Intracellular trafficking can be described as a dynamic exchange between membrane-bound organelles and/or the plasma membrane of eukaryotic cells. This exchange concerns transmembrane or membrane-bound proteins, such as channels or receptors and their bound ligands as well as macromolecules that are taken up or secreted by the cell. Trafficking is generally classified into two major pathways: the secretory pathway from the endoplasmic reticulum (ER) to the Golgi complex and plasma membrane for neosynthesized macromolecules, and the endocytic pathway from the plasma membrane to endosomes and lysosomes or to the Golgi complex and ER (Scott et al., 2014). Several additional trafficking pathways have been identified, e.g., unconventional secretion from endosomal compartments (Rabouille, 2017) or the various exchanges between different organelles (López-Doménech et al., 2018; Quidwai et al., 2021; Baltrusaitis et al., 2023) Some cell types additionally reveal different types of secretory granules in the cytoplasm that originate from the same pathway as the secretory vesicles however are sorted differently at the Golgi complex.

Trafficking relies on transport carriers that are characterized by vesicular-tubular intermediates and is supported by a complex machinery made of several families of proteins (Figure 1A; Box 1). The budding of vesicular-tubular transport carriers from a donor membrane requires the recruitment of membrane-deforming proteins, membrane-constricting proteins and coating proteins. Once the scission with the donor-membrane is completed, vesicular-tubular transport carriers are transported along the cytoskeleton with the help of motor proteins and fuse with a receiving membrane. Different types of adapter molecules connect motor proteins to transport carriers during trafficking. The fusion with a receiving membrane is initiated by tethering molecules and facilitated by a fusion machinery. The different steps of trafficking are regulated by small GTPases of different families that by themselves are regulated by Guanine Nucleotide Exchange Factors (GEFs) and GTPase Activating Proteins (GAPs).

Many genes coding for the trafficking machinery are deregulated in cancer. Yet, despite decades of study, there is no general description of their contribution to carcinogenesis: indeed whereas some members of the trafficking machinery can be downregulated in some cancers, they are upregulated in others, e.g., Caveolin-1 expression is upregulated in pancreatic and prostate cancer while downregulated in lung and breast cancer (Chatterjee et al., 2015; Thompson et al., 2009; Volonte et al., 2017; Wang et al., 2020). The last decades have shown that trafficking can adopt to different perturbations and change in response to cues from the environment and due to genetic or epigenetic modifications. The adaptability of trafficking pathways supports cell plasticity, a major characteristic of cancer cells (Muthuswamy and Bin, 2012). Very little trafficking proteins have been therapeutically targeted due to their critical physiological roles (Di Marco et al., 2020). To gain a deeper understanding of the role of trafficking in cancer, we propose to zoom out to a higher level: Trafficking pathways shape intracellular patterns, which we recognize as organelles. Indeed, the ability to create patterns spontaneously is one outstanding conserved feature of life. Jacques Monod even included this criterion in his definition of life under the name of autonomous morphogenesis (Monod, 1970). Pattern formation in living matter appears at many scales and has been well described on the organism level, e.g., during development (Belintsev et al., 1987). Intriguingly, much less is known about intracellular pattern formation of organelles through trafficking processes. Here, we will discuss organelle patterns as a superior scale of trafficking in the context of cancer.

Although widescale efforts have been made to establish atlases of genomic and proteomic alteration in cancer, no atlas of organelle-level changes has been profiled till date. Systematic studies on intracellular organelles are difficult, because, on the one hand, in vivo approaches are limited by the access of the samples and the lack of subcellular resolution, and on the other hand, in vitro cultured cells display a dynamic shape and strong morphological cell-to-cell variation. To address these limitations we and others have employed novel bioengineering and image analysis approaches: normalized cell culture conditions on adhesive micropatterns have been combined with spatial statistics approaches based on probabilistic mapping (Schauer et al., 2010; Jerabkova-Roda et al., 2023), or novel artificial intelligence (AI) methodology (Wang et al., 2023) has been implemented in order to quantify organelle spatial organization and topology. Using these technologies, the hypothesis was tested that organelle patterns in cancer cells are different from those of normal cells and change during aggressiveness (Mathur et al., 2023; Jerabkova-Roda et al., 2023; Wang et al., 2023). The rationale of this was that the observed vast alterations in trafficking processes in cancer cells will give rise to altered patterns of organelles as organelle emerge from trafficking.

