Atlas of tissue-specific and tissue-preferential gene expression in ecologically and economically significant conifer Pinus sylvestris

Plant material and RNA sequencing

We used the RNAseq data (trimmed reads, BioProject PRJNA531617) previously used to assemble multiple reference transcriptomes of P. sylvestris by Ojeda et al. (2019). The plant material was obtained during the growing season of 2016 (May 26th–27th) and consisted of needles, phloem, vegetative buds and seeds from six non-related adult Pinus sylvestris trees. We used data from trees growing in a natural forest population to ensure that the observed patterns of tissue preferential expression and specificity were robust to genotypic and environmental variation. The material was collected at the Punkaharju Intensively Studied Site (ISS) in Southern Finland (Table S1). Samples were collected in collaboration with Natural Forest Research Institute Finland LUKE that has an agreement on the forest research use with the owner Metsähallitus (The Finnish Forest Administration). In Finland P. sylvestris is not endangered and Finland does not regulate its genetic resources under Nagoya Protocol so CITES was not applied and prior informed consent was not needed. The samples were collected and identified by Tanja Pyhäjärvi. There is no voucher available for the specimens.

Relevant experimental procedures of Ojeda et al. (2019) are briefly summarized here. Megagametophyte and embryo tissues were obtained by dissecting mature seeds collected from the same mother trees from which the vegetative tissues were obtained. Seeds were stored in the dark at 4 °C until germination was induced by exposure to moisture and continuous light for 48 h. Total RNA was extracted from needle, bud and phloem using the Spectrum Plant Total RNA Kit (Protocol B; Sigma, Kawasaki, Kanagawa, Japan), followed by mRNA capture with the NEBNext® Poly(A) mRNA Magnetic Isolation Module (New England Biolabs Inc., Ipswich, MA, USA). For embryo and megagametophyte, mRNA was directly extracted from the whole tissues with Dynabeads mRNA Direct Micro Kit (Thermo Fisher Scientific, Waltham, MA, USA). RNA concentration was quantified with Qubit RNA HS Assay kit (Thermo Fisher Scientific, Waltham, MA, USA). The quality and integrity of the RNA was visually assessed with a 2100 Bioanalyzer using the RNA 6000 Pico kit (Agilent, Santa Clara, CA, USA). A total of 30 libraries were prepared by using the NEBNext Ultra Directional RNA Library Prep Kit for Illumina (New England Biolabs Inc., Ipswich, MA, USA). Sequencing (2 × 150 bp) was conducted with an Illumina NextSeq 500 at the Biocenter Oulu Sequencing Centre (Oulu, Finland).

Transcript quantification and abundance matrices construction

We followed the Trinity Post-Transcriptome Assembly Downstream Analyses pipeline (Trinity v.2.6.6) (Haas et al., 2013, https://github.com/trinityrnaseq/trinityrnaseq/wiki/Trinity-Transcript-Quantification) to generate quantification files at isoform level, and raw counts and normalized count matrices at putative-gene level (hereafter referred as gene level matrices). For transcript quantification we used the trimmed reads from Ojeda et al. (2019) and as reference we used the Trinity_guided transcriptome reported in the same work (Data S1). We chose the non-reduced redundancy Trinity_guided transcriptome as reference (instead of the redundancy reduced assembly) to avoid mapping of reads from different paralogous genes to the same contig. P. sylvestris has a large amount of paralogous and repetitive regions, and with this we reduced the amount of false mapping across paralogs due to sequence similarity.

To obtain independent transcript abundance estimates of each of the six individuals in each of the five tissues we used Salmon 0.9.1 (Patro et al., 2017) as implemented in the Trinity pipeline, with the --SS_lib_type (strand specific) and --trinity_mode options. The –trinity_mode option generated a transcript-to-gene map that allowed the estimation of counts from isoforms to generate counts at a putative gene level during the count matrix generation step. Before any further analysis, we checked for the presence of possible contaminants by searching contigs that had hits to the keywords ‘alveolata’, ‘metazoa’, ‘fungi’, ‘bacteria’, and ‘archaea’. We search for exact matches to these keywords from the results of a translated blast (BLASTX) of the transcriptome annotation file (Ojeda et al., 2019; Ojeda, 2020). We then combined our list of putative contaminants with the contaminants and organelles contigs lists reported in Ojeda et al. (2019), and excluded them from the isoform quantification files and the gene_trans_map. Contaminants were removed after the transcript quantification stage to avoid the false mapping of contaminant reads to non-contaminant contigs in the reference transcriptome.

