Summary
The CCCTC-binding factor (CTCF) is widely regarded as a key player in chromosome organization in mammalian cells, yet direct assessment of the impact of loss of CTCF on genome architecture has been difficult due to its essential role in cell proliferation and early embryogenesis. Here, using auxin-inducible degron techniques to acutely deplete CTCF in mouse embryonic stem cells, we show that cell growth is severely slowed yet chromatin organization remains largely intact after loss of CTCF. Depletion of CTCF reduces interactions between chromatin loop anchors, diminishes occupancy of cohesin complex genome-wide, and slightly weakens topologically associating domain (TAD) structure, but the active and inactive chromatin compartments are maintained and the vast majority of TAD boundaries persist. Furthermore, transcriptional regulation and histone marks associated with enhancers are broadly unchanged upon CTCF depletion. Our results suggest CTCF-independent mechanisms in maintenance of chromatin organization.
INTRODUCTION
Mammalian chromosomes reside in separate chromosome territories in interphase nuclei and are partitioned into topologically associating domains (TADs) characterized by strong intra-domain interactions and comparatively weak inter-domain contacts (Dixon et al., 2012; Nora et al., 2012). TADs are generally invariant across different cell types and evolutionarily conserved in related species (Dixon et al., 2012; Schmitt et al., 2016; Vietri Rudan et al., 2015). Cell-type specific chromatin interactions between distal cis regulatory elements and their target genes are largely constrained by the TADs. Some TADs are further divided into structures such as sub-TADs and insulated neighborhoods (Hnisz et al., 2016; Phillips-Cremins et al., 2013). TADs have been shown to allow enhancers to act on genes within the same domain (Symmons et al., 2016; Symmons et al., 2014), providing a mechanism for long-range control by distal regulatory elements (Dixon et al., 2016; Schwarzer and Spitz, 2014). Disruption of TADs and sub-domains has been shown to result in ectopic enhancer/promoter interactions and altered gene expression programs in cancers (Flavahan et al., 2016; Hnisz et al., 2016) and developmental disorders (Andrey et al., 2013; Franke et al., 2016; Lupianez et al., 2015). Understanding the molecular mechanisms that establish and maintain TADs and other features of chromatin organization will have significant implications in the study of a wide variety of human diseases.
The CCCTC-binding factor (CTCF) is widely believed to play a critical role in genome organization in bilaterian animals (Dekker and Mirny, 2016; Denker and de Laat, 2016; Ghirlando and Felsenfeld, 2016; Hnisz et al., 2016; Ong and Corces, 2014; Vietri Rudan and Hadjur, 2015). This ubiquitously expressed DNA binding protein is associated with a large number of genomic regions via a consensus sequence in the mammalian genome (Kim et al., 2007; Schmidt et al., 2012; Vietri Rudan and Hadjur, 2015; Xie et al., 2007). The CTCF binding sites are highly enriched at boundaries of TADs, sub-TADs, insulated neighborhoods, and at loop anchors (Dixon et al., 2012; Hnisz et al., 2016; Phillips-Cremins et al., 2013; Rao et al., 2014). Furthermore, the CTCF binding motifs at loop anchors and TAD boundaries preferentially adopt a convergent orientation, highlighting a role for the protein in the formation of TADs and chromatin loops (Guo et al., 2015; Rao et al., 2014). In support of this model, genomic deletions encompassing TAD boundaries which contain CTCF binding sites lead to a gain of interactions between neighboring TADs (Lupianez et al., 2015; Narendra et al., 2016; Narendra et al., 2015; Nora et al., 2012). Targeted disruption of specific CTCF binding motifs alters chromatin loops (de Wit et al., 2015; Guo et al., 2015; Narendra et al., 2016; Narendra et al., 2015; Sanborn et al., 2015). Moreover, knockdown of CTCF using RNA interference leads to an increase in contacts across TAD boundaries, and a reduction in contacts between putative CTCF-bound loop anchors (Zuin et al., 2014). Finally, comparison of chromatin domain organization in four mammalian species has revealed a close association between divergent CTCF binding sites and the gain or loss of TAD domains during evolution, further highlighting a conserved role of CTCF in genome organization (Vietri Rudan et al., 2015).
