An efficient method to isolate Kupffer cells eliminating endothelial cell contamination and selective bias

Abstract Multicolor flow cytometry and cell sorting are powerful immunologic tools for the study of hepatic mϕ, yet there is no consensus on the optimal method to prepare liver homogenates for these analyses. Using a combination of mϕ and endothelial cell reporter mice, flow cytometry, and confocal imaging, we have shown that conventional flow‐cytometric strategies for identification of Kupffer cells (KCs) leads to inclusion of a significant proportion of CD31hi endothelial cells. These cells were present regardless of the method used to prepare cells for flow cytometry and represented endothelium tightly adhered to remnants of KC membrane. Antibodies to endothelial markers, such as CD31, were vital for their exclusion. This result brings into focus recently published microarray datasets that identify high expression of endothelial cell‐associated genes by KCs compared with other tissue‐resident mϕ. Our studies also revealed significant and specific loss of KCs among leukocytes with commonly used isolation methods that led to enrichment of proliferating and monocyte‐derived mϕ. Hence, we present an optimal method to generate high yields of liver myeloid cells without bias for cell type or contamination with endothelial cells.


INTRODUCTION
Liver Kupffer cells (KCs) are one of the largest populations of resident m in the body. KCs are located in the sinusoids of the liver where they scavenge and phagocytose apoptotic cells and damaged erythrocytes, 1 contribute to maintenance of immunologic tolerance by priming Foxp3 + T-regulatory cells, 2 and capture gut commensal bacteria that enter the circulation. 3 Under homeostatic conditions, KCs proliferate in situ and persist with relatively little input from conventional hematopoiesis in adult mice. [4][5][6] However, during liver inflammation, stress, or injury, Ly6C hi monocytes are recruited to the liver and subsequently mature into monocyte-derived hepatic m .
Both KCs and monocyte-derived m have been attributed prorestorative or proinflammatory roles in models of acute and chronic liver damage. [7][8][9] To better understand the function of KCs and monocyte derived m , it is important not only to accurately identify these cells but also to ensure a comprehensive portrait of the in vivo population is generated.
Multiparameter flow cytometry is a powerful tool for evaluating changes in number, frequency, and phenotype of diverse monocyte and m populations and is the basis by which these cells are purified for subsequent functional and genomic analyses. Many protocols have been used to isolate leukocytes from the liver, but there is no consensus on which method is most valid, particularly for KCs. Furthermore, although the definition of KCs by flow cytometry is widely accepted as F4/80 hi CD11b lo cells, the reliability of this approach has not been fully investigated. For example, several microarray and RNA-seq datasets identified Cdh5, a gene typically associated with endothelial cells, 10  Here, we demonstrate that the population of KCs conventionally defined by their F4/80 hi CD11b lo phenotype contains a significant proportion of contaminating CD45 + CD31 hi endothelial cells, and that this contamination was present regardless of the method used for isolating leukocytes from the liver. Inclusion of endothelial markers rather than additional surface markers of m was critical for excluding these cells from analysis. Furthermore, quantity and quality of isolated KCs varied significantly dependent on the purification method used. We therefore present a comprehensive protocol for faithfully isolating leukocytes from the liver and a modified gating strategy to effectively eliminate contaminating endothelial cells.

CSF1-Fc and BrdU administration
An Fc conjugate of porcine CSF1 (CSF1-Fc) was prepared as described. 18 Analysis of KC and endothelial cell proliferation in response to administration of CSF1-Fc was performed on cells from a larger unpublished study aimed at assessing the effect of chronic CSF1 delivery on KCs origin in tissue-protected bone marrow chimeric mice made as described previously. 17 Tissue protected chimeric mice were given

300 g centrifugation
Cells were washed in 50 ml, then 30 ml RPMI, and centrifuged at 300 g for 5 min, maximum break and accelerator. RBC lysis buffer (Sigma; 2 ml) was added for 2 min, followed by 2 ml FACS buffer (PBS supplemented with 0.5% BSA and 2 mM EDTA). Cells were pelleted (300 g, 5 min) and the supernatant discarded.

33% Percoll TM gradient
Cells were washed twice in 50 ml liver wash buffer (PBS/2% FCS) by centrifugation at 443 g for 6 min, maximum break and accelerator. The pellet was resuspended in a room-temperature 33% Percoll gradient (25 ml per sample) and spun at 693 g for 12 min, with minimum break and accelerator. The cell pellet was washed in 30 ml liver wash buffer at 300 g for 5 min. RBC lysis buffer (5 ml) was added for 5 min, then 30 ml liver wash buffer and cells spun at 300 g for 5 min.

