Since I was just writing about proteomics, this paper would probably be a good thing to highlight. Over the years, I’ve mentioned an effect that is constantly in the background of all chemistry and biology experiments, one that occasionally raises itself up and causes some major trouble. Small molecules and proteins alike can stick to the sides of the containers we use to work with them, and the smaller the sample, the more you have to worry about this effect. As proteomics moves to single-cell levels, that could be a real concern – some of the proteins of interest are going to look artificially poorly expressed because they’re being lost in transit during the workflow.
The problem is, we really don’t have good numbers on just how severe the problem is, since there’s a lack of quantitative data. What seems certain, though, is that there will be (are already) errors in the measurements we have. Proteins vary widely in adsorption properties with different materials, since they vary so widely in hydrophobicity, charge, and other factors. The single-cell genomics/transcription measurement have come further in dealing with this problem, and the paper linked suggests that the proteomics community learn from their experiences. That’s going to involve benchmarking with known concentrations of a wide range of pure proteins, to get an idea of where the biggest difficulties might be, and it would also be good to have some orthogonal techniques to provide reality checks on each other.
A lot of the literature on this topic comes from the fill-and-finish end of the industry, because stability and delivery of therapeutic biomolecules obviously will depend on the packaging. That link will show you another possible source of error in single-cell proteomics: even if an adsorbed protein does come back off a surface, it might be denatured by that contact, and this could lead to a different proteolysis behavior in that stage of the workflow. All of these weird possibilities on top of the general issues in single-cell assays, which are nicely summarized here. For starters, when you’re picking out single cells, you have an extra burden of picking out the right cells (i.e. the ones that will provide you with the best look at the effects you’re interested in), and how do you know if you’ve done that or not? Cells are constantly changing under pressure from their environments, including the rather odd environment of being yanked out and sampled, and it could be hard to tell if you’re measuring things in the state that you’d like. Ideally, you’re going to sample a lot of single cells, one at a time, and see what the differences might be, but that very sampling program could be skewed by the sort of systematic errors typified by the container-wall issue.
Finally, consider the topic of that earlier blog post, the application of machine learning to those proteomic fragmentation patterns. Robust ML depends on robust data going in, of course, and if different data sets are collected under different conditions, down to the materials in the sample tubes, well, things could get fuzzier. We have a lot of work to do! It’s a good thing that we have so much automation to try to power through it. . .
