Nucleocytoplasmic large DNA viruses (NCLDVs) are the largest viruses known today in both genome and virion size. They infect most major lineages of eukaryotes across diverse habitats (1–4), especially in the marine environment (5, 6). The discovery of these giant viruses revolutionized our understanding of the evolution of viruses and their arms race with protist hosts, yet our knowledge of their infection strategies and transcriptome regulation remains limited. Among the NCLDVs of special ecological importance are members of the family Phycodnaviridae, which infect a wide range of key algal hosts (1). The calcifying eukaryotic alga Emiliania huxleyi forms massive annual blooms in the oceans that have a profound impact on the carbon and sulfur biogeochemical cycles (7). E. huxleyi blooms are frequently terminated by a large double-stranded DNA (dsDNA) virus—coccolithovirus (E. huxleyi virus, EhV) (8), which regulates nutrient cycling and carbon export to the deep ocean (9, 10). This host-virus model provides a trackable system for understanding viral life cycle strategies and host responses.
High-throughput bulk RNA sequencing (RNA-seq) has been used for whole-genome expression profiling of NCLDVs during infection, shedding light on gene prediction, transcript structure, and changes in metabolic pathways (11–13). However, bulk RNA-seq profiles average gene expression levels across many cells, whereas infection states can be variable among single cells. To overcome this limitation, single-cell RNA-seq (scRNA-seq) approaches have been developed to probe the transcriptomes of individual cells in a highly parallel manner. These methods have substantially advanced our understanding of various developmental and immunological processes (14, 15), including host-virus interactions in mammalian systems (16, 17). Here, we use scRNA-seq to investigate the transcriptomic profiles of viral and host genomes in EhV infection at the single-cell level. Specifically, we aim to have a comprehensive understanding of the temporal dynamics and regulation of viral genome expression and the associated changes in the entire cellular transcriptome.
RESULTS AND DISCUSSION
Cell-to-cell variability in overall viral expression
To characterize the temporal dynamics and regulation of viral and host transcriptomes, we infected E. huxleyi CCMP2090 cultures with the lytic virus EhV201 (Fig. 1A, table S1, and fig. S1) (13, 18). Individual cells were isolated into 384-well plates by fluorescence-activated cell sorting (FACS) and processed using the MARS-seq protocol (19), which uses cellular barcodes to tag RNA molecules from individual cells and enables transcript quantification by using unique molecular identifiers (UMIs). By mapping reads to the reference E. huxleyi transcriptome and to the EhV201 genome, we were able to profile both viral and host transcripts in each individual cell. The mean of total UMI counts of host or viral transcripts in a single cell is 187.7, while the median is 148.5. These numbers reflect the relatively low RNA content of E. huxleyi cells, which are only 3 to 5 μm in diameter for naked cells (20).
We compared the total expression levels of all viral genes in 5179 cells from infected cultures, control cultures, and cultures mock infected by ultraviolet (UV)–inactivated viruses (Fig. 1B). While the average viral gene expression increased over time of infection, high variability in viral expression levels between single cells was observed despite a high virion-to-cell ratio (multiplicity of infection 5:1) that was used to reduce the possible variation due to encounter rates. Throughout the first 10 hours of infection, two coexisting groups of cells were observed—one with clear detection of active viral expression (≥10 viral UMIs) and the other with viral transcript abundance at the level of noise (cf. control or mock infection). The observed bimodal distribution of viral transcripts implies the existence of an all-or-none switch that rapidly turns on viral expression (Fig. 1B). The proportion of cells with at least 10 viral UMIs increases from 0% to nearly 90% within 10 hours (Fig. 1C). The mRNA pool in each individual cell was mostly dominated by either host or viral transcriptome, with very few cells in intermediate states, forming a sharp decline in host-virus transcript ratio (Fig. 1D and fig. S2), indicating that EhV massively transforms the cellular transcriptome by taking over almost the entire mRNA pool (fig. S3).
