Subcellular omics: a new frontier pushing the limits of resolution, complexity and throughput

Abstract

We argue that the study of single-cell subcellular organelle omics is needed to understand and regulate cell function. This requires and is being enabled by new technology development.

A eukaryotic cell’s phenotype — its morphology, physiology and function — is an emergent property of its complex network of organelles and subcellular structures. A mechanistic understanding of phenotype requires quantitative analysis of individual organelles and their interactions, facilitated by new technological innovation. Such advances will bring transformative insights into the intricacies of subcellular regulation of a cell’s biology.

Single-cell biology has highlighted that individual cells have greater molecular heterogeneity than previously realized and that this variability has important consequences for tissue-level function¹. Various single-cell approaches are revealing new cell types, differentiation pathways and cell–cell interactions and, in so doing, revealing how complex dynamics elicit emergent function at the cell, tissue and organism scale. We propose that the multiomic complexity within a cell is as significant as between cells and that there is a need to understand subcellular multiomic interactions as they lead to the emergent functional phenotype of individual cells. In our view, the next frontier in the study of the cell is the omics of subcellular biology, asking the question: “How do substructures in the cell interact to create whole-cell function in widely varying contexts?”

Many of the cell’s biological processes are partitioned into organelles and other subcellular structures where proteins, nucleic acids and chemicals including metabolites are concentrated, isolated and modified. Organelles can vary in their physicochemical properties and even possess a distinct subcellular-compartment genome or transcriptome that synergizes with the nuclear genome and other organelles to enable cellular function. These organelles interact in a complex yet coordinated manner to control cellular homeostasis and allostasis². When integrated into the cellular framework, the thousands of individual organelles work in concert to create the ‘systems biology’ of an individual cell.

Of note, the subcellular organizational and regulatory mechanisms that orchestrate normal cell function are altered in an array of human pathologies. Altered function of the mitochondrion, endoplasmic reticulum and lysosome can cause disease directly, but also appear as a downstream consequence of cellular pathophysiology. Tissues heavily dependent on aerobic metabolism (for example, skeletal muscle, heart³ and brain⁴) are most sensitive to primary mitochondrial disease. Similarly, primary lysosomal defects as implicated in Tay-Sachs, Fabry and Batten diseases⁵. Understanding how altered subcellular biology influences disease will be improved by systemically detailing and quantifying subcellular structure and organelle biology but, importantly, also requires advances in technologies, as we discuss below.

We propose that the first step in a comprehensive understanding of genome-scale subcellular biology is the development of technologies allowing the study of the multiomics complexity of the entire complement of individual organelles in a cell. We focus on four key challenge areas in need of technology development to drive the field: (i) increased resolution of assays to isolate organelles, (ii) multiplexed assays at the increased resolution for a full assessment of multiomics heterogeneity and functional variability, (iii) improved throughput of assays to allow complete census of all organelles in a single cell, and (iv) improved sensitivity of the assays to allow quantitative analysis of the subcellular organelles.

Prospects for increasing the resolution and throughput of organelle-scale assays

There are many challenges associated with subcellular characterization at the individual organelle level isolated from a single cell. (i) Organelles are often too small (<1 μm) for identification using technologies developed for single-cell analysis (for example, flow cytometry). (ii) There are a large number of organelles within each cell (for example, >300 mitochondria per cell). (iii) Organelles exist in cells within a complex network of interactions with surrounding organelles, making it problematic to isolate any one structure.

Physiochemical and different membrane organizations of organelles necessitate that multiple approaches be used in their isolation and purification. Extant technologies for organelle isolation each have their strengths and weaknesses. For example, centrifugation methods for organelle purification are simple and can separate a large number of structures. However, their yield varies between 20 and 80%, with relatively low fractionation accuracies due to spread across fractions^6–8. Micro- and nanofluidic devices can manipulate and store biological objects on size scales appropriate for subcellular components. By streamlining organelle preparation and subsequent analysis, microfluidic devices hold great promise for providing the throughput, purity and sensitivity needed to fully understand subcellular organization and physiology. The integration of all processing steps onto a single chip would reduce material loss⁹ and improve sensitivity. Multiplexing of microfluidics approaches can allow multiple organelles of interest to be analyzed from one or more cells^10,11.

