Single-cell transcriptomics identifies CD44 as a new marker and regulator of haematopoietic stem cells development

The endothelial to haematopoietic transition (EHT) is the process whereby haemogenic endothelium differentiates into haematopoietic stem and progenitor cells (HSPCs). The intermediary steps of this process are unclear, in particular the identity of endothelial cells that give rise to HSPCs is unknown. Using single-cell transcriptome analysis and antibody screening we identified CD44 as a new marker of EHT enabling us to isolate robustly the different stages of EHT in the aorta gonad mesonephros (AGM) region. This allowed us to provide a very detailed phenotypical and transcriptional profile for haemogenic endothelial cells, characterising them with high expression of genes related to Notch signalling, TGFbeta/BMP antagonists (Smad6, Smad7 and Bmper) and a downregulation of genes related to glycolysis and the TCA cycle. Moreover, we demonstrated that by inhibiting the interaction between CD44 and its ligand hyaluronan we could block EHT, identifying a new regulator of HSPC development.


Introduction
Understanding the developmental origin of haematopoietic stem and progenitor cells (HSPCs) is of critical importance to efforts to produce blood and blood products in vitro for medical applications. Haematopoietic stem and progenitor cells (HSPCs) originate from endothelial cells in the aorta gonad mesonephros (AGM) of midgestation embryos 1,2 . This process known as the endothelial to haematopoietic transition (EHT) requires drastic morphological changes that have been directly visualised through time-lapse imaging studies both in vitro and in vivo [3][4][5][6] . EHT is a highly conserved process that has been studied across vertebrate models from xenopus and zebrafish to mice 7 . Importantly, the human definitive blood system also has an endothelial origin 8 .
The best tools so far to detect endothelial cells with haemogenic capabilities rely on using fluorescent reporters under the control of Runx1 9 or Gfi1 10 regulatory elements, two keys transcription factors in the process of EHT. Cells expressing these transcription factors already co-express blood and endothelial genes. However, we still do not know the nature of endothelial cells which will acquire the expression of The early haematopoietic hierarchy has been described as a three-step process based on phenotypic characteristics. Specifically, Pro-HSC, Pre-HSC type I and type II populations have been defined based on their expression of the cell surface markers CD41, CD43 and CD45, as well as the time taken mature into definitive HSCs in OP9 co-culture 13 . Recently, an in depth transcriptional investigation was performed on the 4 type I and type II Pre-HSC populations at day 11 of mouse embryonic development, however, the earlier stages of EHT remain largely uncharacterised 14 . Indeed, gaining a solid understanding of HE and the initial steps that endothelial cells must take to become HSPCs has proved difficult in the absence of a robust marker.
Through antibody screening and single-cell RNA sequencing (sc-RNA-seq) we identified CD44 as a novel marker of EHT, enabling the isolation of key cellular stages of blood cell formation in the embryonic vasculature. CD44 is a cell surface receptor principally involved in the binding of the extracellular matrix molecule hyaluronan 15 . Its cell surface expression has been used to identify cancer stem cell populations and has been strongly linked to the metastatic potential of many cancers [16][17][18][19][20] . While previous research has revealed the importance of CD44 in HSPCs migration to the bone marrow, the role of the receptor in early embryonic haematopoiesis has not been characterised 21 . Using CD44 expression in conjunction with VE-Cadherin (VE-Cad) and Kit we could clearly differentiate between vascular endothelium, HE, Pre-HSPC-I and Pre-HSPC-II more accurately than using the combination of VE-Cad, CD41, CD43 and CD45 markers. This has allowed us to perform extensive transcriptional profiling making it possible to characterise the very earliest changes in haematopoietic differentiation from endothelial cells. Moreover, by disrupting the interaction of CD44 and its ligand, we could inhibit EHT, demonstrating an unexpected role for CD44 in the emergence of HSPCs.

