To collect research papers containing information on TF–target interactions, we searched articles in PubMed with the queries “(nidulans[Title/Abstract]) AND transcription [Title/Abstract],” “(crassa[Title/Abstract]) AND transcription [Title/Abstract]” and their variants (e.g., replacing “transcription” by “transcriptional”). In addition, we retrieved the lists of predicted TFs in the two species from FTFD, a database computationally predicting fungal TFs at genome scale (Park et al., 2008). For A. nidulans, the curated papers associated with each TF in the database AspGD (Cerqueira et al., 2014) were also collected. For N. crassa, we directly searched articles in PubMed with the names of characterized TFs. In total, more than 1,000 papers were collected and checked for the presence of information on TF–target interactions. It should be noticed that new research papers providing TF–target interactions (e.g., Virgilio et al., 2017) have been continuously published during the preparation of this manuscript, which need to be collected in the future.

For each regulatory interaction, the ID and name (if applicable) of TFs and targets, the type of experimental evidence, the regulatory function (repression or activation), and the PubMed ID of reference literature were recorded. Perturbations of TFs through gene deletion, knock-down of gene expression, protein inactivation, or similar manners were recorded as “TFdown” manipulations, while those through gene overexpression or introduction of gain-of-function mutants were classified into “TFup” manipulations. The names and IDs of proteins were completed and occasionally corrected according to the protein sequences and annotations in AspGD (A. nidulans version s10-m04-r06) and BROAD ftp site (N. crassa version 12, http://www.broadinstitute.org/ftp/pub/annotation/fungi/neurospora_crassa/). Protein IDs in N. crassa were also linked to the accession numbers in GenBank (BioProject PRJNA132). In TF-perturbation studies, the gene encoding the perturbed TF is often included in the differentially-expressed gene set in transcriptome data, but should not be considered as the target of its product. The cultivation conditions and strengths of regulatory interactions, although important for the understanding of transcriptional regulations (Geistlinger et al., 2013), were not curated due to the difficulty in developing standard formats. The process of data collection and curation was summarized in Figure 1A.

For the curated regulatory interactions supported by low-throughput experiments, the IDs of TFs, their targets, and the type of supporting experiments (“Binding,” “TF-perturbation,” or “Both”) were saved in a tab-delimited text file. The file was imported to Cytoscape 3.5.1 (Shannon et al., 2003), where the columns containing TF IDs, target IDs and experimental types were selected. The Cytoscape built-in plugin NetworkAnalyzer (Assenov et al., 2008) was subsequently applied to calculate the parameters of the networks (treated as directed edges) using default settings. The average number of neighbors was shown in the tab “Simple Network Parameters” of the analysis result.

Putative orthologous proteins between pairs of fungal species were acquired from the Fungal Orthogroups Repository (https://portals.broadinstitute.org/regev/orthogroups/). The Saccharomyces cerevisiae protein IDs were searched in the YEASTRACT database (Teixeira et al., 2014) to check if there are documented regulatory relationships using the “Search for Associations” tool with default parameters.

Our manual curation of TF–target interactions resulted in data associated with 81 and 66 TFs in A. nidulans and N. crassa, respectively (Table 1 for basic statistics, all interactions listed in Table S1). Considering that a 608 and 459 TFs were predicted in the two species, respectively (Park et al., 2008), only a small fraction of TFs were covered. The result suggests our knowledge about the transcriptional regulatory networks in filamentous fungi lags far behind other model organisms like S. cerevisiae, in which almost all the TFs have been identified for their targets and about 163,000 regulatory interactions have been curated (Teixeira et al., 2014). Compared with the status in some “non-model” organisms such as pathogenic bacteria Pseudomonas aeruginosa (Galán-Vásquez et al., 2011) and Mycobacterium tuberculosis (Guo et al., 2009), the fractions of TFs known for targets in the two fungal species were also lower. Most of the interactions had one experimental evidence, while some others were supported by more than 10 records (Figure S1), revealing research preferences for specific regulatory interactions. As an example, regulation of alcohol dehydrogenase gene alcA by AlcR in A. nidulans was supported by 24 experiments from 20 papers, all published before the year 2005.

The experimental evidences for the regulatory interactions were classified into two types. “Binding” evidences included those obtained with chromatin immuneprecipitation (ChIP), electrophoresis mobility shift assay (EMSA), DNA footprinting, and other methods detecting physical binding of TFs to target DNA. Some of the literature reported the exact binding sites of TFs, which were not covered in this study but are worth curating in the future. Interactions with “TF perturbation” evidences indicated those detected through the identification of differentially expressed genes after perturbing the abundance or activity of TFs. In total, 161 and 704 interactions were supported by both types of evidence in A. nidulans and N. crassa, respectively (Figure 1B). That is to say, these interactions involve direct binding of TFs to their target promoters which result in functional regulation of the transcription of the corresponding genes. For example, AreA, a nitrogen-responsive TF in A. nidulans, activates or represses seven of its targets by directly binding to their promoters (Figure 1C).

In addition, the experimental evidences were classified according to the throughput of methods. Microarray- or deep sequencing-based methods (e.g., ChIP-chip and RNA-seq) allow genome-wide mapping of the targets of TFs, while the confidences of identified targets are not necessarily high. In contrast, classical low-throughput studies (e.g., EMSA and Northern blot) only illustrate interactions between individual TFs and targets, but the information are generally of high confidence. Not surprisingly, only 6.7 and 2.9% of all the interactions had evidences from low-throughput studies in A. nidulans and N. crassa, respectively (Figure 1B). Particularly, a total of 20,118 interactions in N. crassa were supported by high-throughput experiments reported in just 27 reference papers.

