Landscape of Epstein-Barr virus gene expression and perturbations in cancer

Abstract Epstein-Barr virus (EBV) is the causative agent for multiple neoplastic diseases of epithelial and lymphocytic origin1-3. The heterogeneity of the viral elements expressed and the mechanisms by which these coding and non-coding genes maintain cancer cell properties in vivo remain elusive4,5. Here we conducted a multi-modal transcriptomic analysis of EBV-associated neoplasms and identified that the ubiquitously expressed RPMS1 non-coding RNAs support cancer cell properties by disruption of the interferon response. Our map of EBV expression shows a variable, but pervasive expression of BNLF2 discerned from the overlapping LMP1 RNA in bulk sequencing data. Using long-read single-molecule sequencing, we identified three new viral elements within the RPMS1 gene. Furthermore, single-cell sequencing datasets allowed for the separation of cancer cells and healthy cells from the same tissue biopsy and the characterization of a microenvironment containing interferon gamma excreted by EBV-stimulated T-lymphocytes. In comparison with healthy epithelium, EBV-transformed cancer cells exhibited increased proliferation and inhibited immune response induced by the RPMS1-encoded microRNAs. Our atlas of EBV expression shows that the EBV-transformed cancer cells express high levels of non-coding RNAs originating from RPMS1 and that the oncogenic properties are maintained by RPMS1 microRNAs. Through bioinformatic disentanglement of single cells from cancer tissues we identified a positive feedback loop where EBV-activated immune cells stimulate cancer cells to proliferate, which in turn undergo viral reactivation and trigger an immune response.


Main
Epstein-Barr virus (EBV) is estimated to cause 120,000-357,900 cases of neoplasms worldwide each year and 1.8% of all cancer deaths are attributed to EBV-associated malignancies 1,2 .
Originally identified in the neoplastic cells of endemic Burkitt's lymphoma (eBL), EBV is the first discovered human tumor virus and infection is extremely common with more than 95% seropositivity among adults worldwide 3,6 .EBV was subsequently implicated as the causative agent in other hematological neoplasms including a fraction of sporadic Burkitt's lymphoma (sBL).The vast majority of nasopharyngeal carcinoma (NPC) and approximately every tenth gastric adenocarcinomas (GAC) are associated with EBV infection and these epithelial malignancies constitute more than 80% of all known EBV-associated cancer cases 7 .Cell lines have been derived from primary tumors or by immortalization of primary B-cells by EBV (lymphoblastoid cell lines) [8][9][10] .
In recent years bulk transcriptome (RNA-Seq) and single-cell sequencing data (scRNA-Seq) from virus-associated neoplasms have become publicly available [11][12][13][14][15][16] .Viral gene expression in neoplasms have been shown to be associated with the respective known viral oncogenes, for example E6 and E7 in human papillomavirus associated cancer and T-antigen in polyomavirus associated cancer 17 .Multiple RNA-Seq studies have shown that EBV mRNA predominantly originate from the BamHI-A/I region, in which none of the known EBV-oncogenes are encoded.In a single gastric adenocarcinoma, it was initially suggested that the rightward transcribed long non-coding RNA RPMS1 was expressed in the tissues 5,17 .However, subsequent analysis of larger cohorts of EBV-expressing neoplasms suggested that the transcripts were encoded by the overlapping leftward transcribed genes 4,18 .EBV genomic fragments containing these regions have been shown to have transformative properties 19,20 .

