A mannitol-based buffer improves single-cell RNA sequencing of high-salt marine cells

Abstract

Single-cell RNA sequencing (scRNA-seq) enables discovery of novel cell states by transcriptomic profiling with minimal prior knowledge, making it useful for studying non-model organisms. For most marine organisms, however, cells are viable at a higher salinity than is compatible with scRNA-seq, impacting data quality and cell representation. We show that a low-salinity phosphate buffer supplemented with D-mannitol (PBS-M) enables higher-quality scRNA-seq of blood cells from the tunicate Ciona robusta. Using PBS-M reduces cell death and ambient mRNA, revealing cell states not otherwise detected. We additionally validate PBS-M in a second tunicate species. This simple protocol modification could enable or improve scRNA-seq for many marine organisms.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12864-025-12370-7.

Background

Single-cell RNA sequencing (scRNA-seq) represents several methods for measuring the mRNA composition of individual cells at genome scale. These methods enable identification and characterization of previously unknown cell types with minimal prior knowledge. scRNA-seq has led to new discoveries in well-studied animals, but it is particularly useful when studying non-model organisms where the wide diversity of cell types is unexplored [1]. At present, the highest-quality scRNA-seq methods require live cells, prepared in suspension after dissociation from tissues [2–5].

Salinity introduces a challenge for droplet-based scRNA-seq of marine organisms. In droplet-based methods, the buffer in which cells are loaded must be compatible with the first step in scRNA-seq, a reverse transcription (RT) of mRNA to make complementary DNA (cDNA). This reaction is carried out using variants of the Moloney Murine Leukemia Virus (MMLV) reverse transcriptase, which is efficient at the osmolarity of vertebrate cells (~300 mOsm/L) [4]. Most marine organisms — invertebrates, protists, and some vertebrates — are osmoconformers, with an internal osmolarity close to that of sea water (~1000 mOsm/L) [6–10]. RT enzyme efficiency drops below 10% at salt concentrations above 200 mM (400 mOsm/L) [11], making high-salinity buffers incompatible with droplet-based methods. Cells isolated from marine osmoconformers thus cannot be introduced into droplet-based scRNA-seq using native-salinity buffers.

Currently, a standard approach for scRNA-seq of marine cells is to dilute live cells to a lower salinity (≤ 300 mOsm/L) before droplet-based encapsulation and subsequent RT. Datasets from multiple marine organisms have been collected in this way, including from tunicates, echinoderms, arthropods, cephalopods, and cnidarians [12–18]. This strategy is premised on cells surviving the decreased osmotic pressure over the time required to complete encapsulation, but some cells might react to, or die from, the osmotic shock. Dying cells could be missing from the dataset and contribute ambient mRNA to other cells, and other cells may exhibit altered transcriptional states.

We report on a simple method to prepare cells for droplet-based scRNA-seq that satisfies both the osmolarity requirements of marine cells and the salinity requirements of the RT. We benchmark the method with blood cells from the tunicate Ciona robusta, which we show are sensitive to osmotic shock, and we additionally validate the approach in a second tunicate species. We expect that our approach will be broadly useful for single-cell study of marine organisms.

Methods

Blood cell collection

Adult Ciona robusta supplied by M-REP (Carlsbad, California, United States of America) were kept at 17–18 °C in artificial seawater (ASW, Instant Ocean SS15-10) and fed daily with 15 mL of Phyto-Feast (Reef Nutrition). All animals were analyzed at 1–2 days after arriving in the lab.

Adults were relaxed by incubating for 3–3.5 h at 4 °C (for scRNA-seq, RT efficiency assay, and two replicates of the cell recovery assay) in about 400 mL ASW with approximately 5 g of 99% L-menthol crystals (Thermo Scientific A10474) (3 adults per 400 mL volume). Two additional replicates of the cell recovery assay were carried out following incubation of the donor animals in menthol for 16 h at 4 °C. Relaxed adults were dissected to expose the endostyle and large blood vessel on the ventral side of the animal. Blood was drawn from this vessel using a zero dead space tuberculin syringe with a 25G needle, then transferred to a microcentrifuge tube on ice which had previously been coated with bovine serum albumin (BSA) by washing with 0.5 mL of 10% (v./v.) BSA (Sigma 9048-46−8) in Dulbecco’s phosphate-buffered saline (DPBS, Gibco 14200075).

