Single-cell RNA sequencing of bronchoscopy specimens: development of a rapid minimal handling protocol

The airway mucosa plays a critical role in maintaining local and systemic homeostasis. Its protective functional properties include intercellular epithelial junctions, mucociliary clearance, soluble intraluminal immune factors, active immune surveillance and initiation of innate and adaptive cellular immunity [1]. Profiling the transcriptome of the airway mucosa has revealed insights into normal airway physiology as well as the pathophysiology of numerous diseases, including asthma [2–4], chronic obstructive pulmonary disease [5–8] and lung cancer [9,10]. However, bulk transcriptomics (either microarray or RNA sequencing) is unlikely to have sufficient resolution to capture the complexity of the airway mucosal cellular landscape.

Single-cell RNA sequencing (scRNA-seq) allows analysis of the transcriptome at single-cell resolution. However, the majority of scRNA-seq data published to date have been derived from invasive airway biopsy, surgical resection or postmortem specimens. Such specimens are difficult to obtain and highly invasive and carry inherent safety risks. This presents a limitation to future large-scale scRNA-seq investigation of the airways, particularly in participants with comorbidities who have a higher risk of complications.

Bronchial brush (BB) sampling of the airway mucosa is a less invasive and relatively safe procedure performed in routine clinical practice. A cytological brush is inserted via a bronchoscope into the airways and gently scraped against the airway wall to capture cells. In clinical practice, the specimens are assessed by cytology, microbiology or nucleic acid tests. In research settings, BB specimens have been used for profiling the airway transcriptome [2,3,6–9] and microbiome [11]. More recently, a number of studies have used BB specimens for scRNA-seq analysis of the airways [12–17].

Despite many years of experience with research bronchoscopies and airway transcriptomics using BB specimens, our early attempts at preparing BB specimens for scRNA-seq were hampered by long processing times, low cell recovery, poor cell viability and low RNA quality. A review of the literature revealed significant heterogeneity among the previously published protocols used for this type of specimen (Table 1). We therefore aimed to develop a novel protocol for processing BB specimens for scRNA-seq based on the following guiding principles: minimize the handling and manipulation of cells, create a single-cell suspension with negligible cell clumping or debris, maximize cell recovery and viability and minimize processing time. Here we provide a qualitative description of the development of our optimized protocol for research. The resulting optimized protocol has facilitated a rapid expansion of our program, with over 40 specimens processed in the past 20 months. Given the inherent variability when working with clinical samples and to emphasize that, subsequent to protocol refinement, our methodology has yielded promising results across diverse disease conditions, we have provided the coefficient of variation for all sequenced samples to date in Supplementary Figure 5.

Materials & methods

Human research ethics

The specimens used for this study were collected under protocols that were approved by the Human Research Ethics Board of the University of British Columbia/Providence Health Care (approval nos. H11-02713, H19-02222 and H21-02149). All study participants provided informed consent.

BB protocol

The details of our standardized research bronchoscopy protocol have been published previously [18]. After receipt of informed consent and under conscious sedation, a fiberoptic bronchoscope (Olympus Corporation, Tokyo, Japan) was passed through the mouth of participants and into the trachea. With the bronchoscope positioned in one of the subsegmental bronchi of the right or left upper lobe, a cytological brush was inserted through a bronchoscope channel into a sixth- to eighth-generation airway, from where BB samples were collected. The cytological brush was then withdrawn from the bronchoscope, and using a pair of stainless steel scissors, the brush was cut into a microcentrifuge tube containing 1000 μl medium and kept on ice until further processing. We evaluated the following media for sample quality (Table 2): Dulbecco's phosphate-buffered saline (DPBS) (Thermo Fisher Scientific, MA, USA), Roswell Park Memorial Institute 1640 (RPMI1640) (Corning, NY, USA) with and without fetal bovine serum (Thermo Fisher Scientific, MA, USA) and PneumaCult-Ex™ culture medium (STEMCELL Technologies, Vancouver, Canada).

