High-resolution determination of human immune cell signatures from fine-needle liver aspirates.
Multicolor flow cytometry represents an important tool for characterization of immune cell phenotype and function. In humans, this technique is most often applied to studies of cells obtained from peripheral blood. However, in many human disease conditions, peripheral tissues are the preferred sites for investigation. Such is the case, e.g., in questions relating to the role of the immunological milieu in human liver disorders. The human liver contains large numbers of immune cells that contribute to various pathological processes 1, 2. Detailed assessment of the intrahepatic immune environment in the human liver by flow cytometry has been generally hampered by the need for large sample sizes and insufficient software tools that can provide accurate information from the, often complex, data attained. Here, we describe how the technique of ultrasound guided fine-needle liver aspiration biopsy (FNAB) sampling in combination with 22-parameter flow cytometry technology 3 and new software analysis of data 4 readily can provide high-resolution mapping of fine immune cell signatures in the human liver. The described technology significantly advances our ability to investigate intricacies of human liver immunology during physiological as well as pathological processes including drug interventions and other conditions. Most likely the techniques and data analysis described here could be adapted also to studies of immune cells from other human organs in physiological or disease settings. As a point of departure, a 22-parameter flow cytometry panel was developed covering major myeloid and lymphoid cell populations such as naïve, effector, and terminal T cells, naïve, unswitched, and switched memory B cells, immature and mature NK cells, and subpopulations of DCs (Supporting Information Table 1A). Use of this panel enabled us to theoretically collect of up to 4.2 million phenotypes (222). Such high numbers of dimensions and potential phenotypes are, however, difficult to analyze and visualize when using standard flow cytometry software 5. Additionally, multivariate relationships that are present in more than two dimensions are easily overlooked. To circumvent the latter challenge, we demonstrate here the usefulness of stochastic neighbor embedding (SNE) 5, a nonlinear dimensionality reduction analysis method, that identifies two-dimensional representations of data at the single-cell level. This technique preserves local and global geometry and has previously been reported to be superior compared to other dimensionality reduction techniques for flow cytometry data visualization 5. Here, we demonstrate its usefulness in the context of probing the total immune cell populations in human liver specimens retrieved by fine-needle aspiration biopsy (FNAB) and compare data from matched PBMC samples (Fig. 1A and Supporting Information Fig. 1A). Aiding visual interpretation of the present liver-specific traits, individual population intensities of PBMC signatures could be subtracted from FNAB signatures (Fig. 1A). When visually evaluating expression intensities of the single parameters used to generate the signatures, SNE analysis resulted in a robust classification of myeloid and lymphoid immune cells into distinct subsets (Fig. 1B). Sample batching and deferred analysis are often prerequisites when longitudinal samples are collected from human subjects, e.g., when multiple collection sites participate in a clinical study where a central facility is responsible for analysis. Such is the case, e.g., in the collection of rare samples over long periods of time or the acquisition of specimens from resource poor settings where appropriate equipment is not accessible 6. To deal with this issue, we developed a method to cryopreserve small numbers of fixed cells before actual flow cytometry staining and subsequent SNE analysis (Fig. 1C). Remarkably, compared to no fixation, prior fixation significantly improved the yields obtained after thawing of samples, thereby making further flow cytometry analysis if anything more robust. In this context, it is a well-known fact that fixation of cells by many different methods may affect mAb epitopes. Addressing this issue, we found the fixative CytoFix to be superior in preserving staining on cells. With this fixative, results were to a large degree comparable to that of unfixed cells (Fig. 1D and E, and Supporting Information Fig. 2). When analyzing rare and limited samples, it is often only possible to perform staining with a single panel. To maximize data output, the antibody panel used was designed so that it would contain several different stricto sensu lineage-specific mAbs conjugated with the same fluorochrome. This allowed us to move beyond reliance on the number of detectors available in the standard flow cytometry instrument used (Supporting Information Tables 1A and B), a strategy previously described for 8-color flow cytometry 7. No significant differences were detected between staining with each of the mAbs alone compared with staining with different mAb combinations (Supporting Information Fig. 3). Furthermore, no differences in subset frequencies were