Dissection of 3D chromosome organization in Streptomyces coelicolor A3(2) leads to biosynthetic gene cluster overexpression

Significance

This work demonstrates the coordination of global adaptive chromosomal rearrangement and transformation of metabolism in different growth phases, describes the interplay of local chromosomal structure and transcription of biosynthetic gene clusters (BGCs), and applies the potential three dimension (3D) regulation mechanism to realize the high production of target natural products. Our study will provide some insight into the prokaryotic chromosomal organization and transcriptional regulation and offer guidelines to increase the production of natural products.

Abstract

The soil-dwelling filamentous bacteria, Streptomyces, is widely known for its ability to produce numerous bioactive natural products. Despite many efforts toward their overproduction and reconstitution, our limited understanding of the relationship between the host’s chromosome three dimension (3D) structure and the yield of the natural products escaped notice. Here, we report the 3D chromosome organization and its dynamics of the model strain, Streptomyces coelicolor, during the different growth phases. The chromosome undergoes a dramatic global structural change from primary to secondary metabolism, while some biosynthetic gene clusters (BGCs) form special local structures when highly expressed. Strikingly, transcription levels of endogenous genes are found to be highly correlated to the local chromosomal interaction frequency as defined by the value of the frequently interacting regions (FIREs). Following the criterion, an exogenous single reporter gene and even complex BGC can achieve a higher expression after being integrated into the chosen loci, which may represent a unique strategy to activate or enhance the production of natural products based on the local chromosomal 3D organization.

The chromosomal DNA must be compacted thousands of times smaller to fit the confined space of cells. Meanwhile, cellular processes such as transcription, DNA replication, and chromosome segregation are compatible with the structure (1). Unlike the membrane-covered nucleus in eukaryotes, the chromosomal DNA of prokaryotes is constrained. It forms a nonmembrane structure called the nucleoid, a dynamic macromolecular complex where the genetic material and its associated proteins are located (2). Bacterial chromosome folding began to be investigated in the 1970s, but the absence of powerful tools hindered the dissection of high-order chromosomal organization. Initially, the nucleoid was thought to be disorderedly packed into the cell, in contrast to the highly organized chromatin of eukaryotic cells (3) until advances in microscopy, structural biology, and molecular biology such as chromosome conformation capture (3C) (4) and genome-scale deviate circular chromosome conformation capture (4C) (5), chromosome conformation capture carbon copy (5C) (6), and high-throughput chromosome conformation capture (Hi-C) (7), which allow for a more systematic and quantitative measure to characterize genome topology and organization. Recent studies demonstrated that bacterial chromosomes are not only highly compacted but also temporally and spatially organized, and this organization is hierarchical, from large-scale macrodomains to smaller scale loop structures that affect different biological processes (1, 8).

During the last 10 y, the higher-order chromosomal structure of diverse prokaryotic species has been under extensive and profound research, including Escherichia coli (2); Caulobacter crescentus (9, 10); Bacillus subtilis (11–14); Corynebacterium glutamicum (15); Streptomyces ambofaciens (16); and Streptomyces venezuelae (17). The influence of nuclear associated proteins (NAPs) on the establishment and maintenance of global or local chromosome organization (2, 17), the dynamic changes of chromosome structure during the cell cycle progression (9), and the contribution of structural maintenance of chromosomes (SMC) complexes in this procedure (13, 14), and ParABS system for chromosome maintenance and separation throughout the cell propagation (11, 15) were the most prominent features of these studies. However, the underlying mechanism between gene expression and local chromosome organization in prokaryotes is still far from being thoroughly revealed.

Gram-positive bacteria of the genus Streptomyces are renowned for producing bioactive natural products with a wide range of applications in medicine and agriculture (18), including antibiotic, anticancer, antifungal, antiparasitic, and immunosuppressive compounds (19). It benefits from the numerous kinds and amounts of natural product biosynthetic gene clusters (BGCs) in their genome. However, most of these BGCs are veritably silent under laboratory growth conditions (20). With the emergence and prevalence of antibiotic-resistant pathogenic bacteria, discovering novel antibiotics and increasing the products of excellent-performance antibiotics remain pressing tasks (21). Various strategies have been proposed to achieve this goal: cloning and expressing the entire BGCs in heterologous hosts (22, 23), manipulating transcription-associated elements directly in native hosts (24, 25), or testing different fermentation conditions (26, 27). Some researchers also pointed out that the location of reporter genes or BGCs in the genome could considerably affect the expression level in bacteria, which was called the “chromosomal position effect” (28, 29). The highest expression strength of a reporter gene or BGCs can reach six or eight times higher than the lowest in different chromosomal positions. Nonetheless, the intrinsic relevance of how the chromosomal position determines the expression level of an inserted reporter gene or whole BGCs is still far from clear, which greatly limited the rational engineering of host chromosomes to increase the expression of BGCs, hence the yield of natural products.