For enzymatic digestion of intercellular junctions and extracellular matrix, approximately 1000 μl of Accutase™ (Thermo Fisher Scientific) was added directly to brush specimens in the medium followed by gentle agitation. The digestion time was titrated from 3 to 10 min, and the enzyme activity was quenched by directly transferring the combined Accutase™/medium into a separate tube containing either 2 ml (‘small-volume protocol’) or 25 ml (‘large-volume protocol’) of medium.

To wash the cells, the cell suspension was passed through a 70-μm cell strainer (VWR, PA, USA) to remove large debris and cell clumps and centrifuged at 400 × g for 10 min at 4°C, after which the supernatant was discarded. Under the small-volume protocol, the cell pellet was resuspended in 2 ml medium and centrifugation was repeated as described, whereas the large-volume protocol included a single centrifugation step. The final cell pellet was resuspended in medium to a total volume of 100 μl, which was then filtered through either a 40-μm cell strainer (VWR) or a 40-μm Flowmi™ strainer (SP Bel-Art, PA, USA).

Single-cell RNA sequencing

Visual assessments and cell counts of the final cell suspensions were performed using a hemocytometer with trypan blue stain (Thermo Fisher Scientific). When necessary, the cell suspensions were diluted to a concentration of 1000 cells/μl in their respective media prior to single-cell sequencing. For further examination, aliquots of the cell suspensions were applied to histological slides by cytospin and stained using a modified Wright–Giemsa stain.

Cell capture & library preparation

Cell suspensions were transferred on ice to our sequencing facility (approximately 25 min away) and loaded onto the Chromium Controller or Chromium X using the Chromium Next GEM Single Cell 3′ Kit v3.1 and Chip G (10x Genomics, Inc., CA, USA). Samples were handled and libraries prepared according to the published Chromium Single Cell 3′ Reagent Kits User Guide [19]. The quality of the cDNA libraries and final DNA libraries was analyzed using the 2100 Bioanalyzer instrument with the High Sensitivity DNA Kit (Agilent Technologies, Inc., CA, USA).

Sequencing

Final libraries were sequenced on the NextSeq 2000 (Illumina, Inc., CA, USA) as recommended by 10x Genomics (read 1: 28 bp; read 2: 90 bp; i7 index: 10 bp; i5 index: 10 bp) at a loading concentration of 650 pM with a 2% PhiX spike-in (Illumina, Inc.). All sequencing was initially performed at a shallow depth of 20,000 reads/cell and assessed, and then further sequencing was performed to a depth of 60,000 reads/cell. Fastq files were generated with Cell Ranger v6.01 (10x Genomics) using the default options, and reads were aligned to the human reference genome (hg19).

Bioinformatics

A detailed description of the data processing is provided in Supplementary Figure 1. Analysis of scRNA-seq data can be challenging because of the presence of technical artifacts, such as ambient RNA. Such contamination would inadvertently impact the downstream analysis of the samples. The SoupX package (version 1.5.2) [20] in R was used prior to performing individual quality control tasks to correct for ambient RNA, whose expression is projected from the empty droplet pool. Following ambient RNA correction, using the Pegasus package (version 1.5.0) [21,22] in Python, we performed additional quality control steps, in which cells with high mitochondrial genes (>20%) and <200 genes per cell were filtered out. This step was followed by log normalization, batch effect correction using the harmony algorithm and doublet removal. We then performed principal component analysis prior to applying Uniform Manifold Approximation and Projection dimensionality reduction. Before performing any further downstream analysis, we annotated the data set using automatic annotation in the Pegasus package in conjunction with the markers noted in the literature.

