Christian Hennig and Gesine Hansen developed the methodology for iterative cytometry; Nico Adams and Christian Hennig developed the model ontology; Christian Hennig developed and programmed the software used for automated microscopy, image recognition, and data mining.
Flow cytometry (FCM) is currently the gold standard for the characterization of cells in suspension using fluorescent antibodies and other dye-tagged probes. In today's laboratory routine, five to eight markers can be detected simultaneously. However, it has become clear that understanding the high complexity of our immune system requires the differentiation of many more cell subtypes and activation states, naïve, memory and effector cell types as well as different regulatory cells. Complicating this, many of these cells belong to rare side populations or are prepared from sample sources that yield only very low numbers of cells. Therefore, simple and robust systems that can distinguish 30 and more markers on a small sample are highly desirable. Because of the combinatorial explosion of possible detectable marker combinations, a less hypothesis-driven, more explorative method that allows a stepwise exploration of a sample would be very useful. The hypothesis-free analysis of a cytometric sample is the fundamental concept of ‘Cytomics’ as introduced by Valet (1). Additionally, ex vivo manipulation on living cells that allow observation of regulation of proteins before and after stimulation of the very same cell are preferred features of future cytometric technologies.
Previous work has demonstrated that it is possible to analyze up to 17 fluorescent markers per cell using FCM (2). Translation of this technology into laboratory routine proved to be difficult, since the availability of ‘exotic’ antibody–dye combinations that are necessary for this multicolor approach are very limited, and the establishment of such a setup is time-consuming because of the multiple compensation steps for the spillover of the different fluorescence spectra of the dyes. Furthermore, stepwise manipulation and explorative investigation of samples is not possible using FCM. Slide-based cytometry (SBC) is an emerging complementary technology, which overcomes some of these limitations by analyzing cells bound to a solid surface (3, 4). Iterative cycles of staining, imaging, and bleaching of dye-tagged probes have the potential to enhance the applications of this approach even further by enabling detection of a high number of parameters per cell (5, 6). We were interested in evaluating whether such an approach could be used to manipulate and quantitatively and qualitatively analyze immune cells with results comparable to current flow-based methods while additionally allowing the analysis of small samples with an unlimited set of markers. Ideally, an implementation of iterative chip-based cytometry (iCBC) for comprehensive immunophenotyping should fulfil the following requirements:
1Simple sample preparation and immobilization of cells, using even minute amounts of sample (e.g., material obtained by biopsies or lavages).
2No modification of cellular composition or quantitative and qualitative expression of markers due to the immobilization procedure.
3Short cycle times for iterative analysis.
4Unlimited set of surface and intracellular markers detectable.
5Low cell-activating property of the surface for functional assays.
6High correlation of results obtained by crossplatform validation (iCBC vs. FCM).
We developed and evaluated an implementation of iCBC for the comprehensive immunophenotyping and even functional analysis of small numbers of immune cells immobilized within microfluidic chips by automated conventional epifluorescence wide field microscopy that achieves the listed requirements. Because of the potentially large space of possible marker combinations associated with each cell, we furthermore discuss the combination of our experimental methodology with up-to-date informatics tools for knowledge capture and knowledge generation.
MATERIALS AND METHODS
Microfluidic chips or slides containing cell-adhesive surfaces for unselective self-immobilization of cells and microfluidics that allow the application/exchange of buffers and staining solutions were obtained from NanoSky (Germany) or Carl Roth (Germany) and used according to manufacturer's recommendations.
Preparation of Cells
For the detection of antigen-specific cells, BALB/c mice were immunized twice intranasally with ovalbumin /LPS (12.5 mg/ml, 40 μl within 3 days). Three days later, spleen cells were harvested and restimulated in culture plates with ovalbumin (200 μg/ml) or mock control for 72 h. For intracellular staining of cytokine production, 1 μl/ml Brefeldin A (GolgiBlock, BD Biosciences, Heidelberg, Germany) was added 2 h before harvesting the cells. For analysis of BAL fluid of asthmatic and nonasthmatic mice, experimental asthma was induced and BAL fluid was obtained using an established protocol (7). For the functional human B-cell assay, an established protocol for CD40L- and IL-4/IL-10-induced germinal center reaction of human PBMC in a cell culture system was used (8). For iCBC, cells were subsequently harvested from the culture, washed in phosphate-buffered saline (PBS), and resuspended in 1–10 μl of PBS. Upon introduction of the suspension into the microfluidic channel, cells distributed randomly over the channel surface. After an incubation time of 5 min for binding of the cells to the surface, the chip was rinsed with PBS. Cells were either fixed in CellFix (BD Biosciences) or subjected directly to the respective staining protocol.