In this study, we applied the high-resolution Hi-C method to characterize the chromosome three-dimensional (3D) organization of the model actinomycete Streptomyces coelicolor A3(2) strain M145 and found that the spatial rearrangement of the linear chromosome was associated with the expression change of BGCs during the growth-phase shift. Especially, the transcriptional level of native genes was highly correlated with the local chromosomal interaction frequency as quantified by the value of the frequently interacting regions (FIREs) (30). Furthermore, a β-glucuronidase-encoding reporter gene gus(a) (31) and a natural tetronate RK-682 BGC (32) were inserted into different genomic contexts, and it was discovered that the activity of β-glucuronidase and the production of RK-682 compound were proportional to the FIRE value. All of these results indicate that the frequency of local chromosomal interaction will significantly affect the expression level of the local gene, whether endogenous or exogenous. With this discovery, we finally realized the overproduction of natural products comparing the traditional strategy.

Results

Higher-Order Chromosome Organization in S. coelicolor.

Since the expression levels of BGCs will dramatically change, accompanied by the transformation of different growth phases [i.e., almost silent before the mid-exponential phase (M phase) and high-expression after the late-exponential phase (L phase)] (33), to ensure whether the chromosome 3D organization will synchronously change to fit this biological procedure, we initially investigated the S. coelicolor genome organization during the M phase (14 h) and the L phase (22 h, SI Appendix, Fig. S1) in liquid R5- medium through Hi-C. The genome is divided into 5-kb bins, and interaction frequencies for each restriction fragment are assigned to corresponding bins. The normalized interaction frequencies are visualized as a two-dimensional matrix for M phase (Fig. 1A and SI Appendix, Fig. S2 A and B) and L phase (SI Appendix, Figs. S2 C and D and S3A). Some notable 3D features from Hi-C data are summarized.

Fig. 1.

Chromosome organization of S. coelicolor in M phase. (A) Normalized genomic contact map obtained from asynchronous population (14 h for M phase, R5- medium, 5-kb resolution). The x and y axes represent the genomic coordinates of the reference genome (S. coelicolor A3(2) strain M145, 8.67 Mb), and the color scale reflects the frequency of contacts between two genomic loci, from dark blue (rare contacts) to dark red (frequent contacts). The orange arrow indicates the region interacted with the whole chromosome in L phase. The horizontal axes in B to F represent the chromosomal position corresponding to A. (B) Top panel: Directional index analysis of the corresponding genomic regions. Downstream (red) and upstream (green) biases are indicated. Down panel: Identified CIDs are represented as blue triangles. Dotted gray vertical lines crossing A to F highlight the barriers between CIDs. (C) RNA-seq profile mapped onto S. coelicolor genome in 1-kb resolution. (D) Genomic annotations of 14 putative laterally acquired insertions (purple block, Top), six rRNA coding genes (light blue block, Middle), origin of replication (oriC, brown block, Middle), and 28 BGCs (red block, Down). (E) Genetic partition in S. coelicolor. Red rectangle for the 1.5 Mb left arm, blue rectangle for the 4.9 Mb core region, and yellow rectangle for the 2.3 Mb right arm. (F) 3D compartmentalization in S. coelicolor. The red and blue biases represent the A and B compartments in M phase, respectively. (G) The comparation of CIDs in M phase (Top) and L phase (Down). The identical CIDs are pointed out by bold lines and the identical CID boundaries are connected by dotted vertical gray lines. (H) Circos plots show the significant interaction among 16 CID boundaries (blue rectangle, Left) or 14 putative laterally acquired regions (purple rectangle, Right). The colors of circle correspond to the genetic partition in E. Each connection/arc represents a significantly enriched interaction. (I) 3D representation of the contact map from A. The chromosome is represented as a chain of beads (20 beads travel across pair of 5-kb bins; see Materials and Methods). The color of beads corresponds to the genetic partition in A, and the gray beads represent the telomeres of the chromosome.

First, both contact maps exhibit a solid and global main diagonal, which reflects the frequent local contacts resulting from interactions among adjacent loci. A secondary diagonal perpendicular to the main one extends from the origin of replication (oriC) adjacency to all rest regions of the chromosome, which shows the alignment of the two chromosomal arms (Fig. 1A and SI Appendix, Fig. S3A). The width and color saturability of the secondary diagonal are increasing over time, which indicates the considerable and gradual chromosomal compaction during growth phases shift (SI Appendix, Fig. S4). Some researchers have proved that the structural maintenance of chromosomes (SMCs) complexes, accompanying with ScpAB, loaded via ParB at parS sites adjacent to oriC, mediate chromosomal interarm contacts (11, 15, 17). As a most functionally conserved DNA-organizing protein among all domains of life (34–36), we assume the same protein complex is also responsible for the observed interarm contacts in S. coelicolor.

Meanwhile, in both phases, general interaction signals are detected between the telomeres of the liner chromosome, except for terminal inverted repeats (TIRs) for which the sequencing signals have been eliminated (Fig. 1A and SI Appendix, Fig. S3A), and this is consistent with previous cytological evidence (37). Interactions among covalently bound terminal proteins mediate the association between the telomeres and make the chromosome exhibit genetic linkage, which would close the genetic map into a circle (38, 39). A similar character has also been independently demonstrated in S. venezuelae (17).