Despite the weight of evidence linking CTCF to TADs and chromatin loops, how CTCF shapes chromatin structure remains incompletely understood. The prevailing model postulates that CTCF acts together with the Cohesin complex, a ring-shaped structure consisting of Smc1, Smc3, Rad21 and SA1/2 subunits that mediate cohesion of sister chromatids during mitosis (Dekker and Mirny, 2016; Denker and de Laat, 2016; Ghirlando and Felsenfeld, 2016; Hnisz et al., 2016; Merkenschlager and Odom, 2013; Nativio et al., 2009; Ong and Corces, 2014; Vietri Rudan and Hadjur, 2015). This model is supported by the observations that the Cohesin complex nearly always co-localizes with CTCF in the genome (Hadjur et al., 2009; Parelho et al., 2008; Stedman et al., 2008; Wendt et al., 2008), that CTCF can physically associate with Cohesin (Xiao et al., 2011), that association of Cohesin with DNA depends on CTCF (Wendt et al., 2008), and that depletion of Cohesin leads to loss of long-range chromatin interactions (Nativio et al., 2009; Seitan et al., 2013; Sofueva et al., 2013; Zuin et al., 2014). The preferential convergent orientation of CTCF binding sites at loop anchors has led to the “extrusion” model in which a pair of tethered Cohesin complexes traversing the chromatin fiber in opposite directions pause at CTCF binding sites, leading to extrusion of the intervening DNA to form a chromatin loop and presumably chromatin domains (Fudenberg et al., 2016; Sanborn et al., 2015).
While the above model can explain the phenotypic consequences after disruption of CTCF binding motifs or inversion of CTCF binding sites (de Wit et al., 2015; Guo et al., 2015; Narendra et al., 2016; Narendra et al., 2015; Sanborn et al., 2015), many exceptions to the model have been reported. Notably, there is not a straightforward correspondence between the presence of CTCF binding sites and formation of a TAD boundary. In fact, roughly 15% of all TAD boundaries show no evidence of CTCF binding (Dixon et al., 2012). Further, nearly a third of chromatin loops anchored on CTCF binding sites show tandem CTCF motifs arrangement (Tang et al., 2015). In addition, although CTCF knockdown increases the frequency of contacts between neighboring TADs, overall TAD structure remains largely unchanged in these experiments (Zuin et al., 2014). The effects of CTCF knockdown become clear only after careful quantitative analysis. In these experiments RNA interference achieved about 80% depletion of CTCF protein levels, and the mild effect on chromatin organization was attributed to incomplete loss of CTCF protein after knockdown. Thus, a major question about the role of CTCF in chromatin organization remains unanswered: what happens to TADs and chromatin loops when cells are fully depleted of the CTCF protein?
Numerous attempts have been made to knock out the CTCF gene in mammalian cells, but the early embryonic lethality of CTCF null animals or growth inhibition in conditional knockout mouse ES cells have hindered further studies (Heath et al., 2008; Sleutels et al., 2012). In order to study chromatin organization after loss of CTCF, we employed the auxin-inducible degron (AID) system to acutely and nearly completely deplete the CTCF protein (Holland et al., 2012; Nishimura et al., 2009; Nishimura and Kanemaki, 2014). To our surprise, we found that the TAD structure is largely maintained after CTCF depletion. There is mild overall reduction of chromatin interaction frequencies, loss of Cohesin complex occupancy, and weakening of the strengths of TAD boundaries, but a large majority of TAD boundaries is preserved. The active and inactive chromatin compartments are also unaltered in the absence of CTCF, along with transcription profiles and histone modifications. Interestingly, TADs in the Lamina Associated Domains (LADs) exhibit a higher degree of weakening than in non-LAD regions. These results provide new insights into mechanisms that regulate chromatin architecture in mammalian cells.