Staining Protocol iCBC
Prior to staining, the chip was scanned for autofluorescence of dead cells or macrophages. Subsequently, 10 μl of the respective PE- or FITC-labeled antibody was added to the microfluidic chip at concentrations of 2 μg/ml to 200 ng/ml (dependent on labeling strength of the respective antibody). After an incubation time of 5 min at room temperature, the chip was rinsed with 200 μl PBS and applied to the imaging system for data acquisition. For the staining of intracellular molecules, immobilized cells were fixed and permeabilized using FixPerm solution (BD Biosciences). For viability testing, the chip was filled with 10 μl of trypan blue and immediately scanned on the imaging system. Afterwards, the chip was rinsed again with PBS to avoid any cytotoxic effects caused by trypan blue itself.
Staining Protocol FCM
For crossvalidation with FCM, harvested cells were resuspended in 50 μl PBS/1% BSA and incubated with 1 μl of the respective dye-labeled antibodies (0.2 μg/ml) for 15 min at RT in the dark. After washing, the cells were subjected to flow-cytometric analysis or permeabilized for intracellular cytokine staining using FixPerm solution (BD Biosciences) and a staining protocol as supplied by the manufacturer.
The optical setup we used for iCBC consisted of an upright microscope (Zeiss axioplan 2e, motorized z-drive, electronic control of fluorescence light- and transmitted light-shutter, Goettingen, Germany), equipped with a mercury lamp (HBO 100) for excitation, suitable filter-sets for the used dyes (PE or FITC), a Plan-Neofluar 16×/0.50 immersion objective (Zeiss), and a motorized scanning table (Merzhaeuser, Wetzlar, Germany). Signals were detected using a sensitive monochromatic 12-bit CCD device (Zeiss AxioCam MRm). Typically, 4–10 different positions were scanned and two pictures were taken for every position (picture 1: 7-s-exposure per position in the fluorescence light mode, picture 2: 100-ms-exposure in the transmitted light mode). The cumulative spatial resolution was 0.3 μm/pixel.
Iterative Staining–Imaging–Bleaching cycles
After staining, the cells bound within the microfluidic chip were scanned by the imaging system as described before. To enable iterative staining of the cells, remaining signals were bleached using the same excitation filter as for imaging by extending illumination by 30 s per position. After bleaching, a second fluorescent-light image was acquired from the same position for background correction. Afterwards, the chip was removed from the microscope and stained for the next cycle. Precise (re-)positioning, which is a prerequisite for tracking each cell as an individual over the cycles, was accomplished by using the transmitted light picture of the pattern formed by the cells at the respective position.
Image Recognition and Processing
Applying Laplace filtering and two-fold smoothing of the picture, 95% of the cells could be automatically separated from the background in the transmitted-light-mode picture. The remaining 5% of cells like agglomerates were tagged manually or excluded from further analysis. For each of the cells, coordinates of its center (corrected by the offset calculated by the readjustment procedure) and size were recorded.
Mean fluorescence intensity per cell was calculated by collecting the gray-scale values of each pixel within a cell divided by its cell area [expressed in arbitrary units (AU)]. Since illumination by the fluorescence lamp is relatively constant during the restaining cycles but not totally homogeneous for every part of the scanned position, we decided to subtract the calculated fluorescence intensity of the cell after the bleaching procedure from the unbleached cell to account for possible local inhomogeneity of illumination.
The information associated with each cell (e.g., cell size, x–y coordinates on the chip, relative fluorescence intensity) was stored in list-mode format. For further analysis, a bivariate gating was performed manually. Subsequently, each cell was grouped by hierarchical clustering by our in-house software. The resulting groups were then compared with a model-cytometry-ontology for the purposes of cell identification, knowledge capture, and structuring.
There is currently no agreed-upon methodology for the development of ontologies and a number of methods have been described in the literature (9). We used a methontology approach (10), which consists of the phases requirement specification, conceptualization, integration, and implementation. The integration stage in the methontology approach normally aims to identify existing ontologies, which already describe some of the required concepts and to integrate these into the new ontology. At this stage, we considered several ontologies such as the cell ontology (11), the gene ontology (GO) (12) and the foundational model of anatomy (13), but found that these ontologies were either at the wrong level of granularity or contained concept definitions, which were unsatisfactory for our purposes and hence decided against integration of these resources. During the implementation phase, the ontology was constructed with the official W3C standard web ontology language (OWL) [McGuiness D, van Hamelen F, OWL Web Ontology Language Overview, http://www.w3.org/TR/owl-features/ (last accessed June 2008)] using the Protégé Ontology Editor and Knowledge Acquisition System [The Protégé Ontology Editor and Knowledge Acquisition System, http://protege.stanford.edu (last accessed June 2008)] (Fig. 6).