We built three count matrices at the gene level based on the clean independent transcript quantification estimates with the script abundance_estimates_to_matrix.pl from the Trinity pipeline. For this, we generated a gene level raw counts matrix (Table S2), which was then used to construct a transcript per million length normalized gene count matrix (TPM escalated matrix) (Table S3). The TPM escalated matrix accounts for differences in isoform lengths that otherwise could inflate FDR due to differential transcript usage (Soneson, Love & Robinson, 2015). Finally, the TPM escalated matrix was used to construct a gene counts matrix normalized using the Trimmed Mean of M values (TMM) method (Table S4), which accounts for differences in the distribution of transcript expression that could lead to an increase in false positive rates, and decrease the power to detect truly differentially expressed genes (Robinson & Oshlack, 2010). Before doing the differential expression analyses and the estimation of tissue specificity, we evaluated the quality of our samples by doing a principal component analysis (PCA) and a Pearson correlation matrix using the gene raw count matrix, according to the Trinity QC samples and biological replicates pipeline (https://github.com/trinityrnaseq/trinityrnaseq/wiki/QC-Samples-and-Biological-Replicates). The intention of these analyses was to look for the presence of batch effects or sample outliers, and to verify that biological replicates clustered within each tissue type and not among sampled individuals.

Differential expression analysis and identification of tissue preferentially upregulated genes

Differentially expressed genes (DEG) and preferentially up-regulated genes (PUR) were identified using the Trinity Differential Expression and Sample-Specific Expression pipelines (Bryant et al., 2017, https://github.com/trinityrnaseq/trinityrnaseq/wiki/Trinity-Differential-Expression). Briefly, we first identified DEG using the gene raw counts matrix with edgeR 3.28.0 (Robinson, McCarthy & Smyth, 2010; McCarthy, Chen & Smyth, 2012). The differential expression analysis was based on pairwise comparisons of each of the five tissues, using the six samples per tissue as biological replicates, then for each pair of DEG identified we obtained their associated false discovery rate (FDR). Next, we obtained a normalized mean value of expression for each tissue by averaging and log2 transforming the counts for each gene across the six replicates for each tissue on the TMM gene matrix. Afterwards, pairwise comparisons of the averaged log2 counts values per tissue were done and a logFC was assigned to each gene. DEG with a maximum FDR of 0.05 for differential expression, and with positive logFC in each pairwise comparison of the averaged log2 TMM normalized counts was then classified as PUR. A summary of pairwise expression differences between tissues based on the logFC of the log2 transformed gene counts in the TMM matrix is provided in Data S2.

Tissue-specific expression

As an alternative approach to quantitatively assess the tissue-specific expression of the genes we calculated the τ index based on the TMM gene counts matrix. The τ index ranges between 0 for widely expressed genes and 1 for exclusively tissue-specific genes (Yanai et al., 2005). As the τ index considers tissue specificity independently of the level of expression, we set as “not expressed” genes with expression values <1 from our TMM matrix in order to exclude genes with low support for true expression and low signal to noise ratio. To do this, we first log2 transformed the matrix in order to normalize the distribution of the expression values. We set all negative values in the matrix to zero, as this represented values <1 before log2 transformation. We excluded contigs that had no expression values or that had expression in just one out of the 30 samples. Then, the τ index was computed separately for each gene across all tissues and replicates following according to the following equation (Yanai et al., 2005; Kryuchkova-Mostacci & Robinson-Rechavi, 2017, https://github.com/severinEvo/gene_expression/blob/master/tau.R):

Singular enrichment analysis

To further characterize the gene expression in the five tissues, we identified the biological pathways for both tissue-specific and tissue preferentially upregulated gene sets with independent singular enrichment analysis (SEA) (Huang, Sherman & Lempicki, 2009; Du et al., 2010). First, we retrieved the UniProt IDs corresponding to our putative genes from the blastx field from our reference annotation file (Ojeda et al., 2019). Then we uploaded the list of UniProt IDs to the uniprot retrieve/ID mapping tool (https://www.uniprot.org/uploadlists/) and restricted the result to GO terms only. We repeated this procedure with the genes used as a background list for the SEA: all the contigs in the gene raw counts matrix for the PUR genes (Data S3), and all the contigs in the filtered TMM matrix in the case of the tissue-specific genes (Data S4).