Results & discussion

The central motivation for our work was to investigate the pathology of airway diseases using scRNA-seq. Bronchial brushing is an attractive method for sampling the lung because it is relatively noninvasive and safe and is routinely performed in clinical practice. A number of investigators have reported scRNA-seq using BB specimens [12–17]. However, a review of these published protocols revealed key differences in the way the specimens were processed (Table 1). Additionally, the number of cells recovered per participant was highly variable, and none of the published protocols were accompanied by detailed validation of their results. We therefore sought to develop our own protocol that maximized cell recovery and quality. The final BB protocol and, for the sake of completeness and in the interest of other investigators, the optimized protocol for preparing bronchoalveolar lavage specimens as well as a Uniform Manifold Approximation and Projection of one of the bronchoalveolar lavage samples we submitted for sequencing using this protocol are provided in Supplementary Figure 2.

The choice of medium is important not only for maintaining cell viability but also for supporting single-cell suspension. The 10x Genomics standard protocol [23] recommends a calcium/magnesium-free balanced salt solution spiked with protein, which minimizes cell losses and aggregation. Indeed, the published BB protocols used either fetal bovine serum or bovine serum albumin (Table 1). However, we found that the use of these solutions for BB specimens was associated with poor cell viability and excessive clumping of cells, which led to low concentrations of viable cells at the end of processing. We found that the best medium was PneumaCult-Ex™, a serum- and bovine pituitary extract-free cell culture medium specifically formulated to support the growth of airway epithelial cells. We used this medium at all stages of processing BB specimens in the optimized protocol. To the best of our knowledge, this is the first time this medium has been used for scRNA-seq experiments.

In order to generate a single-cell suspension from BB specimens, it was necessary to incubate the brushes in a protease solution. Enzymatic digestion of whole tissue and biopsy specimens is routinely used to break down extracellular matrix and create single-cell suspensions. Although BB specimens contain little extracellular matrix, the epithelial cells scraped from the airways were connected by intercellular junctions and therefore tended to remain in sheet-like clumps (Figure 1). The use of protease treatment of BB specimens was common to all previously published protocols, though most employed nonspecific proteases, such as trypsin (Table 1). We elected to use Accutase™, an enzyme mixture with proteolytic and collagenolytic properties, since it is associated with efficient cell dissociation without interrupting cell surface antigens or inducing cell cycle changes [24] and leads to improved viability when culturing fragile cell types [25]. Digestion for 10 min, as used by Chua et al. [12], resulted in widespread cell destruction and fragmentation, with occasional single cells that lacked the expected morphology of bronchial epithelial cells (Figure 2A & Supplementary Figures 3 & 4). After single-cell capture and cDNA library preparation, the Bioanalyzer indicated high levels of short sequences suggestive of either high ambient RNA or fragmented RNA in the specimens (Figure 2B). The processed data confirmed high levels of ambient RNA, a low number of cells (far below the expected number based on the sample input) and only three identifiable cell clusters (Figure 2C). Notably, the proportion of epithelial cells was very low (Figure 2D), suggesting that this cell type is particularly sensitive to the effects of prolonged protease activity. By contrast, Accutase™ digestion for 3 min produced a single-cell suspension with readily identifiable bronchial epithelial cells, including ciliated and goblet cells, as well as immune cells, such as monocytes, and the resulting cDNA Bioanalyzer output showed superior quality of RNA (Figure 3A). The processed scRNA-seq data demonstrated cell numbers closer to the expected count, with a diverse population of epithelial and immune cells consistent with the expected composition of the airway mucosa (Figure 3B & C). We therefore concluded that 3 min was the optimal duration of Accutase™ incubation.