New techniques are emerging for manipulation and analysis of large numbers of droplets and their contents. Examples include trapping and releasing droplets from microarrays and coupling microfluidic devices to analytical instruments such as mass spectrometers^12,13. Microfluidic methods that take advantage of laminar flow can create discrete chemical microenvironments that surround a single adherent cell. Such methods could be further extended to subcellular extraction or analysis of surface-adherent cells while maintaining spatiotemporal information without the need for barcoding¹⁴. Concatenating microscale cellular separation modalities with downstream organelle analyses will allow single-organelle heterogeneity to be studied in the context of single-cell heterogeneity. Promising microscale approaches include microscale filters, inertial focusing, hydrodynamic filtration, deterministic lateral displacement, acoustophoresis, dielectrophoresis, electrophoresis and optical methods^6,7,15–17.

Throughput is an important criterion for comparing existing and prospective subcellular analysis technologies. Throughput is the rate at which useful biological data can be extracted from a biological material and converted into digital information that can be stored, processed and analyzed on a computer. The original Drop-seq¹⁸, in which single cells are lysed and their mRNA barcoded in nanoliter-scale droplets, can process approximately 1,000 cells per second with recovery of approximately 5,000 unique mRNA transcripts per cell¹⁸, resulting in a total data transfer rate of 5 × 10⁶ assays per second up to the stage of isolation and labeling. For organelle isolation, we need 100- to 1,000-fold greater linear size resolution, from the 10-μm scale to the 10- to 100-nm scale. Assuming cellular contents could be assessed without additional volume dilution, existing methods might encapsulate larger organelles of a single cell (for example, mitochondria) within an hour whereas smaller organelles such as ribosomes would require ~11 days. Of course, to isolate components, suspension in some larger volume would be required, greatly increasing requirements for speed. A possible approach is the use of parallelized microfluidics, which can generate droplets at rates 1,000-fold greater than single-droplet devices¹⁹. Alternatively, organelles can be coupled to larger particles (for example, microbeads), thus facilitating isolation. Additional throughput gains could arise from integrating optical detection of the protein and nucleic acid cargo of individual subcellular components, which can be incorporated directly on chip. The challenge will be creating an optical readout that can be multiplexed to detect these many different chemical species. We anticipate that direct optical assays detected by fast, multiplexed sensors will enable a full census of organelle components from multiple cells.

Untangling chemical complexity of organelles as examined by mass spectrometry

Chemical complexity within a cell is staggering: each cell contains hundreds of thousands of distinct compounds consisting of metabolites, lipids, peptides, proteins and polynucleotides. Fortunately, the complexity of specific organelles is significantly lower. For example, an individual Escherichia coli, which is the size of a eukaryotic mitochondrion, may possess only a few thousand metabolites^6,20. Nevertheless, the resolution scale at the single-organelle level is challenging. Organelle volumes are in the attoliter (10⁻¹⁸ L) range. For example, a mammalian cholinergic secretory vesicle 40–50 nm in diameter contains just a few thousand molecules of the neurotransmitter acetylcholine — an amount close to the limits of detection for many methods. Yet measurements performed near the analytical limits of detection may generate irreproducible and unreliable results when only a small number of organelles are analyzed. As an example, comprehensive characterization of a population of single cholinergic secretory vesicles in the mouse striatum requires limits of detection of <10 zeptomoles and the ability to process hundreds to thousands of vesicles. Therefore, single-organelle analysis requires development of methods to minimize losses during sample preparation and optimize single-molecule detection limits.

Several strategies have the potential to allow comprehensive mass spectrometry (MS) evaluation of single-organelle chemical composition, as well as determination of heterogeneity of organelle populations. Examples of methodologies that are on track to provide such capabilities include optically guided MS⁷; lossless sampling of individual organelles⁸; preparation of optimally diluted samples²¹, especially based on micro-enclosed environments²²; and reduction of sample chemical complexity. Further enhancement comes from approaches to increase analyte ionization efficiency, including detection of often dominant neutral molecules²³, and to improve MS instrumentation for higher mass resolution and detection sensitivity¹⁵. Future technological advances promise to provide routine MS detection of a single-molecule analyte using cryogenic detectors^24,25. The potential for such optimization of MS has been demonstrated for ~300-nm-diameter lipid vesicles²⁶ and >500-nm neuroendocrine vesicles²⁷, which were successfully analyzed to quantify various metabolites and peptides.