To resuspend blood cells in alternate buffers, pooled blood was split into several tubes coated with 10% (v./v.) BSA in DPBS. Cells were pelleted by spinning at 800 g for 10 min at 4 °C in a swinging bucket centrifuge (Eppendorf Centrifuge 5810R). The supernatant was gently removed, and cells were resuspended in the appropriate buffer.

Cell recovery assay

We define cell recovery fraction as the number of live cells present after washing, normalized by the initial number of cells present before washing. Pooled blood cells from 9 adults were counted and their viability was assessed using an automatic cell counter (Bio-Rad TC20) with Trypan Blue staining. The cell suspension was evenly split and resuspended in one of 5 buffers: PBS, Ca²⁺- and Mg²⁺-free artificial seawater (CMF-ASW; 0.5 M NaCl, 9 mM KCl, 5 mM HEPES, pH 7.4), or PBS-M with varying mannitol concentrations (0.7 M, 0.6 M, or 0.4 M D-mannitol [Sigma Aldrich M4125] in PBS). Cell concentration and viability were measured with Trypan Blue at 1, 30, and 60 min post-resuspension.

In Fig. 1a, the cell recovery fraction was calculated by dividing the live (Trypan Blue negative) cell count at each time point by the live cell count before pelleting and resuspension. The data in Fig. 1a includes a total of 3 replicates of CMF-ASW and 4 replicates of all other buffer conditions. Samples were collected across two separate days (1–2 replicates per condition per day).

Fig. 1.

PBS-M buffer improves cell recovery and does not inhibit reverse transcription. a The cell recovery fraction, i.e. the number of live cells post-wash normalized by the number of cells present before washing, in CMF-ASW, PBS, or PBS-M buffers. Data is averaged over 3–4 technical replicates per condition across two experimental days. Shaded regions indicate 95% confidence intervals as calculated from standard error of the mean. b Capillary electrophoresis traces of amplified cDNA from L1210 cells prepared in CMF-ASW, PBS, or PBS-M buffers. Peaks at 35 bp and 10380 bp are size markers

Reverse transcription efficiency assay

L1210 cells (ATCC CCL-219) were grown in suspension in DMEM (Thermo Fisher 10566016) supplemented with 10% FBS and 1X Penicillin-Streptomycin (Thermo Fisher 15140122). The cells in culture media were pelleted in a swinging bucket centrifuge (Eppendorf Centrifuge 5810R) at 400 g for 4 min at 4 °C, then washed 3 times with ice-cold DPBS. The cells were pelleted and resuspended at equal cell concentrations in one of 5 buffers: PBS, CMF-ASW, or PBS-M with varying mannitol concentrations (0.7 M, 0.6 M, or 0.4 M D-mannitol in PBS). Cells were diluted in the same buffer to a concentration of 1,000 cells/µL.

Each cell suspension was mixed 1:1 with a 2X reverse transcription (RT) mix, such that the final solution consisted of 500 cells/µL, 1X Maxima RT buffer (Thermo Scientific EP0751), 0.5 mM of each dNTP (Thermo Scientific R0192), 2.5 µM poly-T primer (5′-CTCACTATAGGGTGTCGGGTGCAGNNNNNNTTTTTTTTTTTTTTTTTTTV-3′), 25 µM template switching oligo (5′-AAGCAGTGGTATCAACGCAGAGTACATrGrGrG-3′), 1 U/µL RiboLock RNase inhibitor (Thermo Scientific EO0381), 0.3% IGEPAL NP-40 detergent (Sigma I8896-50ML), and 10 U/µL Maxima H minus RT enzyme (Thermo Scientific EP0751). The RT reaction was carried out at 50 °C for 60 min followed by inactivation at 85 °C for 5 min. RT samples were diluted by 1/4 in nuclease free water to reduce the salt concentration, then purified with a 1.2X volume of AMPure (Beckman Coulter, A63881) following the manufacturer’s instructions.

cDNA was amplified by PCR using KAPA HiFi HotStart ReadyMix (Roche Sequencing, KK2602) and 0.5 µM each of forward (5′-CTCACTATAGGGTGTCGGGTGCAG-3′) and reverse (5′-AAGCAGTGGTATCAACGCAGAGTACAT-3′) amplification primers. Amplified cDNA was purified once more using a 1.2X volume of AMPure. Finally, cDNA was diluted to approximately 3 ng/µL and run on the 2100 Agilent Bioanalyzer using the High Sensitivity DNA Kit (Agilent 5067 − 4626). The measured fluorescence units in Fig. 1b have been corrected for the dilution factor.