First, using a bladder cancer model and well-controlled culture conditions, lysosomes of normal human urothelium (NHU) were compared to invasive and non-invasive bladder cancer cell lines that represented low-grade and high-grade bladder cancer, respectively. Lysosomes are dynamic, acidic organelles for cell clearance and recycling of macromolecules and act as cellular hub for metabolism and signaling. While in NHU cells lysosomes were positioned centrally, they were peripherally dispersed in bladder cancer cells with a stronger phenotype in ‘high-grade’ cell lines (Mathur et al., 2023) (Figure 1B). Similarly, lysosome positioning was altered in patient-derived melanoma cells and patient biopsies, scaling with and supporting melanoma aggressiveness (Jerabkova-Roda et al., 2023). Interestingly, the team of M. Barroso recently integrated artificial intelligence (AI) and imaging quantification to analyze organelle spatial distribution in the breast cancer model. They found that organelle topology allows for a highly precise differentiation between cell lines of different subtypes and aggressiveness (Wang et al., 2023).

Together, these results indicate that alterations of organelle patterns correlate with transformation and aggressiveness of cancer cells. Moreover, when lysosomal perinuclear clustering was induced experimentally in patient-derived melanoma cells, we observed significant reduction of invasive outgrowth in mouse and zebrafish models (Jerabkova-Roda et al., 2023). This study provided a direct demonstration that lysosomal positioning controls cell invasion, illustrating the importance of organelle adaptation in carcinogenesis. Thus, we argue that the patterning of organelles, which represents trafficking collectively, could potentially not only be used as a biomarker in the future, but should also to be considered to better understand, describe and fight cancer. In the next sections, we focus on a physics description of the organelle landscape to identify critical parameters for a stable and non-random organelle organization characteristic of healthy cells.

Based on experimental data, we argue that organelle organization has three key properties: non-random distribution, Non-Equilibrium Steady-State (NESS) and self-organization.

One main challenge is understanding how cell compartments’ non-random steady state self-organization is achieved and maintained from intracellular trafficking. In this respect, no single gene, or not even one gene regulatory network, organizes the cell, instead, organelle organization emerges from many genes. Thus, a comprehensive understanding of intracellular organization requires the investigation of the interactions between genes and the resulting emerging laws. In the next section, we propose to consider biophysical models that describe pattern formation. We review how these models were applied in the context of trafficking proteins to explain pattern formation of organelles. Due to their bacterial origin, mitochondria are special organelles that are distinct in their participation in the cellular trafficking pathways (Baum and Baum, 2014) and can undergo fusion/fission. Similarly, membrane-less organelles such as stress granules, P-bodies, etc., which are biological condensates formed by liquid phase separations, rely less on classical trafficking machinery described in Box 1. Thus, their pattern formation could be described by different models than whose used for instance for endosomes or Golgi complex.

The different ways to create patterns are not mutually exclusive, but rather are working in combination with a strong interplay between them. Indeed, particles such as protein complexes or organelles could shuttle between heterogeneous diffusion in the cytosol, tethering to the cytoskeleton due to a chain of interactions, and convective motion driven by molecular motors walking on cytoskeleton. Building on the idea that particles can switch from one state to another, models have been proposed that couple different states leading to the addition of reaction terms. These seminal models for spatial distribution of particles are reaction-diffusion-convection models. Often they can be reduced to a simpler diffusion-convection model thanks to an approximation on the different time scales of these processes. The simplification relies on the hypothesis that the reaction timescales are small compared to convection/diffusion timescales. For example, an organelle distribution model with three states: i) diffusion ii) microtubule (+) transport and iii) microtubule (−) transport that can be modeled by a set of Partial Differential Equations (PDEs), (Figure 2D). The first part is driven by thermal fluctuations and characterized by its diffusion coefficient. In contrast, the second and third parts are mainly achieved by energy-consuming processes relying on molecular motors that create an emerging convective flow characterized by a velocity. As an illustration, simulated organelle distributions are shown based on a convection-diffusion equation in 2D disk cells for different Péclet numbers (Figure 2E). This simulation illustrates that the clustering increases with the Péclet number. Some other scholars rely on reaction-diffusion models, since the seminal Turing paper of 1952 (Turing, 1952). The Turing model can reproduce complex self-organized morphogenesis patterns (Kondo and Miura, 2010) in a wide variety of situations (Kondo and Asai, 1995; Jacobo and Hudspeth, 2014; Blagodatski et al., 2015; Sekimura et al., 2015; Fofonjka and Milinkovitch, 2021). It has been recently proposed that the Turing model is a general principle of cellular self-organization at the molecular scale (Halatek et al., 2018). However, the Turing model has a robustness problem, because diverse variations in parameters, such as initial conditions, presence of noise, or delay can vanish the pattern (Bard and Lauder, 1974; Murray, 1982; Maini et al., 2012) while experimental work indicates that organelle landscape is robust (Duong et al., 2012; Guardia et al., 2019). Moreover, neglecting the convective flow seems accurate for chemical systems, but is too simplistic for organelle patterning that clearly relies on motor-driven transport.