50 g centrifugation
Cells were washed in 15 ml RPMI containing 10% FCS and centrifuged at 50 g for 10 min with minimum break. The supernatant was collected and spun at 340 g for 10 min, minimum break. The pellet was lysed for 5 min in 2 ml RBC lysis buffer on ice, topped up with RPMI + 10% FCS and spun at 340 g for 10 min, minimum break.

Collection of discarded fractions
For the 300 g spin and Percoll gradient methods, the supernatant or both the hepatocyte layer, and the supernatant between the hepatocyte layer and the leukocyte pellet, respectively, was collected into a fresh tube and centrifuged at 400 g for 5 min. The resultant pellets were counted and stained. For the 50 g slow-spin method, the pellet generated following the 50 g spin was counted and stained.

Isolation of leukocytes from lung
Perfused lungs were collected into RPMI, homogenized using scissors and digested in 2 ml of the enzyme mix detailed above, for 45 min at 37 • C. Digests were filtered through a 100 m strainer, washed with FACS buffer and RBC lysed in 3 ml RBC lysis buffer (Sigma) for 3 min.
After washing, cells were passed through a 40 m strainer and counted.

Flow cytometry
2 × 10 6 liver cells, or 20 l of whole blood was incubated with Zombie Aqua fixable viability dye (Biolegend, London, UK) for 10 min at RT and then with 0.025 g anti-CD16/32 (2.4G2; Biolegend) in 10% normal mouse serum (Life Technologies, Paisley, UK). Cells were then incubated with antibodies (Supplemental Table 1). Cells were washed, spun at 300 g for 5 min and, where necessary, incubated with fluorescently labeled streptavidin. 7-AAD solution (Biolegend) was added to samples 10 min before acquisition when comparing isolation protocols.
DAPI was used as a viability marker for FACS. Liver cells were gated as shown, whereas alveolar and interstitial m were identified as

F4/80 hi CD11b lo KCs identified by flow cytometry contain a subset of Cdh5 hi CD31 hi cells
KCs have been traditionally defined by their F4/80 hi CD11b lo phenotype, which distinguishes them from F4/80 lo CD11b + bonemarrow-derived myeloid cells. 19 Global transcriptomic analyses have demonstrated that, like all tissue m , KCs have a unique transcriptional signature, [11][12][13] with CLEC4F and Tim4 emerging as markers that aid their discrimination from other hepatic monocytes and m . 12,20 However, these same analyses also identified Cdh5 (which encodes cadherin-5), a gene strongly associated with endothelial cells, 10 (Fig. 1D). In comparison, less than 0.52% of lung interstitial m and 0.53% of alveolar m were CD31 hi GFP hi suggesting this phenomenon was not common to all m populations (Fig. 1D). Back-gating of the CD31 hi GFP hi and CD31 lo GFP lo populations revealed they could not be discriminated based on size, granularity, and expression of other leukocyte markers such as CD45, Ly6C, and MHCII (data not shown).
However, whereas tdTomato expression was significantly diminished in the CD31 hi GFP hi fraction of KCs (Fig. 1E), the CD31 lo GFP lo cells retained identical levels to untreated control mice, indicating that Cre-mediated excision of the tdTomato cassette had not occurred ( Fig. 1F). A similar minor increase in GFP fluorescence without loss of tdTomato expression was also observed in B cells, T cells, neutrophils, and monocytes following tamoxifen treatment (data not shown).
Hence, although F4/80 hi CD11b lo -defined KCs contain a subset of CD31 hi Cre-expressing cells, it seems likely that the low GFP fluorescence of CD31 lo KC from tamoxifen-treated mice does not represent meaningful expression of Cdh5-driven Cre.

