Data Processing and Decoding
At CARTANA, we have a ISS Starter Program. Every new user of the technology is enrolled in this program to start successfully with the technology. This program provides individual technical support based on your specific tissue samples and experimental design.
If you have any questions, please contact email@example.com
We estimate that CARTANA ISS can detect about 10% of all expressed transcripts in cells. Therefore if your transcript is expressed at 100 copies/cell, we expect to detect about 10 signal dots/positive cell.
ISS technology is quantitative as one RNA transcript is detected as an ISS spot and therefore, it is possible to quantify gene expression levels and compare different experimental conditions. However, an absolute quantification is always challenging, due to some bias in detection. For example, factors like RNA structure and sequence GC content do affect it, but changes in expression levels can be observed and measured using CARTANA ISS. Furthermore, we can observe various expression levels between different genes and we estimate that a 5-fold (or higher) variation in expression levels can be quantified.
With regards to sensitivity, what are CARTANA's typical statistics for biologically interesting genes in terms of dots/cell, and how are error bars generated based on those detection counts?
The dots/cell varies with the gene of interest and typically, the error bars describe cell-to-cell variability. Depending on the experimental setup, variation between biological replicates can also be considered. One major advantage of the high-throughput feature of ISS is, that it results in a higher statistical power since it can assay many cells at once.
Yes, ISS and scRNAseq have been performed on the same samples.
Please refer to this publication for additional information:
Grundberg, et al. In situ mutation detection and visualization of intratumor heterogeneity for cancer research and diagnostics. Oncotarget (2013). DOI:10.18632/oncotarget.1527
What is the dynamic range of transcript detection, i.e. how few and how many copies can be detected on the same section?
The number of ISS spots that can be resolved depends directly on the microscope hardware settings. In-house, we use a 20x objective with 0.75 numerical aperture. ISS spots are between 0.5-1 µm big, so the question comes down to how many ISS spots can be resolved with that optical settings. An average number of a cell of 20 µm of diameter is in the 103 orders of magnitude.
When using 20X magnification, we estimate that we can detect 100-200 dots/ cells. This number increases to 500-1000 dots/cell when using 40X magnification.
We also have customized solutions in the sequencing pipeline that increase the resolution and that we can discuss with you on a case to case basis. We also highly recommend to inform us when you know that some of your targeted genes have very high expression levels, which allows us to anticipate potential imaging and analysis challenges and use customized solutions (please send us an email at firstname.lastname@example.org)
CARTANA ISS is a reproducible assay. To improve consistency and reproducibility between samples, we highly recommend to prepare and process sections as one batch. Reproducibility is good between CARTANA ISS experiments, but in order to merge data, some batch corrections might still be required.
Yes, one signal dot represents one RNA molecule as it is one ISS spot from the RCA reaction. The process of matching colour combinations in a specific order sufficiently removes many noises arising from fluorescence imaging like NGS and reads alignment. If imaging rounds are sufficiently long and optimally set up, it helps to find true signals versus noise.
Traditionally, segmentation has been done based on nuclear counterstain and/or membrane stain. This provides moderately accurate results. For cell typing analysis, where knowledge about cell types exists, the prior information can be used in junction to more accurately define cell boundaries: http://insitu.cortexlab.net/.
Qian, X. et al. Probabilistic cell typing enables fine mapping of closely related cell types in situ. Nature Methods (2020).
Recently, there have been some segmentation-free methods that purely use the relative spatial information of transcripts, with or without cell-type information.
Partel, G. & Wählby, C. Spage2vec: Unsupervised detection of spatial gene expression constellations. bioRxiv (2020)
Park, J. et al. Segmentation-free inference of cell types from in situ transcriptomics data. bioRxiv (2019)
For quantity of transcripts/cell, do you use number of puncta only or intensity as well? How quantitive is it, for instance detection of 2 fold/3 fold increase in transcripts?
For quantity of transcript/cell we use the number of puncta.
CARTANA ISS technology is quantitative as one RNA transcript is detected as an ISS spot, therefore it is possible to quantify gene expression levels and compare different experimental conditions.
Brain cells have different shapes and sizes, which depends on brain region. How confident can you be in assigning genes to individual cells, particularly in dense regions?
This is indeed very hard to achieve, not only because of size, shape, density of cells, but also because when assaying a section, we are looking at a plane of a 3D structure. Essentially, a compromise has to be reached between efficiency and accuracy. If we draw a parallel to single-cell sequencing, which also has high noise and suffers from problems like doublets and incomplete RNA content, what helps biological discovery is the statistical power that arises from profiling millions of cells, rather than few individual cells having “perfect” profiles.
For in-house analysis and mapping, we use MATLAB but CARTANA is currently developing a user-friendly platform for data exploration.
We are currently working on the development of a user-friendly software for CARTANA ISS results analysis.
The CZI starfish platform can be used to analyse CARTANA ISS Data analysis. You can find the pipeline and example dataset here :
However, CARTANA is actively integrating starfish into its pipeline. We are currently translating the MATLAB script into Python and also working on the development of a user-friendly platform for data analysis.
For tissues that are very dense, like tumor samples, how reliable is the segmentation to get single cell resolution?
Segmentation on tumours is difficult to perform since cells are growing in a disordered way and different cell types can be intermixed. One must check the performance of the segmentation to evaluate how well it worked. Many segmentation algorithms exist and one needs to test which one that works best on the tumour sample of interest.