Single-cell barcoding, library preparation, and sequencing

Pooled blood cells from 5 adults were collected as described above, split into two aliquots, and then resuspended in either CMF-ASW or PBS-M (0.7 M D-mannitol in PBS) at a concentration of 1,000–2,000 cells/µL. Cell viability was measured prior to dilution and encapsulation using a LUNA-FL Automated Fluorescence Cell Counter (Logos Biosystems L20001) and found to be 77% in CMF-ASW and 86% in PBS-M. This measurement was prior to dilution with nuclease-free water or additional PBS-M, which reduced the osmolarity of the dCMF-ASW sample but not the PBS-M sample. Therefore, a further drop in viability may occur in dCMF-ASW prior to encapsulation.

Cell encapsulation, RT, cDNA amplification, and library preparation was done using the Chromium Next GEM Single Cell 3′ v3.1 Kits (10X Genomics 1000123, 1000127, and 1000190), following the manufacturer’s guidelines except for one step. When preparing the RT mix prior to cell encapsulation, the manufacturer protocol instructs to add nuclease-free water, then to add the cell suspension. For the cells in CMF-ASW, we followed these directions (diluted CMF-ASW, or dCMF-ASW); for the PBS-M sample, we added PBS-M instead of adding nuclease-free water.

Libraries were sequenced with Illumina NovaSeq 6000 SP, and a cell by gene count matrix was generated from FASTQ files using 10X Genomics Cell Ranger 6.1.2. For a reference genome, the HT2019 assembly with KY21 gene models from the Ghost Database [19, 20] was used in combination with mitochondrial genes from the Ensembl KH genome assembly [21] (gene IDs listed in Table S3). The available GFF3 file of the KY21 gene models was reformatted as a GTF file using GffRead [22] in order to be compatible with 10X Cell Ranger.

Cell filtering and data quality measures

Viable cells were filtered from background based on (i) the number of unique mRNA molecules (UMI-filtered mapped reads, or UMIs) detected, (ii) mitochondrial fraction, and (iii) scoring low for likelihood of being a doublet. For (i), a UMI count above a threshold was required. These thresholds were set based on a clear separation from background as seen by visual inspection of the plots in Fig. 2d. The threshold values used were 2000 UMIs for dCMF-ASW and 800 for PBS-M, except in Figure S1: in Figure S1a–d, the threshold was set to 800 UMIs for both conditions, and in Figure S1e–f, the list of barcodes determined to be cells by Cell Ranger was used. For (ii), a mitochondrial fraction () below 10% was required for both samples. For (iii), likely doublets were identified using the SCRUBLET package [23]. All cell barcodes are included in Fig. 2d, and only cell barcodes passing the UMI threshold are plotted in Fig. 2e.

Fig. 2.

PBS-M buffer improves scRNA-seq data quality and recovers missing cell states in C. robusta blood. a Experimental set-up for testing the effect of cell suspension buffer on scRNA-seq data quality. Blood from five animals was used. b, c Preparing cells by standard dilution of high-salt buffer (dCMF-ASW) or PBS-M renders differences in: b the number of single cell barcodes passing the scRNA-seq data filters, and c the fraction of mapped reads from non-cell barcode. Cells are filtered by the number of transcripts detected and mitochondrial fraction. d The number of mRNA molecules detected per bin, with cell barcodes binned based on their transcript count. The dashed line shows the median number of transcripts in non-cell barcodes (571 transcripts/barcode in dCMF-ASW and 194 transcripts/barcode in PBS-M). These values quantify the background resulting from free-floating mRNA, which is reduced in PBS-M. e Histogram of the mitochondrial fraction per cell, an indicator of cell lysis. f Joint UMAP embedding of cells prepared in dCMF-ASW or PBS-M buffers, with cells colored by sample. Left: dCMF-ASW cells are marked in blue with the rest of the cells in grey. Right: PBS-M cells are marked in pink with the rest of the cells in grey. Cell states strongly depleted in the dCMF-ASW sample are circled in black. g Joint UMAP embedding with cells colored by the log-ratio of the relative local sample density. The coloring represents the amount of local enrichment of cells from PBS-M (pink) or dCMF-ASW (blue) samples. h Bar plot showing the log fold change in the proportion of cells contributing to each cell state in the PBS-M vs. dCMF-ASW samples. Error bars are calculated based on the standard error of a proportion, and 95% confidence intervals are shown. One cell state, ND-6, had only 3 cells in PBS-M and 0 in dCMF-ASW and was excluded from this analysis

The median UMIs/barcode in empty droplets for each sample, as marked with a dashed line in Fig. 2d, was calculated across empty droplets. Empty droplets were defined as barcodes which did not pass the UMI threshold and which had ≥ 10 UMIs/barcode.

