New Protocol Tracks Natural Killer Cells Attacking Tumors in 3D, One Cell at a Time

Patient-derived tumor organoids have transformed cancer research by preserving the architecture and behavior of primary tumors far better than cell lines or mouse models, yet most organoid systems still miss a crucial ingredient: the immune cells that decide whether a tumor lives or dies. A team led by Yanting Zhu, Bodong Liu and Jue Shi, publishing in BMC Methods, has now laid out a detailed, step-by-step protocol for building co-cultures that bring tumor organoids face to face with natural killer cells, the innate immune system’s fastest-acting killers, and for filming the resulting battle in three dimensions at the resolution of single cells.

The protocol centers on a three-component system combining patient-derived gastric tumor organoids, primary human natural killer cells, and endothelial cells, the vessel-lining cells that form a physical barrier around tissues in the body. A simplified two-cell-type version, omitting the endothelial layer, is also described, and the authors note that fibroblasts can be substituted for endothelial cells at the corresponding steps. The workflow was designed to fill a recognized gap: while organoid-T cell co-cultures are relatively well represented in the literature, standardized methods for incorporating natural killer cells and other key non-tumor components of the tumor microenvironment have remained limited.

Building the model requires several stages of genetic and fluorescent engineering. Tumor organoids are transduced with a lentivirus carrying a histone H2B protein fused to a red fluorescent protein, giving every tumor cell a glowing nucleus that can be segmented and identified in 3D images. Endothelial cells, in this case the HUVEC line, receive a green fluorescent nuclear reporter of their own. Primary natural killer cells, purified from human blood through negative selection and pre-activated for a few days with interleukin-2, are instead labeled with a cytoplasmic live-cell dye called CellTracker. The authors explain this choice on safety grounds: genetic modification often alters the phenotype and function of primary immune cells, whereas the dyes show minimal toxicity and preserve the cells’ killing activity. Nuclear dyes, by contrast, tend to be cytotoxic and can arrest the cell cycle, making them unsuitable for imaging sessions longer than about five hours.

Assembly of the three-cell-type co-culture follows a carefully timed sequence. Fluorescently labeled tumor organoids and endothelial cells are mixed and embedded in a basement membrane matrix for two days, allowing the endothelial cells to adhere and form stable interfaces over the tumor organoid surface. Meanwhile, the natural killer cells are stained, rested for at least half an hour to recover their polarized morphology, which the authors note is crucial for motility and cytotoxic activity, and then added to the mix. A typical imaging droplet contains roughly 20 to 30 organoids of 100 to 150 cells each, together with 30,000 to 50,000 natural killer cells, embedded in Geltrex at a final concentration of 6 to 6.5 milligrams per milliliter in a glass-bottom plate.

The imaging itself pushes confocal microscopy to its limits. Primary immune cells, and natural killer cells in particular, proved considerably more sensitive to phototoxicity than conventional cancer cell lines, so the authors used a highly sensitive spinning-disk confocal system, an Andor Dragonfly, to keep laser excitation low. With their setup, continuous imaging runs for up to 15 hours at a frame rate of one image every four minutes, with each 3D frame built from 27 to 30 z-stacks spaced three micrometers apart, covering a total depth of 80 to 90 micrometers. Those parameters are not arbitrary: natural killer cells move so quickly that frame intervals longer than four minutes compromise single-cell tracking, while too few z-stacks fail to capture the full structure of large organoids.

Turning the resulting movies into numbers requires dedicated software. The authors developed SiQ-3D, a deep-learning-based, Python-based image analysis tool that segments individual cells, classifies their phenotypes as live or dead, and tracks each cell’s trajectory through time. The software can reliably identify single cells when the signal-to-noise ratio is at least 2, and the code, along with example datasets and a user manual, has been deposited on GitHub, Code Ocean and Zenodo. From the segmented trajectories, co-localization analysis identifies which natural killer cells are in contact with the tumor at each moment, revealing interaction frequency, contact duration, motility patterns and directional movement toward the tumor.

Tumor cell death is read out directly from the movies. Because every tumor cell carries the nuclear reporter, lethal strikes can be spotted as nuclear fragmentation, allowing the researchers to measure the kill time, the interval between a natural killer cell’s first contact with a tumor cell and the onset of that cell’s death. Conventional endpoint cytotoxicity assays capture only the final tally of survivors; this approach exposes the intermediate events and rate-limiting steps in between, from migration and infiltration to conjugate formation and the killing itself.

The example results illustrate how strikingly different tumor subtypes can behave. In the absence of endothelial cells, one gastric tumor organoid line, GX068-TO of the intestinal subtype, strongly resisted natural killer cell killing, forming far fewer stable contacts with the immune cells, and those contacts were mostly transient. Another line, GX048-TO of the microsatellite instability subtype, was highly vulnerable: more than 30 percent of its tumor cells were eliminated within eight hours of co-culture, and its natural killer attackers frequently formed long-lasting contacts exceeding 30 minutes, some producing multiple tumor cell kills. The authors classify the first as an NK cell-resistant tumor and the second as NK cell-sensitive.

The most dramatic finding emerged when the endothelial layer was added to the sensitive tumor. An endothelial sheet forming over the surface of GX048-TO dramatically suppressed natural killer cell cytotoxicity, reducing NK cell infiltration and contact formation with the tumor by more than fivefold. In the three-cell-type model, most natural killer cell interactions with tumor and endothelial cells lasted less than ten minutes, and not a single one led to tumor cell death. The endothelial layer thus acts as a physical or signaling barrier that recapitulates an NK cell-excluded phenotype, mirroring the immune resistance observed in living tumors, and suggesting that therapies capable of breaching such barriers, including engineered or activated natural killer cells combined with immune checkpoint blockade or antiangiogenic drugs, might convert excluded tumors into sensitive ones.

The authors are candid about the method’s constraints. Fluorescent labeling is essentially mandatory, since label-free alternatives such as holotomography still lag behind fluorescence confocal microscopy in spatial resolution and sensitivity, complicating the distinction and tracking of different cell types. Normal organoids are harder to transduce with lentivirus than tumor organoids and may need electroporation instead. Photobleaching gradually dims the cytoplasmic dye signal in natural killer cells, though it remains adequate for tracking, and imaging parameters must be tuned to each experiment’s goals. The imaging analysis is optimized for organoids of roughly 100 to 150 cells, and although gastric tumor organoids serve as the representative model here, the authors state the workflow can be adapted to organoids from other tissues. Positioned as a preclinical platform, the protocol offers researchers a patient-specific system for profiling natural killer cell responses within a more complete tumor microenvironment, and for screening immunotherapies designed to overcome the barriers tumors erect against them.

Subject of Research: A protocol for 3D single-cell imaging of natural killer cell interactions with multi-cell-type tumor organoids

Article Title: Three-dimensional single-cell imaging analysis of tumor-natural killer cell interaction dynamics in multi-cell-type tumor organoid models

Article References: Zhu, Y., Liu, B., Li, X., Yan, H. H. N., Leung, S. Y., Yang, Z., & Shi, J. (2026). Three-dimensional single-cell imaging analysis of tumor-natural killer cell interaction dynamics in multi-cell-type tumor organoid models. BMC Methods, 3(1), Article 27. https://doi.org/10.1186/s44330-026-00081-8

Image Credits: AI Generated

DOI: 10.1186/s44330-026-00081-8

Keywords: tumor organoids, natural killer cells, 3D live-cell imaging, tumor microenvironment, co-culture, immunotherapy, endothelial cells, single-cell analysis, confocal microscopy, patient-derived models, cytotoxicity, Three-dimensional