Mapping the continuum of viral infection states
To further explore the variation among infected single cells, we applied the MetaCell method using the k-nearest neighbor graph partitions (21) for organizing single cells into cohesive groups (metacells) of cells with similar viral gene expression profiles. Rather than grouping cells into discrete clusters, the two-dimensional (2D) projection of the k-nearest neighbor graphs revealed a continuum of infected cells, where the seven metacells defined by the algorithm correspond to different phases of infection (Fig. 2A). The infected cells were spread along an infection continuum, reaching maximum heterogeneity at 6 hours post infection (hpi) (Fig. 2B). In contrast, cells pretreated with cycloheximide (1 μg/ml), an inhibitor of eukaryotic translation (22), were mainly localized in the initial phase (metacell 1) of infection at 6 hpi (Fig. 2B). This suggests that newly synthesized proteins are required for later viral transcription programs, but not for the initial phase. The continuum of cells allows us to map infection progression on a pseudotemporal scale of 0 to 1 that was calculated as the relative distance to the origin (earliest state) of the continuum (Fig. 2A), providing a new dimension to determine infection states of individual cells within heterogeneous populations.
Highly resolved viral gene expression programs
By reordering cells on the basis of the newly defined infection states, we found that viral genes were largely organized in five discrete kinetic classes of genes herein termed as immediate-early (73 genes), early (136), early-late (49), late 1 (22), and late 2 (40), which formed partially overlapping sequential waves of expression (Fig. 3 and table S2). Among the 447 predicted coding sequences (CDSs) in EhV201, 320 can be unambiguously assigned to one of the five distinct kinetic classes. Of the other 127 genes, 36 were barely detectable (<20 UMIs in total across all cells in Fig. 3; table S2) and are not shown in Fig. 3, while the others generally have low (<100 UMIs in total) or sparse expression levels with class assignment undetermined (16), early-undetermined (17; clustered with immediate-early, early, and early-late), or late-undetermined (58; clustered with late 1 and late 2) (Fig. 3 and table S2). Most genes within each of the five discrete classes show highly similar expression profiles, but outliers do exist. For example, three immediate-early genes (AET98025.1, AET98171.1, and AET98238.1) were semiconstitutively expressed even at later infection states.
We also observe the correlation between the expression dynamics and known functions of genes. The early kinetic class encompasses most of the genes involved in information processing and metabolism, including DNA replication (DNA polymerase delta subunit, proliferating cell nuclear antigen, and DNA topoisomerase II), nucleotide metabolism [ribonucleoside-diphosphate reductase, dUTP (deoxyudirine triphosphate) pyrophosphatase, and thymidylate synthase], and lipid metabolism (Fig. 3C). In addition to the unique virus-encoded sphingolipid biosynthesis enzymes dihydroceramide synthase, dihydroceramide desaturase, sphingosine 1-phosphate phosphatase, and serine palmitoyltransferase (13, 23–25), fatty acid elongase and sterol desaturase are also in the early class (Fig. 3C). The early genes further encode four RNA polymerase II subunits (RPB1, RPB2, RPB3, and RPB6), which probably enable transcription independent of the host machinery. Two small RNA polymerase II subunits (RPB5 and RPB10) are encoded in the early-late class. Together with three transcription factor genes in the early-undetermined, early, and early-late, the RNA polymerase genes may be key regulators of the kinetic classes. Most of the genes assigned to the late 1 and late 2 kinetic classes have unknown functions, but two genes are noteworthy: the packaging adenosine triphosphatase (ATPase) for packing viral DNA into virions (late 1 class) and the major capsid protein, which is the main component of the virion capsid (late 2 class).