Of the 715,398 putative genes in the raw counts matrix used for the differential expression analysis, 17,227 have a unique UniProt ID and represent 108,947 GO terms. The background list for the tissue-specific genes data set consisted of 177,075 contigs of which 14,079 have a unique annotation and represent 90,198 GO terms. For both data sets only uniquely annotated genes and their corresponding GO terms (Data S5–S14) were used for running the singular enrichment analyses to avoid inflating the number of GO terms falsely, and creating a bias in the analysis.

We used the GO terms along the UniProt IDs as input for the SEA using the agriGO (http://systemsbiology.cau.edu.cn/agriGOv2/index.php) platform (Du et al., 2010; Tian et al., 2017). We used the custom background list option, applied a hypergeometric test as statistical test method with a minimum of five mapping entries per term, and Hochberg FDR as multi-test adjustment method with a significance level of 0.05. As the significant enrichment of child terms can inflate the enrichment significance of parental terms, after the SEA we used REVIGO (Supek et al., 2011, http://revigo.irb.hr/) to reduce the redundancy of the GO terms and highlight the unique and non-dispensable terms per tissue for both PUR and tissue-specific genes. We used the list of enriched GO terms found in the SEA and their respective p-value as input, selected the small output setting for redundancy, the whole UniProt as database and SimRel as measure of semantic similarity.

Transcript quantification and abundance matrices construction

We mapped a total of 707,063,773 trimmed and adapter removed reads from five different tissues (embryo, megagametophyte, needle, phloem and vegetative bud) and six biological replicates (six different genotypes) per tissue type to P. sylvestris TRINITY_guided transcriptome (Ojeda et al., 2019). On average 23,568,792 reads originated from each tissue, ranging from 29,591,629 reads for needle to 20,469,80 reads for phloem. On average 76% of the reads per replicate were successfully mapped to the reference (Table S5). After mapping 1,307,500 contigs had aligned reads at the isoform level. Of those, 119,882 contigs were removed from the downstream analyses as they were identified as contaminants (Data S15). The final set consisted of 1,187,460 contigs at isoform level and were used to construct raw counts and normalized matrices at gene level for downstream analyses (see Materials and Methods section). The total number of putative genes with expression signal in the gene level matrices was 715,398, much higher than the number of annotated genes in any conifer (Nystedt et al., 2013; Neale et al., 2014; Gonzalez-Ibeas et al., 2016). This magnitude, albeit probably an overestimate, is typical to transcriptome studies (Little et al., 2016). This is likely a result of single genes being present in multiple fragments, isoforms split into multiple genes, and different alleles originating from heterozygous material identified as separate genes during assembly and classification as genes by Trinity (Ojeda et al., 2019). However, part of the genes originate from gene families and since clustering similar genes is possible in downstream analysis, we chose to err on the side of potentially over splitting the genes rather than imperfectly clustering similar transcripts as a single gene, as over clustering will inherently lead to loss of information. We believe that providing expression data with minimum clustering will be most versatile for later use of the transcriptome and expression data in genome annotations and other studies.

Quality assessment of biological replicates

As we used different genotypes as biological replicates, we first verified that the replicates clustered by tissue type and not by genotype, and checked for the presence of potential outliers in the dataset. We used the raw counts matrix data (Table S2), a principal component analysis (PCA) and a Pearson correlation to verify this. The PCA separated the tissue samples into five distinct clusters without any overlap, indicating that among-tissue variation is the main factor of among-sample variation (Fig. 1). Hence, our approach captures the differentiating gene expression profiles of the five tissues. In the PCA, the seed-derived megagametophyte and embryo samples clustered closest to each other, suggesting similarity in their gene expression profiles. Also phloem and bud samples clustered close to each other, whereas needle samples showed the most unique gene expression profile. In the hierarchical clustering analysis, based on the correlations of gene expression profiles, the differences among tissues are relatively shallow. But, similarly to the PCA, all replicates are clustered according to their tissue type and not according to their genotypes, corroborating the PCA results (Fig. S1).

