Кількісна оцінка життєздатності клітин органоїда за 3D-флуоресцентним зображенням — optical sectioning vs. single-plane assessment
An organoid is not a monolayer on a dish — it is a self-organized, roughly spherical, multicellular tissue that can span 100–600 μm in diameter. A conventional widefield or single-confocal-plane image only ever samples a thin optical slice (often 1–5 μm thick) somewhere through that volume. If viability differs by depth — which it almost always does, because nutrients, oxygen, and drugs diffuse in from the surface — a single plane can radically over- or under-represent the true state of the tissue, depending purely on where the focal plane happened to land.
Organoids grown in Matrigel or suspension culture lack a vascular system. Oxygen, glucose, and other nutrients reach interior cells purely by passive diffusion from the surrounding medium through the outer cell layers. Diffusion-reaction modeling (and decades of tumor-spheroid literature) shows that passive oxygen diffusion through densely packed tissue is reliably sufficient only to a depth of roughly 100–200 μm before intracellular O₂ tension falls below levels needed for oxidative metabolism.
Below this radius, cells experience progressive hypoxia, nutrient starvation, and metabolic waste accumulation — conditions that can trigger reversible quiescence, apoptosis, or outright necrosis depending on severity and duration. Critically, this means viability in an organoid is not spatially uniform: it is a function of local depth from the surface, often forming a shell of healthy, proliferating cells around a progressively compromised or necrotic interior as the organoid grows past the diffusion limit.
A 2D image — whether a brightfield photo, a single widefield fluorescence exposure, or one confocal optical section — samples only whatever plane the microscope happened to be focused on. If that plane passes near the organoid’s equator but misses the core, or if it merely grazes the surface, the resulting viability estimate reflects that one slice, not the organoid as a whole.
Consider a 350 μm diameter organoid with a healthy 150 μm-thick outer shell and a necrotic core. A focal plane through the equator, tangent to the surface, will be dominated by live, well-stained peripheral cells and will report viability far higher than the tissue actually contains once the buried core is accounted for. Conversely, a plane accidentally centered directly on a small necrotic pocket could underestimate viability by overweighting a local defect that is not representative of the bulk tissue.
Both failure modes share the same root cause: a 2D image has no information about anything above or below the focal plane. There is no way, from a single slice, to distinguish “this organoid is uniformly 85% viable” from “this organoid has a healthy shell over a necrotic core that happens to average to 85% at this particular depth.” These are biologically very different states — one predicts stable long-term culture, the other predicts imminent shedding of necrotic debris and eventual organoid collapse — yet a single 2D image cannot tell them apart.
This sampling bias becomes more severe, not less, as organoids grow larger and rounder, exactly the regime in which core hypoxia is most likely to occur.
The core insight motivating this entire simulation: viability is a volumetric property of a 3D object. Any assessment method that samples less than the full volume is making an assumption — implicit or explicit — that the sampled region is representative of the whole. For organoids beyond roughly 200–300 μm in diameter, that assumption frequently fails.
Before any imaging can quantify viability, cell state must be converted into an optical signal. The standard approach uses two fluorescent probes with complementary selectivity: one that only fluoresces inside cells with intact membranes and active esterases (marking live cells green), and one that only enters cells with compromised membranes and binds nucleic acids (marking dead cells red). Together, they turn an invisible physiological state into a directly quantifiable color-coded image.
Calcein-AM (acetoxymethyl ester) is a non-fluorescent, membrane-permeant molecule. It freely diffuses into all cells regardless of viability state. Inside a living cell, intracellular esterases — enzymes present only in metabolically active cells — cleave the AM ester groups, converting the molecule into free calcein: a polyanionic, strongly green-fluorescent compound that is well retained inside cells with an intact plasma membrane (it cannot easily leak back out).
This makes calcein a dual-requirement live indicator: a cell must have (1) active esterase enzymes (indicating ongoing metabolism) and (2) an intact membrane (to retain the cleaved product) to appear green. Dead or dying cells typically fail one or both conditions — esterase activity collapses quickly after cell death, and membrane integrity is lost, allowing any calcein that did form to leak out.
