High-dimensional multi-pass flow cytometry via spectrally encoded cellular barcoding

Advances in immunology, immuno-oncology, drug discovery and vaccine development demand improvements in the capabilities of flow cytometry to allow it to measure more protein markers per cell at multiple timepoints. However, the size of panels of fluorophore markers is limited by overlaps in fluorescence-emission spectra, and flow cytometers typically perform cell measurements at one timepoint. Here we describe multi-pass high-dimensional flow cytometry, a method leveraging cellular barcoding via microparticles emitting near-infrared laser light to track and repeatedly measure each cell using more markers and fewer colours. By using live human peripheral blood mononuclear cells, we show that the method enables the time-resolved characterization of the same cells before and after stimulation, their analysis via a 10-marker panel with minimal compensation for spectral spillover and their deep immunophenotyping via a 32-marker panel, where the same cells are analysed in 3 back-to-back cycles with 10–13 markers per cycle, reducing overall spillover and simplifying marker-panel design. Cellular barcoding in flow cytometry extends the utility of the technique for high-dimensional multi-pass single-cell analyses.

is fit well with a single power-law function ( − 1) ).+ , which yields  , () ≈ -) ( − 1) $.+  (5) This shows that the increase of SS by adding -th fluorophore increases with .This is likely due to use of non-ideal fluorophores in high-marker panels.In practice, this is driven by limited antibody availability, need to use tandem fluorophores with broader emission and absorption linewidths and instrument constraints.The increment of SS by a fluorophore may be referred to as the spillover cost of the fluorophore (analogous to the chemical potential of a molecule to the free energy of the system).For example, the spillover cost is 0, 10 , 24.6 , and 41.6  for the 1 st , 11 th , 21 st , and 31 st fluorophores, respectively.The solution of Eq. ( 4) is  ≈ ( − 1) where  = -+.+ .The curve fit to the simulation data is excellent.
We find that the power-law dependence,  ≈ ( − 1) ., describes the characteristics of high-marker panels quite well.Figure B below shows the analysis of several other published panels, OMIP-060, OMIP-064, OMIP-067, OMIP-069, and OMIP-084.All of the results are fit reasonably well with a single power law function with an exponent in range of 2.8 to 3.5.A detailed interpretation of this finding is beyond the scope of this document.For comparison, we also calculated  for the LASE's 3-cycle panels.The results are fitted with indices between 3 and 4, as shown below. .Since there are  cycles, the total SS is given by . We find the ratio of the SS between the -cyclic and non-cyclic cases to be Let us consider the above 28-marker case with Where we use  = 3.3.For  = 3 cycles, ( & ( ≈ 0.08.This means that the SS in 3-cycle cytometry is 12.5 times lower than the SS in non-cyclic case, and this ratio is constant independent of the total number of markers. Using () =  / ( 2 ), we get ( − 1) .= ( 2 − 1) ./ (.#$) and find For  = 3.3 and  = 3, we get  2 = 2.15 .This means that using 3 cycles one can measure 2.15 times more markers with the same SS.
This simulation data of Eq. ( 7) obtained from the 28-marker panel appears in Fig. 6(e) in the main paper.
The number of fluorophores that can be used would be limited by the limited availability of fluorophores and undistinguishable spectral overlap with an existing fluorophore.Currently, the record experiment used  = 40 (OMIP-69, see Fig. B above).3-cycle cytometry can extend this limit to 86.In principle, more cycles can further push the limit.With  = 5, up to 122 markers should be possible, requiring 24-25 fluorophores, with the same SS as the non-cyclic, 40-marker cytometry using 40 fluorophores.

Figure
Figure A: Simulation result of the OMIP-068 panel.

Figure B :
Figure B: Simulation results of various OMIP panels in the log scale.Lines, curve fit with  = ( − 1) .; the best-fit  values are indicated.
Figure C: Simulation results of the LASE 3-cycle panels in the log scale.The best-fit  values are indicated.