Second, the self-interacting regions termed chromatin interaction domains (CIDs) previously described in B. subtilis and other species (2, 9, 11, 16) are also detected in S. coelicolor. Following the directional preference analysis method used in C. crescentus (9), 17 obvious CIDs are identified in M phase, ranging in length from 85 to 1,470 kb (510 kb on average, Fig. 1B), and 13 CIDs are identified in L phase, ranging from 220 to 1,470 kb (667 kb on average), respectively. (SI Appendix, Fig. S3B). Particularly, while only five intact CIDs are exactly the same, 10 boundaries remain, which indicates this dynamic process (split of a CID to smaller scale two or more and fusion of adjacent CIDs to larger scale one) (Fig. 1G). It suggests the adaptive dynamic adjustment during the growth phase transformation.

CID boundaries with lower guanine and cytosine (G+C) content than the chromosomal average account for 93.8% (15 of 16) and 91.7% (11 of 12) in M phase and L phase, respectively. Additionally, G+C content of CID boundaries is statistically lower than the interior (SI Appendix, Fig. S5 B and E), which is similar to the cases in other species (11, 40, 41). In these locations, higher AT content might facilitate the formation of CID boundaries, for it has been previously documented that the DNA sequence component would influence its physical properties, such as DNA curvature, etc. Meanwhile, NAPs with AT content binding preference might maintain these barriers to some extent (40, 42, 43). Actually, numerous binding motifs of known transcription factors and NAPs are found within these boundaries (Dataset S1).

To investigate the relationship between 3D structure and gene expression, we also sampled and sequenced the mRNA in cells culturing to both two growth phases and observed the dramatical difference between them (Fig. 1C and SI Appendix, Figs. S3C and S6). Meanwhile, we found that 50% (40 of 80) and 56.6% (30 of 53) genes in CID boundaries are highly expressed (within the top 30% of all genes in the genome) in M phase and L phase, respectively (Dataset S2), and the transcriptional strength of these genes within CID boundaries is higher than in the interior (SI Appendix, Fig. S5 C and F). Considerable parts of high-expression genes at CID boundary are correlated with substance and energy metabolism required for the growth and development of the cells (Dataset S2 and SI Appendix, Figs. S7 and S8). A high proportion of these highly expressed genes hold the same expression orientation with the direction of continuous replication [accounting for 77.5% (31 of 40) in M phase and 76.7% (23 of 30) in L phase, respectively, Dataset S2]. Together with the recent report on the interplay between elongating RNA polymerases (RNAPs) and condensin (44), these results hint that the evolution of the gene’s high expression is partly through the directional convergence between transcription and chromosomal replication. CID-bound but low-expression genes mainly code for the membrane protein, transcriptional regulator, and some postmodification genes for natural product biosynthesis, which are responsible for signal transduction as well as regulation of secondary metabolism (Dataset S2). Interestingly, only 40% (16 of 40) in M phase and 47.8% (11 of 23) in L phase of these genes share the same direction between transcription and chromosomal replication (Dataset S2), which indicates the directional convergence in another aspect.

Significant interactions (Materials and Methods) among the CID boundaries are detected in M phase (Fig. 1H). However, these interactions are obviously decreased in L phase (reduced by 65% of the significant interaction number among the identical 10 CID boundaries, SI Appendix, Fig. S3H). A similar phenomenon occurs among 14 putative laterally acquired insertions through horizontal gene transfer (45) (reduced by 58% of the significant interaction number, Fig. 1 D and H and SI Appendix, Fig. S3 D and H). A putative regulatory mechanism that adapts to the transition of the metabolic situation may be implied by the colocalization and dispersion between these locations in different growth phases.

According to the essentiality and persistence of genes, the genome of Streptomyces could be divided into a central core region, comprising genes with unconditionally essential and two terminal chromosomal arms with genes coding for dispensable function (45), which was called “genetic partition”. In S. coelicolor, the core region is approximately 4.9 Mb, with two uneven arms of 1.5 Mb (left arm) and 2.3 Mb (right arm) in length, separated by the first and last rRNA coding gene (45) (Fig. 1E). The concept of “chromosome compartments” was first proposed in eukaryote (7) and being a general term to measure the interactions among topologically associating domains (TADs). Two types of compartments, A and B, were classified based on their preferential interaction with each other (46, 47). Compartment A was characterized as transcriptionally active and open genomic regions (7); on the contrary, compartment B was associated with the silent expression regions such as nuclear lamina and nucleolar (48, 49). We compartmentalized the chromosome of S. coelicolor following the protocol of Zheng (50) and detected three independent compartments in both phases (Fig. 1F and SI Appendix, Fig. S3F). Compartment A extends from around 2.3 Mb to 6.2 Mb, giving two compartment B lengths of approximately 2.3 Mb (left-compartment B) and 2.4 Mb (right-compartment B) in M phase. The right-compartment B almost overlaps with the right arm, exhibiting the correlation between genetic partition and 3D compartmentalization. However, the left-compartment B is 0.8 Mb longer than the left arm (exceeding 56.7% of the left arm length), presenting the first case that the genetic partition is not well consistent with 3D compartmentalization. Only minor differences in 3D compartmentalization exist between L phase and M phase, which shows the relative stability of the large-scale chromosomal structure. The organization of chromosome 3D compartmentalization can be visualized by converting the contact map to a 3D topological diagram, where the whole left arm (red beads) and a small part of the core region (blue beads) together constructed left-compartment B (Fig. 1I and SI Appendix, Fig. S3I).