A lot of work on the experimental and theoretical part of the analysis of organelle distribution has been performed in hyphae of the filamentous fungi Ustilago maydis (Lin and Steinberg, 2017; Gou et al., 2014; Higuchi et al., 2014; Lin et al., 2016). The efficiency of modeling in hyphae is due to its constant and simple cylindrical morphology. Some modeling has also been performed on animal cells. For example, the distribution of endosomes (Pangarkar et al., 2005) and lipid droplets in embryonic Drosophila cells (Maly, 2002), as well as endocytosed virus (Lagache and Holcman, 2008; Lagache et al., 2009) was efficiently modeled by the Fokker Planck equation. Thus, convection-diffusion models are able to reproduce the spatial organization of organelles in a large diversity of situations, and a general model of organelle distribution has been proposed (Dinh et al., 2006). According to the chosen parameters, this model generates different kind of organizations. Surprisingly, this simple model can produce patterns similar to a good range of experimental data.

Some papers use the full reaction-diffusion-convection model (Smith and Simmons, 2001; Dinh et al., 2005). For example, 4-states models using reaction-diffusion-convection propose that melanosome distribution is controlled mainly by microtubule binding rate (Slepchenko et al., 2007). Another paper proposes that endosome clustering is controlled by a Péclet number and an organelle interaction coefficient (Pangarkar et al., 2012). The common point of these models is that they do not generate instability or oscillations, confirming the stabilization effect of convection.

Based on recent observations of organelle changes in cancer we argue that carcinogenesis could be seen as a loss of a reference steady state organization. In this context, the increasing mutations in diverse genes, including trafficking proteins, either could support the loss of the reference state or could compensate this loss. Interestingly, organelle steady state dynamically changes during cell division (see Table 1): it seems to oscillate along the cell cycle when observed at longer time scales (Carlton et al., 2020). Notably, daughter cells re-find a comparable steady state as the mother cell after cell division under homeostasis (Figures 3A-D). Contrary, cancer cells seem to change at these time scales (Figure 3C). Thus, whereas healthy, differentiated cells could be described as being in an oscillatory state at long time scales, cancer cells can neither be considered as steady nor oscillatory (Figure 3D). Cancer cells seem instead to diverge versus a dynamic state that we define as plasticity.

Notably, a chaotic regime seems to be easily achieved in biological systems, as both oscillations and chaos can emerge in a non-linear system with time delay (Mackey and Glass, 1977). Oscillations/chaos have been observed in many biochemical reactions (Sel'kov, 1968; Aon et al., 2011; Benincà et al., 2015; Olsen and Degn, 1977; Decroly and Goldbeter, 1982; Olsen and Degn, 1985; Baier and Sahle, 1998). However, chaotic regimes have not yet been observed for the organelle landscape. Future investigations could reveal whether cells have mechanisms to avoid chaotic regimes altogether, or if these regimes exist but have not been observed due to the lack of quantitative measures at the appropriate time scales. This question is especially relevant for cancer cells that randomly accumulate mutations, making their organelle landscape potentially unstable. Alternatively, intra-tumoral natural selection could stabilize whose organelle landscapes that correlate with the most aggressive cell behavior.

Modeling results illustrate that cell organization can be described by general physical principles. Thanks to this biophysical description, loss of organization could be considered as a stability problem of physical models. Interestingly, modeling allows to identify properties that contribute to stability. For instance, it is well known that the Turing model has a robustness problem (Bard and Lauder, 1974; Murray, 1982; Maini et al., 2012), but can be stabilized by the addition of convection. As convection often is represented by molecular motor activity, we speculate that the increased expression of motor proteins seen in melanoma could be interpreted as a compensation for the loss of stable organelle positioning at long time scales that is observed in aggressive melanoma cells (Jerabkova-Roda et al., 2023). Alternatively, active supply with ATP due to increased energy production, as often observed in cancer cells, could increase the convective flow indirectly.

Further, from the theoretical part, boundary conditions can also have an impact on the stability of the system. For example, it has been reported that Dirichlet boundary condition (i.e., fixed concentrations at boundaries) enhances stability of the Turing model (Dillon et al., 1994). It will be interesting to explore whether cancer cell interactions with their microenvironment including other cell types could represent some boundary conditions for organelle distribution. Indeed, the effects of boundary conditions have not yet been extensively studied in the context of organelle distribution.