For subcellular MS analysis, sample throughput is also a critical bottleneck. To gain the most biological insight, direct analysis of organelles is preferable to measurements that pre-separate analytes. Depending upon the MS system, organelles can be delivered to the MS ion source (for example, electrospray ionization^28–30) in a stream of liquid or placed onto a surface and then probed. In liquid stream-based technologies, hundreds of cells per second can be measured — for example, ~250–500 cells per second for cytometry by time of flight (CyTOF)³¹. Similar or even higher throughput in single-organelle analysis should be possible, since the integrity of the subcellular structure would no longer be a concern. Surface-based organelle MS analysis can be performed in several modes, including mass spectrometry imaging, pioneered by the Caprioli group³², and variations thereof (for reviews, see refs.^9,33). Mass spectrometry imaging should be capable of subcellular analysis in individual cells, including those in tissue sections. Nevertheless, as we noted above, for a complete census, we are likely to need at least another 1,000-fold increase in acquisition rate in cases of most frequently used MS approaches.

Proteomics of single organelles

While protein detection in organelles is complicated by the multiple isoforms, tertiary structures and compartmental environment of the organelles, preselected protein detection in single organelles has historically been possible through multiple approaches, including immunogold transmission electron microscopy. Alternatively, single speckle detection fluorescence microscopy³⁴ allows protein detection in single organelles without the need for electron microscopy. These approaches using affinity-based reagents are, however, indirect; current direct approaches do not provide complete protein coverage or quantitative information at the single-organelle level. An issue in protein detection at single-organelle resolution is that many proteins are highly dynamic and post-translational modifications can be selectively localized to subcellular sites. Therefore, the comprehensive analysis of the protein complement of single organelles must take into account spatiotemporal factors including where the organelle is located and when sampling occurs.

Development of a multimodal approach combining advanced sample preparation and analysis technologies would allow qualitative characterization of a proteome at lower throughput and quantitative assessment of organelle chemical heterogeneity and variability with high throughput. This approach requires solubilized or extracted proteins including hard-to-isolate macromolecular complexes (for example, transmembrane proteins). Depending on experimental goals, the extraction procedures may need to preserve 3D molecular structure including labile post-translational modifications. Structural characterization of isolated proteins can be done with several detection approaches (summarized in ref.¹⁶), such as individual ion mass spectrometry¹⁷ and nanopore electrospray³⁵. These approaches target single molecules and can provide partial structural information but will not be 100% efficient.

Analyses of single organelles with relatively small volumes require subnanometer to low-nanometer spatial resolution. Rapidly developing imaging technologies that have the potential to perform such measurements include high-resolution cryo-electron tomography¹⁸ and single-particle cryo-electron microscopy³⁶. Both approaches have demonstrated subnanometer spatial resolution. Organelle analysis in live cells is nascent, with ultra-fast coherent diffractive imaging with X-ray free-electron lasers detecting proteins at 4-nm resolution and theoretically capable of reaching subnanometer levels³⁷. However, both improved resolution and better structural contrast are needed to distinguish the 3D conformations of these molecules, including proteoforms.

Signal amplification to reveal omics complexity of organelles

Single-molecule methods and MS demonstrate great capabilities for detection of certain molecules, but other approaches are better suited for detection of other subcellular entities (for example, nucleic acids or proteins), especially when signal amplification would be beneficial. Enhancing the sensitivity of nucleic acid detection can be achieved through use of more efficient amplification enzymes with high fidelity. Also, dyes with higher quantum yields or ways of incorporating more dyes when using nonamplified detectors will be helpful. Underappreciated detection challenges include the frequent association of cellular chemicals with intracellular molecules that can hinder analyte access and create variable in situ tertiary structure and composition of RNA and other molecules, preventing detection³⁸. New reversible fixatives, higher sensitivity probes, and/or enzymes that function at different temperatures (with higher temperatures that melt nucleic acid structures facilitating in situ hybridization-based approaches) may help to normalize biological complexities associated with specific organelles and will help to harmonize multimodal subcellular data collection³⁹.