Antibody screening and single-cell RNAseq identifies CD44 as a potential marker of early haematopoietic fate
It is well established that HSPCs have an endothelial origin within the embryo [4][5][6]22 . In order to better characterise the transition between endothelial and haematopoietic identity, we performed both an in vitro antibody screen and an in vivo single-cell RNAseq experiment to identify new markers allowing the identification of subpopulations within VE-Cad + cells ( Supplementary Fig. S1). Antibodies against 176 cell surface markers 23 were tested against the VE-Cad + population generated from our in vitro ESC differentiation system into blood cells which recapitulates faithfully the early blood development process 24 . Forty-two of these markers were expressed on VE-Cad + cells (Supplementary Table S1). We looked for bimodal expression to separate distinct endothelial populations and identified a short-list of sixteen candidates to test in vivo including CD41 and CD117 (also known as Kit, a marker of HSPCs) already known to split VE-Cad + cells 13 . Eight of these markers (CD44, CD51, CD55, CD61, CD93, Icam1, Madcam1 and Sca1) were found to split the VE-Cad + endothelial population of the AGM in two (Fig. 1a). In parallel, VE-Cad + cells were isolated from the AGM region of E10.5 embryos and their transcriptional profiles analysed sc-RNA-seq ( Fig. 1b-d). Clustering analysis identified a population with both haematopoietic and endothelial gene expression, distinct from the other endothelial population (Fig. 1d). Bioinformatics analysis showed that Cd44 is one of the best marker genes for this population of transitioning cells co-expressing endothelial and haematopoietic genes (Fig. 1c). The expression of Cd44 was also positively correlated with other known haematopoietic markers such as Runx1, Gfi1 and Adgrg1 (Gpr56) (Fig. 1d).
Given the association of Cd44 with endothelial cells undergoing EHT at both the protein and mRNA level we decided to further investigate its role in embryonic haematopoiesis.

CD44 marks transitioning cells with differing morphology
To validate our screening results and investigate the identity of CD44 + cells, we performed immunofluorescence and more detailed flow cytometry analysis on the 6 AGM region of mouse embryos at embryonic day 9.5 (E9.5) and 10.5 (E10.5) (Fig. 2).
Immunofluorescence of cross-sections of mouse AGMs revealed that CD44 marked cells that were part of the vascular wall and cells that were incorporated in haematopoietic clusters (Fig 2a). Flow cytometry revealed that CD44 expression significantly increased in the VE-cad + endothelium of the AGM between E9.5 and E10.5 when cells are undergoing EHT (Fig. 2b-c). Furthermore, by staining with an antibody against Kit (a marker of intra-aortic haematopoietic clusters), we found that the majority of cells with lower levels of CD44 expressed little or no Kit (Fig 2d).
By grouping the cells based on their cell surface expression of CD44 and Kit, we found these populations to be significantly different in terms of cell size, suggestive of cells undergoing a morphological transition (Fig. 2e). Altogether, our results indicate that CD44 marks a subset of endothelial cells and cells in the haematopoietic clusters of the AGM during the key window of HSPCs development in the mouse embryo.