As shown in Figure 1a, iCBC generates raw data images (I) that are subsequently processed by image recognition methods: cells are separated by software from the background (II) and mean fluorescence intensity for each cell is calculated for every tested marker (histogram: III). This was repeated for each tested marker in a cyclic setup (Fig. 1b). After bivariate gating (positive/negative), each cell is assigned automatically to a ‘fingerprint’ of the tested marker set. To further analyze the quality of the experimental data, extensive validation of the system was performed:
Stability of the imaging system/background correction
In laser-based cell scanning systems, fluctuations of the laser beam may result in instability of the readout (14). Although illumination by the wide-field epifluorescence approach used in our setup does not lead to strong pixel-wise fluctuations, any local inhomogeneity of illumination could also lead to the generation of artifacts. Although local inhomogeneity was already acceptable, we subtracted pixel-wise a background image of the same position after bleaching. This led to a further improvement of data quality (Fig. 2a).
Generation of ‘artificial’ phenotypes
The validity of a chip- or slide-based system also depends on the generation of ‘artificial’ immunophenotypes either by the detachment of cells from the surface during subsequent cycles of analysis or by software-induced merging of phenotypes from two or more adjacent cells. We therefore analyzed chips with a medium cell density (700 cells/position, 1 position = 0.15 μm2). Cell loss was measured by setting a trigger-point in Cycle 1 (100%), afterwards following these cells in transmitted light mode over six cycles (immobilized living cells) or 21 cycles (fixed and permeabilized cells), respectively. With immobilized living cells, 4% of all cells were lost, mainly occurring between Cycles 1 and 2, whereas fixed cells showed no detectable cell loss (<0.05%). Erroneous software-induced merging of phenotypes did not occur in this setting (see Table 1, CD4+CD19+ cells), since software recognizes and excludes cells that show a strong nonhomogenous staining pattern (moon-phase like).
Table 1. Crossvalidation FCM versus iCBC
Comparison of iCBC versus FCM with regard to specificity of stainings. Mouse spleen cells (n = 5 animals) were either stained iteratively with CD3-PE, CD4-PE, CD8-PE, and CD19-PE for analysis by iCBC or stained with CD3-PerCP, CD4-PE, and CD8-FITC or CD19-PerCP, CD4-PE, and CD3-FITC for flow cytometric analysis. Relative numbers of cells in each compartment show comparable results for both technologies.
60 ± 1
60 ± 0.7
39 ± 2.5
40 ± 3
24.5 ± 2
27 ± 1.5
15 ± 2
14.5 ± 1.5
17 ± 3
19 ± 2
One of the most desired features of improved cytometric methods is the comprehensive analysis of small samples containing only few cells. In slide-based systems, these cells, when immobilized on the surface, should form a homogeneous layer with a high cell density although allowing an unambiguous separation between single cells. To evaluate this, two geometries of microfluidic channels (2 and 10 μl sample volume) were tested for cell densities on the chip surface resulting from different concentrations of cells within the sample volume. As shown in Figure 2b, maximum surface cell density was about 2,500 cells/scanned position (1 position = 0.15 μm2). Optimum cell density ranges between 400 and 1,800 cells/position. We therefore concluded that 10,000–100,000 cells should be applied in a final sample volume of about 2 μl, whereas larger number of cells should be resolved in 10 μl final sample volume.
Analysis of living cells
Since it is known that some markers show significant differences when measured with immunofluorescence methods prior- versus postfixation (15), some sensitive markers like CD62L or CD27 have to be measured on living cells. Viability of immobilized cells was therefore evaluated by repeated staining of the cells using trypan blue and recording the number of living cells. Eight hours after binding of the cells to the chip surface, the number of viable cells was above 90%. To evaluate changes of markers due to on-chip activation, we tested repeatedly for CD62L shedding, one of the earliest activation markers (16), during this 8-h period. The number of CD62L-positive mouse spleen cells was 42% directly after preparation from blood as well as after immobilization on the surface. Without further stimulation, the number slowly decreased reaching 30% after 8 h. In contrast, additional stimulation with 250 pg/ml PMA and 250 pg/ml Ionomycin on the chip surface led to a rapid activation and CD62L-shedding of the cells (Fig. 2c).
To evaluate comparability of iCBC with FCM, we performed a crossplatform validation using a standard flow cytometer (FACScan, Becton Dickinson).