RESULTS
Removal of CTCF binding on the genome by the auxin-inducible degron system
We genetically engineered a hybrid mouse embryonic stem cell (mESC) line F123 (129/CAST) (Gribnau et al., 2003) by inserting a sequence encoding the AID domain into 3’ end of the endogenous Ctcf gene using CRISPR/Cas9 and microhomology-mediated end-joining techniques (Nakade et al., 2014). The resulting cell clones were further engineered to stably express Tir1, which binds to CTCF-AID upon auxin exposure to induce rapid degradation of the CTCF-AID protein (Figure 1A: development of CTCF-AID mESCs). In this system, degradation of protein can be observed within few hours, and we confirmed the virtually complete depletion of CTCF within 24 hours of auxin exposure by Western blot (Figure 1B). The mESCs with AID-tagged CTCF can expand normally and auxin itself has no toxicity (Figure S1A). We observed that CTCF depleted cells could form colonies and were viable, but their growth was severely retarded (Figure 1C and Figure S1B). To ensure that the cells have enough time to go through several cell cycles without CTCF, we investigated the CTCF depleted cells for up to 4 days after auxin treatment. Additionally, to overcome potential clonal variations, we carried out in-depth molecular characterization using two independently derived cell clones.
We first performed CTCF ChIP-seq in each cell clone to confirm that CTCF occupancy is lost after auxin treatment at each time point (Figure 1D and 1E). The distribution of CTCF (p-value < 1 }10−5) in untreated CTCF-AID cells resembles that of normal ES cells (Shen et al., 2012), indicating that AID fusion does not affect its DNA binding (Figure S1C). The number of peaks dramatically decreased to less than 7% and 2 % of that in untreated CTCF-AID cells after 24 hours of auxin treatment in clone 1 and 2, respectively, and reached the minimum of less than 1% in both clones by 48 hours of auxin treatment (Figure 1F and Figure S1D-E). The remaining CTCF peaks in the CTCF depleted cells also exhibited significantly decreased enrichment levels. These results indicate that CTCF is nearly completely removed from the genome after 24 hours of auxin treatment and the penetrant depletion efficiency was maintained in both clones.
CTCF depleted cells grew significantly slower than untreated cells, but otherwise exhibited normal cycling behavior (Figure S1F), in contrast to previous observations of cell cycle arrest in T cells (Heath et al., 2008) and inability to proliferate of mouse ES cells after complete KO (Sleutels et al., 2012). Interestingly, the two CTCF-AID cell clones exhibit different growth rate both prior to and after CTCF depletion (Figure 1F). Nevertheless, judging from the Western blots and ChIP-seq analysis results, severe enough depletion of CTCF was achieved in both clones that allowed us to investigate the contribution of CTCF to chromatin architecture on a genome-wide scale.
TADs and compartments are largely maintained upon CTCF depletion
To examine the effect of CTCF depletion on chromatin organization, we performed in situ Hi-C experiments (Rao et al. 2014) with cells treated with auxin to trigger CTCF degradation for 0, 24, 48, and 96 hours. Hi-C reveals regions of genomic DNA that are in close spatial proximity in a genome-wide fashion using high-throughput sequencing. Surprisingly, TAD structure was maintained for several days after CTCF depletion and genome-wide loss of CTCF binding (Figure 2A). The vast majority of TAD boundaries persisted days after the near-complete loss of CTCF protein (Figure 2B), and the frequency of interaction within TADs (intra-TAD) as well as interaction across TADs (inter-TAD) (Krijger et al., 2016) was also preserved (Figure 2C). We did observe a slight weakening of TADs after CTCF depletion as evidenced by blurred edges in Hi-C heatmaps (Figure 2A), marginal decrease of insulation score (Figure 2D), increased interaction frequency across TAD boundaries (Figure S2A), and reduced Directionality Index (DI) adjacent to TAD boundaries (Figures 2E and S2B). However, despite this quantitative weakening of TADs, the degree to which TADs remain intact in the absence of CTCF is unexpected. These data demonstrate that CTCF is not strictly required for the maintenance of TADs.
We next investigated whether the genome-wide loss of CTCF occupancy affects the higher order genome structure by principal component analysis of the Hi-C datasets, which reveals spatial segregation in compartments A and B harboring active and inactive chromatin, respectively (Lieberman-Aiden et al., 2009). Consistent with the above finding of no significant difference in the TAD structure (Figure 2A), the distributions of compartments A and B were remarkably similar before and after CTCF depletion in every time point (Figure S2C & S2D).
Previous studies have suggested that CTCF has critical roles in organizing genome (Ghirlando and Felsenfeld, 2016; Ong and Corces, 2014). Surprisingly, our results showed that the genome segmentations in TADs and compartment A/B were generally maintained in the absence of CTCF.