To further characterize the viral protein-encoding genes, we purified virions from lysed cultures and analyzed their protein composition using liquid chromatography–mass spectrometry (LC-MS). In total, 76 virus-encoded genes had at least 10 peptide-spectrum matches (table S3) and were shown as present in virion proteome in Fig. 3C. Comparisons with the kinetic class assignments suggest that most components of EhV virions are encoded by the late classes (late 1, late 2, and late-undetermined). Although many early genes have high mRNA levels, only 3 (with unknown functions) out of 136 early genes were detected in the virion proteome. The packaging ATPase encoded by a late 1 gene was also absent in the proteomic data (table S3). These observations clearly suggest that EhV uses transcriptional kinetic classes to temporally regulate production of virion components in a highly selective manner.
To elucidate the regulatory mechanisms underlying the viral gene kinetic classes, we looked for enriched sequence motifs de novo in the promoter regions [±100 base pair (bp) from the first base of the start codon] of genes belonging to different kinetic classes (Fig. 4A and table S4). The most enriched motif in the promoter region of immediate-early genes ends mostly at the ATG start codon (positions 12 to 14) and was previously identified as a putative promoter element associated with initial expression in another EhV strain (26). We also revealed short, highly enriched elements, including a 6-bp motif around the ATG of early genes and a 7-bp motif mostly upstream of the start codon in the early-late genes. An AT-rich 20-bp motif was detected, with a degenerated center, usually immediately upstream of the start codon of late 1, late 2, and other late genes. We further found that these putative promoter elements are found to be conserved across EhV genomes (fig. S4), which share most of their gene contents (fig. S5 and table S5). The late motif bears notable similarity to the late promoter element of Mimivirus, a distantly related NCLDV that infects amoebae, which is composed of two AT-rich decamers separated by a degenerated tetramer and is also present in the genome of the Mimivirus virophage (11). Although genes in most kinetic classes are scattered throughout the EhV201 genome, the immediate-early genes are mainly concentrated in two genomic regions (Fig. 4B), as also seen in the strain EhV86 (26). Such organization could facilitate the rapid activation of the immediate-early class once the host cell is infected, allowing transcription initiation of the entire cascade of the various kinetic classes. Together, our findings of newly defined viral kinetic classes are strongly supported by the discovery of potential regulatory mechanisms that are conserved across EhV genomes.
Host transcriptional profiles along infection progression
Last, our single-cell dual RNA-seq approach allows quantification of not only transcripts encoded by the viral genome but also respective host transcriptomes including those of the mitochondria and chloroplasts (Fig. 5). As depicted in Fig. 1D, there was a sharp shift of the cellular transcriptome from host-encoded to virus-encoded transcripts. Reduction in host transcripts has also been observed at the single-cell level for the non-NCLDV dsDNA Herpesvirus (27), but EhV causes more extensive host shutdown at a faster pace. By mapping the expression levels across infection states, we found differential shutdown of nucleus-encoded and organelle-encoded genes (Fig. 5A). Compared with the control (0.12 UMIs per transcript per cell), mock-infected (0.12), and infected culture with fewer than 10 viral UMIs (0.10), most nuclear transcripts were rapidly shut down at the onset of active viral expression, leaving only 0.02 UMI per transcript per cell in the early phase of infection. On the contrary, levels of the mitochondrial and chloroplast transcripts at the same stage of infection were at similar or slightly higher levels compared with the other cultures. Moreover, the two types of organelles show distinct expression patterns, with mitochondrial transcripts persisting at lower yet relatively stable levels throughout the infection progression and chloroplast transcript levels that were high in the beginning but then mostly soon declined (Fig. 5, B and C).