EBV in primary tissues and cell lines
We started with bulk transcriptome data from publicly available datasets originating from four malignancies with known EBV-association, NPC, GAC, eBL and sBL, as well as EBVassociated tumor-derived cell lines (ECL) and EBV-transformed lymphoblastoid cell lines (LCL) 10,[21][22][23][24] .The datasets were mapped against the EBV reference genome (Figure 1a, Extended Data Table 1).The EBV fraction was calculated as parts per million (ppm) and datasets with lower than 10 ppm EBV reads were classified as EBV-negative and were not processed further 25 .In accordance with known epidemiology, the majority of NPC and eBL tumors contained more than 10 ppm EBV reads and conversely, the majority of GAC and sBL did not contain any EBV reads (Figure 1b) 2 .The EBV fraction in NPC varied between 14-1,131 ppm, in GAC 15-331 ppm, in eBL 18-470 ppm, and in sBL 149-502 ppm, likely reflecting the purity of tumor cell in the biopsy (Extended Data Fig. 1a).Of the fourteen ECL, eight cell lines contained less than two ppm EBV reads, while the remaining six contained 135-286 ppm EBV reads.All LCL datasets contained, as expected, high levels of EBV reads ranging from 816-16,946 ppm.The EBV gene expression of 156 tumors with minimum 10 ppm EBV-reads originating from 106 NPC, 30 GAC, 16 eBL, and 4 sBL were further processed.The detected EBV-reads were then aligned to the viral genome on coverage plots (Figure 1c) (Supplementary Information; EBV RNA).Of the average EBV coverage in these primary tumors, RNA from the adjacent BamHI-A and I region constituted 88% (standard deviation ±15%) in NPC, 92% (±6%) in GAC, 85% (±19%) in eBL, and 92% (±7%) in sBL (Extended Data Fig. 1b).In contrast, with the exception of the ECL C666-1 (90% BamHI-A/I RNA), on average 6% (standard deviation ±7% and ±4% respectively) of the EBV RNA, in the ECL and LCL datasets, aligned to the BamHI-A/I region.