Endothelial cells contaminate the traditional F4/80 hi CD11b lo KC gate
We further investigated the identity of the CD31 hi GFP + fraction of F4/80 hi CD11b lo cells. Analysis of all CD31 hi cells within the liver preparations revealed binding of CD45 and F4/80 antibodies to be specific when compared with their respective isotype controls ( Fig. 2A).
Although these could be simple doublets, they may also represent transmigrating hematopoietic cells described in human liver sinusoidal endothelial cells (LSECs) in vitro. 22 Either way, we find no evidence that CD45 + CD31 hi cells represent single cells with both hematopoietic and endothelial characteristics akin to those described in rat liver. 23

Endothelial cell contamination is present irrespective of liver digestion protocol
Multiple methods for generating single cell preparations of murine liver leukocytes have been published and while most protocols involve enzymatic digestion, the separation of leukocytes from hepatocyte debris is more inconsistent. Hence, to exclude the possibility that the observed endothelial cell contamination was an artifact specific to our protocol for isolation of leukocytes from the liver, we compared the frequency of CD31 hi cells in the KCs population retrieved using 3 methods representative of commonly published protocols. In brief, all methods used the same enzymatic digestion step but employed either two 300 g centrifugation steps (as used for Figs. 1 and 2), 17 an initial 50 g prespin to first pellet and discard hepatocytes, 24,25 or a 33% Percoll gradient to remove the majority of hepatic debris. 26,27 The relative frequency of all CD31 hi endothelial cells recovered as a proportion of all live cells was equivalent between the 300 g and the Percoll gradient methods, though reduced with the 50 g prespin method, suggesting a proportion of endothelial cells are removed by this step (Fig 3A). However, there was a clear population of CD31 hi and 3C), and most pronounced with the 50 g method. Notably, KC-specific Cdh5 expression was observed in microarray datasets generated with 11,13 or without 12 Percoll gradient purification and low-speed centrifugation. Hence, although it remains possible that KC express Cdh5 at higher levels than other resident tissue m , it is highly likely that contaminating endothelial cells contributed significantly to the high expression of this gene by KCs. The exclusion of CD31 hi cells ( Fig. 3D) is therefore a critical step for the faithful identification of KCs in mouse liver irrespective of isolation protocol.

Isolation method can lead to selective loss of KCs
After removal of CD31 hi endothelial cells, we observed clear differences in the relative abundance of KCs amongst live CD45 + cells between isolation methods, with much lower frequencies using both the Percoll and 50 g prespin methods, whereas the frequency of F4/80 lo CD11b hi cells was more consistent (Fig. 4A). To determine whether the difference in relative abundance in KCs between methods was due to selective loss of these cells with the Percoll gradient and 50 g prespin methods or enrichment with the 300 g method, we analyzed the hepatocyte pellet from the 50 g prespin, the supernatant from the first 300 g spin, and the hepatocyte layer from the Percoll gradient alongside the normal isolates. KCs were clearly identifiable in the discards from all protocols but at an elevated frequency in the Percoll and 50 g spin methods (Fig. 4B). Comparison of the frequency of KCs of total CD45 + CD31 − cells within the isolate compared directly with the frequency of KCs in the discard for each sample showed that KCs were significantly enriched in the discard of both the Percoll and 50 g spin methods, but not in the 300 g spin method (Figs. 4C-4E). Thus, KCs were selectively lost using both the Percoll and the 50 g spin methods, whereas the ratio of all leukocytes including KCs was not altered in the discard versus the isolate with the 300 g spin method. The Percoll method also yielded far fewer CD45 + cells compared with the 300 g method (Fig. 4F), that together with the lower abundance of KCs, corresponded to over a 7-fold reduction in yield of KCs using this method ( Fig. 4G). It was not possible to obtain accurate cell counts from the 50 g method due to large amount of debris seemingly retained with this method, nor were many KCs present in the pellet extracted from the discarded Percoll (Fig. 4G) For simplicity, only 1 slow prespin (50 g) and 1 density gradient (33% Percoll) method were compared in our study. However, we acknowledge that there is considerable variation reported in number and speed of centrifugations, as well as gradient concentrations used in protocols to isolate liver leukocytes. 11,13,28,29 Nevertheless, our data suggest that careful optimization of these methods will be required for studies where KCs are the major cell of interest.

SUMMARY
It is imperative that populations isolated for flow cytometric, gene comparison, and functional analyses are representative of those in vivo. We have demonstrated that the established CD45 + F4/80 hi CD11b lo gating strategy used to identify KCs in liver preparations also contains a population of endothelial cells. These cells could not be excluded using antibodies to KCs surface antigens and because endothelial cells were also found to form aggregates with other CD45 + cell types, we propose the simplest method to exclude them from analysis is by inclusion of antibodies to CD31 or other endothelial markers. We suggest that this strategy be adopted universally (Fig. 3D), particularly as endothelial contamination was apparent using all commonly used methods for the preparation of