SEARCH

SEARCH BY CITATION

Conventional flow cytometry (CFC) is one of the most powerful and widely applied cytological tools with users able to generate statistically robust data sets with relative ease due to its multiparameter, high throughput nature. A key limitation of CFC in certain experimental settings is that it provides little or no spatial resolution of the fluorescent signals under investigation. This issue has been addressed by the development of imaging flow cytometry (IFC) technology that is able to provide a compromise between the multiparameter, high-throughput of CFC and the spatial information afforded by traditional fluorescence microscopy (1). As the use of IFC as a research tool continues to grow and evolve, it is imperative that the cytometry community addresses the need for a standardized approach to report IFC-derived data for peer review and publication. The overriding aim for any scientific publication should be to provide sufficient methodological detail to allow the reader to fully comprehend and evaluate the study and failure to do so seriously compromises any of the scientific conclusions reached from the experimentation. The published MIFlowCyt guidelines for reporting CFC-derived data for publications provide an excellent framework for exactly this purpose (2, 3). At present no such guidelines exist for IFC data and the real danger is that reporting and subsequent evaluation may fall into a “void” between those who excel in flow cytometry and those that specialize in imaging techniques. While publications exist in which IFC-derived data has been presented with sufficient detail for full methodological and scientific evaluation (4–7) there is clearly a requirement for a more formalized framework. In this brief communication, we highlight the fact that most, if not all existing MIFlowCyt requirements for CFC may also be applied to the initial steps of IFC analysis. Furthermore, in the interim before “MIFlowImageCyt” guidelines for IFC-derived data are published, the inclusion of a minimum of two additional criteria covering the image analysis component (Table 1) would greatly aid methodological reproduction and scientific evaluation.

Table 1. A summary of the key considerations for IFC data presentation listed to correlate with the existing MIFlowCyt requirements for CFC data
MIFlowCyt designationsCFC reporting considerationsIFC reporting considerationsSuggested IFC controls
  1. Suggested additions specifically for IFC data are shown in italics. Abbreviations used in the table are: CFC: conventional flow cytometry. IFC: imaging flow cytometry. CCD: charge-coupled device. PMT: Photomultiplier tube. QC: quality control. FMO: fluorescence-minus one. MRB: morphometrically relevant biological.

1.1–2.4: Experimental overview  and flow sample/specimen  detailsAs outlined in MIFlowCyt checklist.  To include  experimental details/optimization, description of reagentsAs for CFC, pay attention to reagent titration.  Need to balance signals from same excitation  laser and avoid saturation on CCD due to  limited dynamic range. Saturated pixels will  impact compensation and all elements of the  analysisTitrate fluorescent reagents for IFC using  CFC and/or IFC. Quality and  reproducibility of staining affects mask  performance and therefore quality of  measurements made
3.1–3.3: Instrument detailsAs outlined in MIFlowCyt checklist.  To include laser configurations, filter and  detector voltagesAs for CFC. One or two camera system. Can  report which lasers were on and the powers  used. Any acquisition classifiers set and  justification as to why they were implementedReport exact QC methods for CFC and  IFC and when they were done in relation  to sample acquisition
4.1: List-mode data fileUpload FCS files to  http://flowrepository.org/Upload raw image files, compensation matrix,  and analysis template files to http://  flowrepository.org/To include examples of fully stained  samples, as well as FMO and MRB  controls for full analysis reproduction
4.2: CompensationAs outlined in MIFlowCyt checklist  to include how compensation has been  performed and with which reagentsAs for CFC. Report how compensation  has been done with values. Report laser  power settingsSingle fluorescence controls stained  with the matched fluorochrome. Often  preferable to use broadly expressed  antigens while keeping fluorochromes  the same as in experimental samples
4.3–4.4: Gating (e.g., single,  live cells, and total  fluorescence profiles)As outlined in MIFlowCyt checklist.  To include information  on data transformation and gatingAs for CFC, but also use morphometric  parameters for defining single cell populations,  typically the area and aspect ratio of a default  channel maskSet gates using well-described controls  such as FMO. Use confidence intervals  to set gates as in CFC
4.5 Masking details  (default or adapted)N/AReport exact mask used using software naming  string for auditing and reproduction purposes.  Include images for each masking adaption.  Include software versionMRB controls may assist in the design  of a mask if they exemplify  or accentuate the biology to  be measured
4.6 Description of  morphometric/  spatial feature extractionN/ALikely to generate a continuous variable due to  the heterogeneous nature of the biology being  measured. Describe feature used, how it was  arrived at, controlsPositive and negative MRB controls  to set any gating limits where possible
5.1–5.3: Data presentationAs outlined in MIFlowCyt checklist. Label  axes, provide graphical example of gating  strategy. Biological controls and/or FMOAs with CFC. Show example of gating (Fig. 1).  Clearly label axes where morphometric  parameter is plotted out. Ensure a clear audit  trail between axis label, feature, and maskSee 4.6 above. To also include details  of the image display settings/pixel  scaling as this will affect visual  interpretation of images

Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data

  1. Top of page
  2. Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data
  3. Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations
  4. What is the Future of IFC Analysis and Data Reporting?
  5. Acknowledgements
  6. Literature Cited

Both technologies record emitted light from cell intrinsic and/or extrinsic fluorescently tagged agents as they are presented in solution for laser excitation. In the case of CFC, emitted photons are recorded using photomultiplier tubes (PMT) whereas IFC uses a spatially registered charge coupled device (CCD) camera to the same end. The final result is essentially the same as the IFC analysis software identifies pixels above background within the image frame using the default channel mask and can then generate a sum of these pixel values for display on a linear or logarithmic scale akin to the pulse area derived from a PMT (4, 8). Because of these commonalities, there is no reason why IFC-data reporting should not also include detailed and validated descriptions of how spectral compensation has been performed as well as how any gating thresholds have been set on the total object fluorescence in an analogous fashion to CFC-derived data (Fig. 1). The existing MIFlowCyt checklist for CFC data outlines these requirements in full (3) and controls for setting gated thresholds on total object fluorescence that have been reported elsewhere for both clinical and nonclinical CFC (9). With respect to compensation, signal saturation from IFC can be problematic particularly as the ability to regulate intensities is dependent on laser power rather than PMT voltages; this becomes especially relevant when one laser excites two or more fluorochromes. We have circumvented this issue by carefully titrating fluorescence reagents using an LSRII CFC system to position the signal median at ∼103 with a PMT voltage of 400–450 V (4). The process of compensation in IFC is practically analogous to CFC and can thus be reported using the existing MIFlowCyt criteria to include fluorochrome-specific spill over matrix coefficients and laser configurations (Table 1). Once any required compensation has been performed total object fluorescence in specific channels can be used to set gated thresholds to refine our populations in an analogous fashion to CFC (see Fig. 1). We have noted that due to the per-pixel nature of how total fluorescence is measured by IFC there is often higher variation in values affecting population resolution in comparison to PMT-derived signals from the same samples. As such, fluorescence minus one (FMO) controls becomes particularly important for setting gates for positive identification in IFC (see Fig. 1). FMOs must be collected with the same laser power and configuration as the fully stained samples. Moreover, other controls such as isotypes, secondary alone or full staining with absence of biological target (KO) are also required to validate the signal and the acquisition settings. In essence, we need to ensure that before we ask spatial or morphometric questions of IFC-derived data, the fluorescence has been acquired with appropriate settings, compensated and validated for setting gated thresholds using well-established CFC controls. As such all of the MIFlowCyt requirements for CFC-derived data can easily be met from IFC-derived data reporting up to this stage.

thumbnail image

Figure 1. A side-by-side comparison of likely strategies for setting threshold gates using total object fluorescence from the same sample acquired on a 6-channel, 5-laser ImageStream x (ISx, Amnis Corp) IFC system and a 4-laser, 16-detector LSRII (BD Biosciences) CFC system. Murine thymocyte preparations were stained with anti-GL3 APC and incubated with Near Infra-red LIVE/DEAD (NIRLD) dye prior to fixation (4% formaldehyde) and permeabilization (0.1% triton-100) to then stain intracellularly with an anti-NfκB2antibody and a species reactive AF488 labeled secondary antibody. Data analysis was performed using FlowJo v9.x (CFC) and IDEAS 4.0 (IFC). 1. LDNIR fluorescence exclusion was used to gate on live cells. 2. Side scatter width (SSC-W) was used to eliminate doublets by CFC whereas a combination of the default bright-field (BF) channel mask area and aspect ratio were used to define single cells by IFC. 3. Intact cells were identified by CFC using side scatter area (SSC-A and forward scatter area (FSC-A). Uniquely to IFC analysis, poorly focused events were eliminated using a measure of pixel value variation across the default BF channel mask. 4. A GL3-APC FMO was used to set the gate for GL-3 APC positivity on the fully stained sample at a 99.9% confidence interval. 5. Finally, an AF488 FMO was used to determine sample positivity at 99.9% confidence interval. CFC acquisition settings were: 530/30 blue = 411v, 660/20 red = 681v, 780/60 red = 797v. Compensation values were: APC: 780/60 red—2.7%. NIRLD: 660/20 red—4.8%. IFC acquisition settings were: Cell classifier on BF (CH3) of >15 U. Blue laser = 100 mW, Red laser = 120 mW. Compensation values were: AF488: CH1 = 13%, CH3 = 20%, CH4 = 7%. APC: CH4 = 24%, CH6 = 23%. NIRLD: CH6 = 10%.