As shown in Table 1, readout of main cellular components of spleen cell homogenizates of Balb/C mice was nearly identical between the two methods. All other markers tested in murine and human cells showed comparable results and were not dependent on the time-point of staining after binding of the cells to the chip (data not shown).
As shown in Figure 2d, sensitivity of iCBC is enhanced 10-fold in regard to the used flow cytometer. This mainly depends on signal artifacts of the detector of the flow cytometer that broadens the signal of the unstained population (background signal). Since image recognition of iCBC is not influenced by such artifacts, the signal of negative cells is more homogeneous enabling more sensitive separation of weak signals from the background. Furthermore, dynamic range of both FCM and iCBC allows separation into high, medium, and low expression signals with comparable results (Fig. 2d).
To demonstrate the usefulness of this method for both research and clinical applications, we chose three different settings: (1) identification of rare cells, (2) comprehensive analysis of a small sample size, and (3) immunophenotyping of human B-cell differentiation. Additional example data and a data-viewer can be downloaded (see Material and Methods).
Research I: Rare cells—Detection of antigen-specific memory T-cells in mouse spleen homogenizates
Rare antigen-specific circulating memory cells (e.g., <0.01% of total spleen cells) producing cytokines after immunization and in vitro restimulation were detected and phenotyped in spleens of nasally immunized mice and control mice using the following marker set: IFNy, IL-4, IL-10, TCR2, TCR1, CD19, B220, CD49b, CD3, CD4, CD8, CD11c, CD11b, CD44, CD62L. As shown in Figure 3, nasal application of antigen resulted in the generation of IFN-gamma producing, CD3+CD4+CD11c-IL4- Th1 memory T-cells (additionally CD62L-CD44+ TCR2+CD8-CD11b-B220-, data not shown). The frequency of those cells was about three in 10,000 cells, analyzed in five independent experiments with five mice per group in each experiment. No such cells were observed in controls (AG-specific restimulated spleen cells of nonimmunized mice or nonrestimulated spleen cells of immunized mice).
Research II: Comprehensive analysis of small sample sizes—Bronchial lavage (BAL) of mouse lungs
Cells obtained by bronchial lavage (100 μl) of lungs of ‘asthmatic’ mice that had been immunized to ovalbumin by systemic and local antigen application (7) were stained using the following markers: Autofluorescence, CD3, CD4, CD8, TCR2, TCR1, CD49b, CD44, CD62L, CD19, B220, MHCII, CD11b, CD11c, F4/80, Gr-1, CD103, CD27, CD28, CTLA, B7-1, B7-2, B7-H1, Fox-P3, GITR, CXCR4, CCR3, CCR5, CCR7, IFNy, IL-4, IL-10 (all PE labeled). As shown in Figure 4, a variety of different cell types and differentiation states that are not accessible by conventional FCM can be observed by iCBC.
Clinical setting: Immunophenotyping of human B-cell differentiation in an artificial lymph-node model
Primary humoral immunodeficiencies (PHID) can be grouped by their ability to produce specific antibodies or by the activation of B-cells after antigenic stimulation (e.g., Hyper-IgM syndrome, CVID, ALPS). By using CD40L and IL-4 as stimulus, we performed a comprehensive phenotyping of differentiation steps of in vitro-activated human B-cells of a patient with humoral immunodeficiency and lymphoproliferative syndrome and a control (Fig. 5). In healthy individuals, all stages of B-cell activation (naïve B-cells, extrafollicular activated, germinal center B-cells, plasmablasts, and memory cells) are distinguishable. In the patient, B-cells can be activated and enter the germinal stage but are unable to proceed to formation of plasmablasts or memory cells. This is in congruence to the clinical appearance of the patient with hypogammaglobulinemia and generalized lymphadenopathia, thus underlining the usefulness of iCBC for diagnostic purposes.
The growing knowledge of the immense complexity of the immune system is challenging and calls for enhanced technologies that allow for a multiparameter analysis of its components under different conditions. Until now, more than 100 different subpopulations of cells and activation states can be detected within a single sample of blood, lymph node, spleen, or bone marrow (17). None of the existing flow cytometric technologies is able to perform a comprehensive analysis of this complexity in every-day laboratory routine (deep immunophenotyping), especially with regard to small sample sizes and stepwise, explorative cytometry.
Immobilizing cells on a two-dimensional support and iterative staining–imaging–bleaching cycles are evolving as an alternative concept for comprehensive analysis of surface and intracellular markers of cells (5, 6). Until now, however, the implementation of this approach for extensive immunophenotyping of small samples was hampered by relatively long sample preparation and analysis times as well as the requirement of relatively large amounts of sample. Furthermore, the choice of (often harsh) sample preparation and fixation methods (spin fixation, formaldehyde treatment, or drying) led to modification of cellular composition and/or changes in quantitative and/or qualitative expression of markers within the sample.