Single-cell q-RT-PCR analysis identifies three homogeneous CD44 + populations with an increasing haematopoietic profile
Using the Biomark HD single-cell qPCR platform, we analysed the expression of 95 genes associated with both endothelial and haematopoietic cell types 24 . We performed extensive transcriptional profiling on the CD44 Neg , CD44 Low Kit Neg , CD44 Low Kit Pos and CD44 High populations identified (Fig. 2d) between E9.5 and E11.5 (342 cells in total). We found that the CD44 Low Kit Pos and CD44 High populations expressed numerous haematopoietic genes (Fig. 3a). The CD44 Low Kit Pos cells also expressed many endothelial genes as in our sc-RNA-seq analysis (Fig. 1d), however the CD44 High cells appeared to be more advanced in the EHT process and lacked endothelial gene expression (Fig. 3a). Of note, the CD44 Low Kit Pos population expressed specifically Gfi1 and Itgb3 (coding for CD61) as well as the heptad of transcription factors (Gata2, Runx1, Lyl1, Erg, Fli1, Lmo2 and Tal1) whose simultaneous co-expression is responsible for the dual endothelial-haematopoietic identity of Pre-HSPCs ( Supplementary Fig. S2) 24 .
Conversely, the CD44 Neg and CD44 Low Kit Neg populations both showed specific endothelial gene signatures and lacked haematopoietic gene expression (Fig. 3a).
Despite their different CD44 expression patterns, these cell populations clustered together ( Supplementary Fig. S2). We repeated this experiment using a new selection 7 of 96 genes based on our sc-RNA-seq experiment (Supplementary Table S2). With our new gene list, we were able to confirm the dual endothelial-haematopoietic and haematopoietic identities of the CD44 Low Kit Pos and CD44 High populations, respectively. Surprisingly, the CD44 Neg and CD44 Low Kit Neg populations formed two distinct clusters (Fig. 3b). In addition to Cd44, we found the genes Adgrg6, Emb, Fbn1, Pde3a, Plcg2, Serpinf1, Smad6, Smad7, Sox6 and Stxbp2 to be up-regulated in the CD44 Low Kit Neg cells compared to CD44 Neg . In contrast, Bmp4, Kit, Hmmr and Pde2a were more expressed in CD44 Neg endothelial cells (Fig. 3c).
Although the four groups defined by VE-Cad, CD44 and Kit were confirmed to be distinct transcriptionally, our clustering analysis found a fifth population (SC_3) composed of cells from both CD44 Low Kit Neg and CD44 Low Kit Pos groups (Fig. 3a). Its transcriptional profile was found to be intermediary between SC_2 (CD44 Low Kit Neg ) and SC_4 (CD44 Low Kit Pos ) e.g. it expressed Adgrg1, Runx1, Itgb3 and Spi1 like SC_4 but was still expressing Adgrg6 and Pcdh12 like SC_2 (Fig. 3c).
This suggests a developmental link between the CD44 Low populations where CD44 Low Kit Neg cells would be the direct precursors of the CD44 Low Kit Pos population which would then go on to generate CD44 High cells. Interestingly, it is within this transitional SC_3 population that we saw the up regulation of Runx1, Spi1 and Gfi1, which are three of the four transcription factors used to reprogram adult mouse endothelial cells into HSCs 12 .
VE-Cad, CD44 and Kit could segregate the earliest stages of haematopoietic development more accurately than the combination of VE-cad, CD41, CD43 and