Finally, it could be considered that alterations of specific gene families contribute in different ways to instability. For instance, Ras alterations could potentially contribute to organelle distribution changes through the induction of a strong chain of interactions, whereas motor proteins could contribute through changes in convection. Changes in families that regulate lipid composition of membranes or cell crowding could alter the diffusion behavior of membrane proteins or cellular macromolecules, and could contribute to organelle alteration through the diffusion term.

Different from cancer, genetic diseases based on finite mutations, such as lysosomal storage disorders or Huntington’s disease (Erie et al., 2015), could be characterized by a transition from one steady state organization versus a new pathological organization sustaining harmful functions (Figures 3B, D). In this case, the loss of the physiological steady state organization could be described as a modification in the parameters of these models. For example, mutations in motor proteins would change the Péclet number or their kinetic rate changing the reaction terms in the models without leading to instability.

Here, we zoomed out from the molecular functions of trafficking proteins and looked at one level above: the intracellular patterns of organelles that represent the integrated behavior of cellular trafficking activity. The rationale of this is that cancer cells reveal an altered intracellular organization of organelles at long time scales, and we argue that these changes in organelle patterning need to be considered to better understand, describe and fight cancer. Based on the fact that organization can be described by biophysical models, we propose to consider physical emerging properties such as instability of organization as an interesting but unexplored property of cancer cells. In the future, it could be propitious to perform rigorous stability analyses of different models to further identify sources of loss of the stable organization as seen in cancer cells at long time scales.

Of course, there are several limits of the discussed Partial Differentiation Equitation models due to several assumption that are made for simplification. These include the i) well-mixed system approximation that cytosol is homogeneous, ii) linear kinetics hypothesis that binding/unbinding processes can be modeled by first order reactions and iii) the choice of boundary conditions that are often under-estimated. Alternatively, some authors proposed an agent-based approach to model organelle dynamics (Mayorga et al., 2018). In agent-based modeling, the system is modeled by a set of discrete entities called agents. Each agent behaves according to a set of predefined rules. Emerging effects can appear from individual behavior of agents. These models are stochastic and more apt to describe local regulations (Dalmasso et al., 2017; Dinh et al., 2007) or feedbacks (Mayorga et al., 2017). Unfortunately, these models have not been used extensively to investigate the spatial distribution of organelles. Future work should show if they are consistent with models based on PDE, how they compare to these and which properties of complex systems we can learn from them. These models should be investigated and compared in normal versus pathological cell conditions such as cancer.

Investigating the dynamic intracellular landscape of organelles, showing fascinating pattern alterations, could allow the establishment of novel biomarker. Potentially, alterations in organelle patterns could show less heterogeneity than gene/protein expression profiles of different cancers. This could potentially facilitate the prediction of cancer progression or cell responses to drug treatment and resistance.

Finally, organelle patterns could also help reveal the cancer microenvironment and identify cancer cell addictions that make them more vulnerable to treatment than healthy cells. Intriguingly, organelle positioning strongly correlates with environmental cues (see Table 1). Therefore, changes in organelle landscapes observed in cancer could indicate which alterations are found in their microenvironment. For instance, peripheral lysosome trafficking in tumor cells was found to result from acidic extracellular pH, inducing cathepsin B secretion and tumor invasion. This behavior was reversed by inhibitors of sodium-proton exchangers (NHE) that induced a time-dependent retrograde aggregation of lysosomes (Steffan and Cardelli, 2010). More recently, in Glioblastoma stem-like cells (GSCs), organelle alterations have been linked to cholesterol addiction that revealed a vulnerability of glioblastoma to cholesterol-lowering drugs such as statins, particularly in conditions in which organelle alterations were aggravated (Maghe et al., 2024). Thus, interfering with cancer cell addictions, revealed by changes in cellular organization, could emerge as an effective strategy for cancer cell elimination.

The contribution of organelle dynamics in cell homeostasis and disease is still understudied due to lack of resolution, accessibility and cell-to-cell heterogeneity observed in classical cell models. Therefore, novel tools from mathematics, physics and engineering are urgently needed to study organelles under controlled culture conditions in higher dimensions. The focus on intracellular organelles constitutes a novel effort addressing cell perturbation beyond individual gene modification that will open exciting perspectives in the understanding, diagnostics and therapeutics of multifactorial diseases such as cancer.