Despite the rapid shutdown of most nuclear transcripts, 106 transcripts still had a relative abundance of more than 10% for the bulk average of cells assigned to metacell 1, the early phase of infection progression (Fig. 5A). Analyses of the Gene Ontology (GO) (28) terms assigned to these transcripts (table S6) suggest that they are relatively enriched (P < 0.05) in GO terms that include mitochondrial protein complex, mitochondrial membrane part, monosaccharide metabolic process, and nucleotide metabolic process (table S7). Notable annotations from the Kyoto Encyclopedia of Genes and Genomes (KEGG) (29) include NADH (reduced form of nicotinamide adenine dinucleotide) dehydrogenase subunits NDUFV2 (KEGG number K03943) and NDUFA13 (K11353), as well as succinate dehydrogenase iron-sulfur subunit SDHB (K00240), which are nucleus-encoded components of mitochondrial protein complexes involved in the tricarboxylic acid (TCA) cycle and oxidative phosphorylation. This suggests that the expression of both the nuclear (Fig. 5A) and mitochondrial genes [e.g., nad1–6 encoding NADH dehydrogenase subunits (Fig. 5B)] was maintained at a high level to sustain these key mitochondrial metabolic processes. Other putative gene products are TOM40 (K11518), part of the mitochondrial outer membrane complex that imports proteins, phosphoglycerate kinase (PGK; K00927), a major glycolytic enzyme, and phosphoglucomutase (PGM; K01835), which channels glucose to glycolysis or the pentose phosphate pathway. While most nuclear genes are shut down, the high expression of these genes likely prepares the cell to produce enough energy for viral protein synthesis, DNA replication, lipid biosynthesis, virion assembly, and other processes. The high metabolic demand of EhV and other giant viruses with large burst size points to high dependence of viruses on their host metabolic state for optimal replication (13, 30). Previous single-cell analyses revealed down-regulation of most of the examined E. huxleyi metabolic genes in the stationary phase, concomitant with low viral gene expression, as compared with infected exponential phase cells (31).
The delayed shutdown of chloroplast-encoded genes suggests the need for active chloroplasts in early stages of infection, which is in agreement with the recent finding that viral infection depends on light in both the coccolithophore E. huxleyi (32) and the pelagophyte alga Aureococcus anophagefferens (33). Consistent with protein immunoblot analysis showing significant decrease in the level of RbcL [RuBisCO (ribulose-1,5-bisphosphate carboxylase-oxygenase) large subunit] compared with PsbD (photosystem II protein D2) at 24 hpi (32), the transcript encoding PsbD persisted until the late phase of infection while most transcripts, including the one for RbcL, were shut down earlier (Fig. 5B). This transcriptional difference among chloroplast-encoded genes could be a possible mechanism to divert photosynthetic energy to nucleotide synthesis via the pentose phosphate pathway (32), which coincides with the high level of phosphoglucomutase despite the systemic shutdown of nuclear transcripts. Although it is unclear how genes are regulated differently between the chloroplast and mitochondrial genome, the organellar membranes might play a role in preventing transcript shutdown at least early during infection. Unlike mRNA transcripts that were either quickly shut down or maintained, ribosomal RNA (rRNA) transcripts, especially that comprising cytosolic ribosomes, increased in quantities until the middle stage of infection (fig. S6), which could suggest a corresponding boost in translation.
scRNA-seq has proven to be a very powerful tool for comprehensive understanding of complex host-virus dynamics, disentangling cells from different sampling times and infection states. Using this approach, we dissected the transcriptomic dynamics and regulation of viral infection in a globally important microalga and showed rapid selective shutdown of host genes induced by an NCLDV. The comprehensive high-resolution mapping of infection states and kinetic classes allowed identification of conserved regulatory sequences, which likely orchestrate the sequential expression programs during the viral life cycle. Functional analyses of nuclear, chloroplast, and mitochondrial transcripts along the infection progression further shed light on the potential strategies used to optimize viral replication. The applicability of scRNA-seq provides new resolution to track host-virus dynamics and holds great potential for quantification of active viral infection in natural populations and identification of additional ecologically important host-virus systems. This, in turn, will facilitate the assessment of the impact of viruses on the function and composition of microbial food webs in the marine environment.