Chromatin loops are weakened but persist in CTCF depleted cells
Previous studies have identified roughly 10,000 chromatin loops in various mammalian cell types, and a majority of them are anchored on CTCF binding sites preferentially organized in a head-to-head orientation (Guo et al., 2015; Rao et al., 2014; Sanborn et al., 2015). These CTCF-anchored chromatin loops have been shown to depend on CTCF binding sites, as deletion of the CTCF binding sites disrupts the loops (Narendra et al., 2016; Narendra et al., 2015; Sanborn et al., 2015). To examine the impact of CTCF depletion on chromatin loops we first used HICCUPS (Rao et al., 2014) to identify 4284 chromatin loops in untreated mESCs (FDR < 0.1%). As reported before, the majority of them (3199 peaks) were anchored at CTCF binding sites on both sides (Rao et al., 2014). We then performed Aggregate Peak Analysis (APA) (Rao et al., 2014) on the set of CTCF-anchored long-range loops (genomic distance > 100 kb) to compare the strength of these loops before and after CTCF depletion. The aggregated interaction frequency between loop anchors was noticeably reduced after auxin treatment but not completely abolished (Figure 3A). Surprisingly, the loop strength appeared to be regained after 96-hour treatment, when the number of CTCF peaks is only 1-2% of the initial number of peaks (Figure 3A). These results indicate that, while CTCF may be required for the robustness of chromatin loops, additional factors are likely involved in maintenance of the chromatin loops.
To confirm persistence of long-range chromatin interactions after CTCF depletion, we further investigated interactions around the Sox2 gene using 4C-seq (Simonis et al., 2009). In mESC, Sox2 is regulated by a super-enhancer located ~ 130 kb downstream of the gene (Li et al., 2014; Zhou et al., 2014). CTCF binding could be detected at the super-enhancer, and it was lost after the auxin treatment. The 4C-seq experiment showed that there is robust interaction between Sox2 gene and the super-enhancer in untreated mESC, and the interaction persisted after days of auxin treatment (Figure S3A), supporting the idea that a CTCF-independent mechanism maintains the chromatin interactions between Sox2 and its downstream enhancer.
Recently, we reported frequently interacting regions (FIREs) as a feature of chromatin organization that are characterized by unusually high levels of local chromatin interactions and enrichment for active enhancers (Schmitt et al., 2016). We analyzed FIREs at each time point of auxin treatment (Figure 3B and Figure S3B). As for the case of TADs and compartments, the distribution and the number of FIREs were largely unchanged in CTCF depleted cells. The observations that FIREs are maintained after CTCF depletion are consistent with the results of the 4C-seq that revealed preservation of chromatin interactions between the Sox2 gene and its downstream enhancer.
Cohesin complex remains on chromatin temporarily after CTCF depletion
To explore the factors that may maintain TADs and chromatin loops in the absence of CTCF, we investigated the genomic distribution of Cohesin complex before and after CTCF depletion. Consistent with previous reports, ChIP-seq analysis of the Cohesin subunit Rad21 revealed a striking reduction of Cohesin binding genome-wide upon CTCF depletion (Figure 4A-B). Surprisingly, the rate at which Cohesin binding was lost was substantially slower than the loss of CTCF, and this difference was most apparent in Clone 1 (Figure 4B). The majority of Rad21 peaks in Clone 1 were still preserved after 24 hours of auxin treatment even though CTCF peaks had already been nearly completely lost. Rad21 occupancy then decreased progressively and reached the lowest number at the 96-hour time point. In Clone 2, the removal of Rad21 peaks was more rapid, with 95% of Rad21 occupancy lost after 24 hours of Auxin treatment (Figure 4C). The difference between the two clones might be caused by the different speeds of cell division (Figure 1C). These observations indicate that while the Cohesin complex requires CTCF for its genomic localization, as previously reported, it can remain associated with chromatin, albeit temporarily, after the removal CTCF protein. Since Cohesin occupancy is eventually lost from most TAD boundaries (Figure S4A), it is unlikely that Cohesin is responsible for maintaining TADs in the absence of CTCF (Figure S4B-C).