Tissue preferentially upregulated and tissue-specific gene expression

We defined a gene as tissue PUR when there was a significant log fold change in the expression value compared to the other tissues. To identify tissue PUR we first did a differential expression (DE) analysis. For this we included all the genes in the raw count matrix (Table S2). We decided not to apply any minimum number of counts per gene as a filtering threshold to run the analysis, as we later applied a 5% false discovery rate (FDR) threshold for the identification of PUR genes. Out of the 715,398 genes initially included in the DE analysis, 198,413 genes had a maximum 5% FDR for differential expression and were further included in the analysis to identify PUR genes. We identified a total of 48,001 genes with tissue preferential expression, and out of the five tissues needle has the highest number of PUR genes (Table 1)

Quantification of tissue specificity allows a powerful statistical analysis of correlation between tissue-specific expression and for example, evolutionary rate or other dependent or explanatory variables and factors. We identified the tissue specifically expressed genes by calculating the τ score per gene. The score ranges from zero to one, with a zero given to genes expressed in all tissues and one given to completely tissue specific genes. For this analysis we retained a set of 177,075 genes (Table S6) after applying the filtering criteria described in Methods. We considered a gene as tissue specifically expressed only if its τ = 1. We identified a total of 3,899 genes with a tissue-specific pattern of expression. Similarly, the PUR analysis results, needle has the highest number of tissue-specific genes (Table 1). To obtain the annotation of the genes identified as tissue PUR and tissue specific, we retrieved the corresponding UniProtKB identifiers (Ojeda, 2020) from the Trinotate for the 715,398 putative genes in the TMM count matrix, out of which 97,435 (14%) had a Swiss-Prot (Bairoch & Apweiler, 2000) protein match based on BLASTX (Ojeda et al., 2019). Most of the Swiss-Prot annotations (67%) originated from Arabidopsis thaliana (65,214 genes). Other common annotation sources were Nicotiana tabacum (9,794; 10%) and Oryza sativa (8,946; 9%). Only 1,663 genes (1.7%) had an annotation to other Pinus species, of which 177 (10.6%) were hits to P. sylvestris, and 608 (36.5%) genes had Swiss-Prot annotation to Picea. Note that Swiss-Prot is a manually curated database that does not currently have a comprehensive set of annotated gymnosperm proteins and therefore the best matches are often obtained from the model plants such as A. thaliana. A proportion of our putative genes share the same gene identifier (annotation) (Table 1). This probably reflects the incomplete collapse of different isoforms in the assembled transcriptome used as reference, or the presence of gene families (Wegrzyn et al., 2014). Also, a high number of the genes identified as PUR or tissue specific lack annotation altogether, which is not surprising as genes with higher tissue-specific expression have less conserved sequences and are less likely to find orthologs among other species (Lemos et al., 2005; Raherison et al., 2012). A summary of the 715,398 genes indicating their normalized expression level (TMM), τ score, tissue specificity status, PUR status, and annotation can be found in the Supplementary information (Table S7).

Cursory inspection of annotations of highly expressed tissue PUR and tissue-specific genes are congruent with some of the already known functions of the tissues. These results confirm that our analyses capture biologically meaningful characteristics of the tissues. For example in megagametophytes, enzymes related to seed storage lipid mobilization and germination were upregulated and specifically expressed. Similarly, in needles, several chlorophyll a–b binding proteins are upregulated. In embryo, multiple ribosomal proteins and other proteins indicating active protein synthesis were upregulated. In vegetative buds, expression of genes involved in defense against insect attack, like (−)-alpha-pinene synthase and dirigent (Ralph et al., 2006) that take part in oleoresin synthesis, were highly expressed and specific to this tissue. In phloem, the two genes annotated as metallothionein-like protein EMB30, an aquaporin and a thioredoxin-like protein were highly expressed, similarly to Quercus suber phellem (cork) where metallothionein reacts to oxidative stress (Mir et al., 2004) or in Pinus taeda xylem where the same proteins were among the most highly expressed genes (Lorenz & Dean, 2002).