Ethidium homodimer-1 (EthD-1) is a large, highly charged dimeric molecule that cannot cross an intact plasma membrane. It is therefore excluded from live cells entirely. When a cell dies — by necrosis, late apoptosis, or mechanical/chemical injury — its plasma membrane loses integrity, and EthD-1 can enter the cytoplasm and nucleus, where it intercalates into double-stranded DNA. Binding to DNA increases its fluorescence quantum yield roughly 40-fold and shifts its emission into the red, producing a bright, unambiguous red nuclear signal specifically in dead or dying cells.
Because the two dyes require opposite membrane states to produce signal, they are almost mutually exclusive at the single-cell level: a healthy cell fluoresces green only, a dead cell fluoresces red only (calcein having leaked out and esterase activity having ceased), and the ratio of green-positive to red-positive objects across the full imaged volume becomes a direct, cell-resolved viability metric.
This live/dead assay (commercialized as kits such as LIVE/DEAD™ Viability/Cytotoxicity Kit) is chosen specifically because its two channels are spectrally well separated and mechanistically orthogonal — it reports two independent biological readouts (metabolic activity and membrane integrity) rather than a single proxy, making the resulting classification far more robust than either dye alone.
Once the organoid is stained, generating a true 3D viability map requires capturing not one image but a full stack of optical sections spanning the entire depth of the structure. Confocal laser scanning microscopy rejects out-of-focus light at each depth using a pinhole, while light-sheet (selective plane illumination) microscopy illuminates only a thin plane at a time from the side — both approaches build up a complete z-stack, plane by plane, that together reconstructs the organoid’s full three-dimensional structure.
A confocal microscope scans a focused laser spot across the sample and places a pinhole aperture in front of the detector, conjugate to the focal plane. Light originating from above or below the focal plane is defocused at the pinhole plane and largely blocked, so only in-focus light from the current depth reaches the detector. By stepping the focal plane through the sample in small increments (the z-step) and recording an image at each depth, a full z-stack of optically-sectioned images is built — each one a clean cross-section through the organoid at that specific depth, free of blur from other planes.
For a 300 μm diameter organoid imaged at a 3–4 μm z-step, this typically means acquiring 75–100 individual optical sections per fluorescence channel to cover the full volume — each one independently in focus, together forming a complete depth-resolved dataset.
Light-sheet (selective plane illumination) microscopy takes a different approach: rather than scanning a point and rejecting out-of-focus light with a pinhole, it illuminates the sample from the side with a thin sheet of laser light that is only a few micrometers thick, coincident with the focal plane of a perpendicular detection objective. Because only the in-focus plane is ever illuminated, there is no out-of-focus light to reject — the entire plane is captured in a single camera exposure, and the sample or light sheet is then stepped through the depth to build the z-stack.
This geometry dramatically reduces both phototoxicity and photobleaching (since regions outside the current plane are never exposed to laser light) and can acquire full volumes an order of magnitude or more faster than point-scanning confocal — a major advantage for large or light-sensitive organoid samples, or for live time-lapse imaging of viability over days.
Whichever modality is used, the output is the same in principle: a stack of dozens to hundreds of individual 2D optical sections, each one a clean slice through a known depth, which together span the organoid’s entire volume rather than one arbitrary plane through it.
A raw z-stack of fluorescence images is not yet a viability measurement — it is hundreds of megabytes of pixel intensities. Automated computational reconstruction stitches the optical sections into a coherent 3D volume, segments individual cells or nuclei in that volume, classifies each as live or dead from its fluorescence signature, and sums the results into an objective, reproducible viability score for the entire organoid.
Reconstruction software (e.g. Imaris, Arivis, napari, or open-source pipelines built on ImageJ/Fiji and scikit-image) aligns the sequential 2D optical sections along the known z-axis spacing to form a single 3D voxel volume, with each voxel carrying an intensity value for every fluorescence channel acquired (typically a live channel and a dead channel, sometimes alongside a nuclear counterstain such as Hoechst or DAPI for total cell counting).