Finally, we analyzed some abnormal lines in the contact matrix of L phase. The faint and cross-shaped line in 4.3 Mb exhibited the interaction between the ori region and the rest of the chromosome (SI Appendix, Figs. S2 C and D and S3A), and this signal may be covered by the significant long-range contact background in M phase. The separation and translocation of the ParB-oriC segregation complex along the nucleoid might be responsible for the phenomenon, consistent with the cases in C. crescentus (9) and C. glutamicum (15). This interaction intensity was almost the same as the average interaction frequency between bins separated in 2.0 Mb, which showed the physical proximity and relative insulation of the segregation complex and the nuclear region it passed by. We noticed the evident line located in 6.9 Mb, which means the global and apparent interaction between this region and the rest of the chromosome (SI Appendix, Figs. S2 C and D and S3A). It was even sharper and clearer than the one of 4.3 Mb. Interestingly, this region comprised three precursor supply genes (SCO6269 to SCO6271) for the synthesis of malonyl-CoA (51), the assembly unit for the polyketide synthase (PKS), and well-characterized regulation genes (SCO6265 to SCO6266) for γ-butyrolactone (GBL) signaling circuits, which hints the correlation between natural product biosynthesis and chromosomal interaction.

Remodeling of Local Chromosome Organization Accompanying Nature Products’ Biosynthesis.

The Streptomyces species are capable of synthesizing numerous natural products to achieve a competitive advantage in the complex soil environment. This progress generally starts at late-exponential and stationary phases. To understand the interplay between natural product biosynthesis and the rearrangement of local 3D chromosome organization during metabolic differentiation, we further examined the changes in local 3D organization and transcription of all 28 BGCs encoded in S. coelicolor genome in two growth phases (SI Appendix, Fig. S9 and Dataset S3). The results showed that 10 BGCs have significant changes at the transcription level and two BGCs have a synchronized obvious change in both aspects (Fig. 2 B and C and SI Appendix, Fig. S10).

Fig. 2.

The coordination between local chromosomal organization and transcription of BGCs. (A) The map of S. coelicolor chromosome with the same genetic partition as Fig. 1E. The relative location of dissected BGCs is represented by a black rectangle. (B and C) The different local Hi-C maps in the 2 kb resolution compare the contact enrichment in M phase versus L phase. The chromosomal ranges comprise the undecylprodigiosin BGC (Left) or coelimycin P1 (Right) and their 150-kb flanking regions, respectively. The color scale reflects the different contact frequency in two growth phases from dark blue (frequent contacts in L phase) to dark red (frequent contacts in M phase). The nested CIDs around red BGC are partly outlined by black lines, and the independent interactive domain covering cpk BGC is highlighted by a dark triangle. The orange arrow indicates the contact signal with the whole chromosome as SI Appendix, Fig. S3A. (D and E) The RNA-seq profile (reads per kilobase per million mapped reads, RPKM) of the single gene mapped onto the same S. coelicolor genome scope as B and C, respectively. The RNA-seq profiles in two growth phases are exhibited in mirror symmetry.

In all BGCs of S. coelicolor genome, undecylprodigiosin (red) BGC has been extensively studied in previous works (52, 53). This red pigment enables quick visual discrimination of mutations that affect its production, partly making S. coelicolor the model strain for the genus among thousands of isolated Streptomyces species (54). In the results, the transcriptional level of the red cluster increased by 143 times from M phase to L phase (Fig. 2D and Dataset S3). It is accompanied by the relocation of the BGC between two nested domains (Fig. 2B) analyzed by the directional index method, which has been developed in other organisms to help define the CID hierarchical organizations (9). The arising and dissipating of nested domains and topological barriers may be an adaptive dynamic change to different growth conditions or reflect the heterogeneity within the population (11, 55). Similar observations have recently been reported in S. ambofaciens, in which the high transcriptional expression of a 29-kb BGC corresponds to the formation of an additional sharp boundary after culturing for 48 h (16).

BGC of coelimycin P1 (cpk) is responsible for producing a yellow-colored alkaloid compound under specific fermentation conditions (56). The transcriptional strength of this BGC is dramatically increased 541-fold from M phase to L phase (Fig. 2E and Dataset S3). Intriguingly, a newly formed independent interactive domain spanning through the whole cluster was observed in L phase (Fig. 2C and SI Appendix, Fig. S3G). It is similar to the case in the eukaryotes species Arabidopsis thaliana, where the thalianol BGC will form different kinds and scales of cluster-associated interactive domains accompanying the activation or silence of it (57), which could suggest a comparable characteristic of 3D regulation mechanism between the prokaryote and eukaryote. Meanwhile, this independent topological structure coincided with the particular DNA region in the 6.9-Mb loci, which extensively interacted with the whole chromosome, hinting an associated biological procedure for secondary metabolism.

Taken together, we focused on the synchronized variation between the local chromosomal structure of different BGCs and their expression level in different growth phases and found two typical cases. One generates a new nested domain boundary and the other forms an independent topological domain, and both cases coincide with the significant enhancement in transcription. These results emphasize the close correlation between the transcription of secondary metabolite gene clusters and their local 3D structure in our experimental conditions.