We developed and evaluated a microfluidic-chip-based system for comprehensive analysis of immune cells (iCBC) using extremely small sample sizes. To avoid ex vivo modification of the sample composition or marker expression due to sample preparation, we used commercially available cell-adhesive surfaces within microfluidic chips. This allowed the unspecific self-immobilization of statistical numbers of living cells (e.g., lymphocytes, granulocytes, monocytes/DC, platelets, or bacteria) in a homogenous manner and subsequent iterative analysis by simple fluid exchange. Crossvalidation of the same samples between iCBC and FCM revealed a high degree of correlation in specificity (Table 1). In contrast to parallel marker detection by FCM, no spillover/compensation problems occur, since only one antibody is detected per cycle. It is noteworthy that, because every marker can be used with the same tagged dye (e.g. FITC, PE), it is possible to easily extend the set of usable commercial available antibodies. Using a different signal detection approach, sensitivity is enhanced 10-fold compared to FCM (Fig. 2d). The quality of results is further enhanced as more markers (negative and positive) can be combined to ensure validity of the observed cell type or activation state.
Several methodological approaches have been developed that in principle allow in combination the detection of a wide variety of markers per cell using SBC (18). In comparison to previously reported slide-based immunophenotyping methods, where the number of staining cycles was limited to four (6), or where markers can only be measured after fixation of cells destroying fixation-sensitive markers (5), we developed a simple, robust, and reproducible setup that extends the number of possible manipulations to a theoretically unlimited number of parameters. In the first cycles, fixation-sensible markers like CD62L, CD27, and chemokines are detected. As shown, even in vitro activation of cells can be performed and monitored on-chip (Fig. 2c). Afterwards, the cells are fixed and stained for fixation-stable or intracellular markers. This generates the possibility of storing the chips with fixed cells. After initial analysis of the results, these chips can then be used later for evaluating additional markers. This explorative cytometry in turn leads to shorter periods of knowledge generation, since an experiment can be carried out once and then analyzed successively by repeated rounds of staining over several weeks for several parameters as new questions arise from initial data mining.
The method presented in this paper currently has two bottlenecks: Although we have considerably simplified and shortened sample preparation and staining time (5 min per marker), there is still a relatively long bleaching time (30 s per position) to remove the remaining fluorescence prior to the subsequent staining. This is acceptable for research, because the usefulness and uniqueness of the obtained results outweigh the relatively long analysis time of cumulative 10–20 markers/8 h. However, for routine diagnostic purposes, it would be helpful to develop novel dye-tags that allow a shortening of the signal-removal procedure.
The second bottleneck is the handling of the combinatorial explosion of information resulting from the use of large marker sets: The large and even increasing number of differentiable cell types requires the detection of a larger number of markers per cell for unambiguous cell identification and classification. Moreover, the experimental setup and the nature of the sample source have a significant influence on the composition, activation- and differentiation states of cells within the sample, thus increasing the complexity of the system even further. For this reason, the combination of high content cytometry methods with modern informatics tools such as ontologies for knowledge capture and structuring as well as for comparison of obtained data with other experiments and the published state of the art is mandatory. Ontologies are fundamental tools for knowledge management in a machine-comprehensible format. Although there is ongoing work directed toward extending existing ontologies like the gene ontology (12) for immunological work (19), currently no publicly available ontology exists, which, in our view, could be used or extended for our purposes. We have therefore developed a draft for the cytometry domain, which covers the concepts and workflows needed for the evaluation of our example experiments and models selected concepts such experimental protocols and cell definitions (Fig. 6). It is important to note that this workflow also represents a general model of the way in which cytometric information is communicated in immunological research papers and allows the extraction of knowledge from such papers and its storage in a unified format. Our initial experiences with using this ontology approach suggest that a cytometry ontology is a valuable extension of cellular analysis methods that produce high content cytometry information.
Taken together, the new platform method for cytometry presented here combines the advantages of simple sample preparation and immobilization of cells, using even minute amounts of sample, detectability of an unlimited set of surface and intracellular markers, and low cell activating properties of the surface enabling functional assays for an explorative, less hypothesis-driven approach for cytometric analysis.
We believe that explorative cytometry by combination of iCBC with appropriate informatics tools is very useful for further unrevealing the complexity of our immune system. Besides the presented examples, future applications of this technology are phenotyping procedures and especially functional studies in immunological research, oncology, pathology, pharmacology, and hematology. Therefore, chip-based cytometry approaches could become valuable tools for achieving the goals of a human cytome project and systems biology (20).