Among the five tissues analysed, the needle had the highest number of genes with tissue-specific expression and embryo the lowest (Table 1). Except for two genes, one in megagametophyte and one in needle, all the genes with tissue-specific expression were also among the PUR genes. However, as tissue specificity does not require a high expression level, genes with τ score equal to one are not necessarily the most upregulated genes in their respective tissues. Comparison of our findings to other studies is not straightforward as there are very few transcriptomic studies in P. sylvestris. But in comparison to a previous study (Merino et al., 2016), where they focus on the comparison between megagametophyte and embryo tissues at different developmental stages, we identified less megagametophyte and embryo specifically expressed genes. One of the reasons for this difference could be that the identification of unique genes in Merino et al. (2016) was based only on the comparison between embryo and megagametophyte tissues. As the identification of tissue specific genes is contingent to the number of tissues used for the analysis, it is expected that the higher the number of tissues used in the comparison, the lower the number of tissue specific genes that will be identified. In contrast, we found a higher number of tissue specific genes in embryo, bud and needle compared to a previous study in conifers (Raherison et al., 2012), where several tissue types were used. One notable difference between this (Raherison et al., 2012) and ours was the higher number of tissue-specific genes for megagametophyte found in P. glauca. Raherison et al. (2012) found the highest number of unique genes in the megagametophyte in comparison to other tissues analyzed. The low number of megagametophyte specific genes identified in our study could be due to the use of mature embryos as starting material as previous research suggests that the number of unique transcripts in the megagametophyte varies during the developmental stages of embryogenesis (Merino et al., 2016).

One caveat of our analyses is that, unlike other studies, we did not use microdissection in order to obtain the tissue samples (Cañas et al., 2017). Hence, some of the “tissues” are a mix of tissue types. Needles, for example, include several tissues (phloem among them) (Pongrac et al., 2019), and mature embryos contain the shoot and root meristems as well as cotyledons (Singh, 1978). In contrast, the mature megagametophyte is a quite uniform storage tissue consisting of cells packed with starch protein and lipids (Simola, 1974; Vuosku et al., 2015). Another limitation of the dataset is that it represents only one point in time and space, although gene expression is a dynamic process and quantitative and qualitative variations exist over spatial and temporal scales. Instead of sampling across several developmental stages or across a spatial gradient our dataset represents a wider set of tissues, which increases the power to identify tissue PUR and tissue specifically expressed genes. The added value of the dataset lies in the unexpected functions and connections discovered among biological pathways and genes with previously unidentified signals of tissue-specificity or up-regulation.

Functional characterization of tissue preferentially upregulated and tissue-specific genes

GO enrichment analysis allows the identification of gene functions enriched with certain functional roles. The number of enriched functions was of the same magnitude across tissue types, ranging from 253 to 452 for PUR genes and from 58 to 169 for tissue-specific genes (Tables S8–S17). The total number of GO terms, the number of significant enriched terms, and the number of terms after the reduction of redundancy are shown in Table 2. Most of the genes (86%) with expression signals in our study lacked annotation from the Trinotate pipeline. Thus, they did not contribute to functional analysis or GO enrichment results. A summary of the most highly expressed genes per tissue, and the most enriched, non-redundant GO terms in the biological processes category are shown in Fig. 2. The complete lists of gene identifiers and their corresponding GO terms per tissue and per each set of genes (Data S5–S14), along with tables with the results of the SEA showing each GO terms, its p-value, and FDR (Tables S8–S17), and lists with levels of uniqueness or indispensability for each significantly enriched term in the five tissues (Data S16) are provided in supplementary information.

In needles the significant GO terms reflected the exposure of trees to various stresses and interactions with other organisms, whereas in embryos, buds and the phloem the GO terms were mainly connected to different development-related processes. In needles the enriched biological process GO terms among tissue-specific genes were related to immune response (GO:0006955) as well as response to stress (GO:0006950) and other organisms (GO:0051707) such as oomycetes (GO:0002229), bacteria (GO:0042742) and fungi (GO:0009817). Moreover, terpene synthase activity (GO:0010333), which may play a key role in the defense against herbivores (Achotegui-Castells et al., 2013), was an enriched molecular function among tissue-specific genes in needles, but also in embryos and vegetative buds. For example, reactive oxygen species (ROS) related biological processes (GO:0006800 and GO:0042743, GO:0034614) and molecular functions (GO:0004601, GO:0004364) were enriched among the GO terms in the tissue-specific genes of embryos, which is consistent with an active ROS protection in developing tissues. In the phloem, a special differentiation process, syncytium formation (GO:0006949), indicating the interconnection of phloem sieve elements to generate a transport route (Geldner, 2014) was an enriched biological process among the tissue specific genes.