TAD structure within the Lamina-associated domains is sensitive to CTCF loss
A fraction of the genome is positioned near the nuclear membrane through association with the nuclear lamina and these genomic regions are referred to as lamina-associated domains (LADs) (Yanez-Cuna and van Steensel, 2017). LADs have been mapped in various mammalian cell types, and they generally correlate with inert transcription state and late replication timing (Gonzalez-Sandoval and Gasser, 2016; Yanez-Cuna and van Steensel, 2017). The constitutive LADs (cLADs) shared by all tested cell types are highly conserved in mouse and human and contain few genes (Meuleman et al., 2013; Peric-Hupkes and van Steensel, 2010). LADs and TADs often share the same boundaries (Dixon et al., 2012) and are also enriched for CTCF binding sites (Guelen et al., 2008; Meuleman et al., 2013). To investigate whether chromatin organization in cLADs are affected by loss of CTCF, we inspect the TAD structure within cLADs in the mouse ES cells. Interestingly, we found that TADs within these regions were frequently disrupted, in contrast to the TADs located outside LADs (Figure 4D). The insulation scores of the TAD boundaries in cLAD are more severely affected by loss of CTCF than TAD boundaries outside the LADs in both clones (Figure 4E). There was a small difference between the two cell clones with regard to the disruption of TADs by loss of CTCF. Clone 2 exhibited a more dramatic reduction of insulation score in TADs within cLADs than clone 1. This might be due to the faster growth rate of clone 2, which undergoes more cell divisions than clone 1 during 48 hours of auxin treatment (Figure 1C). Nevertheless, in both clones, the insulation scores at TAD boundaries outside the LAD regions are essentially unaffected by CTCF loss (Figure 4E). This result suggests that TAD structure in the LADs and outside LADs are differentially maintained, with the former likely more dependent on CTCF and the Cohesin complex than the latter.
Gene expression profiles and chromatin state remain largely unaltered after CTCF depletion
In order to determine whether re-organization of TADs in the LADs leads to changes in gene expression, we performed RNA-seq with both clones before and 24hrs, 48hrs or 96 hrs after auxin treatment. Surprisingly, we identified fewer than 10 genes that were differentially expressed after 24 or 48 hours of auxin treatment (FDR < 0.1) (Figures 5A, S5A and Supplementary table 5). After 96 hours of auxin treatment, 253 genes were differentially expressed. We also performed ChIP-seq for three histone modifications marking active chromatin (H3K4me1, H3K4me3 and H3K27ac). In agreement with the lack of overall gene expression changes upon CTCF depletion, there was no significant change in chromatin modification patterns (Figure S5B). Consistent with a role for CTCF in gene regulation, the genes with CTCF binding at the transcription start sites (TSS) showed slightly more down-regulation in transcription than genes without CTCF binding sites at the TSS (Figures 5B and S5C). We also observed that genes in the LADs were affected by CTCF depletion to a higher degree than genes outside the LADs (Figure 5C). This finding, combined with the observation that TAD structure within LADs is more severely affected after loss of CTCF, points to a role for CTCF in regulating genome architecture and gene regulation within the specific context of LADs and not genome-wide, as has been widely assumed.
Discussion
The role of CTCF in chromatin organization in mammalian cells has been generally accepted, providing a framework for the understanding of long-range control of gene expression by cis regulatory elements such as enhancers and promoters. However, we demonstrate in this study that maintenance of higher order chromatin organization such as TADs and compartments does not strictly depend on CTCF. Using the auxin-inducible degron system, we acutely depleted CTCF protein in mouse embryonic stem cells, and monitored the impact of CTCF loss on genome architecture using Hi-C. We observed reduced chromatin looping strengths at CTCF-anchored chromatin loops but no overt changes in TAD structure and compartments, except for a small number of TADs located in constitutive LADs. We also failed to detect significant changes in chromatin states and transcription profiles. Our results suggest that the CTCF is not generally required for maintaining the TADs and compartmental organization of the genome, and additional factors must be involved in their demarcation.