Washing with medium followed by centrifugation not only removes ambient (i.e., extracellular) RNA but also eliminates residual proteases from the cells. The previously published protocols for BB specimens as well as the 10x Genomics standard protocol [23] employed a minimum of two wash–centrifuge cycles (Table 1). However, we found that multiple wash cycles led to excessive cell loss. We therefore employed a single large-volume wash of 25 ml PneumaCult-Ex™ for BB specimens, which greatly improved the cell yield while minimizing ambient RNA. We also removed debris and cell clumps using mechanical filtration. The presence of cell clumps not only leads to multiplets in the sequencing output (i.e., more than one cell being captured in an oil droplet) but also interferes with the microfluidic technology and generation of the gel emulsion [23]. We found that a 40-μm cell strainer was adequate to remove large clumps and debris, such as mucus, as an initial step. However, the cell strainer required adequate rinsing with clean medium to minimize the loss of single cells captured by the strainer. Since cell retrieval from BB specimens was typically low, this use of extra medium further diluted our cell concentrations. In order to minimize the volume and therefore increase cell concentration, we used a 40-μm Flowmi™ cell strainer as the final filtration step. This filter attaches directly to the pipette tip and can be used to filter volumes as low as 50 μl with minimal volume loss. In our experience, we were able to filter volumes of 100–120 μl, with a typical volume loss of 20–40 μl.

To optimize the BB specimen protocol, there were several other considerations. The 10x Genomics standard protocol [23] and some of the published protocols [16,17] employed a dead cell removal step (e.g., flow cytometry) in order to maximize the viable cell concentration. However, we elected to forgo this step for three reasons. First, our cell recovery was typically low, and dead cell removal would further reduce the final cell concentration. Second, we considered our cells fragile and needed to avoid further mechanical stress. Third, the quality of our cells was highly dependent on the time since collection, meaning delays related to this step would further compromise the overall quality of the cells. We also did not employ red blood cell lysis techniques since our bronchial epithelial cells were sensitive to the typical lysis reagents (e.g., ammonium chloride). Instead, we avoided using macroscopically contaminated specimens and removed any red blood cell ‘contaminants’ in silico using bioinformatics tools. We kept specimens and reagents at 4°C rather than at room temperature (as recommended by the 10x Genomics standard protocol) [23] because, in our experience, this minimized cell clumping. Finally, we used low retention materials (conical and microcentrifuge tubes) wherever possible to maximize cell recovery. With each of these measures, our BB specimen protocol produced high-quality single-cell suspensions with increased and reliable cell counts (Table 2). The processed sequencing data showed that all of the expected cell types in the airway epithelium were present (Figure 4). Demographic information pertaining to the data presented in Figure 4 can be found in Supplementary Figure 6.

Limitations

At the time of this protocol development, we faced multiple challenges, including operating in the midst of a once-in-a-generation pandemic, which severely limited our access to human samples; the scarcity of samples available for replication/validation since each participant was enrolled under a set approved protocol for specimen collection; the extremely poor quality of cells in the early stages, which limited our ability to provide even the most basic quantitative metrics (e.g., cell count and viability); and the prohibitively high cost of sequencing suboptimal specimens. These challenges contributed to some important limitations. First, the limited sample availability meant that we were unable to systematically vary the experiments one variable at a time. It also restricted our ability to obtain technical replicates for each of the experimental conditions. Taken together, we were not able to robustly assess the effects of varying each condition and instead relied on qualitative descriptions of single trials. We included quantitative measures such as cell counts and sequencing output wherever possible. We purposefully avoided the use of cultured cells to optimize the protocol since, in our experience, primary BB cells do not behave in the same way as cultured cells, especially with regard to their fragility. Second, the bronchoscopy samples were processed immediately following bronchoscopy, which kept the overall time from cell collection to capture short. Having the bronchoscopy suite and the processing lab in the same building might not be the case for all research centers, and as such our protocol might not be generalizable to laboratories with different setups. Nevertheless, we believe that overcoming such logistical difficulties to maximize efficiency and minimize processing time should be a guiding principle for all laboratories performing scRNA-seq of BBs regardless of the setup.

Conclusion

We developed a protocol for preparing BB and bronchoalveolar lavage specimens for scRNA-seq that maximizes cell viability and recovery by minimizing specimen handling and processing time. The use of this optimized protocol has allowed us to build an scRNA-seq data set from over 50,000 cells, representing the airways of 26 participants. This large data set will be used to explore the pathology of various airway diseases and how changes in the airways at single-cell resolution may relate to clinically important outcomes.