Megagametophyte-specific genes have crucial functions in seed germination and energy conversion

Gymnosperms are characterized by the haploid female gametophyte tissue, the megagametophyte, which surrounds the embryo in developing and mature seeds. The megagametophyte can be considered a functional homolog of the endosperm in angiosperms due to its role as a nourishing tissue (King & Gifford, 1997; Costa, Gutièrrez-Marcos & Dickinson, 2004). However, the megagametophyte develops from a haploid megaspore before the fertilization (Singh, 1978) and is therefore entirely maternally inherited unlike the diploid or triploid endosperms of biparental origin (Williams & Friedman, 2002; Baroux, Spillane & Grossniklaus, 2002). To give an example of the potential uses of the dataset, we provide a more detailed description of the megagametophyte expression profile, but leave the in-depth analysis of the other tissues for later investigations.

Among highly expressed and up-regulated genes in the megagametophyte were malate synthase (EC 2.3.3.9) and isocitrate lyase (EC 4.1.3.1) that are essential in glyoxylate cycle converting lipids into carbohydrates in seeds (Ching, 1970), as well as other glyoxysomal proteins like Acetyl-CoA acyltransferase (EC 2.3.1.16), ABC transporter and peroxisomal fatty acid beta-oxidation multifunctional protein AIM1 (Graham, 2008). Seed storage related genes such as 2S seed storage-like protein, 11S globulin seed storage protein 2 and 13S globulin basic chain and some isocitrate lyase copies were completely megagametophyte specific (τ = 1). Antimicrobial and antifungal protein coding genes were the most highly expressed among annotated megagametophyte-upregulated genes.

The enriched GO terms of biological processes and molecular functions in the megagametophyte tissue-specific genes included seed germination and the mobilization of nutrient reserves. Nutrient reservoir activity (GO:0045735) indicated the mobilization of energy sources from the megagametophyte for seed germination and early seedling growth, as well as lipid catabolic processes (e.g. GO:0016042, GO:0044242). Malate dehydrogenase activity (GO:0016615) and heme binding (GO:0020037), which mostly originated from the cytochrome P450 enzymes containing heme cofactors (Xu, Wang & Guo, 2015), reflected the resume of active metabolism. Also, response to ROS (GO:0034614) and antioxidant activity (GO:0016209) suggested active metabolism and signaling. ROS are natural by-products of metabolism and may be detrimental to seed viability because they can cause oxidative stress. However, in the seed ROS also work as signals which underpin the breaking of dormancy and provide protection against pathogens (Jeevan Kumar et al., 2015). Megagametophyte cells showed responses to hormone stimulus (GO:0032870) and the function of hormone-mediated signaling pathways (GO:0009755) including abscisic acid (GO:0009738), auxin (GO:0009734) and ethylene (GO:0009873) which also belong to the molecular networks regulating seed dormancy and germination (Seo et al., 2009; Guangwu & Xuwen, 2014; Miransari & Smith, 2014; Shu et al., 2016). Cellulose biosynthetic process (GO:0030244) and primary cell wall biogenesis (GO:0009833) suggest that cell walls in the megagametophyte may participate in water retention and give mechanical support to the germinating embryo (Otegui, 2007). Similarly to previous findings in P. sylvestris (Merino et al., 2016) megagametophytes, we found enrichment for processes involved in the response to chemical and endogenous stimuli (GO:0042221, GO:0071495 ). Merino et al. (2016) suggested that the megagametophyte could also be involved in the regulation of the embryo development through the induction of signaling pathways triggered by sensing environmental signals in a similar way the angiosperms’ endosperm does (Yan et al., 2014). Altogether, our findings show that the megagametophyte is not just a reserve nutrition for the germinating embryo, but a metabolically active tissue contributing in multiple ways to seed germination and, thus, underline the importance of the haploid stage in P. sylvestris life cycle.

Several enzymes widely used in allozyme-based population genetic studies ((Szmidt & Muona, 1989) and references therein) such as aconitate hydratase (EC 4.2.1.3), malate dehydrogenase (EC 1.1.1.37) and aspartate aminotransferase (EC 2.6.1.1) were megagametophyte-specific and among the top 50 expressed genes in the tissue. As they may be more prone to natural selection against recessive deleterious variants when expressed at the haploid stage, early population genetic analyses may have bias in for example estimates of the overall genetic diversity based on these loci as highly expressed genes are known to be under strong purifying selection.

Supplementary information

All the supplementary files can be found at DOI 10.6084/m9.figshare.c.5264255.v1.