MATERIALS AND METHODS
Algal culture and viral stock
The coccolithophore E. huxleyi strain CCMP2090 was grown in K/2 (34) filtered seawater medium at 18°C with an irradiance of 100 μmol photons m−2 s−1 and a 16:8–hour light-dark cycle. The viral strain EhV201 (35) was propagated on CCMP2090 culture until the culture lysed to clearance, and viral lysate was filtered using a 0.45-μm PVDF (polyvinylidene difluoride) Stericup-HV vacuum filtration system (Merck). We used the plaque assay as previously described (36) to quantify infectious virions, and filtered viral lysate generally contained ~108 infectious virions/ml. Cultures at the exponential phase (~106 cells/ml−1) were used for viral infection at a high multiplicity of infection (five infectious virions to one cell) at 2 hours after the light cycle began. Algal cells and virion particles were enumerated using an Eclipse flow cytometer analyzer (iCyt) during an infection time course (fig. S1). UV-inactivated virus was obtained by exposing viral lysate in a plastic petri dish to UV in a transilluminator system for 15 min. For inhibition of protein synthesis, cycloheximide was added to culture for a final concentration of 1 μg/ml immediately before viral infection.
Single-cell isolation by FACS
Single cells were isolated into individual wells in 384-well plates on a FACSAria II or III sorter (BD Biosciences) using a 488-nm laser for excitation and a nozzle size of 100 μm. Cells were identified on the basis of chlorophyll red autofluorescence (663 to 737 nm) and separated under the “Single Cell” mode for maximal purity.
Massively parallel scRNA-seq
The scRNA-seq method is based on the MARS-seq2.0 protocol (19). Before sorting, each 384-well plate was prepared using a Bravo liquid handling platform (Agilent) to transfer into each well a 2-μl lysis buffer containing 2 or 4 nM reverse transcription (RT) primers with cellular and molecular (UMI) barcodes. Plates with sorted cells were immediately frozen on dry ice and later stored at −80°C. After thawing, the plates were heated to 95°C for 3 min to evaporate the lysis buffer, cooled, added RT reagent mix, and placed in a Labcycler (SensoQuest) for RT reaction. The unused primers were then degraded with Exonuclease I (NEB). cDNA from each set of wells with 384 (one pool per plate) or 192 (two pools per plate) unique cellular barcodes were then combined into pools, which underwent second-strand synthesis and in vitro transcription to amplify the sequences linearly. The RNA products were fragmented and ligated with single-stranded DNA adapters containing pool-specific barcodes. A second round of RT was carried out, followed by polymerase chain reaction with primers containing Illumina adapters to construct DNA libraries for paired-end sequencing on a NextSeq or MiniSeq machine (Illumina).
Read processing and mapping to reference sequences
The FASTQ files were processed using the analytical pipeline of MARS-seq2.0 (19), which mapped the reads to viral and host reference sequences and demultiplexed them based on the pool, cellular, and molecular barcodes. At least four reads per UMI were obtained per cell. For the virus, the predicted CDSs in the EhV201 genome sequence (JF974311) (37) were used as reference. For the host, an integrated transcriptome reference of E. huxleyi (38) was used, where both forward and reverse (labeled with “_rev”) strands of each sequence were included for mapping because the CDS directionality was often unknown. In addition to the nuclear sequences, the transcriptome reference contains chloroplast and mitochondrial transcripts, which were identified by the basic local alignment search tool (BLAST) (39) searches against the respective organellar genome sequences (40, 41).
Viral and host transcript abundance
To avoid empty wells or those with low transcript capture during library preparation, wells with fewer than 20 total UMIs (including all viral and host transcripts) were removed for most downstream analyses. Wells with more than 1000 transcripts were also removed to prevent wells with doublet or multiplet cells from confounding single-cell infection state analyses. The vast majority (>99%) of control or infected E. huxleyi cells had fewer than 1000 total UMIs (mean and median UMI counts, 187.7 and 148.5), which reflects their low RNA content due to the small cell size [3 to 5 μm in diameter for naked cells (20)].