EBV genes expressed in cancer
The majority of the EBV RNA mapped to the BamHI-A/I regions in primary tumors aligned to areas in which multiple genes overlap (Figure 1d).In order to discern the RNA elements within the BamHI-A/I regions we conducted RNA peak (Extended Data Fig. 2a,b), strand-specificity (Figure 1d third panel above/below baseline) (Supplementary Information; EBV RNA), transcription start site (Figure 1e) (Supplementary Information; RPMS1 transcription start site), RNA splicing (Figure 1f) (Supplementary Information; BamHI-A/I splice-junctions), and polyadenylation signal analyses (Figure 1g) (Extended Data Table 2).The results all supported that the major transcript in the four EBV-positive tumor types originated from the long noncoding RNA RPMS1.However, RNA located in the RPMS1 introns not coupled to the constitutive exons (Figure 1d, BALF5 mid-region) suggests the presence of other RNA elements.Co-transcriptional activation of these elements was observed in an inducible RPMS1 promoter mutant Burkitt's lymphoma cell line (Supplementary Information; EBV RNA, BamHI-A/I splice-junctions and Namalwa modified cell lines).An adapted full-length singlemolecule sequencing of RPMS1 (Extended Data Fig. 2c-e) 26 , allowed us to identify three new rightward elements which we named BamHI-A rightward elements, BARE1-3, (Supplementary Information; BamHI-A rightward elements).We amended the EBV reference genome (NC_007605.1) with the new annotations for BAREs and aligned the EBV reads to the new genome.Non-aligned EBV reads were plotted and additional gene segments were added to increase the aligned fraction producing an >95% EBV mappability for all except three datasets (Extended Data Table 1).In order to produce an unbiased quantification of the global EBV expression in the different tumor types and cell lines, we applied the length-adjusted tpm-values of house-keeping genes and EBV genes by normalizing them to the content of the entire dataset (Extended Data Table 3).Calculation of tpm-values for EBV genes required additional modifications due to overlapping regions exemplified by RPMS1 and BAREs, for which we calculated the tpm-values relative to their unique regions and the overlapping regions as a fused gene, RPMS1/BAREs (Figure 2a-f) (Extended Data Table 4).The tpm-values of RPMS1 were thus calculated relative to its first four exons and BAREs relative to respective first exons, while the fused RPMS1/BAREs was calculated based on RPMS1 exons V-VII which all rightward BamHI-A/I genes share.This division creates a bias with artificially lower values for the RPMS1 and BAREs unique regions, due to 5'-degradation of mRNA which is overrepresented in polyA-enriched libraries (Figure 2g).The similar division was also employed for the LMP1/BNLF2 and LMP2A/B genes (Figure 2h-i).
Although no absolute threshold can be set, we chose to mainly consider genes with tpm-values over 5 (Supplementary Information; Tpm-threshold) (Extended Data Fig. 3a-c).The normalized tpm-values of EBV genes in tumors are diluted due to the inclusion of stromal cells 27 .Conversely, using a low tpm-cutoff will include viral genes that are likely to originate from few cells undergoing reactivation, which are responsible for the high viral background.A high degree of EBV reactivation can be observed in three tumors, NPC3, eBL1 and eBL5, in which global viral transcription including oriLyt RNA (eBL1) can be observed (Extended Data Fig. 4) (Supplementary Information; EBV gene expression).With the exception of tumors with EBV reactivation, high expression of viral genes was only observed for the genes RPMS1/BAREs (77% of tumors), LMP1/BNLF2 (10% of tumors) and LMP2A/B (1% of tumors).Intermediate expression of RPMS1/BAREs, LMP1/BNLF2 and LMP2A/B were detected at 15%, 14% and 8% of tumors respectively.Low expression of RPMS1/BAREs were detected in the remaining tumors (8%) and LMP1/BNLF2 and LMP2A/B were observed in 22% and 31% of tumors.Thus, RPMS1/BAREs were expressed in all tumors, on average 77 tpm.
The most abundant and common protein coding RNA originated from the LMP1/BNLF2 gene.
LMP1 has a 2 kb unique 5'-region separated from BNLF2, compared with 840 base pairs for RPMS1 and BARE1, and therefore less likely to be false negative due to RNA degradation.Amongst the NPC, 61 datasets had >5 tpm LMP1/BNLF2, but only 29 datasets had LMP1 expression >5 tpm (Figure 2h).The majority (46/61) NPC had at least two-fold higher tpmvalue of LMP1/BNLF2 compared with only LMP1.This indicates that the RNA originated from BNLF2 and not LMP1 in the majority of neoplasms 28,29 .In contrast, BNRF1 which is located within the last intron of LMP2A/B and shares 448 base pairs 3'-UTR with LMP2A/B was not expressed in the neoplasms (Figure 2i).In contrast, in all six EBV-expressing cell lines (ECL), BHRF1 can be detected at low or moderate levels in the lymphoma cell lines.Multiple EBNAs were expressed, as well as LMPs.However, compared with primary neoplasms RPMS1/BAREs were only expressed at low levels in two lymphoma ECL and at high levels in the NPC cell line C666-1 (Figure 2e).In LCL the EBV expression encompassed almost the entire viral genome in all datasets, which likely reflect the different cell/virus cycle stages in the in vitro culture and a distinct proliferative drive compared with the tumor samples (Figure 2f).
Comparison between EBV-positive and their EBV-negative counterparts has previously described that EBV-tissues have an enrichment of genes correlating to an upregulation of proliferative and immune signaling pathways 4,22,30 .To identify EBV-induced perturbations we applied a variance stabilizing transformation normalization using all tumors in each respective cancer category to find the largest differences between the tumors irrespective of EBV-status 31 (Extended Data Fig. 5a).The principal components with largest Euclidean distance between the midpoint of the EBV-positive and EBV-negative datasets showed the genes that most likely influenced by the presence of EBV (Fig. 2j, Extended Data Fig. 5b-e).A pathway enrichment analysis for each of the principal components showed that all the EBV associated cancer types contained perturbations of MYC and E2F targets, G2M cell cycle progression and interferon response.However, using bulk RNA sequencing, perturbations may arise from interindividual differences and stromal cell composition.Also, cell pathways activated/down-regulated in both EBV-positive and EBV-negative tumors would not be detected.We therefore extended our analyses to scRNA-Seq datasets, which allows for the identification of perturbations in specific cell populations.