Combining sensitive nucleic acid amplification techniques with protein detection methods (initially developed in the immunodetection amplified by T7 (IDAT) procedure^10,11) or endogenous biotin labeling of proteins using proximity-based labeling (APEX) approaches¹² has the potential to greatly enhance sensitivity of proteomics methods (Fig. 1). Proximity protein detection approaches have the added advantage of detecting two or more closely associated proteins simultaneously. The combination of antibodies or RNA aptamers with nucleic acid detection methodologies can be enhanced to detect single proteins¹³ or protein–RNA complexes (for example, membraneless stress granules¹⁴) in subcellular environments. An important benefit of combining high-sensitivity nucleic acid amplification with antibody detection is that low-affinity yet specific antibodies will attract new use through the development of more sensitive nucleic acid amplification to detect the smaller amount of antibody–antigen interactions. Such approaches will enhance the proteomics toolbox for the study of organelle biology. A continuing challenge will be controlling the amplification to maintain quantitative linearity and bounding the noise that is inevitable when detecting the small absolute abundances of proteins in individual organelles.

Fig. 1 |. Analysis of mitochondria as an exemplar of single-cell organelle chemical and biology detection methods.

The subcellular site of action for many of the approaches highlighted in this Comment is depicted here. Single-mitochondrion analysis began with morphological analysis of cells and observations of organelle morphology. Proteins can be detected using a variety of indirect methods that often rely on signal amplification after antibody labeling (gray Y) of specific mitochondrial membrane antigens (yellow trapezoids). DNA (green circular mitochondrial DNA) and RNA (red lines) can be detected using direct hybridization methods as well as other synthetic procedures. Live organelle biosensors, both genetic (which require transfection) and nongenetic, permit visualization of organelle dynamics and aspects of their physiology in live cells. MS imaging is a direct laser interrogation of organelles (lightning bolt) that releases detectable metabolites (blue spheres). Applications of microfluidics methodologies are anticipated to perform many of these biochemical analyses on large numbers of organelles. Computational multimodal machine learning seeks to synthesize these biochemical data into testable models that can be used to understand how organelles work individually and in synergy to facilitate cellular function. Blue line, oligonucleotide probe; stars, label; green lines, double-stranded DNA oligonucleotide; red protein, enzymatic activator; blue oval, proximity-labeled membrane antigen. aRNA, amplified antisense RNA; APEX, ascorbate peroxidase proximity labeling; ATAC, assay for transposase-accessible chromatin; BioID, biotin proximity labeling; CEPIA, calcium-measuring organelle-entrapped protein indicators; EM, electron microscopy; FACCT, fluorescent amplification catalyzed by T7 RNA polymerase technique; HRP, horseradish peroxidase; SABER, signal amplification by exchange reaction; scFV, single-chain variable fragment; TIVA, transcriptome in vivo analysis; TurboID, rapid biotin proximity labeling.

In situ subcellular spatial analysis

In single-cell analysis, in situ spatial techniques are a rapidly expanding area of development⁴⁰. For both single-cell biology and subcellular biology, molecular characterization with spatially localization is critical to understanding the component interrelations and emergent properties. Spatial addresses can be encoded at the time of component isolation, but the most common approach is to directly measure in situ or in situ label measured molecular species. Advances have been made in spatial transcriptomics as a result of the ease of bar-coding and amplification⁴¹, and we envision this modality also being the most approachable assay in subcellular biology, but again challenges remain with respect to resolution and capacity. For example, technologies based on single-molecule multiplexed smFISH can now probe up to 10,000 genes and have the potential to be scaled to the whole transcriptome. However, optical crowding of fluorescent spots presents a problem in resolving the localized transcriptome of, say, a single mitochondrion. Physical expansion has been adapted to smFISH⁴², but for organelles, the spatial extent of individual organelles will need to be expanded; that is, matrix expansion will need to start at within-organelle scales.

In situ spatial barcoding has much higher throughput, delivering barcodes to spatially defined locations. Barcode spot size determines the resolution with new techniques (for example, Stereo-seq, with 0.22 μm diameter of rolling-circle-amplified DNA nanoballs, yielding 0.5 to 0.7 μm separation⁴³). However, enhancing resolution by reducing molecular components that bar-code or index captured mRNA results in reduced capacity such that each spot might recover a few hundred molecules out of hundreds of thousands of molecules in a cell. Direct in situ sequencing⁴⁴, while having greatly progressed from the original methods, also suffers from similar problems of optical crowding, limited capacity and limited speed. The most important challenge for in situ spatial approaches is resolving the ~10–1,000 mRNA molecules associated with a single organelle at a spatial resolution of <1 μm. While organelles at this scale can be isolated and assayed, maintaining precise spatial indexing by in situ assays will require using approaches that overcome optical size-resolution limits.