Data reanalyzed in Figure S1i was processed in the same way, except UMI/barcode thresholds were adjusted manually for each library. C. robusta collection 2 data from Scully et al. [24] used a 2000 UMI threshold for both libraries. Botryllus schlosseri data from Lebel et al. [25] used thresholds of 500 UMI (for the A1 library) and 700 UMI (for the B2 library).

Dimensionality reduction

Preprocessing and dimensionality reduction were done using Scanpy [26] (version 1.8.1) using default parameters except where noted below. Genes expressed in fewer than 5 cells were excluded. Counts were normalized (pp.normalize_total) and log transformed (pp.log1p), and highly variable genes were identified (pp.highly_variable_genes) as described in [27]. Principle component (PC) analysis was performed (tl.pca, use_highly_variable = True) on z-scored data (pp.scale). A k nearest neighbor (k-NN) graph (pp.neighbors, n_neighbors = 10) was generated based on Euclidean distance in PC space using the first 50 PCs. This k-NN graph was used for UMAP visualization (tl.umap) [28].

Log-Ratio of sample density

The log-ratio of sample density is a measure of enrichment or depletion of one sample relative to another in gene expression space. We adapted the measure defined in [29] as follows: the 200 nearest neighbors of each cell were identified in PC space using the first 50 PCs (Euclidean distance). A cell’s “neighborhood” includes itself and its nearest neighbors. Let and be the number of cells within the neighborhood of cell from the dCMF-ASW and the PBS-M samples, respectively. We used =200 and drop the argument below. For each cell , a numerical representation of the log-scaled relative sample density, shown in Fig. 2g, was calculated as

These values were plotted in a UMAP generated by Scanpy (pl.umap, sort_order = False).

Cell state annotations

Cell state annotations based on cell morphology for cell clusters in the PBS-M sample were determined in [24]. Cell state labels for the PBS-M sample were copied from the annotated counts matrix available on NCBI GEO, Series GSE296253 (GSE296253_combined_processed_data.h5ad). Labels were also copied for a set of cell barcodes previously determined to be likely doublets of two cell states (hyaline amoebocytes 3 [HA-3] and signet ring cells [SRC], labeled “HA-3/SRC doublets”) [24].

The remaining unlabeled cells were annotated based on k-nearest neighbors classification. Specifically, for each unlabeled cell, the k = 10 nearest labeled cell neighbors were determined based on Euclidean distance in PC space using the top 50 PCs (implemented with sklearn.neighbors.KNeighborsClassifier, n_neighbors = 10). Unlabeled cells were labeled with the cell state name shared by the most neighbors (sklearn.neighbors.KNeighborsClassifier.predict).

The doublet HA-3/SRC cell state is shown in Fig. S1j, but excluded from log fold change calculations in Fig. 2h.

Differential gene expression analysis

Genes that are broadly differentially expressed between dCMF-ASW and PBS-M are listed in Table S2. Specifically, genes are considered differentially expressed within a cell state based on a Wilcoxon rank-sum test (scanpy.tl.rank_genes_groups, method=‘wilcoxon’), requiring log fold change absolute value > 1 and FDR < 0.05. Genes which are differentially expressed in at least one third of all cell states (at least 11 out of 33) are shown in Table S2.

For GO analysis, human homolog(s) of C. robusta genes in Table S2 were determined based on [24]. GO enrichment analysis was performed on this list of homologs using the Gene Ontology knowledgebase [30, 31].

Results

This study was motivated by an interest in carrying out scRNA-seq on C. robusta blood. As a standard control prior to scRNA-seq, C. robusta blood cells were harvested and the fraction of live cells recovered was evaluated by Trypan blue staining after resuspension in PBS (~330 mOsm/L) [32], a low-salt buffer routinely used with droplet-based scRNA-seq. We observed an 81% loss of live cells in PBS, as compared to a 40% loss after resuspension in buffered Ca²⁺- and Mg²⁺-free artificial seawater (CMF-ASW, ~1000 mOsm/L). This loss mostly occurred within 1 min of resuspension (Fig. 1a).

