations that have occurred over much greater timescales.
I’ve always been curious about whether this approach could help us better understand emerging Alzheimer’s therapies, such as comparing control and treatment groups – or even to better map chemical alterations that accompany side effects like ARIA. In short, the future of multimodal MSI is bright and full of opportunity!
Timothy J. Trinklein is a Postdoctoral Researcher in the Department of Chemistry and the Beckman Institute for Advanced Science and Technology at the University of Illinois Urbana-Champaign, USA.
Newsletters
Receive the latest analytical science news, personalities, education, and career development – weekly to your inbox.