CD45 markers
To place our results in context of known AGM subpopulations, we performed transcriptional analysis of the Pro-HSC, Pre-HSC type I, and Pre-HSC type II populations defined according to the combination of VE-cad, CD41, CD43 and CD45 markers (Fig. 4a) 13 . Following hierarchical clustering, we found three clusters: the first mostly composed of Pro-HSCs, a second being a mix of Pro-HSCs and Pre-HSCs type I and a third one composed only of Pre-HSCs type II ( Supplementary Fig. S3).
We then performed a clustering analysis in conjunction with the populations defined by CD44 (Fig. 4b). This revealed that the SC_5 (CD44 High ) population closely associated with the Pre-HSC type II and the SC_4 (CD44 Low Kit Pos ) population with Pre-HSC type I and part of the Pro-HSCs. SC_2 (CD44 Low Kit Neg ) clustered closely 8 with the remaining Pro-HSCs. Finally, only five cells with Pre-HSCs type I and Pro-HSCs phenotype clustered with the SC_1 (CD44 Neg ). Ninety-seven per cent of Pro-HSCs, Pre-HSCs type I, and Pre-HSCs type II were CD44 positive ( Supplementary   Fig. S3 and Fig. 4c).
Overall, we have demonstrated that the phenotypes based on CD44 expression could allow us to isolate all key populations in the process of HSCs formation more accurately than the phenotypes previously described.
Bulk RNA-seq analysis further distinguishes CD44 Neg and CD44 Low Kit Neg endothelial populations and identifies early changes in the differentiation process To further compare the two endothelial populations found in the AGM, we performed RNAseq on 25-cell bulk samples from these populations across three different time points (E9.5, E10 and E11). We analysed as well the more advanced stages in EHT: CD44 Low Kit Pos at E9.5 and E10 and CD44 High at E11. The bulk RNAseq approach allowed us to detect low abundant genes (such as genes encoding transcription factors) more efficiently than sc-RNA-seq and also to measure smaller changes in gene expression between populations.
The samples clustered according to their marker expression, despite the difference in developmental time, confirming our previous experiments (Fig. 5a). We identified several haematopoietic genes switched on in the CD44 Low Kit Neg population including, Ctsc, Nfe2, Runx1 and Ifitm1, suggesting that these cells are subjected to the EHT process (Fig. 5b). The expression pattern of endothelial genes fits with our previous observations -more highly expressed in CD44 Neg and CD44 Low Kit Neg , moderately expressed in CD44 Low Kit Pos and absent in CD44 High (Fig 5b). Moreover, we used this dataset to check the expression of genes corresponding to proteins we found in our antibody screen (Fig. 1a). Cd93 and Madcam1 showed an endothelial expression pattern and could be used to separate endothelial from blood cells in the AGM. In contrast, Itgb3 marks specifically the CD44 Low Kit Pos population while Ly6e marks both CD44 Low Kit Pos and CD44 High populations as shown previously in Supplementary Fig. S2 and Fig. 3.
Interestingly, this transcriptome analysis showed strong differences between the two endothelial populations of the AGM. We found 1605 genes differentially expressed between these two populations (p-value < 0.01, Wald test). Among them, several genes from the Notch pathway (Hey2, Jag1, Dll4, Hey1 and Notch1) were 9 significantly more expressed in the CD44 Low Kit Neg population compared to CD44 Neg .
Similarly, antagonists of the TGFbeta/BMP pathway, including Smad6, Smad7 and Bmper, were up-regulated in CD44 Low Kit Neg cells compared to CD44 Neg ones. In contrast, target genes of the Wnt pathway (Lef1, Ccnd1 and Myc) were more highly expressed in CD44 Neg compared to CD44 Low Kit Neg . From this analysis we could also identify specific markers for each of the endothelial populations (Fig. 5c). Remarkably, a large number of the differentially expressed genes between the two endothelial populations belonged to metabolic processes (395 out of 1605 genes; 1.32-fold enrichment, p-value <0.05, Fisher's exact test). We therefore identified specific metabolic pathways distinguishing the two populations ( Fig. 6a and Supplementary Table S3). Notably, the CD44 Low Kit Neg population showed a pronounced down regulation of genes coding for enzymes involved in glycolysis, TCA cycle and respiration, suggesting reduced ATP generation. Furthermore, amino acid and nucleotide biosynthesis genes were also down regulated. Altogether it suggests that the CD44 Low Kit Neg is a non-proliferative, metabolically rather quiescent state which is in line with the smaller size of this population compared to the CD44 Neg 10 ( Fig. 2e). Given that endothelial cells are known to obtain most of their energy from glycolysis 25 , this change in metabolic status supports a loss of endothelial identity.
These cells also show a marked increase in the expression of pathways leading to lipids with regulatory function: glycerolipids, glycerophospholipids, phosphatidylinositol and sphingolipids (Fig. 6a).
Interestingly, we observe that several genes involved in autophagy, a process known to be regulated by phosphatidylinositol and sphingolipids 26-29 , are also highly upregulated in the CD44 Low Kit Neg population (Fig. 6b). Concordantly, two key processes accompanying autophagy, ubiquitylation and proteolysis, are also upregulated. As autophagy has been shown to play a key role in embryonic development and haematopoiesis [30][31][32] , this provides further support to the CD44 Low Kit Neg cells being in transit from endothelial to haematopoietic cells.