Our results call for a revision of the current models of CTCF’s role in chromatin organization. One possible refinement is that there is a distinction between the mechanisms that establish chromatin loops at CTCF binding sites and the organization of chromatin into TADs and compartments. We observed that loss of CTCF in the mESC was indeed accompanied by significant reduction of chromatin looping anchored on CTCF binding sites, but the lack of substantial changes in TADs and compartments suggests that CTCF-mediated chromatin loops are not entirely responsible for the formation of TADs and compartments. This conclusion is consistent with previous findings that disruption of CTCF motifs led to loss of chromatin loops at certain CTCF binding sites but without changing the local chromatin domains (Sanborn et al., 2015). Also consistent with this conclusion is that CTCF binding sites are found more frequently within TADs than at TAD boundaries (Dixon et al., 2012). Furthermore, removal of the Cohesin complex in mammalian cells did not disrupt TADs (Seitan et al., 2013; Zuin et al., 2014). Thus, chromatin loops mediated by CTCF are insufficient to maintain TADs boundaries, and other mechanisms must be responsible. We note that in our experimental analysis, cell number at least doubled between 24h and 96h after auxin treatment (e.g. cell number increased ~4-fold for clone 2 between these time points). As TADs are disassembled in mitosis and reassembled during the transition to G1 (Dileep et al., 2015; Naumova et al., 2013), our results suggest that significant CTCF reduction also does not prevent TAD formation, although addressing this point will require additional future effort.
A second way to reconcile our observation and previous studies is that distinct mechanisms may be responsible for establishment of TADs in the late replicating LADs and early replicating non-LAD regions. As noted above, previous studies have shown that TADs are lost during mitosis and are re-established in early G1 (Dileep et al., 2015; Naumova et al., 2013). It is possible that different factors in different nuclear compartments might be involved in establishment or maintenance of TADs in the regions that replicate early and regions that replicate late during cell cycle. This could explain why the disruption of CTCF binding sites leads to the elimination of certain loops or TAD boundaries in LADs but not in non-LAD regions.
What mechanisms could be responsible for maintenance of TADs and chromatin compartments in the non-LAD regions? One potential factor is Topoisomerase II beta (Top2b), which has been found to interact with CTCF and Cohesin, and colocalizes with them at TAD boundaries (Uuskula-Reimand et al., 2016). Since Top2b can modulate the supercoiling of DNA, it is conceivable that TAD borders would have distinct topological property from inside the TAD, and such topological property is maintained after removal of CTCF. Another factor may be ZNF143, which like Top2b is colocalized with CTCF and Cohesin complex genome wide and at TAD borders (Bailey et al., 2015; Heidari et al., 2014). However, ZNF143 binds preferentially at promoters and may be primarily involved in mediating promoter-centered chromatin interactions. It is also possible that non-coding RNAs might be involved in establishing or maintaining TADs, as CTCF has been demonstrated to interact with many RNAs including the steroid receptor RNA activator SRA (Kung et al., 2015; Saldana-Meyer et al., 2014; Yao et al., 2010). Clearly, the role of CTCF in chromatin organization is not as clear-cut as previously proposed. Our study will help direct future efforts that promise to further elucidate the mechanisms of chromatin organization.
While this manuscript was under preparation, a similar work appeared in BioRxiv (Nora et al., doi:https://doi.org/10.1101/095802). With similar experimental approaches employed, the authors made similar observations with regard to the role of CTCF in chromatin loop formation and insulation scores at TAD boundaries. However, important distinctions exist between the two works with regard to TAD maintenance and transcriptional profiles after CTCF depletion. Future experiments are needed to address whether these differences were due to experimental data, data analysis methods or interpretation.
Methods
Detailed methods and data analysis are described in the supplemental method.
Accession numbers
Sequencing data have been deposited in Gene Expression Omnibus (GEO) under accession number GSE94452, and can be accessed at https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=olctqyiktlyrhkf&acc=GSE94452.
Author contributions
N.K., H.I. and B.R. conceived the project. H.I., X.X. and H.Z. engineered cell lines. N.K., T.L. and Z.Y. carried out library preparation. N.K., F.M. and A.D. performed cell cycle analysis. N.K., D.G., R.F. and B.L. performed data analysis. J.D. contributed to experimental design. N.K., H.I. and B.R. wrote the manuscript. All authors edited the manuscript.
Acknowledgments
We would like to give special thanks to Samantha Kuan for operating the sequencing instruments. We would like to acknowledge the help of Ming Hu (NYU) for sharing software code. We would also like to give special thanks to Anthony Schmitt and Sora Chee for sharing protocols and giving numerous helpful advice, as well as the additional members of the Ren laboratory. This work was supported by the Ludwig Institute for Cancer Research (B.R.), NIH (1U54DK107977-01) (B.R.), NIH (1U54DK107965-01) (H.Z.) and a Postdoc fellowship from the TOYOBO Biotechnology Foundation (N.K.).