EBV expression in scRNA-Seq NPC samples
We processed scRNA-Seq data of 532,122 cells originating from four NPC studies consisting of 63 primary nasopharyngeal samples including 52 NPC and 11 non-tumor (NT) tissues (Figure 3a, Extended Data Table 1) [11][12][13]16 . Amogst the NPC datasets, the single cell  5).In Study 2 (0.2-69.2%) the epithelial cells were enriched and the results were therefore not representative of an unperturbed tissue.The variation of epithelial cell content likely reflects the biological differences, but also the efficiency of epithelial cell dissociation.The distribution of cell composition showed that T and B-lymphocytes were the most abundant stromal cell types in both NPC and non-tumor tissues (Extended Data Fig. 6).
EBV reads were detected in the stromal and epithelial cells (Figure 3a).However, a high variation of the fraction of infected cells was observed between patients.No EBV was detected in the non-cancerous samples (scNT), with the exception of scNT5 and scNT11 where one EBV-positive B-lymphocyte was found in each dataset.EBV status for the NPC tissues was clinicopathological analyzed by EBV encoded RNA in situ hybridization (EBER-ISH) (Study 1-3) or using an EBV specific antibody (Study 4) (Figure 3b).When comparing the results of scRNA-Seq with the experimental assays, four samples in each group had a discordant EBV status.EBER-ISH had the highest sensitivity for the detection of EBV, and the inability of scRNA-Seq to detect EBV RNA in three tumors could be due to the limited number of input cells, low levels of EBV polyadenylated gene expression and/or the low capture rate of the scRNA-Seq technique 32 .In contrast, EBV was detected in four scRNA-Seq datasets which were negative in EBER-ISH or EBV antibody staining (Figure 3b, marked in grey).The proportion of EBV RNA positive epithelial cells in the EBV positive tumors was highly variable, ranging from 0.4-98.3%.
When we considered the EBV expression in 18 samples that contained more than 100 EBV positive epithelial cells (Figure 3c), RPMS1/BAREs was detected in every tumor at high proportions (>46% of EBV-positive cells).Considering the capture rate of the methodology this implies that all cells expressed RPMS1/BAREs.LMP1/BNLF2 and LMP2/BNRF1 were also detected in every tumor, but at highly variable proportions.The variability was most prominent in LMP1/BNLF2 ranging from 3.4% to 99%.Even though scRNA-Seq has a lower sensitivity in terms of RNA capture compared to bulk sequencing (Extended Data Fig. 7), the absence of viral RNA background originating from reactivated cells observed in bulk sequencing allows for every viral transcript in scRNA-Seq to be considered.A few reactivated cells as defined by expression of the immediate-early genes BZLF1/BRLF1 was detected in half of the tumors (Supplementary Information; EBV-positive cells) 33 .Low levels of EBNA1/3B/3C can be observed in the majority of tumors.This supports that all NPC expresses RPMS1/BAREs at high levels, LMP1/BNLF2 and LMP2/BNRF1 at variable levels and possibly EBNA1/3B/3C at low levels (Extended Data Table 6).The EBV expression in stromal cells mostly mirrored the expression in epithelial cells, but specific tumors displayed a high degree of reactivation in stromal cells.
The classification of cancer cell status solely based on EBV RNA has its limitations.A high fraction of EBV false negative cells is expected considering the low capture rate of the technology.A proportion of EBV false positive cells caused by indiscriminate uptake of apoptotic bodies from cancer cells by healthy cells would also reduce the correct classification 34 .Analysis by inference of copy-number variants, a pseudo-marker for chromosomal aberrations, allows for cancer cells assignment based on transcription from entire segments of chromosomes instead of a few EBV reads (Figure 3d) (Supplementary Information; Cancer cell identification).With minor exceptions, the pattern of malignant cells within the same tumor displayed a high homogeneity reflecting the clonality of the cancer cells.As expected, the assignment of cancer cells overlapped with the EBV-positive cells, but with an increased sensitivity (Figure 3e-g).