Runx1 is not required for the formation of CD44 Low Kit Neg endothelial cells
CD44 has allowed us for the first time to clearly define the key VE-Cad + populations in the AGM. The transcription factor RUNX1 is a key driver of HSPC development and is known to down-regulate endothelial identity through its target genes GFI1 and GFI1B 10,33 . Next, we decided to evaluate the impact of Runx1 loss of function on the different CD44 + cells. Using a Runx1 knockout mouse model 34 , we stained for VE-Cad, CD44 and Kit expression and performed transcriptional profiling on the sorted cells (Fig. 7). We found that in the absence of Runx1 there is a loss of CD44 High and CD44 Low Kit Pos cells and we observed a concomitant increase in the frequency of the CD44 Low Kit Neg population ( Fig. 7a and 7b). Interestingly, we found no obvious transcriptional differences between the CD44 Low Kit Neg populations derived from Runx1 +/+ versus Runx1 -/embryos indicating that Runx1 is not necessary for the formation of these endothelial cells but for the promotion of the transition into CD44 Low Kit High cells ( Supplementary Fig. S5).

All CD44 + populations display haematopoietic potential
To understand the haematopoietic potential of the different populations defined by VE-Cad, CD44 and Kit expression, we performed ex vivo assays using an OP9 co-culture system. No colonies were formed at the single-cell level from either the CD44 Low Kit Neg or the CD44 Neg populations. However, by plating cells at a density 11 of 300 cells per well, we could observe round cell colony formation from the CD44 Low Kit Neg population but not from the CD44 Neg (Fig. 8a).
In contrast, using single-cell sorting, we found that the CD44 Low Kit Pos population was the most potent with an average of 43% of single cells forming round cell colonies after three days of growth. Similarly, the CD44 High population produced haematopoietic colonies but with a lesser frequency, on average 11% of single cells showed the ability to form haematopoietic colony on OP9 (Fig. 8b-c). To uncover the differentiation potential of the cells generated by CD44 Low Kit Pos and CD44 High , we performed CFU assays following the OP9 co-culture. Both populations readily generated both erythroid and myeloid colonies with the CD44 Low Kit Pos population demonstrating a significantly higher capacity than the CD44 High population ( Fig. 8d-e).
However, the four-fold increase in the number of CD44 Low Kit Pos colonies on OP9 did not correspond to a four-fold increase in CFU colonies suggesting a higher replating efficiency of the CD44 High cells compared to CD44 Low Kit Pos (Fig 8f). We further tested the lymphoid potential of the CD44 High population by growing cells for 21 days on either OP9 or OP9-DL1 with lymphoid promoting cytokines. We demonstrated that this population could give rise to both B and T cells ex vivo (Fig. 8h).
Overall, we found that all populations expressing CD44 displayed haematopoietic differentiation capacity including CD44 Low Kit Neg reinforcing the differentiation link between them as suggested by the transcriptome analyses described previously.

Interrupting hyaluronan binding to the CD44 receptor inhibits the endothelial to haematopoietic transition ex vivo and in vitro
So far, we demonstrated that CD44 was a very useful marker to distinguish the different stages of EHT. Although the CD44 knock-out mice do not have a severe haematopoietic phenotype 35 , compensatory mechanisms through other Hyaluronan receptors (e.g. Hmmr 36 expressed by some CD44 Low Kit Pos and CD44 High cells in Fig.   3a) may be at play to diminish the consequences of CD44 loss of function. In order to explore the functional role of CD44 in EHT we employed a pharmacological approach. By treating CD44 High sorted cells with a CD44 blocking antibody, we found that round cell colony formation could be inhibited in a dose dependent manner ( Fig.   9a-b). The blocking antibody inhibited not only the number of colonies deriving from the ex vivo sorted cells but also the size of the colonies generated (Fig 9c). 12 To further investigate the role of CD44 in EHT we used an ESC in vitro differentiation system that mimics embryonic haematopoiesis. We performed a haemangioblast culture and analysed CD44 expression at day 1, 2 and 3. Endothelial In conclusion, these results for the first time demonstrate a regulatory role for hyaluronan and its receptor CD44 in the formation of HSPCs. 13