We evaluated buffers for compatibility with scRNA-seq and for the ability to maintain cell viability. One such buffer consisted of 0.7 M D-mannitol dissolved in 1X PBS (PBS-M). D-mannitol is used as an osmotic agent in medical settings and several electroporation protocols [33–36]. At 0.7 M concentration, it contributes 700 mOsm/L, thus PBS-M has comparable osmolarity to sea water (~1000 mOsm/L total) with an ionic strength similar to PBS. When we resuspended C. robusta blood cells in PBS-M, there was no significant increase in cell loss as compared to resuspension in CMF-ASW (Fig. 1a). This was consistent at reduced D-mannitol concentrations of 0.4 and 0.6 M.

PBS-M is compatible with the RT reaction required for scRNA-seq. To show this, we performed RT and subsequent PCR amplification on mouse leukemia L1210 cell lysates in either PBS, CMF-ASW, or PBS-M buffers with varying mannitol concentrations (0.4, 0.6, and 0.7 M). Capillary electrophoresis of the amplified RT products showed that reaction efficiency was poor in CMF-ASW, as expected due to high salinity [11]. However, the quantity and size distribution of the resulting cDNA was similar between PBS and all tested concentrations of PBS-M (Fig. 1b).

We next tested PBS-M for use in scRNA-seq. Blood cells from five adult animals were collected, pooled, and split equally into two samples that were prepared in tandem and jointly analyzed on a 10X Genomics Chromium scRNA-seq system (Fig. 2a). The first sample was prepared by standard dilution as in [12]: cells were pelleted and resuspended in CMF-ASW, then diluted >4.3-fold with water prior to droplet encapsulation. For the second sample, cells were instead resuspended in PBS-M, then diluted with PBS-M instead of water prior to droplet encapsulation. A target of ~6,000 cells was collected for each condition. scRNA-seq libraries were sequenced (see Table S1 for sequencing quality metrics) and analyzed to evaluate evidence of cell death and cell state changes resulting from sample preparation.

Several metrics indicate that PBS-M led to fewer dying cells than diluted CMF-ASW (dCMF-ASW). (i) Cell counts: we detected 6,108 viable cells in the PBS-M sample as compared to 4,443 for dCMF-ASW, even though we prepared the same number of cells for each condition (Fig. 2b). (ii) Ambient mRNA: dying cells release their mRNA, leading to increased ambient mRNA levels that are detectable in empty droplets. The PBS-M sample had considerably lower levels of ambient mRNA as compared to dCMF-ASW. This is seen from both the decreased fraction of reads from empty droplets (Fig. 2c, 25.8% for PBS-M vs. 66.2% for dCMF-ASW), and the decreased median number of transcript counts in empty droplets (Fig. 2d, 194 transcripts/droplet for PBS-M vs. 571 for dCMF-ASW). (iii) Mitochondrial transcripts: cells dying during scRNA-seq sample preparation lose nuclear-derived mRNA but retain mitochondrial-derived mRNA, leading to an increased fraction of mitochondrial transcript counts () [37]. The PBS-M sample has 0.5% cells with >20%, compared to 3.6% for dCMF-ASW (Fig. 2e).

Overall, these results demonstrate that basic data quality metrics are greatly improved by using PBS-M. This improvement is robust to strategies for filtering cell barcodes during scRNA-seq analysis (Fig. S1a–h). Normal ranges for measures of ambient mRNA and mitochondrial transcripts were also reproducible with a second PBS-M collection of C. robusta blood, as well as with blood from a distantly related tunicate species, Botryllus schlosseri (Fig. S1i).

In C. robusta blood, the cells prepared in PBS-M also reveal transcriptional states which are depleted or entirely absent when using dCMF-ASW. To visualize this, we generated a joint UMAP embedding and colored cells by preparation method (Fig. 2f) and by a measure of local sample enrichment or depletion (the log-ratio of the number of nearest neighbors that belong to the PBS-M sample vs. the dCMF-ASW sample) (Fig. 2g). These plots reveal cell clusters that were under-represented in the dCMF-ASW sample. To determine these clusters’ identities, cell state annotations documented in [24] were transferred to this dataset (Fig. S1j). We identified the dCMF-ASW-depleted clusters to be large granules hemocytes and morula cells (LGH/MC), progenitor cells (candidate multipotent progenitors, cMPP, and lineage-restricted progenitors, cLRP), and a cell state whose morphology is not determined (ND-2) (Fig. 2h). These cell states’ depletion in dCMF-ASW may be caused by increased sensitivity to osmotic shock.