Host perturbations in cancer cells
Comparison of cancer cells with healthy epithelial cells from the same tumor removes interindividual bias and the shared microenvironment allows for detection of perturbations in tissue-specific conditions.Furthermore, removal of stromal cells increases the signal of cancer cell specific perturbations.A gene set enrichment analysis based on comparison of equal number of cancer and healthy cells from 42 tumors shows a distinctive pattern shared amongst the EBV-positive tumors (Figure 4a, Extended Data Fig. 8) 35 .EBV-positive cancer cells from the four studies shared eight upregulated pathways compared with corresponding healthy cells for each tumor.Upregulation of five proliferative pathways were observed in the EBV-positive tumor cells as well as basal cells compared with corresponding differentiated cells in seven non-tumor biopsies (Supplementary Information; Epithelial cell classification).Cancer cells from the EBV-negative tumors displayed significantly upregulation of the pro-proliferative mitotic spindle pathway.In the EBV-negative HK1-cells transfected with RPMS1 miR-BARTs, but not RPMS1 long non-coding RNA (data not shown), the upregulation of four of the proliferative pathways were reconstituted.
The EBV-positive cancer cells displayed downregulation of immune response 36 ; a result not observed in the basal cells, demonstrating that the perturbation observed in the cancer cells is not due to differences in epithelial cell types.Downregulation of interferon response was further confirmed in both HK1-cells transfected with RPMS1 miR-BARTs as well as in a Burkitt's lymphoma cell line, Namalwa with an induced RPMS1 promoter, but not in cells expressing ectopic RPMS1 long non-coding RNA (Extended Data Fig. 9) (Supplementary Information; Namalwa modified cell lines).These findings indicate that the RPMS1 miR-BARTs induces pro-proliferative pathways and inhibits immune response in cancer cells 37 .
In order to identify EBV-induced changes we sorted the genes, based on the number of tumors in which they are perturbed in the same direction, and identified significantly enriched ones (Figure 4b, Extended Data Table 7).The genes were most strongly correlated with downregulation of oxidative phosphorylation in the EBV-positive cancer cells, likely to be due to the Warburg effect 38 .A strong correlation was also observed with the downregulation of interferon response.The genes involved in immune response downregulation were compiled in order to discern the various pathways (Figure 4c).More than two-thirds of these genes, including MHC1 39 , were also shown to be downregulated in HK1 cells transfected with miR-BARTs and Namalwa with an induced RPMS1 promoter (Extended Data Table 7).The largest proportion of immune genes were regulated by cytokine response.Thus, we analyzed bulk sequencing data of NPCs to detect the gamut of expressed cytokines 40 .The origin of the expressed cytokines was then determined in the scRNA-Seq datasets.The epithelial cells expressed the majority of cytokines of which a few are known to be induced by interferon stimulation (Extended Data Fig. 10a).Amongst the two cell types found in all tumors, B-and T-cells, B-cells produced few cytokines at low levels in most studies.The costimulatory cytokine CD70 was expressed at highest levels in B-cells.Throughout the four studies, the chemokine CCL4 was expressed in a large fraction of T-cells and interferon gamma was expressed at high levels in a smaller subset of cells.The scRNA-Seq data shows that healthy epithelial cells and lymphocytes expresses interferon stimulated genes.Considering that these cells, especially the lymphocytes, constitutes a large proportion of the tumor, this would explain the interferon response upregulation found in the bulk sequencing results (Figure 2j-m).Conversely, the cancer cells exhibit a dysregulated response to interferon.To determine whether this contrasting response we looked at downstream effector genes of interferon gamma (Extended Data Fig. 10b).
No unified difference can be observed in the expression of interferon gamma receptor and downstream kinases between the cancer cells and healthy cells.Multiple genes were expressed at too low levels to allow for a proper comparison.However, the highest expressed STAT3-gene were upregulated in cancer cells from all four studies (significantly in three), but not in the EBV-negative tumors (Extended Data Fig. 10b).A weaker trend for STAT1 downregulation can also be observed.Both STAT1 and STAT3 are known to be bound by miR-BARTs 37,41 .The downregulation of interferon and p53 pathways as well as upregulated proliferative pathways are known hallmarks of STAT-dysregulation 42 .We therefore propose that the ubiquitously expressed EBV RPMS1 gene induces interferon response dysregulation through viral microRNA perturbations of STAT-expression (Figure 4d).LMP2A/B and EBNA1 in NPC, GAC, eBL, sBL, ECL and LCL.Three datasets containing additional EBV genes expressed at more than 5 tpm are marked with an asterisk.These datasets contained EBV genes indicative of lytic replication within the neoplasm.g-i, tpm-distribution of genes with overlapping 3'.The fused RPMS1/BAREs 3' end, compared with the