found when high and low cell numbers were compared, showing that analysis of as few as 50,000 cells is enough to characterize major and minor immune cell subsets from the samples analyzed (Fig. 1F). Taken together, we illustrate here how clinical material from FNAB, followed by fixing and cryopreserving of cells, subsequent 22-parameter flow cytometry analysis and advanced software analysis can yield results on liver immune signatures at a level previously unattainable. To illustrate the usability of the presently described strategy for high-resolution analysis of a limited numbers of cells, we evaluated the presence of distinct lymphocyte populations in FNAB samples (Fig. 2A, Supporting Information Fig. 1B). SNE generated a signature that accurately distinguished immature and mature CD4 and CD8 T cells, γδ T cells, mucosal-associated invariant T cells, subsets of NK cells, as well as mature and immature B cells (Supporting Information Fig. 1C). A residual plot display readily portrayed differences between the PBMC and FNAB samples (Fig. 2A). In this context, although SNE allows for extraction of phenotypic signatures from identified subsets 8, it is not designed to statistically determine which of the alterations between datasets significantly contribute to differentiating the sets. To do this, one can apply a recently published method that combines hierarchical clustering with group lasso regression analysis for selection of clusters distinguishing FNAB from PMBC samples 9. A similar approach has recently been used to determine peripheral blood immune signatures correlating with clinical recovery after surgery 10. Here, clustering of 17 parameters was performed and a total of 52 clusters in a hierarchical tree was generated (Fig. 2B and Supporting Information Fig. 4A and Table 1). In the subsequent group lasso regression analysis, three distinguishing clusters were identified, two out of which were more abundant in the FNAB samples as compared with PBMC samples (Fig. 2C and D and Supporting Information Table 2, Fig. 4D). The parameters used for generating the hierarchical cluster signature identified phenotypes of the three distinguishing clusters (Fig. 2E). Cluster I represented NKG2A+/−CD56bright NK cells, and clusters II and III signified discrete subsets of CD4 T cells (Fig. 2E). The three clusters were also recognizable in the SNE signature (Supporting Information Fig. 1C and 4B), indicating a robustness of the findings. Finally, and interestingly, with conventional flow cytometry analysis, only cluster I could be assessed and confirmed (Fig. 2F). In contrast, the two CD4 T cell clusters were present in multivariate space not allowing identification by standard flow cytometric analysis (Fig. 2E). Taken together, we illustrate how state-of-the-art methodologies, technologies, and software can be applied to fine human liver immunophenotyping at a level not previously attainable, based on small number of cells obtained with a minimally invasive procedure (i.e. FNAB). With the complexity of flow cytometry data continuously increasing new approaches to handle data, here exemplified by the use of SNE and hierarchical clustering with automated identification of distinguishing immune cell subsets, likely will facilitate unbiased analysis and find previously unrecognized multivariate cellular phenotypes. These techniques, and other techniques such as the imaging methods histocytometry and laser-scanning microscopy, which allow for investigation of the spatial segregation of cells in tissue 11, 12, should expand the horizons of human liver immunology research and advance our ability to interrogate the intricacies of immune responses in this organ. Finally, the strategies described above are most certainly not limited to the liver, but could readily be extended to other human organs and tissues. We thank M. Holm, UPPMAX, Uppsala for writing data pre- and postprocessing code, which enabled usage of Barnes-Hut SNE. This work was funded by the International Research Training Group 1273 supported by the German Research Foundation (to H.W.), the Center Research Grants 738 (project B2) and 900 (project A5) supported by the DFG (H.W.), the Deutsches Zentrum für Infektionsforschung (H.W.), the Swedish Research Council (N.K.B. and H.G.L.), Swedish Cancer Society (N.K.B. and H.G.L.), the Karolinska Institutet (N.K.B. and H.G.L.), the Swedish Society for Medical Research (N.K.B.), the Cancer Research Foundations of Radiumhemmet (N.K.B), the Swedish Society of Medicine (N.K.B.), Novo Nordisk's Foundation (N.K.B.), Jeansson's Foundation (N.K.B.), Bengt Ihre's Foundation (N.K.B.), Åke Olsson's Foundation (N.K.B.), Groschinsky's Foundation (N.K.B.), Hedlunds' Foundation (N.K.B.), KI-DKFZ Joint Project Grant (N.K.B. and H.G.L.), and Julin's Foundation (N.K.B.). The authors declare no financial or commercial conflict of interest. As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. Table 1A. Instrument settings and staining panel Table 1B. Lineage restriction of markers used in double Table 2. Cluster sizes (frequency) of FNAB and PBMC samples for the 52 clusters identified with Citrus Figure s1 Figure s2 Figure s3 Figure s4 Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.
Read more