The Local Chromosomal Interaction Frequency Can Prominently Influence the Expression Level of Endogenous and Exogenous Genes.

We next investigated how the local chromatin organization would affect the expression of embedded genes. “FIRE score” was used to normalize the total raw local cis contacts for each bin in a 200-kb scale in eukaryotes (30). Superenhancers and human tissue-specifically expressed genes were enriched in such regions with high FIRE scores (30). Similarly, we calculated the “FIRE value” (Dataset S4) using minor modified methods (Materials and Methods). This value and transcription profile were plotted along the genome at 5-kb resolution (Fig. 3 A and B). The obvious correlation [Pearson correlation (PC) = 0.72, 0.65 in M phase and L phase, respectively] between the two signals pinpoints the underlying links between gene expression and local chromosomal interaction frequency.

Fig. 3.

The relativity between the expression of endogenous or exogenous genes with local chromosomal interaction frequency. (A and B) The correlation between the transcription of endogenous gene and local chromosomal interaction frequency in 5-kb resolution for M phase (A) and L phase (B). The contact between each bin and its 200-kb region upstream and downstream (FIRE signal, blue curve) accompanying with associated transcription profile (RNA sequencing signal, red curve) were plotted along the genome. (C and D) The relative location of selected HCR windows in the genome for M phase (C) and L phase (D). The color of chromosome corresponds to the genetic partition in Fig. 1E. The black triangles represent the locations of the inserted reporter gene within HCRs. (E and F) The correlation between the expression of exogenous reporter gene with local chromosomal interaction frequency in M phase (E) and L phase (F). The horizontal axis represents the FIRE value of HCR, and the vertical axis represents the expression strength of reporter gene. Each point from right to left represents HCR-M1 to M10 (E) or HCR-L1 to L10 (F), respectively. The numbers above each point mean the ratio of expression strength between each HCR and attB site integrated control indicated by a red line. The error bars are 1 SD. The image embedded in the penal indicates the color of fermentation broth in two phases.

To further verify the causality between gene expression and local chromosomal interaction frequency, we inserted a β-glucuronidase-encoding reporter gene gus(a) into a series of chromosomal positions with the following selected criteria. Each of the 11 continuous 5-kb bins constituted a window, and the PC between the FIRE value and associated transcription profile in each window was measured independently (Dataset S4). We set PC = 0.8 as the threshold, and these windows with PC ≥ 0.8 were defined as high correlative regions (HCRs). These HCRs were then sorted from high to low by the average FIRE value of the comprised eleven 5-kb bins. The difference between the highest and lowest FIRE value in HCRs was equally allocated to achieve a series of standard aliquots, and 10 HCRs with the lowest deviated FIRE value compared with the standard one would be selected for further investigation (Fig. 3 C and D and SI Appendix, Table S1). These twenty selected windows were named “HCR-MN” and “HCR-LN” for M phase and L phase, respectively (N = 1 to 10). In this study, the expression of gus(a) was under the control of a constitutive promoter ermEp1*, derived from the most used promoter ermEp for Streptomyces (58). To prevent the potential transcriptional read-through with adjacent genes, two bidirectional fd-phage terminators (59) enable up to 85% transcriptional isolation efficiency in Streptomyces (28) were set flanking the ermEp1*-gus(a), and their opposite direction avoids the possible elimination by homologous recombination (SI Appendix, Fig. S11B). Then the reporter cassette was introduced into each HCR-MN and HCR-LN window using Zeng’s protocol (60). To maximally prevent the potential interference between the gus(a) cassette and local genes, the reporter cassette was preferred to be inserted in the opposite orientation between two operons with the same transcriptional direction (SI Appendix, Fig. S11A). After culturing to a specific growth phase, the cells were sampled to measure the activity of β-glucuronidase, and an optimized measuring and calculation method based on Myronovskyi’s case (31) was adopted to minimize the intergroup deviation.

The FIRE value of HCRs and β-glucuronidase activity in both growth phases were highly correlative (Fig. 3 E and F. PC = 0.98, P-value = 3.23e⁻⁷ in M phase and PC = 0.97, P-value = 5.75e⁻⁶ in L phase), which meant the local chromosomal interaction frequency could significantly influence the expression level of the comprised single gene. The highest activity in HCR-M1 was about 22 times higher than HCR-M10 in M phase, and the difference between HCR-L1 and HCR-L10 was as significant as 13 folds.

To measure the potential application value of our study, a control strain was constructed with the same reporter cassette integrated into the ϕC31 attB site (61) (SI Appendix, Fig. S11C), a most used heterologous gene expression method in a broad range of Streptomyces species. As a result, in M phase, the β-glucuronidase activity in four windows (HCR-M1–M4) was higher than the attB control, and the highest activity in HCR-M1 was 1.95 times higher. While in two windows (HCR-M5 and M6) was almost equal to the attB control, and in the rest four windows was lower than it (Fig. 3E). In L phase, four windows (HCR-L1–L4) had higher, and six windows (HCR-L5–L10) had lower β-glucuronidase activity than the attB control. The highest activity in HCR-L1 was 1.92 times higher than the control (Fig. 3F).

