Nature Methods recently hosted a webcast on multimodal single cell analysis, sponsored by Illumina. Peter Smibert from the New York Genome Center and Ranit Kedmi from the New York University School of Medicine each gave presentations about their work in applying a powerful, multimodal single cell analysis method called CITE-seq, developed in Dr. Smibert’s lab.
They then joined moderator and Nature Methods’ Chief Editor Allison Doerr to answer live questions from online listeners about CITE-seq. Dr. Smibert and Dr. Kedmi answered as many questions as they could tackle in the ~40-minute live Q&A session. Below, you can find their answers to additional selected questions posed by live listeners.
If you'd like to listen again, or if you missed the live webcast event, you can still register to hear the recording!
Q&A with Peter Smibert (PS) and Ranit Kedmi (RK)
Q: Is CITE-seq limited to tagging surface markers?
PS: For the moment, yes. Surface antigens are easily accessible to antibody staining. Accessing and specifically probing intracellular antigens with oligo-labeled antibodies while preserving both the cell itself and its RNA content is a challenging problem.
Q: What should I do if I want to tag a cell type without a highly specific surface marker, such as a fibroblast?
PS: This assay is only as good as the antibodies you are using. The advantage of using as many markers as you can is that you may find marker combinations that define different cell phenotypes.
Q: Isn't it easier to isolate cells using flow and then run single-cell RNA-seq?
RK: Those are two different approaches. CITE-seq profiles cell surface markers in an unbiased manner and can discover cell surface markers that represent cell type or cell state.
Q: How can different tissue dissociation methods affect the detection of cell surface proteins?
PS: All dissociation methods, especially those using proteolytic enzymes, have the potential to alter the surface proteome of the cell. I would recommend testing the effect of your dissociation protocol on the markers of interest by flow cytometry before commencing CITE-seq experiments.
Q: Does adding antibodies to a cell modify its properties?
PS: This is a possibility. It is well known that certain antibodies binding to surface receptors can activate those receptors. We try to perform staining at 4°C (or on ice) and keep incubations as short as possible to limit the transcriptional response to the staining.
Q: A downside of this technique could be that some of the antibodies used in the experiment could trigger receptors and activate immune responses. Is it possible to fix the cells to overcome this?
RK: Some fixed epitopes can lose the ability to be recognize by the antibody. I suggest to work on ice.
Q: How much did you have to optimize antibody concentrations for large numbers of antibodies (e.g. via titration)?
PS: We did not do much ourselves. BioLegend have done a lot of this work behind the scenes. For balancing this large panel we did correct for overrepresented sequences by adding unlabeled antibodies to the panel.
Q: We have observed a false negative of CD4 in rat spleen samples. Can your comment some of your experiences with false negatives?
PS: Back when we were conjugating our own antibodies, we would see a false negative (i.e. failure to see enrichment in a cell-type known to express the marker) around 5-10% of the time. We attributed this to the random nature of our conjugation chemistry: attachment of an oligo to an amine in the epitope recognition site would almost certainly destroy the ability of that antibody to bind its epitope. BioLegend's conjugation chemistry is not random and therefore does not have the same issue.
Q: Are you worried about non-specific binding when you have hundreds of antibodies? Did you measure the background?
RK: We use isotype to measure nonspecific background. In addition, the CITE-seq data is normalized and scaled.
Q: You have used 220 antibodies on 10x Genomics single cell version 3 kit. Did you see any reduction in RNA sensitivity using these polyadenylated antibodies?
PS: We see typical numbers of genes and transcripts per cell in these experiments. While we have not done a direct comparison for the 220-antibody experiment (same cells, with and without antibodies), we have previously done these experiments very carefully with cell hashing antibodies (that have very high numbers of epitopes per cell) and see no reduction in sensitivity. You can see this work in Stoeckius et al, 2018 in Genome Biology.
Q: Is it possible to buy panels of CITE-seq antibodies commercially?
PS: BioLegend have just recently made an optimized human TBNK panel of 9 markers available. It is my understanding that there are several other optimized panels in development.
Q: Can you use the CyTOF built panels with CITE-seq? Are they compatible?
PS: You could certainly make panels with the same clones, but they would need to have barcoded oligos attached rather than metal tags.
Q: When using hashing to identify doublets, how far can you push with superloading before the 10X machine starts to clog? How far past the 17k cells per lane that 10x recommends can you push the superloading?
PS: We have never seen a good single cell prep clog the 10x channel. The only time we see clogs are if dissociation is poor. Supporting this, a recent preprint from Christoph Bock's lab (that you should really check out - scifi-RNA-seq) has taken superloading to levels far beyond what we have done, without appreciable clogging.
Q: Because of the oligo-label structure, does CITE-Seq only work with a 3' sequencing approach?
PS: polyA oligos are designed to use with 3' tag approaches. We have demonstrated that specifically designed reagents compatible with 10x Genomics immune profiling solution enable CITE-seq and cell hashing in conjunction with 5' tag cDNA detection. This approach also allows detection of other modalities including direct capture of guide RNAs. Please see Mimitou et al, 2018 for more information.
Q: How does the PCR handle work?
PS: The PCR handle is a common sequence present on all CITE-seq or cell hashing tags that comprises part of one of the standard Illumina read 2 priming sequences. After PCR with the appropriate primers, the resulting library is fully compatible with standard Illumina sequencing conditions.
Q: Could you explain more about the mRNA false negatives? Is their cause dropouts?
PS: Yes. Dropout is a common problem in single cell RNA-seq. While methods are improving, there are far more mRNA molecules in a cell than are captured by high-throughput scRNA-seq methods.
Q: Please comment on artefacts by doublets, especially doublets of two cells of similar subsets.
PS: Doublets are a challenge and it is best to either avoid them or design experiments to enable their detection and removal. This is one advantage of cell hashing and related methods where a majority of doublets can be detected and ignored.
Q: How do you address different staining patterns for the same marker? Does it make sense to use multiple antibodies for the same cell surface marker to make sure all cells with the expression are stained?
RK: From my experience, the pattern of staining intensity by CITE-seq represents the actual stain you can get by flow cytometry.
Q: How much correlation across the different markers between the FACS data and the antibody data from CITE-seq did you see?
RK: Some antibodies didn’t work. But from those that worked, each validation by flow was successful.
Q: Do you see any (quantitative) discrepancies between CITE-seq antibodies and the respective mRNA for the same antigen?
RK: Yes. CITE-seq seems to have much lower false negatives.
Q: I'd like to learn more about the sequential gating strategy software tools that were used on the CITE-Seq data.
RK: For gating we simply load the data on FlowJo, a flow cytometry software tool.
Q: What software for data analysis to you recommend?
PS: Seurat, developed and maintained by our close collaborators in the Satija lab is the tool we most commonly use. This group has put together a lot of tutorials and vignettes demonstrating how to use Seurat for CITE-seq and cell hashing data. 10x also now supports "Feature barcoding" which is a very similar concept to CITE-seq.
Q: How do you group cells which are the same type but show non-identical transcription patterns due to different stages of expression?
RK: UMAP clustering of CITE-seq results can segregate cells based on differential gene expression that later need to be assessed by the scientist (as with any single-cell RNA-seq approach).
Q: Is it possible to determine the polarization of a protein within a single cell?
PK: Not with CITE-seq as described.
Q: Can CITE-seq be applied to smaller cells, like bacteria?
PK: We haven't tried.