Challenges for Quantitative Analysis of Cellular Mechanics
The goal of our discussions at the Chicago 2015 workshop was to identify key challenges facing the development of quantitative cellular mechanics, and to outline possible solutions to some of these challenges.
1. how to integrate mechanical measurements into biological research
A. make the case for newer mathematical modeling tools, faster tools, using those tools to get into integrated system representations.
B. how to combine models for individual components into a larger scale integrated model.
C. do we need detailed models to explain mechanics?
D. emphasize need for actually doing physical measurements instead of just jumping all the way from model to cellular behavior
E. A challenge is deciding what would you want to know, mechanically, about a system? à depends on the system, different parameters of interest depending on system.
F. theory has a role for suggesting what measurements would be useful or interesting.
G. validation for absolute measurements – how do you check if what you are doing makes sense
2. Issues that arise during Experimental measurements of mechanics
A. does a measurement actually measure what you want it to measure? For example: does tissue cutting still give you tension if you have knocked down some other component? interpretation depends on the model. instead of saying that laser cutting reports on tension, say that it reports on recoil velocity which in our model relates to tension…. example: if you knock down something that changes the viscosity so that you think you are seeing a change in tension but actually it's a change in some other parameter that links recoil velocity to tension. One must have a model and assume that noting else besides tension is changing in the model.
B. how to translate mechanical field measurements into molecular entities that can be more directly related to biology/medicine
C. for controls/inputs/knobs, how can we know if we are actually turning multiple knobs at the same time because eof feedback.
D. what is an acceptable level of error in a measurement.
E. assumption that moduli are linear and symmetrical.
F. to go between stress and strain one needs a constitutive law for the rheology ,and then one must state what assumptions go into that law.
G. The hidden assumption in elastic models is not only the moduli but also the reference configuration. But sometimes the reference configuration can’t be measured in biology.
H. The ability to image so many things at single molecule is powerful, but then we don’t have a good way to bridge from that level to the larger scale mechanical measurements. How to bridge molecular scale force sensors to cellular scale mechaniclal measurements. Can one go from molecules to continuum with one type of linkage?
I. Are the systems we study large enough in terms of numbers of molecules so that mean field approaches are appropriate?
J. Even if one have stress tensors for all linkages across an interface, is it possible to derive effective interfacial tension? Or does it also depend on the architecture of the network, clustering, etc. Even if we know all the forces on a subcellular object, can we necessarily infer what the deformation will be?
K. Pessimism for cell mechanics: we don’t currently have the means to make the kinds of measurements that we want. you can perturb for example with laser ablation but you can’t directly measure the response without using a model to interpret and that requires assumptions about unknowns in the model. ability to use models to interpret an indirect measurement becomes critical. need for more direct measurements.
One possible solution could be that with a good model we could find something that we aren’t currently measuring, but that would be much easier to measure, maybe not directly a force or stress measurement but still would be very informative about mechanical behavior.
L. There is a problem with the current climate that wants to have models and experiments together in one paper hard to review, makes it take longer, etc. As a result, the models are often relegated to a secondary status by the authors, or get poorly reviewed by the reviewers.
3. Collaborations between theorists and experimentalists
Testing theoretical models with experiments can have value for everyone, for example models can suggest alternative ways of measuring the same thing, and thus have value in driving new experiments. One obstacle noted is that modeling and experimental colleagues don’t always agree about when to publish how much to publish, where to publish. There is clearly a need to make sure everyone is getting what they need for career progression and for progressing the science.
Testing theory calls for better and more objective, quantifiable way to link up with collaborators. Some possible formats that were cited in discussion:
-Scialog does this so look to that for exploration
-Ideas lab at NSF
-Aspen center for physics
-Fisk-Vanderbilt bridge program
It was discussed that this type of structured interaction might work best if focused on a particular problem that is right for a joint experimental/modeling approach. For example, a hackathon. either based on one biological problem or based on type of approach, might provide a suitable framework to nucleate new collaborations.
Interactions should not just focus on getting things done but also teaching each other. The myth is that things are so hard you can’t do them yourself, but often you can use intuition once you learn about something, and then you would be able to make progress and then later on check in with super experts about details. One suggestion is to start any sort of hackathon or other event with a few tutorials to teach everyone about the problem and approaches.
4. How to study mechanics at different length scales
This was identified as a major challenge, but the only conclusion reached was that there is a need for more thought on this topic. The one cited example where it works is collagen – starting with MD of individual collagens in water, it was then possible to predict fibril mechanical response, and then use this information to build up networks and tissues. So one can really do a multi-scale holistic model. This raises the question of how hard would it be to go from something like that to conditions inside a cell?
If one just counts integrins and the forces on the integrins, they agree to within a factor of two with bulk force measurements at focal adhesions. but does it start to disagree as you get farther away from the integrin/ECM interaction? do we need to understand the dynamics of every molecule to predict how the cell will move etc?
At any level of description there should be a limited set of descriptors, that would then predict function at the next higher level. but how do we come up with those descriptors? how do we decide if we have the right descriptors? can we predict lots of different behaviors under different conditions or contexts (i.e. range of explanatory power), can you perform mutant or perturbation analysis using the exact same model and make accurate predictions?
5. Looking forward – what are some “Five Year Challenges”?
Five year challenge #1: Identifying critical state variables at various levels of description and trying to understand how those link to other scales. For example, a theorist might come up with a small number of state variables that should be important at the level of cell density or stiffness, but those variables are then determined by other variables at a lower molecular scale. Probably only a subset of those lower level variables matter. One five year goal is to define a protocol for arriving at what variables are important (using theory, for example, would be one approach, another would be experimentally testing lots of things to see what is important, evaluate situations in which different approaches have worked, which ones are most efficient and work , etc)
Five year challenge #2. Standardizing techniques that we currently have and making them more robust across different systems.
Examples of common approaches that should be standardized across systems:
Some of these different software works better for different specific cases maybe need a way to know when to use which.
standardizing tools for determining forces
how can we make tool development in to a viable career path. imageJ example -- came from NIH image so paid for by NIH. NCBI example can someone fund an organization to make build and organize tools, organize the tool building effort. and maintain standards. support and nurture people who work on this.
if you make a new tool how do you let people know about it?
Five year challenge #3. How to really use machine learning tools and which ones to use, not just abdicate intellectual responsibility. how to store and manipulate huge datasets. big problem with image data, huge datasets how to deal with it. can we get funding at a high level to set up infrastructure for this.
Action Items for future Workshops
- workshop to focus on integrating multiple scales
- workshop organized as a Hackathon to bring modelers and experimentalists together.