This is a significant improvement (and I’ve written up my notes from the webinar below)…but is this level of doublets low enough for the kind of experiments single-cell users might be planning?
A brief summary of the problem: Earlier this year Fluidigm announced that a significant number of single-cell captures were actually two cells were being taken into the IFC for processing and not one. The medium size C1 96 IFC was capturing around 30% doublets (SD 10%); the HT IFC was capturing around 44% doublets (SD 23%); and the small and large size 96 IFCs were capturing around 10% doublets (SD 4%). The redesigned medium size C1 96 IFCs shows a >4-fold reduction in doublet rate at around 7%.
How did they fix the problem: Most problematic was the high number of chambers containing “stacked doubletsâ€ where two cells were stacked, one on top of the other, in the Z-plane of the nest site making them very difficult to identify by XY imaging alone. Non-staked doublets could already be identified by imaging of the IFC during the recommended protocols (although this is not trivial, and there are companies offering systems to help – see the end of this post). The major effort in redesign has been in the capture site on the IFC where the nest height has been decreased to be more “size appropriate” to the chip cell-size (see images below). This prevents one cell sitting on top of another in the capture site. Fluidigm showed high concordance to previous IFCs with respect to the chemistry so hopefully users should not need to run lots of validation experiments. Similar changes to IFC design to reduce stacked doublets are coming on the HT 800 cell IFCs soon.
|The 3 changes made to IFC design
Getting the best from your C1: Fluidigm also presented recommendations for maximising C1 performance. Their data underscore the importance of using imaging as a QC metric for assessing and monitoring quality in single-cell research (which is tough to do on DROP-seq, In-Drop etc). The current recommendation is to use nuclear staining, which they said was superior to Z-stacking in identifying doublets. Fluidigm suggested that this should add very little time to imaging protocols (5-30 minutes), and they verified that nuclear staining did not affect DGE in their cell types and experiments…however it is unclear to me what the real impact is on users in adding a nuclear staining step to their protocols.
Controls for single-cell: Testing your setup with mixed species experiments is probably one of the most important steps to take, in fact it is the first experiment recommended by the Drop-seq team to new users. Fluidigm assessed doublets with imaging and sequencing, using a 50:50 mix of mouse NIH 3T3 and human K562 cells. These were nuclear-stained and imaged as per normal protocols. Mouse NIH 3T3 cells were stained with Hoechst 33342 and Human K562 cells stained with SYTO 11 to identify species by microscopy prior to sequencing. After imaging, cells were processed according to the C1 mRNA Seq protocol. Correlation of imaging and sequencing data to identify doublets was high but not perfect. A useful observation was that empty chambers have less than 10% the average read-depth than cell containing chamber, making this a potentially useful QC metric for users.
Conclusions and recommendations from the webinar: The medium 96 IFCs now perform as expected (see the “but” at the end of the post). Greater than 90% of occupied sties are single cells, doublet rates have been reduced from 30% to around 7%, and the stacked doublet rate has dropped from ~30% to 3%. Using nuclear staining is recommended as it is more reliable and less time consuming than Z-stacking, but users should consider the impact of cell-cycling on this analysis. The distribution of cell size also affects capture performance (see figure below) and Fluidigm would recommend keeping this as tight as possible (however being very strict on cell size is likely to bias experiments even further in excluding some cell types in a biological system). Perhaps most important to remember is to use an orthogonal method to validate single-cell experiments. The new chips are coming soon and there is an exchange program for unopened chips just contact your sales person.
What should single-cell researchers learn from Fluidim’s woes: first and foremost that single-cell experiments almost always carry a risk of, or a certainty of, some data coming from 2-cells captured together. Poisson loaded droplet-based systems will suffer most as the percentage of cell-containing droplets increases, so you may be better off sticking to 1000-10,000 cells rather than trying to get 40k or 50k+ cells. Imaging your microfluidic chip is key.
Better understanding of the causes of 2-cells capture, and better tools to QC data will help. But the single-cell research community needs to develop tools that can cope with this kind of technical artefact such that it does not intefere too much with downstream analysis and interpretation.
Automated imaging of C1 chips: Many users experience a particular bottleneck when comes to visualising and documenting the cell capture on the C1 chips either the 96 or the newer 800 chips. Fluidigm has published microscope spec’s including the use of an automated stage and automated image acquistion software programmed to screen IFC capture sites. They use a “Leica model CTR 4000 and EXi Blue Fluorescence Microscopy Camera model EXI-BLU-R-F-M-14-C with 10x, 20x, and 40x objectives”.
The Wafergen ICELL8 includes an automated imaging solution to scan and detect single cells
The Cambridge Bioscience JuLi Stage is a system that can scan and image an entire C1 chip using florescent (GFP, RFP & DAPI) and bright-field phase contrast image at 4x, 10x & 20x magnifications in around 6 minutes.