The regulatory interactions supported by low-throughput experiments were then used for the bottom-up reconstruction of transcriptional regulatory networks. The obtained networks of A. nidulans and N. crassa containing only high-confidence interactions were composed of 81 and 58 TFs, 675 and 602 TF–target interactions, respectively (Figure 2).

The networks recaptured some already established regulatory models. For example, a sub-network in A. nidulans clearly revealed the complex regulatory relationships between eight TFs involved in conidiation, with BrlA playing a central role in the regulation (Figure S3, see review by Park and Yu, 2012).

Both regulatory networks were well connected, with only six and five connected components (i.e., sub-network within which any two nodes are connected to each other by paths if considering the network as an undirected graph) detected in A. nidulans and N. crassa, respectively (Figure 2). The nodes in the networks had 3,462 and 3,451 neighbors in average, respectively, in the two species, as shown in the tab “Simple Network Parameters” of the result of NetworkAnalyzer. Specifically, we looked into regulatory interactions involved in four kinds of network motifs (i.e., small set of regulatory associations, Alon, 2007).

Previous genome-scale analysis of the presence of TFs in fungi indicated that only a small part of them is evolutionarily conserved in the filamentous ascomycetes (Todd et al., 2014). For example, the above-mentioned AlcR and BrlA, which are well-studied in A. nidulans, do not have an ortholog in N. crassa. GalR, an activator of galactose utilizing genes, was only found in A. nidulans but no other fungi (Christensen et al., 2011). Further, even for orthologous TFs, their regulatory targets and consequent functions can vary largely between species (Lavoie et al., 2009). Here, the curated transcriptional regulatory interactions were compared between A. nidulans and N. crassa based on protein orthology.

A total of 5,842 putative orthologous relationships were identified between proteins in the two species. These include 27 pairs of TFs collected for targets in both species in this study. Of the 675 regulatory interactions supported by low-throughput experiments in A. nidulans, 371 had putative orthologs for both TF and target in N. crassa. However, only 40 interactions involving 18 TFs were evidenced by low-throughput experiments in both species (Table 2). Interestingly, most of the target genes associated to these shared interactions encode metabolic enzymes, including those in nitrogen metabolism, plant cell wall-polysaccharide degradation and cell redox homeostasis. This result could be linked to the relatively high conservation of metabolic pathways among biological processes (Peregrín-Alvarez et al., 2009). Some of these interactions were supported by different types of experiments in the two species, thus providing complementary knowledge for the regulations. For example, the TF ClrB in A. nidulans might directly binds to the promoter of the gene encoding the cellulase CbhB to activate its expression (Coradetti et al., 2012), as suggested by the interaction between CLR-2 and cbh-1 (putative orthologs of ClrB and cbhB, respectively) in N. crassa (Craig et al., 2015).

In addition, some regulatory interactions supported only by high-throughput experiments in one species were corroborated by low-throughput experiments on putative orthologous TFs and targets in the other species. These include 20 and 6 interactions (supported only by high-throughput experiments) in A. nidulans and N. crassa, respectively (Table S3). As an example, regulation of clock-controlled gene ccg-1 by ATF-1 was supported by EMSA and Northern blot in N. crassa (Yamashita et al., 2008), and the orthologous regulation was also reported in A. nidulans through microarray analysis of wild-type and TF-deletion mutant (Emri et al., 2015), suggesting conserved regulatory mechanisms between the species.

Most of the transcriptional regulatory interactions curated in this study were only described in one of the two species but not in the other, which could be due to the research bias in the selection of TFs and targets, and the quick evolution of regulatory networks. Considering that A. nidulans and N. crassa belonging to Eurotiomycetes and Sordariomycetes, respectively, diverged about 300 million years ago (Hyde et al., 2017), it is not surprising that only a small fraction of regulatory interactions were identified as conserved between the two species. We then looked for the orthologs of TFs and targets involved in the above-mentioned 66 regulatory interactions (supported by low-throughput experiments in at least one of the two species) in the more distant S. cerevisiae. Twelve of the interactions also existed in S. cerevisiae according to the records in YEASTRACT database (Teixeira et al., 2014; Table S3). For example, A. nidulans AcuM, N. crassa AOD-2, and S. cerevisiae Rds2, which are orthologous proteins, all regulate genes in gluconeogenesis. While the regulation of PCK1 and FBP1 by Rds2 in S. cerevisiae is conserved in A. nidulans and N. crassa, AcuM/AOD-2 expand their functions to the control of alternative oxidase gene aodA/AOD-1, which is not present in S. cerevisiae. For Rim101 (Peñalva et al., 2008) and Gcn4 (Hoffmann et al., 2001) as pH and amino acid starvation responsive regulators, respectively, in S. cerevisiae, their auto-regulation was detected in all three species. On the other hand, this comparison indicated remarkable rearrangement of regulatory networks during evolution (Table S3). For example, although XKS1 and GRE1 involved in xylose metabolism in S. cerevisiae have their orthologs in A. nidulans and N. crassa, their regulators in the latter two species (XlnR/XLR-1) could not be found in S. cerevisiae. Taken together, the results suggested that a conserved TF could switch its regulon, and a new TF could arise to regulate conserved genes, during the evolution of fungi (Lavoie et al., 2009; Vaquerizas et al., 2009). The possible evolutionary events resulting in the diverse transcriptional regulatory networks are expected to be elucidated by comparing the regulatory interactions in more fungal species.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer BF and handling Editor declared their shared affiliation.