Discussion
Using antibody screening and sc-RNA-seq to dissect the endothelial populations in the AGM, we discovered that CD44 was a robust marker to distinguish all the main populations in the EHT process. Combining CD44 with Kit and VE-Cad allowed us to discriminate the different stages of EHT more accurately that the method based on VE-Cad, CD41, CD43 and CD45 cell surface markers 13 . In addition, we showed that CD44 had an unexpected regulatory function in the EHT process.
Our work has been instrumental in distinguishing the different types of endothelial cells in the AGM region. The CD44 Low Kit Neg has a gene expression signature strongly compatible with arterial identity (e.g. expression of Efbn2 and Sox17 and upregulation of Notch pathway target Hey2) while the CD44 Neg cells coexpressed genes related to the venous (e.g. Nr2f2 and Aplnr) and arterial cell fates (e.g. Efnb2 and Sox17) (Fig. 5d). This co-expression of venous and arterial genes supports previous work indicating that the dorsal aorta can contribute to the formation of the cardinal vein 37 .
Another interesting finding was the striking metabolic state difference between the two endothelial populations. The CD44 Neg cells appeared much more metabolically active than the CD44 Low Kit Neg suggesting that the latter population is becoming quiescent. It is surprising since the acquisition of a quiescent phenotype in endothelial cells occurs normally after birth following the completion of angiogenesis. Smad7 at the single-cell level suggesting that the quiescence phenotype we observed could also be linked to the co-expression of the two inhibitory Smads.
The acquisition of quiescence in this context could be the first indication of the divergence from an endothelial phenotype. The metabolic distinction of the CD44 Low Kit Neg population was coupled with an increase in genes involved in 14 autophagy. This is in line with the known role of autophagy in embryogenesis, haematopoiesis and stem cell maintenance 30-32,40,41 , such as protein and organelle turnover and protection from reactive oxygen species. Our data thus suggest that these cells mark the resetting of metabolic and regulatory states to initiate the EHT process.
We consequently propose that the CD44 Low Kit Neg population is the source of HSPCs in the AGM, hence the haemogenic endothelium ( . Given the role of CD44 in the metastatic process, it is possible that there is an overlap in its function for these transformations. Therefore, understanding the downstream targets of the CD44-hyaluronan interaction could also have implications also for cancer biology. 15

Timed mating and embryo dissection
For timed pregnancies, C57BL/6 wild-type mice or Runx1 +/mice were mated overnight. Embryos were collected in PBS supplemented with 10% FBS (PAA Laboratories

Single-cell RNA sequencing data analysis
The paired 2 x 101bp Illumina reads from the libraries were quantified using Salmon 43 with the setting -l IU to indicate library topology, and the optional flags --posBias and --gcBias to account for coverage and amplification biases present in sc-RNA-seq protocols. As an index the cDNA annotation of Ensembl release 85 for GRCm38.p4 was used, together with ERCC spike-in sequences. The TPM values were rescaled to not include ERCC expression and only consider endogenous gene expression.
Technical features of the data were compared with manual annotation of samples in C1 chambers through microscopy. Samples with less than 500,000 mapped reads and more than 30% mitochondrial content were discarded from analysis. This Novel markers for hematopoietic cells were identified using a likelihood ratio test, where the alternative model included a binary term for whether the cell was haematopoietic, and the null model just assumed a common mean for all the cells.
The P-values from the likelihood ratio test were corrected for multiple testing by the Bonferroni procedure. The top differentially expressed genes were investigated to find markers which could be used in follow-up experiments, and Cd44 was considered a good candidate (Fig. 1c).