Cliff diagrams of viral and host mRNA
A cliff diagram was plotted for each time point after infection (Fig. 1D). The total viral mRNA transcript counts were calculated by summing up all UMIs assigned to virus-encoded CDSs. The total host mRNA counts were all host UMIs except for those assigned to the three transcript sequences corresponding to nuclear, mitochondrial, and chloroplast rRNA genes. For each sampling time point, the cells were ordered by the ratio of host to all (host + virus) mRNA from high to low. Cells with fewer than 20 total UMIs were excluded from the plots to avoid sampling biases.
Doublet assay and in silico simulations
The EhV201 viral genome has a much lower GC content (40.46%) (37) than the E. huxleyi host genome (65%) (42). This difference in GC content and other factors might bias the capture and amplification of viral and host transcripts during MARS-seq and might lead to the either-or dominance of host and viral mRNA (Fig. 1D). To test if the presence of viral mRNA has effects on the quantification of host mRNA transcripts and vice versa, we conducted a doublet assay. Half of the doublet assay plate consisted of single cells sorted from infected culture at 8 hpi, and the other half plate consisted of two cells per well of which one cell was sorted from infected culture at 8 hpi and the other cell from control culture (table S1). Cliff diagrams were then plotted for each half of the plate (fig. S2, A and B). If the viral mRNA does not interfere with host mRNA quantification, wells with cells isolated from both the infected and the control cultures should show host and viral mRNA UMI counts that are additive, as if the two cellular transcriptomes were sequenced separately and merged together. To simulate this in silico, we randomly sampled sequenced single-cell transcriptomes from the control culture plate with the closest time point to 8 hpi (table S1) using the sample function in R, and they were combined one-by-one with those from the 8-hpi half plate with single cells. Cliff diagrams were plotted for each of the eight simulations (fig. S2, C to J).
MetaCell analysis of infected cells
The MetaCell package is an unbiased approach for cell grouping based on sparse single-cell expression profiles without enforcing any global structure (21). To account for the small number of UMIs per cell and to use more viral genes for grouping infected cells, we used more inclusive criteria for gene marker selection (T_tot = 20, T_top3 = 1, T_szcor = −0.01, T_vm = 0.2, and T_niche = 0.05). A total of 178 variable genes were selected as marker genes for constructing k-nearest neighbor graphs with K = 150, followed by coclustering with bootstrapping based on 1000 iterations of resampling 75% of the cells and an approximated target minimum metacell size of 80. Unbalanced edges were filtered with K = 40 and α = 3.
After plotting single cells in a 2D projection and identifying the directionality of infection progression (Fig. 2), we quantified the infection states of individual cells based on the relative distance to the beginning of the progression (i.e., upper-left). The ad hoc origin was defined as having x and y coordinates of the leftmost and the highest cells, respectively, and the ad hoc end as having those of the rightmost and the lowest cells, respectively. The infection index of an individual cell was calculated as the ratio of its distance from the origin to the distance between the origin and the end. A color gradient was painted on the basis of the scale of the infection indices between 0 and 1.
Hierarchical clustering of genes
We used hierarchical clustering to group viral genes with similar expression patterns across infected cells. Viral genes with fewer than 20 UMIs in total across all the cells with active viral expression were excluded to avoid spurious clusters or groupings that are not well supported. The pheatmap package in R was used for hierarchical clustering of UMI counts in log scale, using Pearson correlation as distance measure and the “average” agglomeration method. The cluster tree of viral genes was manually curated by branch swapping using Archaeopteryx (43) to order the genes by kinetic classes and by expression levels.
Annotations of EhV201 genome
A comprehensive annotation table was prepared for EhV201 genes by ordering them according to the manually curated cluster tree (Fig. 3B). The sequence descriptions were based on the GenBank record (JF974311), with modifications based on relevant publications (44, 45), BLAST searches against the nr database (46), and the nucleocytoplasmic virus orthologous groups (2).