unique regions for RPMS1, BARE1, BARE2 and BARE3, the fused LMP1/BNLF2 3' end, compared with the unique regions for LMP1 and the fused LMP2A/2B 3' end, compared with the unique regions for LMP2A, LMP2B and BNRF1 in NPC, GAC, eBL and sBL.j, Generalized pathway perturbations in principal components for respective cancer type.Principal components correlating with highest EBV-status separation are marked with bold.tpm, transcripts per million reads; NPC, nasopharyngeal carcinoma; GAC, gastric adenocarcinoma; eBL, endemic Burkitt's lymphoma; sBL, sporadic Burkitt's lymphoma; PC, principal component.2 EBV gene expression in bulk RNA sequencing data a-f, Heatmap depicting tpm-values of four gene regions RPMS1/BAREs, LMP1/BNLF2, LMP2A/B and EBNA1 in NPC, GAC, eBL, sBL, ECL and LCL.Three datasets containing additional EBV genes expressed at more than 5 tpm are marked with an asterisk.These datasets contained EBV genes indicative of lytic replication within the neoplasm.g-i, tpmdistribution of genes with overlapping 3'.The fused RPMS1/BAREs 3' end, compared with the unique regions for RPMS1, BARE1, BARE2 and BARE3, the fused LMP1/BNLF2 3' end, compared with the unique regions for LMP1 and the fused LMP2A/2B 3' end, compared with the unique regions for LMP2A, LMP2B and BNRF1 in NPC, GAC, eBL and sBL.j, Generalized pathway perturbations in principal components for respective cancer type.Principal components correlating with highest EBV-status separation are marked with bold.tpm, transcripts per million reads; NPC, nasopharyngeal carcinoma; GAC, gastric adenocarcinoma; eBL, endemic Burkitt's lymphoma; sBL, sporadic Burkitt's lymphoma; PC, principal component.
Nasopharyngeal carcinoma single-cell RNA sequencing datasets a, Proportion of epithelial and stromal cells in the four studies.Striped portion of the bar shows the fraction of EBV-expressing cells in each category.b, Classi cation of each sample according to their origin and EBV-status according to EBER in situ hybridization, antibody detection or UMI in the single-cell data.Samples showing concordant results from two analyses are shown in green, discordant in purple and unknown in grey.c, EBV expression in datasets containing more than 100 epithelial cells (green).The proportion of epithelial cells from each tumor expressing fused EBV gene is shown in the respective column (orange).Genes expressed over 2 cpm were included.d, Inferred chromosomal RNA expression throughout the genome in T-cells (upper panel) and epithelial cells (lower panel), position on x-axis correspond to position in respective chromosome.Epithelial cells divided by unsupervised hierarchical clustering.Areas in red depicts inferred gains and blue loss of chromosomal segment.e-g, Epithelial cells extracted from NPC1 were reclustered in UMAP.Cancer cells classi ed according to EBV expression (blue) showed a lower sensitivity compared to cancer (red) and healthy cell classi cation based on inference of chromosomal copy number variation.NPC, nasopharyngeal carcinoma; EBER, Epstein-Barr virus-encoded small RNAs; UMI, unique molecular identi er; cpm, counts per million reads; UMAP, uniform manifold approximation and projection.Figure 4 EBV-induced host perturbations a, Changes in biological pathways between cancer cells and healthy cells from the same patients.Hallmarks enriched in all four EBV-positive NPCs studies are listed.The same pathways for EBV-negative tumors and healthy controls in which basal cells were compared to differentiated cells in non-tumor samples are shown alongside.Absence of bar indicates no signi cant differences.Induced changes in the EBV-negative nasopharyngeal carcinoma cell line HK1 transfected with RPMS1 miR-BARTs (circles) and Namalwa cells treated with doxycycline (triangles) to upregulate RPMS1 gene are shown in the right column.b, Genes perturbed in the same direction in multiple tumors.Enriched genes are marked with green (upregulated) and magenta (downregulated).The x-axis shows the negative log10 of the false discovery rate q-value (FDRq) for pathways in respective category.c, Immune response genes downregulated in tumor cells categorized according to pathway.Genes in italics are also part of the NF-κB pathway.d, Depiction of factors involved in viral perturbations in NPC epithelial cells.MITSP, mitotic spindle; UVD, UV responsed down; IFNA, interferon alpha response; IFNG, interferon gamma response; OXPH, oxidative phosphorylation; IFN, interferon.

Supplementary Files
preparation in Study 1, 3 and 4 (scNPC1-15 & 32-52, scNT1 & 9-11) was achieved by direct dissociation of primary tissue.In Study 2 (scNPC16-31 and scNT2-8), the epithelial cells were first enriched by flow cytometry and then remixed with stromal cells.Cell type specific clusters of the nasopharyngeal tissue showed that epithelial cell content in the tumor varied among the samples in the different studies, 0.7-37.3% in Study 1, 0.3-18.5% in Study 3 and 0.0-7.5% in Study 4 (Extended Data Table

Figure 1 .
Figure 1.Detection and characterization of EBV gene expression

Figure 2 .
Figure 2. EBV gene expression in bulk RNA sequencing data

Figures Figure 1
Figures

Figure
Figure 2 This is a list of supplementary les associated with this preprint.Click to download.