In vitro ES cell differentiation system
The

Antibody screen, Flow cytometry and cell sorting
The antibody screen was performed using the Mouse Cell Surface Marker  imaged on a Leica SP5 confocal microscope.

Single-cell qPCR data analysis
Analysis of single-cell qPCR data was performed as previously described 24 .
Briefly, initial analysis was performed using the (Abcam). The number of colonies were assessed after 3 days in culture.

Haematopoietic colony forming assay
One hundred cells were initially sorted onto a confluent OP9 stromal layer as per OP9 co-culturing assay. After three days in culture cells were harvested with TrypLE express (Gibco) and colony-forming unit-culture (CFU-C) assays were initiated using Methocult complete medium (Stem Cell Technologies). Cells were grown in 35mm culture dishes and colonies quantified after 7 days.  M  o  r  g  a  n  O  a  t  l  e  y  ,  C  o  n  c  e  p  t  u  a  l  i  z  a  t  i  o  n  ,  F  o  r  m  a  l  a  n  a  l  y  s  i  s  ,  I  n  v  e  s  t  i  g  a  t  i  o  n  ,  V  i  s  u  a  l  i  z  a  t  i Figure S1 and File S1.        sc1_cell1_salmon_out  sc1_cell71_salmon_out  sc1_cell53_salmon_out  sc1_cell61_salmon_out  sc1_cell84_salmon_out  sc1_cell90_salmon_out  sc1_cell10_salmon_out  sc1_cell58_salmon_out  sc1_cell92_salmon_out  sc1_cell17_salmon_out  sc1_cell2_salmon_out  sc1_cell25_salmon_out  sc1_cell36_salmon_out  sc1_cell34_salmon_out  sc1_cell12_salmon_out  sc1_cell65_salmon_out  sc1_cell32_salmon_out  sc1_cell52_salmon_out  sc1_cell87_salmon_out  sc1_cell86_salmon_out  sc1_cell20_salmon_out  sc1_cell57_salmon_out  sc1_cell38_salmon_out  sc1_cell68_salmon_out  sc1_cell50_salmon_out  sc1_cell56_salmon_out  sc1_cell22_salmon_out  sc1_cell33_salmon_out  sc1_cell55_salmon_out  sc1_cell51_salmon_out  sc1_cell19_salmon_out  sc1_cell4_salmon_out  sc1_cell63_salmon_out  sc1_cell94_salmon_out  sc1_cell47_salmon_out  sc1_cell48_salmon_out  sc1_cell96_salmon_out  sc1_cell23_salmon_out  sc1_cell78_salmon_out  sc1_cell79_salmon_out  sc1_cell27_salmon_out  sc1_cell67_salmon_out  sc1_cell44_salmon_out  sc1_cell5_salmon_out  sc1_cell62_salmon_out  sc1_cell80_salmon_out  sc1_cell43_salmon_out  sc1_cell69_salmon_out  sc1_cell64_salmon_out  sc1_cell66_salmon_out  sc1_cell13_salmon_out  sc1_cell49_salmon_out  sc1_cell54_salmon_out  sc1_cell60_salmon_out  sc1_cell40_salmon_out  sc1_cell28_salmon_out  sc1_cell42_salmon_out  sc1_cell76_salmon_out  sc1_cell26_salmon_out  sc1_cell29_salmon_out  sc1_cell18_salmon_out  sc1_cell82_salmon_out  sc1_cell93_salmon_out  sc1_cell70_salmon_out  sc1_cell46_salmon_out  sc1_cell74_salmon_out  sc1_cell16_salmon_out  sc1_cell14_salmon_out  sc1_cell45_salmon_out  sc1_cell35_salmon_out  sc1_cell77_salmon_out  sc1_cell8_salmon_out  sc1_cell11_salmon_out  sc1_cell15_salmon_out  sc1_cell6_salmon_out  sc1_cell30_salmon_out  sc1_cell39_salmon_out