Because the z-step and pixel size are known from the acquisition settings, this reconstruction is metrically accurate — distances and volumes in the reconstructed 3D image correspond to real physical distances and volumes in the organoid, not merely a qualitative depiction.
Individual nuclei or cells are then segmented from the 3D volume using either classical algorithms (3D watershed on the nuclear channel, intensity thresholding, distance-transform-based splitting of touching objects) or deep-learning segmentation models trained specifically for dense 3D biological volumes (e.g. StarDist-3D, Cellpose 3D, or custom-trained U-Net variants).
For each segmented object, the software measures the mean or integrated fluorescence intensity in the live channel and the dead channel within that object’s boundary. Objects are then classified — typically by simple thresholding calibrated against positive and negative controls, though more sophisticated pipelines use per-cell intensity ratios or trained classifiers — as live, dead, or ambiguous. This is done identically and reproducibly for every one of the thousands of cells across the entire volume, not just for cells that happen to lie in one arbitrarily chosen plane.
Once every segmented cell across the full 3D volume has been classified, the overall viability metric is simply:
Viability (%) = (live cell count or live cell volume) / (total cell count or total cell volume) × 100
Because this sum spans the entire reconstructed organoid — surface, mid-depth, and core alike — it is not vulnerable to the sampling bias inherent to single-plane imaging. It also enables spatial viability mapping: viability can be reported not just as one number for the whole organoid, but as a function of radial distance from the organoid surface or centroid, directly revealing whether cell death is uniformly distributed or concentrated in a specific region such as the core.
This shift from a manually eyeballed single image to an automated, full-volume, per-cell quantitative pipeline is what converts organoid viability assessment from a qualitative impression into a reproducible, statistically comparable measurement suitable for drug screening and disease modeling.
The clearest demonstration of why 3D quantification matters is core necrosis: a region of dead or dying cells at the geometric center of the organoid, driven by the same diffusion limits that make large avascular tissues vulnerable to interior hypoxia and nutrient starvation. A surface-only glance, or a single optical plane grazing the periphery, will show a healthy, brightly live-stained exterior — and completely miss a substantial dead core hidden beneath it.
As an organoid grows beyond the effective diffusion range of oxygen and nutrients from the surrounding medium (roughly 150–200 μm in densely packed tissue), cells at the geometric center become progressively starved of oxygen and glucose while simultaneously being unable to clear metabolic waste and CO₂. Initially this drives cells into a quiescent, low-proliferation state; if the deprivation persists or worsens as the organoid continues to grow, interior cells undergo necrotic cell death, and a dead core forms — often visible as a darker, more compact, and eventually cavitating region at the organoid’s center in brightfield images, and unambiguously visible as a red (EthD-1-positive) core in a fully-reconstructed 3D live/dead volume.
This is a well-established phenomenon from decades of multicellular tumor spheroid research and applies equally to many organ and tumor organoid systems: the same physics of diffusion-limited transport that shapes spheroid biology governs organoid viability distribution.
Because the necrotic core is, by definition, buried beneath a shell of healthier peripheral cells, any assessment method that does not sample the interior will systematically overestimate viability. A researcher glancing at the outer surface under a standard microscope sees a bright, uniformly live-stained exterior and would reasonably conclude the organoid is healthy. A single confocal plane through the equator that happens to catch mostly the shell (which occupies more solid angle than the core, geometrically) will likewise report high viability, diluted only slightly by whatever fraction of the core the one plane happens to intersect.
Only a full 3D reconstruction that segments and classifies cells throughout the entire volume — shell and core alike — can correctly weight the necrotic interior in the final viability calculation, and only volumetric or full-stack visualization can reveal the necrotic core’s size, shape, and severity directly.
This is the central practical payoff of full 3D optical sectioning over 2D or surface-only imaging: it is often the only way to detect core necrosis at all, let alone quantify it accurately. For drug efficacy and toxicity screening in particular, missing a necrotic core can mean mistaking a struggling, dying organoid culture for a healthy one — a failure mode with direct consequences for experimental conclusions.