To interrogate transcriptional cell state changes, we identified genes which were differentially expressed between dCMF-ASW and PBS-M in each cell state. Only 24 genes were consistently differentially expressed (in at least one third of cell states, see Table S2), and there is no enrichment for Gene Ontology terms [30, 31] associated with osmotic stress. These transcriptional differences might be due to technical batch effects rather than transcriptional changes induced by buffers.

Discussion

scRNA-seq has been revolutionary in enabling the discovery of new cell types, developmental and regenerative cell states, and complex immune responses. Even in well-studied mammalian systems, scRNA-seq has revealed unappreciated diversity of cell phenotypes [29, 38]. The number of cell states in mammals is dwarfed by the sheer number present across invertebrates. Each invertebrate species hosts a diversified repertoire of cell types, and few invertebrate animal tissues are defined at a cellular level. For instance, invertebrate immune cells are still largely described by cell morphology, with few or no molecular markers available to discriminate between cell types [39, 40]. Thus, scRNA-seq can short-cut years of work to identify and characterize cell types, enabling cell type comparisons across species.

Despite the promise of the technique, scRNA-seq was developed and optimized largely for vertebrates and is not compatible with high salinity buffers that mimic the body fluids of most marine animals [11]. We have shown that using the low-salt, high-osmolarity PBS-M buffer for scRNA-seq of C. robusta blood cells greatly improves data quality by significantly reducing cell death. We saw two types of improvement: First, the baseline data quality obtained was considerably higher. Second, PBS-M let us observe certain cell states which were under-sampled or absent when using the standard dilution approach to scRNA-seq of marine cells. As demonstrated in [24], this data collected using PBS-M has enabled rapid cell type characterization and quantitative cross-species comparisons. Other osmotic agents, such as sorbitol and glycine, could serve a similar purpose.

With these advantages, some caveats are due. First, scRNA-seq may lose cell states independent of the buffers used. This is particularly true for large or fragile cells in tissues that are difficult to dissociate [41]. In our case, 40% of cells were lost during wash steps even when using high-osmolarity buffers, suggesting that some cells are sensitive to pelletting and resuspension. Second, we cannot rule out the possibility that PBS-M may itself induce transcriptional changes in some circumstances. In the tunicate B. schlosseri, for example, D-mannitol and other sugars can inhibit blood cell phagocytic activity, which could over time lead to transcriptional changes in these cells [42]. Nevertheless, in this protocol cells spend only ~30 min in PBS-M on ice, and we observe few genes showing consistent expression differences across cell states between dCMF-ASW and PBS-M.

We expect that PBS-M should be suitable for preparing cells for scRNA-seq in other marine invertebrate species, though validation in other species is necessary. Here, we show that data quality remains high in a second tunicate species, B. schlosseri. Certain species, such as some annelids, may require different D-mannitol concentrations to match their body osmolarity [10]. To this end, we recommend direct assessment of cell viability in PBS-M prior to performing scRNA-seq. Beyond this test, the use of PBS-M is simple. It could be broadly useful for generating high-quality scRNA-seq data from non-model marine organisms, better enabling researchers to explore the diversity of marine cell types.

Abbreviations

BLC

Blebbing-like cell

GA

Granular amoebocyte

HA

Hyaline amoebocyte

ICC

Irregular compartment cell

LGH

Large granules hemocyte

MC

Morula cell

ND

Morphology not determined

RA

Refractile amoebocyte

RC

Round cell

RSC

Round spreading cell

SGH

Small granules hemocyte

SRC

Signet ring cell

URG

Unilocular refractile granulocyte

cLRP

Candidate lineage-restricted progenitor

cMPP

Candidate multipotent progenitor

Authors’ contributions

TDS carried out the experiments and data analysis. AMK supervised the work. TDS and AMK wrote the manuscript. Both authors read and approved the final manuscript.

Funding

TDS was supported by the Harvard Medical School Systems Biology Lynch Fellows Program and AMK is a Mallinckrodt Foundation Scholar. This work was supported by NIH grant R01HD096755.

Data availability

The datasets generated and analyzed in the current study are available in the Gene Expression Omnibus (GEO) repository, accession number GSE292926. Code for scRNA-seq data analysis is available on Github at https://github.com/AllonKleinLab/paper-data/tree/master/Scully_mannitol_2025.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

AMK is a co-founder of Somite Therapeutics, Ltd.