Together, we found that the transcriptional strength of endogenous genes was highly correlated with local chromosomal interaction frequency, which was quantified as the FIRE value in this study. Furthermore, the exogenous gus(a) reporter gene cassette was inserted into a series of HCRs with gradient FIRE value, and the measured β-glucuronidase activity was highly correlated with the FIRE value in both growth phases, which verified the local chromosomal interaction frequency could evidently affect the expression of the located gene.

Exogenous Natural Product Can Achieve Higher Production When the BGC Engineered into HCRs with a Higher FIRE Value.

Numerous strategies have been proposed and applied to activate the silent BGCs or increase the production of well-performing natural products (20, 62, 63). They were dominantly deliberated and manipulated at the one-dimensional sequence level. Inspired by the promising results from a single reporter gene and further verifying the feasibility of enhancing the production of target BGC according to the three-dimensional feature, an 11.8-kb exogenous BGC candidate for natural product RK-682 biosynthesis was integrated into the attB site of S. coelicolor (SI Appendix, Fig. S11D). The entire RK-682 BGC contains nine genes organized into two operons and produces a tetronate compound (Fig. 4A), a potent inhibitor of protein phosphatases and HIV-1 proteinase (32, 64).

Fig. 4.

The correlation between the production of RK-682 and local chromosomal interaction frequency. (A) The chemical structure of RK-682 and its BGC (yellow arrows). (B) LC–ESI–HRMS analysis of RK-682. Two peaks corresponding to RK-682 are due to complexation with divalent metal ions. (C and D) The correlation between RK-682 production with local chromosomal interaction frequency in M phase (C) and L phase (D), respectively. Each point from right to left represents HCR-M1 to M10 (C) or HCR-L1 to L10 (D), respectively. The numbers above each point mean the ratio of RK-682 production between each HCR and attB site integrated control indicated by a red line. The error bars are 1 SD.

To simplify the heterologous expression system and avoid the native regulation to the endogenous promoter P_actII-orf4 (65), the two native operons (rkA-G and rkH-I) of RK-682 BGC were fused into one under the control of one promoter ermEp1* and sandwiched by two fd-phage terminators. This recombinant BGC cassette was embedded into an integrative plasmid (SI Appendix, Fig. S11F) and showed no obvious production difference from the native one under the same fermentation conditions. Then the RK-682 BGC cassette was introduced into the same HCR windows as the gus(a) reporter cassette (Fig. 3 C and D and SI Appendix, S11E and Table S1). The production of RK-682 compound for each mutant was then detected by LC-ESI-HRMS (Fig. 4B). The results showed that the FIRE value in HCRs and RK-682 production were also highly correlated in both growth phases (Fig. 4 C and D, PC = 0.94, P-value = 6.06e⁻⁵ in M phase and PC = 0.90, P-value = 3.61e⁻⁴ in L phase), which further emphasized the influence of local chromosomal interaction frequency on the expression of the embedded gene or even whole BGC. Interestingly, different from the apparent linear correlation between the β-glucuronidase activity and FIRE value in HCRs, the BGC case was curvilinear in both phases. We speculate that it might be caused by the different characteristics of final products and the associated different detection methods of the two cases. In the case of a single reporter gene, the transcription, translation, catalyzation of sufficient substrate, and the measure index were linear variations. However, in the BGC case, the production of RK-682 needs the participation of a series of enzymes with different catalytic efficiencies, and the complex biosynthesis model, together with potential precursor competing and quorum sensing, may synergistically contribute to the curvilinear correlation. The highest RK-682 production in HCR-M1 was about 15 times higher than that in HCR-M10 in M phase, and this difference between HCR-L1 and HCR-L10 even reached 30 folds. Compared with the conventional attB integration method, in M phase, the RK-682 production in five windows (HCR-M1–M5) was found to be higher, while in four windows (HCR-M7–M10) was lower than the attB control site, except that it was almost equal in HCR-M6. The highest production in HCR-M1 could reach 4.36 folds enhancement compared with the control (Fig. 4C). In L phase, two windows (HCR-L1 and L2) had higher RK-682 production than the attB control while eight (HCR-L3–L10) had lower production. The highest production in HCR-L1 was 1.37 times higher than the control (Fig. 4D).

Collectively, the heterologous RK-682 BGC was engineered into each HCR window, and the production of RK-682 was also highly correlated with FIRE value in both growth phases, ensuring our scenario again. Compared with the attB integration control, the increased RK-682 production in some HCR windows with a high FIRE value suggests the application’s potential to increase the production of natural products.

Discussion

In this study, we investigated and compared different 3D features of the S. coelicolor chromosome in two growth phases with different metabolic situations; then, we focused on the influence of chromosomal topology on the expression of embedded genes, and finally applied the potential 3D regulation mechanism to realize the high production of natural products.

