top of page




Many cell types in the human body are characterized by a distinct and particular cell size. Keratinocytes, for example, are 14um in diameter. But, what determines the specific size of each of the different cell types? and why are cells regulated to have one specific size over another?




Cell size is regulated by several signaling pathways including mTOR, Hippo & Wnt. One question that remains unkown is: how can a common set of pathways specify different (and particular) sizes in different cell types? How are the particular cell size values encoded in the chemistry of signaling? 


Regulation of cell size plays role in a variety of important physiological processes and pathologies. Examples include the changes in cell size that accompany differentiation or the cell size abnormalities in cancer tumors.


In healthy tissues, individual cells (of a given type) are almost identical in size. This size uniformity suggests a regulation that specifies a common target size to the numerous individual cells in the tissue. What are the mechanisms responsible for this regulation and why is this regulation aberrant in pleomorphic tumors?


To address these questions we combine single cell microscopy with genetic and pharmacological approaches to measure and perturb cell size. We apply various mathematical approaches to deconvolute the influence of different effectors of cell size, such as growth factors and cell cycle stage.



Cell size in the germline of C. elegans (imaged by Daniel Witvliet in collaboration with Brent Derry)

Different cell types in the human body are quite different in size. While signal transduction pathways that promote increases or decreases in cell size have been discovered, it is not known how these pathways specify one particular size over another.




Drugs are discovered by screening compounds for their influence on targets that are thought to underlie disease. Screen assays are typically bulk measurements - measurements that are averaged over the population of single cells that are exposed to the drug. This approach accepts the simplifying assumption that individual cells respond to a given compound in more or less the same way. Reality, however, does not conform to this simplification.


We are developing mathematical methods that exploit single cell variability to make inference onto the signaling that controls reporter targets. 


Over the past decade, significant pharmaceutical efforts have been directed towards screening for compounds, such as Imatinib (Gleevec), that specifically inhibit the “one” target or oncogene responsible for a pathology. While the approach has not been fruitless, its success is dimmer than its’ anticipated promise. One difficulty is that certain proteins, including oncogenic Ras, were found difficult to effectively inhibit. However, even when potent inhibitors are identified therapeutic success is often limited. In part, this is because current strategies target an oncogene in isolation from the network of signals within which it is embedded. Inhibiting oncogenic BRAF, for example, with the recently developed vemurafenib relives BRAF’s feedback inhibition of HER2, which affectively offsets therapy by giving rise to drug resistance. As a result, there is a growing urgency to identify compounds that target the “channels of communication” with which a pathogenic process communicates with, rather than the “one” specific target. 


Ergodic Rate Analysis (ERA) recovers growth rates from fixed time point measurements.

In 2013 we published a letter in Nature describing ergodic rate analysis (ERA), a method that retrieves a precise account of the intracellular dynamics of a labeled protein or phospho protein from a single snapshot image of fixed cells.  The novelty of ERA is that the temporal profile (dynamics) of a labeled target is obtained without the expensive need for temporal measurements (e.g. time course or time lapse measurements). While the mathematical details of ERA escape the scope of this overview, the intuition underlying it is simple: faster (more transient) molecular events will be observed in fewer cells in an imaged population. The logic underlying ERA can also be utilized to uncover additional details of the regulation of a labeled reporter target. A feedback inhibition on phospho-Akt, for example, would be revealed by “fewer than expected” cells with increased pAkt levels in an imaged population. Again, the latter statement is an intuitive over-simplification of a rigorously developed mathematical method.


In recent years we have been developing mathematical methods of inference to predict a compounds influence on the regulation of a reporter target from high throughput measurements with single cell microscopy. We anticipate that these statistical methods would not only reduce costs of drug discovery but also provide the ability to screen features that were previously "unscreanable".




Single cell measurements of cell size, DNA levels and a reporter of G1-exit (Geminin degron) are used to estimate a joint probability density finction that describes the dependency of cell size on cell cycle progression. The structure of this function is mathematically analysed to make inference on how cell size is regulated in cell cycle. 
Cellular Individuality


Individual cells, in healthy epithelia, are regular in their appearance. In contrast, individual cells in malignant populations are varied in size, shape and drug response. Variability in cell size is particularly noticeable in pleomorphic tumours - a subtype of cancers defined their increased morphological cell-to-cell variability. Presently, the roots of the increased variability in pleomorphic tumors is unknown. Nor is it understood how why variability is correlated with aggressive pathology. 


To study this question we  developed a model of pleomorphic xenograph breast cancer tumors in mouse. We combine single cell imaging and characterization of the cultured cells in vitro with measurements on tumors.

Increased variability in cell size is often a diagnostic of malignancy.

In addition, over the past years the Kafri lab has programmed a pipeline of software dedicated for image analysis to address questions by quantifying the response of single cells. With these, we are able to quantify multiple proteins in single cells that are cultured in vitro, follow changes in protein expression over time, and quantify signaling in tissue sections with single cell resolution. Also, recently, we have expanded our repertoire of tools to include identification of morphological features. Important, since the image processing code is generated in-house, it is adaptable to changes in demands and changing conditions.

The Kafri lab has been perfecting in-house software to automatically classify fibroblasts into distinct, computationally characterized morphologies. Importantly, morphological distinctions are known to classify not only transformed cancer and non-transformed cells, but also other various aspects of cell function. For example, the epithelial to mesenchymal transition is accompanied by distinct changes in morphology (Moreno-Bueno, G. et al., 2009). Similarly, morphological features distinguish senescent cells from non-senescent cells (Sikora, E, et al. 2016) or differentiated cells from cancer cells that become dedifferentiated. In other words, we will “ask the computer” to scan the hundreds of thousands of imaged cells and classify these single cells into groups that are morphologically similar.

bottom of page