Organelles, despite their crucial roles in cellular function, remain poorly understood due to challenges in analyzing their minute volumes and diverse chemical compositions. Traditional methods for single-organelle analysis, like isolating organelles and placing them on a slide for mass spectrometry (MS) measurement, suffer from low throughput and automation difficulties. Mass spectrometry imaging, while useful, is limited in resolution and throughput for organelle analysis. Building upon single-cell MS approaches using matrix-assisted laser desorption/ionization (MALDI), this research adapts the technique to single organelles. This adaptation required improvements in object targeting (using brightfield microscopy to identify organelles without chemical labeling), analyte detection (using high-resolution MS), and data analysis (unsupervised workflows to characterize heterogeneity). The goal was to achieve high-throughput, simultaneous detection of peptides and lipids in individual organelles, specifically DCVs and LVs isolated from the *Aplysia californica* exocrine atrial gland and red hemiduct, respectively. The study aimed to uncover chemical heterogeneity within and between these vesicle types to better understand their roles in neuropeptide and hormone packaging and release.

Previous research has used MS imaging for cellular and subcellular analysis, but this approach has limitations in throughput and spatial resolution for organelle-level studies. Methods involving isolating and manually positioning individual organelles for MS measurement are also hampered by low throughput. Recent advancements in single-cell analysis utilize scattering cells on a slide, locating them via fluorescence microscopy, and then targeting them with MALDI MS. This study builds on this by adapting the method to single organelles, a significant technical challenge given their much smaller size and the need to avoid introducing potentially interfering exogenous chemicals during sample preparation.

The researchers developed a three-step approach for high-throughput organelle preparation. First, they used brightfield microscopy to identify and image DCVs and LVs without the need for chemical labeling, which can interfere with MS analysis. Image processing algorithms (detailed in Supplementary Figure 4) were used to distinguish vesicles from the background, creating a binary image mask identifying the vesicles' locations. This label-free approach circumvents the problems associated with paraformaldehyde fixation, which crosslinks molecules and reduces their ionization efficiency. A volatile, isosmotic ammonium acetate buffer was used to deposit the vesicles on an indium tin oxide (ITO)-coated glass slide. Vesicles were allowed to sediment and adhere before the solution was aspirated. The binary image mask generated from the microscopy images was then used by the microMS software to guide the MALDI MS analysis, ensuring precise targeting of individual vesicles. A 200 µm filter was used to prevent overlapping laser spots. MALDI Fourier-transform ion cyclotron resonance MS was used for high-throughput simultaneous detection of peptides and lipids using an *m/z* range of 150-4500. Data were preprocessed using Compass Data Analysis 4.4.2 (Bruker) and MATLAB 2018b (MathWorks), involving internal calibration using the exact mass of known AG peptides, peak picking, and mass spectral alignment. For DCV analysis, the *m/z* range was truncated at 1100 to focus on AG peptides. For LV/DCV classification, the range 500-1100 was used. CX decomposition was used for feature selection, followed by k-means clustering to group DCVs based on their peptide profiles. For LV analysis, a modified image processing pipeline facilitated targeting of LVs based on their unique morphological characteristics (Figure 3a). A machine learning model (gradient boosting trees) was trained to differentiate LVs and DCVs, with feature importance determined using SHAP values. LC-MS/MS was used to validate peptide identifications.

The high-throughput analysis of 598 DCVs identified over 50 mature peptides from eight known prohormones and one novel prohormone (AG Peptide D). This revealed differential packaging of peptides from the same prohormone into different DCV subtypes. CX decomposition identified three distinct DCV clusters with different peptide compositions. Cluster 1 was characterized by N-terminal peptides AGPB1 [71–81], AGPA1 [22–34], and Peptide D [63–75]; Cluster 2 showed distinct peptide populations; and Cluster 3 contained multiple C-terminal peptides from AGPA1 and AGPB1, including Peptide D [132–162]. The analysis also revealed simultaneous detection of peptide and lipid species within individual DCVs, identifying three phosphatidylcholine (PC) lipid species. Analysis of 123 LVs from the red hemiduct, using a modified image processing pipeline and a machine learning model, revealed significant differences in lipid profiles compared to DCVs. This classification had an accuracy of 98.6% ± 0.78%. The presence of sterol lipids was strongly associated with LVs, potentially contributing to their unique cristae-like inner membrane structures. In summary, this study provides a high-throughput method to profile the peptide and lipid contents of individual organelles, revealing significant heterogeneity within and between different organelle types.

This research significantly advances single-organelle analysis by enabling high-throughput, label-free characterization of both peptide and lipid content in intact organelles. The findings demonstrate the power of combining image-guided MALDI MS with sophisticated data analysis techniques to uncover previously unappreciated heterogeneity within morphologically defined vesicle populations. The differential packaging of peptides from the same prohormone into different DCV subtypes has implications for understanding neuropeptide and hormone signaling. The distinct lipid profiles of DCVs and LVs highlight the importance of considering organelle-specific lipid composition in the context of membrane structure and function. This approach complements existing single-cell and subcellular analysis techniques, offering improved throughput and adaptability to various organelle types.

This study successfully developed and validated a high-throughput, label-free method for single-organelle analysis using image-guided MALDI MS. The method revealed significant heterogeneity in the peptide and lipid content of DCVs and LVs. The approach's versatility and adaptability make it a valuable tool for investigating the chemical complexity of subcellular structures and advancing our understanding of cellular function. Future work could explore the application of this method to other organelle types and cellular systems.

The axial resolution of light microscopy limits the approach to vesicles ≥500 nm in diameter. The study focused on specific vesicle types from *Aplysia californica*; further validation is needed to confirm the generalizability of the method to other organisms and vesicle types. The interpretation of the machine learning model relies on the accuracy of feature annotation, which is subject to limitations in database completeness and mass spectral matching.