In recent years, single-cell technologies have become indispensable tools for understanding the cellular composition of tissues, especially tumors. Techniques like single-cell RNA sequencing (scRNA-seq), ATAC-seq, and multiplexed cytometry all depend on a shared first step: tissue dissociation — the process of enzymatically or mechanically breaking down tissue to isolate individual cells.
This process, while widely used, raises an important and valid concern: to what extent does dissociation alter or destroy the morphology of the cells we aim to study? Specifically, when analyzing dissociated solid tumor samples using high-resolution brightfield imaging, are intracellular features — such as nuclear shape, chromatin texture, and cytoplasmic structure — still preserved well enough to carry biological meaning?
This question is particularly relevant for morphology-based platforms like ours, which rely on visual phenotyping of unlabeled single cells. Here, we examine the evidence.
What Happens During Dissociation?
Solid tissue dissociation typically involves a combination of enzymatic digestion (e.g., collagenase, trypsin, dispase) and mechanical disruption (e.g., pipetting, trituration) to break down the extracellular matrix and liberate single cells. These protocols have become standardized in the context of scRNA-seq and flow cytometry workflows and are used routinely across major tissue atlases such as the Human Cell Atlas and TCGA initiatives.
While dissociation undoubtedly removes cells from their spatial context and can alter certain membrane structures, multiple studies suggest that key intracellular features are preserved, particularly those residing in the nucleus or cytoplasm. Indeed, most single-cell workflows assume — either implicitly or explicitly — that the resulting cells remain viable and representative of their in situ state, albeit outside their original architectural niche [1].
Cytology as a Precedent
A useful reference point is cytopathology, which has long relied on dissociated cells for diagnosis. Pap smears, fine-needle aspirates, and fluid cytology are all based on single cells removed from tissue. Despite lacking tissue architecture, cytopathologists have developed robust diagnostic criteria based on morphological hallmarks like:
- Enlarged nuclei with irregular contours
- Hyperchromatic or coarse chromatin
- Abnormal nuclear-to-cytoplasmic (N:C) ratios
- Cytoplasmic granularity or vacuolation
 |
| Real cell images from the nuclear-to-cytoplasmic (N:C) ratio survey [9]. |
These features — all visible in dissociated cells — remain central to diagnostic workflows across cancers of the cervix, thyroid, breast, and lung [2,3].
Quantitative Morphology and Image-Based Phenotyping
In parallel, image-based phenotyping has emerged as a quantitative counterpart to traditional microscopy. High-resolution brightfield imaging, when coupled with computational models, allows objective extraction of morphological features from single cells. This approach has found increasing application in drug discovery and phenotypic screening [4].
Recent studies have extended this to clinical samples. Recursion Pharmaceuticals, for instance, demonstrated that brightfield images alone can capture ~90% of phenotypic signal otherwise obtained through fluorescent Cell Painting assays [5].
 |
|---|
| The relationships between KOs are maintained between using the original CellPaint images, or the CellPaint images generated from brightfield inputs (Source: Recursion [5]) |
Similarly, machine learning models have been trained to infer gene expression, mutation status, and cell state directly from cell morphology, using dissociated cell images as input [6,7].
Evidence from Dissociated Tumor Cells
A recent collaboration between AbbVie and Deepcell evaluated whether morphology alone could identify drug-resistant phenotypes from dissociated solid tumor samples.
- Classifiers were trained on cell lines rendered resistant to chemotherapeutics and targeted agents (e.g., cisplatin, paclitaxel, binimetinib), using brightfield images captured after dissociation.
- The models achieved >80% accuracy in distinguishing drug-resistant from parent (drug-naïve) cells across all 6 drug–cell line combinations studied.
- Most notably, when applied to a dissociated patient biopsy sample from a lung cancer case, the morphology-based classifier predicted that 35% of the cells were paclitaxel-resistant — a result highly concordant with independent scRNA-seq analysis of the same sample, which estimated 45% resistance.
 |
| UMAP plots show how the DTC sample shares morphological features with both parent (drug-naïve) and paclitaxel-resistant H460 cells. Example images illustrate predicted parent and resistant cells from the DTC sample (Source: [8]) |
These findings suggest that, despite dissociation, morphological signatures relevant to therapeutic response are preserved and can be extracted from label-free images.
Dissociation: A Necessary Tradeoff?
It is true that dissociation introduces changes — cell rounding, loss of polarity, and potential stress responses. However, these alterations do not appear to obscure core morphological traits like nuclear morphology, chromatin texture, or cytoplasmic density — features that are consistently used to distinguish healthy from malignant or drug-sensitive from drug-resistant cells.
Moreover, dissociation is not only compatible with morphology-based methods — it is foundational to most single-cell technologies in use today. Platforms ranging from 10x Genomics to CyTOF all assume that dissociated cells can represent meaningful biological states, despite their removal from tissue context.
What we now know is that this holds true for morphology as well, particularly when imaging is performed with sufficient resolution and analyzed with robust computational tools.
Conclusion
The concern that dissociation may destroy cell morphology is valid — and worth examining. But the evidence points in another direction. From clinical cytology to modern AI-powered imaging, we see consistent proof that key morphological characteristics survive dissociation, and remain highly informative.
For researchers and clinicians alike, this opens the door to new diagnostic and phenotyping strategies that do not rely on tissue sections or molecular labeling — and can instead read biology from what is still visibly present: the shape, texture, and structure of single cells.
References
- Van den Brink et al., “Single-cell and spatial transcriptomics: the tissue-dissociation trade-off,” Science, 2020.
- Baloch & Livolsi, Cytopathology of the Thyroid, Springer, 2007.
- Orell et al., Fine Needle Aspiration Cytology, Elsevier, 2012.
- Chandrasekaran et al., “Image-based profiling for drug discovery: due for a machine-learning upgrade?” Nat Rev Drug Discov, 2021.
- Recursion Pharmaceuticals, 2024 preprint: “Brightfield phenotyping matches Cell Painting in accuracy.”
- Chlis et al., “IFC-seq: Predicting transcriptomes from morphology,” Cell Reports Methods, 2020.
- Herbig et al., “AI-based enrichment of rod photoreceptors from brightfield images,” Commun Biol, 2022.
- Ramathal et al., “Deep learning-driven morphology analysis enables label-free classification of therapeutic agent-naive versus resistant cancer cells,” bioRxiv, 2025. https://doi.org/10.1101/2025.01.22.634357
- Zhang, M.L., Guo, A.X. and VandenBussche, C.J. (2016), Morphologists overestimate the nuclear-to-cytoplasmic ratio. Cancer Cytopathology, 124: 669-677. https://doi.org/10.1002/cncy.21735