Virion proteome composition
Purified EhV201 virion samples were lysed using an 8 M urea buffer and subjected to tryptic digestion followed by LC-MS analysis on a Q Exactive HF instrument (Thermo Fisher Scientific). We used Byonic version 2.10.5 (Protein Metrics) as the software to process the data and to search against the viral CDS database (table S3). In total, 76 viral CDSs had at least 10 peptide-spectrum matches and were shown as present in virion proteome in Fig. 3.
Sequence motif analyses of viral genomes
A 200-bp region surrounding the predicted start codon ATG (A of ATG is base 101) was extracted for each CDS in the EhV201 genome using the Extract Genomic DNA tool of Galaxy (47). Promoter motif discovery was performed on the positive strand of these regions for genes in different kinetic classes using MEME Suite version 5.0.5 (48), Genomatix Genome Analyzer CoreSearch (Intrexon Bioinformatics), and Improbizer (https://users.soe.ucsc.edu/~kent/improbizer/improbizer.html), with the number of motifs to detect set to four. While all three programs agreed on the motifs detected, the results of the MEME analyses are used in this study, where the most common motifs of lengths 6 to 20 are shown (Fig. 4A, fig. S4, and table S4). To test if the short 6-bp sequence motif in early genes is significantly enriched, a differential analysis was performed for early genes against all late (late 1, late 2, and late-undetermined combined) using DREME (49) (table S4).
Circular visualization of viral genes
The UMI counts of individual viral genes were summed over all cells showing active viral expression (≥10 viral UMIs). The log2 values of these total UMI counts were visualized on the 407-kb genome sequence of EhV201 (JF974311) using CCT (CGView Comparison Tool) (50), with a ring for each kinetic class and the height proportional to the log-transformed expression value. Another ring shows the presence/absence of each gene product in the virion proteome (Fig. 4B).
Comparative analyses of other EhV genomes
The best BLAST hits in the CCT analyses were used to find the homologous proteins of the various kinetic classes in other EhV genomes. DNA surrounding the ATG of those genes was extracted for each strain with the GetFastaBed tool in Galaxy (47). Promoter analyses were run on the sequences of the promoter regions of three additional strains (EhV207, EhV86, and EhV202) individually as described above and with all four together (fig. S4 and table S4). These three strains were chosen as representatives of three major clades of EhV strains (51). A CCT plot was drawn to show the sequence similarity between EhV201 and all 12 other EhV strains with whole genome sequences (fig. S5).
Nuclear and organellar transcripts
Host gene expression patterns were analyzed for different categories of single cells, including control culture, mock infected culture, infected culture with fewer than 10 viral UMIs, and different metacells of infected cells with at least 10 viral UMIs. Transcript sequences with fewer than 100 total UMIs across all the cells were excluded to avoid lowly expressed ones, leaving 720 unique transcripts in total (Fig. 5A). The average UMI count was calculated for each transcript in each category, and the sum of the average values of all categories was normalized to 100. Hierarchical clustering of host transcripts was done using pheatmap as described above on the basis of the normalized values. The sliding windows of infected cells (window size = 50 cells, step = 10 cells) were generated for mRNA transcripts from the nuclear, mitochondrial, and plastid genomes (Fig. 5, B and C) and the major transcript from each genome that encodes rRNA (fig. S6) using the rollapply function of the zoo package in R.
Functional analyses of nuclear transcripts
Nucleus-encoded transcripts in Fig. 5A were annotated with GO identifiers (28), as previously described (38). GO term enrichment analyses of transcripts with a relative abundance of more than 10% (Fig. 5A) were conducted using the classicFisher method with a nodeSize of 10 as implemented in topGO version 2.34.0 (52), with P values listed in table S7. Identifiers of KEGG (29) were assigned to the transcripts with KEGG Automatic Annotation Server (53) based on the BBH (bi-directional best hit) method (nucleotide) with the default list of reference organisms.