In both phases, the partial incoordination between genetic partition and 3D compartmentalization was demonstrated in our study (Fig. 1 E and F and SI Appendix, Fig. S3 E and F). The whole chromosome was insulated into a series of CIDs of different sizes by barriers (Fig. 1B and SI Appendix, Fig. S3B), and the genetic features of CID boundaries were primarily consistent with those of other species (9, 11, 15). However, unlike the situation in S. ambofaciens (16), there was no obvious uniformity between the direction of transcription and continuous replication of the genes within CID boundaries. Instead, this consistency mainly existed in the high-expression genes within CID boundaries (Dataset S2). We thus propose that the evolution of the gene’s high expression is partly through the directional convergence between transcription and chromosomal replication, which provides rational design guidance for the heterologous expression of natural products to activate or increase their production. Meanwhile, these CID boundaries were prominently interacting with each other in M phase and decreasing in L phase (Fig. 1H and SI Appendix, Fig. S3H). The colocalization and dispersion between these sites in different phases may imply a potential regulatory mechanism to adapt the transformation from vegetative growth to secondary metabolism. A similar situation also exhibits between 14 putative laterally acquired regions (Fig. 1H and SI Appendix, Fig. S3H).

During metabolic differentiation, the global chromosomal organization experienced an extensive and profound rearrangement, mainly characterized by reshaped CIDs and more condensed chromosomes (Fig. 1G and SI Appendix, Fig. S4), previously described as the switching from a rather “open” to a “closed” conformation (16). As the coexpression model of genes within CIDs in Mycoplasma pneumoniae (40), we thus propose that the rearrangement of CIDs and the arising and dissipating of nested domains might modulate with the local transcriptional regulation.

Given the correlation between the variation of local 3D structure and transcription for BGCs, the red BGC relocates at the boundary of two nested domains with a significant increase in expression in L phase (Fig. 2 B and D). It may mainly be caused by the restraint effect of high local transcription on the diffusion of DNA supercoils or NAP-mediated physically separate domains (9). Along with growth to the subsequent stational phases, the expression strength of the red cluster should be further increased (33), which may result in a sharper and more obvious boundary, as seen with congocidine BGC in S. ambofaciens (16).

The whole cpk BGC forms an independent interactive domain in L phase only, insulating this cluster from the surrounding chromosome environment, along with the dramatic transcriptional enhancement (Fig. 2 C and E and SI Appendix, Fig. S3G). Additionally, we find that the particular DNA region interacting with the rest of the chromosome is just adjacent to this cpk BGC-associated interactive domain (Fig. 2C and SI Appendix, Fig. S3A). Some clues are collected to explain this phenomenon: 1) In prokaryotes, the coupled transcription-translation (CTT) model forming “DNA · RNAP · mRNA · ribosome” expression complexes is a typical dogma for prokaryotic gene expression (66). 2) The nascent peptide of the innermembrane proteins, together with the expression complexes, would be translocated to the membrane by specific proteins in  E. coli (67) and the expression complexes would expand to “DNA · RNAP · mRNA · ribosome · protein · membrane”. 3) The subcellular localization of PKS proteins was found to be highly associated with the membrane (68, 69). We thus propose that, as a typical type I PKS, the translated nascent peptide of the cpk cluster will be translocated to the membrane together with the complete expression complex “cpk BGC · RNAP · mRNA · ribosome · protein · membrane”, and during this process, the relative loose region in the boundary of the cpk cluster-associated domain exhibits the global interaction with the whole chromosome for the population Hi-C method. This translocation and interaction may also benefit the diffusion of malonyl-CoA and γ-butyrolactone, which are biosynthesized by the embedded genes to match the secondary metabolic situation in L phase. Finally, the membrane-located complete expression complex with a high local concentration of positive regulator protein and PKS assembly unit contributes to the high-efficiency production and secretion of coelimycin P1.

Recent research has pointed out the correlation between the transcription and the contact frequency between adjacent bins in some species (2). Considering the potential influence of chromosomal topology on a larger scale on the expression of embedded genes, we optimized FIRE value in our case and found that the transcription strength is highly correlated with FIRE value, i.e., local chromosomal interaction frequency. To further investigate the causality between them, in a pilot experiment, we introduced the gus(a) reporter system to some genomic loci that only consider different FIRE values. The consequence comprised numerous divergent outliers, and subsequent analysis showed that the PC between FIRE value and native transcription strength in regions containing the partly selected loci was very low. We thus propose that some extra factor may dominate the transcription in the low PC region and consequently cause the deviation of the inserted reporter system’s expression. So, in subsequent experiments, the HCR was first defined in which the local chromosomal interaction frequency should be the main determining factor for transcription, and finally the high correlation between the FIRE value in HCR and β-glucuronidase activity in all sites confirmed the scenario (Fig. 3 E and F). This significant correlation is reproduced in the subsequent RK-682 test (Fig. 4 C and D). These data showed that the method had the potential application value to promote the production of target natural products due to the obvious enhancement in both β-glucuronidase and RK-682 production compared with the attB control. However, the genuine cause underlying this discovery needs more research to shed light, and some hypotheses were proposed here. The complicated development and differentiation cycle and exquisite spatiotemporal gene regulation of Streptomyces indicate they are a specific advanced species that somewhat resemble fungi, for which some possible high-dimensional gene regulation mechanisms may have evolved. For instance, the prokaryotic enhancer-like element (70) may be enriched in high FIRE value loci and increase the expression of the embedded gene through frequent cis-interaction. Additionally, for some underlying translocation mechanisms, some chromosomal regions may be translocated to the “transcriptional factory”, which was described as the spatiotemporal colocalization between the high-expression gene and the local hot spot for the high concentrations of the machinery required for mRNA biosynthesis (71, 72). The spatiotemporal colocalization accounts for the high local chromosome interaction frequency, and the special hubs make the embedded genes highly efficient in transcription.