Download figure to PowerPoint

Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations

  1. Top of page
  2. Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data
  3. Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations
  4. What is the Future of IFC Analysis and Data Reporting?
  5. Acknowledgements
  6. Literature Cited

The divergence of CFC and IFC analysis comes when it is time to ask defined questions relating to object morphology and/or the spatial positions of signal x or its spatial relationship to signal y. Without some level of supervised learning and a priori knowledge of the biological question, the task of dealing with an infinite number of masks and feature combinations can be daunting. Even well-defined spatial measurements such as assessing translocation by IFC can be improved upon greatly when complementary microscopy techniques are used to guide and validate the analysis (5). If the question is more complex, then higher-resolution microscopic images are often essential to guide the masking and feature extraction (10). Compared to total object fluorescence, spatial or morphometric measurements by IFC often require more consideration to be paid to how the software identifies specific pixels by masking. The nature of the mask is determined by software algorithms largely based on pixel intensity and variation within an object image frame (1). The mask defines the pixels on which a given measurement will be made with the aim of best resolving one population from another. As such, it may be necessary to optimize the masking rules using a defined set of modification parameters embedded within the IFC analysis software. Current analysis software provides a fully auditable naming string for every mask created and as such can easily be reported as a workflow for publication (4) to satisfy potential “MIFlowImageCyt” considerations (Table 1, 4.5). In essence the resolving ability of any IFC measurement is often only as good as the ability of the mask from which it is extracted to define the most relevant pixels within an image frame. An example of the intimate relationship between mask and measurement is shown in Figure 2A. The nuclear channel images of prophase, metaphase, and anaphase cells are visually distinct; however, the default mask does not reflect this. If we then extract two morphometric parameters from the default nuclear mask (aspect ratio and spot count) we fail to adequately resolve these populations. If we make controlled modifications to the nuclear channel mask to demarcate these differences better, the same features are now able to resolve the populations. In the case of aspect ratio, we have generated a continuous variable that reflects the biological nature of cells transition from prophase to metaphase. To set confidence interval gates for prophase and metaphase distinction, we generated a morphometrically relevant biological (MRB) control by treating a sample with the prophase blocker, nocodazole. It is quite common for the spatial and morphometric components of IFC analysis to generate a continuous variable in line with the biology being quantified, as such MRB controls are a very important consideration to validate and justify the nature of the measurement (4, 5, 10). MRB controls are similar in some regards to classical CFC controls such as FMOs, however they require more than simply the presence or absence of particular fluorescence and must take into account the spatial or morphometric nature of the signal under investigation. In the example of translocation, the appropriate MRB controls would be sets of untranslocated cells and cells with a known translocation event (5). In the case of recent work using IFC to measure antigen (Ag) polarity in B cells we calculated the delta between the default BF channel mask centroid and the intensity weighted centroid of the default mask of the Ag-bead channel. As an MRB control, we measured the delta between the bright-field image and a fluorescent signal that we expected to be nonpolarized and set gates using confidence intervals accordingly (10). While this approach did not require any masking adaptation to be made, it did require the selection of the correct feature to best report the possible spatial arrangements of beads within the B cell (Fig. 2B). Specifically, we found that unless we intensity-weighted the bead centroid, cells with multiple foci where one was clearly more intense was missclassified as nonpolarized (see Fig. 2, event 12407). An alternative approach to feature extraction by IFC analysis is to manually select groups of cells that best describe the visual phenomenon to be resolved and perform a scan of various feature combinations to find the one with the greatest resolving potential as judged by the Fisher discriminatory ratio (Rd) (6). While the Rd metric does not require gating thresholds to be set on the distribution of the final measured parameter, it is still dependent on the input mask and at present this semi-automated approach to IFC data analysis may work best when no masking modifications are needed. Even if no masking adaptations are required, the success of such an analysis is still dependent on the appropriateness of the default feature sets and may require the generation of completely de novo parameters through user intervention. Moreover, to give context to the Rd value, some kind of MRB control is still essential (5). In terms of reporting consideration for IFC-derived data, however the features are extracted, the choice of masking and measurement should be fully reportable and justifiable from an analytical and biological perspective. These could easily beincorporated into any forthcoming “MIFlowImageCyt” guidelines (Table 1, points 4.5–4.6) for IFC-derived data presentation.