Quantitative modeling and data analysis needs

Concomitant with advances in organelle and chemical characterization, accelerating the field of quantitative subcellular biology requires corresponding developments in both data analysis and modeling approaches. Analysis of subcellular data is even more challenging than that of single-cell data, in part as a result of higher levels of technical noise and sparse datasets. Further compounding the challenge, understanding how the complex subcellular interactions lead to emergent cellular phenotypes will generate another layer of modeling challenges in systems biology frameworks⁴⁵.

As mentioned earlier, advanced imaging modalities including chemical imaging will be a crucial platform for assessing intracellular organelle dynamics (that is, size, movement and interactions), and, as such, subcellular image analysis will present new challenges. For example, image segmentation identifies substructures of interest within complex images. Subcellular images will need segmentation at multiple scales: segmenting cells, then organelles and then structures within organelles. There has been rapid recent progress in applying deep learning to automate image segmentation and eliminate manual steps^46,47. Moreover, for both current single-cell spatial omics and future subcellular omics, new methods are needed to move from spatial data to process inferences. For example, machine learning models have been developed to use cellular morphology to infer localized gene expression⁴⁸. Understanding spatial processes involves asking how generative cellular dynamics create the spatial patterns and how the spatial context (that is, the given patterns) influences cellular dynamics. The methodological question is whether we can estimate these interactions from spatial patterns. Similar questions have been raised in the field of landscape ecology⁴⁹, and adaptation of such methods is likely to be a fruitful path.

Subcellular omics data present unique technical problems. Appropriate statistical models will be required to resolve the anticipated low signal-to-noise ratio and scarcity. In single-cell biology, a rich array of approaches has been developed to account for high noise and sparsity. Yet current approaches model the technical problems by parametric probability distributions such as the negative binomial⁵⁰, which may be too limited. Moreover, few, if any, studies attempt to experimentally isolate the noise components to fit the parametric models. Owing to low-abundance signals, subcellular data are likely to be even more sparse and sensitive to technical or batch artifacts than current single-cell analyses. We suspect that it will be important to mechanistically model the assay protocols to more precisely characterize and reduce the noise.

Challenges and opportunities for exploring quantitative subcellular omics

Each cell is a collection of complex biophysical machines that interact with each other to elaborate cell functions. While eukaryotic cell organelles have their own functional capacity, it is interorganelle interactions — either through direct contact or by virtue of chemical signals — that result in the emergent properties of single cells. A hallmark of disease is dysregulation of these properties. Understanding and manipulation of these emergent properties require detailed knowledge of organelle biochemistry and physiology. Development of high-throughput subcellular biology is fundamental to understanding the underpinnings of the emergent biology of a cell.

With the ultimate technical goal being spatiotemporal analysis of organelles within cells that are resident in their natural cellular environment without physical isolation, many of the hoped-for developments will need transformative advances. In particular, analysis and understanding of in situ organelle biology will require advances in imaging, biosensor development, quantitative chemical capture methods and molecular approaches that can be specifically and selectively targeted to the complex cellular and tissue environment in which the organelle resides. For example, a promising initial approach is the use of photoactivable capture molecules that can be selectively activated in individual organelles to interrogate the chemical composition of each organelle^51,52. Advances in in situ MS including enhanced detection sensitivity as well as laser-energy focusing capabilities may also enable more complete analysis of a variety of chemicals in selected in situ targeted organelles.

While much work remains, understanding how cellular function is organized and executed by constituent subcellular structures is the framing step for developing a systems theory of cell type based on the emergent properties of cells. Detection and analysis of the chemical composition of organelles is only one of many steps required to ultimately manipulate the emergent properties of cells. If we may take a page from single-cell transcriptomics analysis, genome-scale detection of RNAs has led to testable hypotheses as to how cells respond to various challenges. We anticipate that the development of models of organelle biology in the context of cell function will lead to hypotheses that are experimentally testable by the manipulation of individual and multiple organelles. When such experimental platforms are developed, a generalizable systems theory of the cell will emerge. Along this scientific journey, insights into organelle biology will undoubtedly provide new understandings of disease mechanisms and spotlight novel targets for therapeutic development.

The methodological challenge is formidable. But, consider that in 2009 one of our co-authors published single-cell transcriptomes of 30+ cells⁵³, one of the largest collections at the time; a mere dozen years later, there are datasets of millions of cells. Graham Green wrote in The Power and the Glory: “There is always one moment in childhood when the door opens and lets the future in.” For the study of subcellular biology, we believe that door is opening now.