Materials and Methods

Bacterial Strains and Growth Conditions.

S. coelicolor M145 strain and mutant strains were grown on solid soy flour-mannitol (SFM) medium for sporulation. Then the spores stocked in 20% glycerol were inoculated to 2×TY medium for germinating. The total cells were collected and inoculated to liquid R5- medium and grown for about 14 h to M phase or 22 h to L phase. E. coli DH10B was used as the cloning host, and E. coli ET12567/pUZ8002 was used for intergeneric conjugation. All E. coli strains were cultured in 2×TY medium at 37 °C. Full information is described in detail in SI Appendix, Materials and Methods and Tables S2 and S3.

The Construction of Plasmid and Streptomyces Mutant.

For plasmid construction, different sgRNA targeted to each HCR window were one-step cloned to NheI and XbaI-digested plasmid pWHU2653 with promoter kasO*p and B1006 terminator by Gibson assembly. Then a series of left and right homologous arms together with the gus(a) reporter cassette or the RK-682 recombinant BGC cassette were introduced to the HindIII site of these sgRNA-born plasmids by Gibson assembly to achieve plasmids pGAM1–10, pGAL1–10, pRKM1–10 and pRKL1–10, respectively. The gus(a) reporter cassette or RK-682 recombinant BGC cassette was integrated into PvuI and XbaI linearized pSET152 to produce pGAattB or pRKattB, respectively. For S. coelicolor mutant construction, each plasmid was transformed from E. coli ET12567/pUZ8002 to S. coelicolor by intergeneric conjugation. All mutants were verified by PCR and sequencing. Full information is described in detail in SI Appendix, Materials and Methods.

Measurement of Glucuronidase Activity and Production of RK-682.

For the glucuronidase activity test, strains were cultured in liquid R5- medium. After reaching the specific growth phase, the whole cells were harvested and resuspended with 40% glycerol. Three-quarters of the cell suspension was separated and centrifugated, then washed twice with distilled water. This part was dried, and then the dry weight was measured. The rest of cell suspension was also centrifugated, the supernatant discarded, and then resuspended with GUS buffer. The cell suspension was then lysed by sonication. Cell lysate with six repeats was transferred to 96-well plates and then supplemented with p-nitrophenyl-β-D-glucuronide. Incubating at 37 °C, the optical density at 415 nm was measured per 5 min for a total of 40 min.

For RK-682 production and analysis, strains were cultured in liquid R5- medium. After reaching the specific growth phase, the whole cells were harvested by centrifugation and then washed twice with distilled water. These cells were dried and then the dry weight was measured. These dry cells were extracted with ethyl acetate, and the extracts were then dried and redissolved in methanol for further analysis. Analysis of RK-682 was performed with a Phenomenex C18 column. LC–ESI–HRMS analysis was carried out on an LTQ XL Orbitrap coupled with an Accela photodiode array detector, Accela PDA autosampler, and Accela pump, using electrospray ionization in the negative-ion mode. Full information is described in detail in SI Appendix, Materials and Methods.

Transcriptome Analysis.

Total RNA was extracted from the tissue using TRIzol^® Reagent according to the manufacturer’s instructions. RNA-seq transcriptome library was prepared following the TruSeqTM RNA sample preparation Kit. After quantifying by TBS380, the paired-end RNA-seq sequencing library was sequenced with the Illumina HiSeq × TEN. htqc was used to filter the sequencing data to make clean data. Then the cleaned reads were mapped onto the S. coelicolor genome using SOAP. Full information is described in detail in SI Appendix, Materials and Methods.

Hi-C Library Construction and Analysis.

For Hi-C library construction, bacterial cells were cross-linked with 3% formaldehyde for 1 h at 4 °C and quenched with 375 mM final concentration glycine for 15 min. The cross-linked cells were incubated with lysozyme at 37 °C for 10 min to lyse the cells. Endogenous nucleases were inactivated with 0.5% SDS, then chromatin DNAs were digested by 50 U Sau3AI, marked with biotin-14-dCTP, and then ligated by 50 U T4 DNA ligase. After reversing cross-links, the ligated DNA was extracted using the QIAamp DNA Mini Kit according to the manufacturers’ instructions. Purified DNA was sheared to 300 to 500-bp fragments, then blunt-end repaired, A-tailed, and adaptor-added before purification through biotin-streptavidin-mediated pull-down and PCR amplification. Finally, the Hi-C libraries were quantified and sequenced on the Illumina Nova-seq platform. Further information about the Hi-C data analysis is described in detail in SI Appendix, Materials and Methods.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Dataset S03 (XLSX)

Dataset S04 (XLSX)

Dataset S05 (TXT)

Data, Materials, and Software Availability

RNA-seq and Hi-C seq data data have been deposited in NCBI BioProject (PRJNA891940) (73).