thumbnail image

Figure 2. IFC-measurements require careful mask and feature selection. (A) The resolving potential of any feature is intimately linked to the input mask it is extracted from. Jurkat cells were fixed in 70% ethanol and stained with an antibody against the serine 10 phosphorylated form of histone H3 (pH3) and propidium iodide (PI) for nuclear visualization prior to acquisition on a 5-laser, 6 channel ISx IFC system. Multispectral images of prophase (P), metaphase (M), and anaphase (A) cells are shown with the default nuclear channel mask [auditable naming string is (M04)] or the adapted nuclear mask [auditable naming string is range (threshold (M04, PI, 45), 45-5000, 0-1)] superimposed over the nuclear channel image in the 3rd column. The aspect ratio and area parameters where then calculated from both masks and plotted as bivariate dot plots. The gate for prophase and metaphase distinction was set using the MRB control of the prophase blocker nocodazole at a 99.9% confidence interval. (B) Even without masking adaptations the extracted feature must reflect the nature of the spatial measurement. Multispectral images of CTV-labeled B cells with internalized Ag-coated beads for three possible intracellular spatial arrangements of beads. Values for the delta between the default BF mask centroid (DC) and either the nonintensity weighted (NIW, Column 2) or the intensity weighted centroid (IW, Column 3) of the default bead channel mask were calculated. As an MRB control for setting the polarity gate, the delta between the BF and CTV default channel masks centroids was calculated (control DC, Column 1). The cut off for polarization was set at 1.5 based on a 99.9% confidence interval using the control DC distribution. [Color figure can be viewed in the online issue which is available at wileyonlinelibrary.com]

Download figure to PowerPoint

What is the Future of IFC Analysis and Data Reporting?

  1. Top of page
  2. Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data
  3. Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations
  4. What is the Future of IFC Analysis and Data Reporting?
  5. Acknowledgements
  6. Literature Cited

It is widely accepted that the human eye is more efficient than a computer at discerning qualitative visual differences but poor at quantifying them. The opposite is true of a computer but there is still a requirement for some human input at various points in IFC-data analysis. One can envisage the future of IFC analysis whereby a similar selective approach as reported by de la Calle et al. is employed (6) but at one press of a button a computer generates all possible masking variations and all possible parameters to produce one single, plottable, and reportable parameter that is in effect the sum of several iterative measurements to best resolves populations in n dimensions. As computer processing power increases, we may even be able to perform totally unsupervised analysis similar to that which is currently possible with multiparameter CFC (11) and now mass cytometry (12). Regardless however, one would still expect detailed explanations of how this parameter in n dimensions was arrived at from an analytical, technical, and importantly biological perspective, particularly as the morphometric and spatial element to IFC-data present a serious challenge. In summary, the authors would like to see more attention to detail with regard to publications using IFC, whereby the MIFlowCyt requirements for CFC are adhered to in combination with full details of the image analysis components such as mask construction, measurements made of the pixels under the mask, justification for the mask used, and the feature. This should also include full details of all fluorescence (FMO) and biological (MRB) controls. Without these details,it is impossible to ascertain the quality of the science presented or to repeat the method successfully. One would expect similar guidelines to be applicable to other forms of imaging cytometry and also both clinical and non-clinical applications of IFC.

Acknowledgements

  1. Top of page
  2. Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data
  3. Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations
  4. What is the Future of IFC Analysis and Data Reporting?
  5. Acknowledgements
  6. Literature Cited

AF and DD acknowledge support from Cancer Research UK. They also acknowledge Dr Melanie Wencker and Dr Adrian Hayday for the IFC and CFC data shown in Figure 1 and Dr Olivier Thaunat and Dr Facundo Batista for the IFC data shown in Figure 2B. Thanks to Prabhjoat Chana from the Flow Cytometry Core Facility at Guy's and St Thomas' NHS Biomedical Research Centre for useful discussions.

Literature Cited

  1. Top of page
  2. Using Standard MIFlowCyt Controls to Set Gating Thresholds on IFC-Derived Data
  3. Creating Masks and Extracting Features: Potential “Miflowimagecyt” Considerations
  4. What is the Future of IFC Analysis and Data Reporting?
  5. Acknowledgements
  6. Literature Cited