Amoeboid motility is a complex cellular process driven by highly-organized cytoskeletal dynamics involving alternating cycles of pseudopodial protrusions and retractions.
In some cases, the subtle differences in motility between different strains may only be revealed by comparison of pseudopod dynamics. However, because of the highly stochastic fluctuations in the behavior of pseudopods, large data sets may be necessary, making manual detection, identification and quantification impractical. In this study we presented an automated method based on skeletonization that detects and characterizes pseudopodial behavior of cells. These algorithms have been illustrated on movies of chemotactic Dictyostelium cells.
Several methods for detecting pseudopods automatically or semi-automatically in microscopic cell movies have been proposed previously. In the 3D Dynamic Image Analysis System (3D-DIAS, Soll Technologies), pseudopods are identified based on manual identification of the "nonparticulate" cytoplasm using DIC images [2
]. In addition to being completely automatic, our proposed method has the advantage that only shape information is needed, thus allowing one to apply this to fluorescent images, making correlation studies like those of Figure possible. Moreover, though all of our examples are illustrated in 2D images, the skeletonization technique can be easily expanded to 3D shapes. Other applications for 3D skeletons have been proposed [36
Pseudopodial activities have also been identified by approximating the local curvature of the boundary using a closed polygon formed from a chain of nodes [4
]. This form of detection may depend on the magnification, image quality, as well as sampling density. In contrast, the skeletonization technique uses topological and geometrical information from the whole shape, and takes into account not only the local curvature, but also its relationship to the complete path of the membrane, thus allowing great flexibility in applications.
Related research, based on level set methods, tracks the evolution of virtual markers to determine the spatial and temporal dynamics of a region of interest on the cell membrane [3
]. In general, this method requires high resolution imaging in both space and time to deliver topologically consistent solutions. In contrast, our method does not need special boundary markers and can be applied to a relatively broader range of resolutions.
For accurate pseudopod detection, the time resolution of the movies should not be too low, so that a considerable part of the cell body overlaps from frame to frame. The valid range should be decided by the size and speed of the specific cell types of interest. In our case, cells move at ~10 μm/min, and the average length of the cell is 10 μm. The minimal frame rate we use is 10 s/frame, which is equivalent to a distance of ~1/6 of their body length between successive frames. At the same time, the spatial resolution must be selected so that desired pseudopodia activities can be identified at least by eye. The spatial resolution for most of our movies is either ~0.3 μm/pixel when using a 40× objective, or ~0.2 μm/pixel when using a 60× objective. The lower the resolution, the more shape features are lost during imaging, and fewer pseudopodial activities, especially subtle ones, can be captured. This will cause a general problem for all detection method, including manual counting. Other parameters in Table are chosen to minimize the discrepancy between results obtained using the automated method and manual observations of pseudopodial activity. These work well with different cell strains, frame rates and magnifications. However, when tracking significantly different cell movements, other parameters may need to be selected.
One possible drawback of relying on topological information is that skeletonization sometimes cannot detect protrusions with bleb-like structures, since the local curvature at these structures is relatively low compared to typical activities. This problem may also exist in other automated methods based on local curvature calculations. However, in most of our studies, bleb-like structures occur rarely and do not influence the results statistically. For example, there is only one bleb in 101 s in the movie [Additional file 1
] at time 59 s, which seems to be a short-lived pseudopod but is missed by the algorithm. Other than the blebs, we do not find shape structures that consistently cause the method to fail in the movies of Dictyostelium
cells we analyzed. Attempts have been made to characterize the extension of pseudopodia by amoeboid cells in the absence of external cues as well as in shallow gradient of chemoattractant, based on the assumption that spatial differences in chemoattractant receptor occupancy gives rise to biases in the direction of pseudopod extensions [4
However, other studies suggest that the bias might come from pseudopod retractions [27
]. Here, we view pseudopodial behavior as a dynamic process that includes both protrusions and retractions, and modeled this dynamic behavior using an autoregressive model - in which the state at a given time depends on the historical activities - for each pseudopod. Our results are consistent with models that suggest that, for a Dictyostelium
control cell moving in a shallow gradient, a considerable fraction of pseudopods experience both protrusion and retraction, and they tend to retract back when shifting far away from the right direction (Figure ). At the same time, the consistent protruding pseudopods may play a more important role in leading the movement of the cell centroid compared to other pseudopods (Figure ).
Published results show a discrepancy regarding the production rate of pseudopods, with manual counting reporting less pseudopods than methods that record pseudopods automatically. Our data illustrates that these differences can be attributed to a counting bias. In manual techniques, only the most prominent and persistent protruding activities are identified as pseudopods, whereas automatic methods are able to detect smaller protrusions and retractions. Thus, our method can be applied to quantify both the dominant deformations as well as subtler dynamic perturbations of shape. Our results suggest a higher ratio of "de novo" pseudopods relative to those from splitting during chemotaxis in shallow gradients, compared to previous published results (Additional file 2
]). We conjecture that this also comes from the fact that not all meaningful boundary activities are captured in these analyses. We note that for cells chemotaxing to a micropipette, the fraction of pseudopods arising from splitting is even lower (Table ).
By using the cross correlation method to analyze the molecular drivers of protrusions and retractions, we find that dynacortin, as a marker of F-actin, colocalizes with protrusion activities, consistent with the notion that actin polymerization drives protrusions. On the other hand, myosin-II is depleted from the front of the cells, and is enriched at the sides of cells. As myosin-II contributes ~20-30% to cortical viscoelasticity and to cortical tension [39
], the localized myosin-II modulated increase in cortical tension at the side of cells may thus help to inhibit lateral pseudopods, as previously suggested [40
]. This argues that myosin-II has a substantial role in enabling of a polarized morphology seen in cells.