The tri-stable-switch circuit was designed to switch between two functionalities in response to environmental changes
The tri-stable-switch circuit in the plasmid pETDuet1-1 (Fig. 1a) was designed based on a tri-stable switch derived from the bacteriophage lambda [23]. The mutant lactose-inducible promoter placm (Additional file 1: Table S1) and pH-responsive promoter patp2 (Additional file 1: Table S1) were cloned into this plasmid to sense the signals. The key enzymes applied within the system were the products of lacZ (β-galactosidase, β-GAL, Additional file 1: Table S1) and the fusion gene ompA-lldD (L-lactate dehydrogenase, L-LDH, Additional file 1: Table S1). The strain E. coli BL21 was chosen because it has been commonly used for stable expression of nontoxic exogenous proteins. The circuit was then transformed into E. coli BL21 to form the engineered strain BL21: pETDuet1-1. In theory, BL21: pETDuet1-1 was able to dynamically switch between two functional states, which are regulated by a lactose signal and a pH signal. The theoretical working principle was as follows.
BL21: pETDuet1-1 accumulated β-GAL after it colonized the colon (Fig. 1b). The average pH in the colon has been reported to be 7.0 [24], which, as a signal, maintained continuous cI gene (Additional file 1: Table S1) expression by inducing the patp2 promoter. The expression of cI, which hindered the transcripts of the gene downstream of the pR promoter (Additional file 1: Table S1), suppressed expression of ompA-lldD gene and cIII gene (Additional file 1: Table S1), thus ceasing the function of the pH rescue. At this moment, the engineered bacteria would focus on the expression of the lacZ and accumulate β-GAL for supplementary lactose digestion when unabsorbed lactose fluxed into the colon.
BL21: pETDuet1-1 gradually switched from lacZ expression to ompA-lldD expression after lactose fluxed into the colon (Fig. 1c). On the one hand, the lactose, as a signal, triggered the placm promoter, thus activating the positive feedback loop of pRE promoter, cro gene, and cII gene (Additional file 1: Table S1). The expression of cro then began to suppress lacZ expression after pRM promoter (Additional file 1: Table S1) via binding to its binding site [25]. Additionally, cro expression has been shown to be strengthened by cII expression, which is inhibited to a degree because cII expression is still suppressed by endogenous Ftsh gene expression [26]. On the other hand, fermentation of lactose by the gut microbiota has been demonstrated to produce lactic acid and other short-chain fatty acids, leading to a pH drop within the colon, which would weaken patp2 and inhibit cI expression. However, previously expressed products of cI would continue to suppress the expression of ompA-lldD and cIII to a certain degree, and the suppression would gradually diminish as these products are degraded. Hence, ompA-lldD expression would gradually recover to a normal condition, producing a signal peptide [27], and L-LDH [28, 29] would be translocated to the cell membrane to convert lactic acid to pyruvate in the periplasm. Additionally, the gradual recovery of cIII expression would unsuppress cII expression by inhibiting endogenous expression of Ftsh [26]. Unsuppressed cII expression would then strengthen cro expression, thus accelerating the inhibition of lacZ expression. Once lactose fluxed into the colon, the entire system was in an intermediate state of double functions.
Once the switch process was completed, the BL21: pETDuet1-1 focused on ompA-lldD expression (Fig. 1d), and the suppression of cIII and ompA-lldD expression was removed. Constitutive expression of cIII eliminated the suppression on cII expression via the endogenous expression of Ftsh, thus allowing the lactose-activated positive feedback loop to inhibit the lacZ expression. Expression of ompA-lldD continued to produce efficient signal peptides, allowing L-LDH to convert lactic acid to pyruvate in order to rescue the pH drop (Fig. 1e). Afterward, the pyruvate would be transported into the cell by its carrier protein [30, 31] for usage in the tricarboxylic acid (TCA) cycle [32]. Once the engineered bacteria complete digesting the lactose and restoring the pH, BL21: pETDuet1-1 would subsequently switch to β-GAL accumulation until the next round of lactose ingestion. Thus, in this manner, BL21: pETDuet1-1 would alternate its function in response to the lactose intake.
The tri-stable-switch circuit was efficient under a range of pH conditions in vitro
The interactions between cII & pRE, cI & pR, cro & pRM, and cIII & cII have been tested using fluorescence detection (Additional file 1: Table S2, Additional file 1: Table S3, Additional file 2). The circuit switch was also confirmed to work in theory using mathematical simulation (Additional file 1: Table S4, Additional file 2). In order to test the circuit in vitro (Fig. 2), we prepared mediums with three pH values. In order to simulate the acidic conditions caused by excess lactose intake in the human colon, which typically has a pH of 7 [24], and the mouse colon, which normally has a pH of 5 [33], we adjusted the pH of these mediums by adding 0.1% lactic acid or 1% lactose. The three pH sets included pH set I (initial pH = 4.54 ± 0.012), pH set II (initial pH = 5.34 ± 0.02), and pH set III (initial pH = 6.25 ± 0.02).
We subsequently cultured various bacterial strains, including the test strain (BL21: pETDuet1-1, Additional file 3) and the control strain (BL21: pETDuet1-0) with an empty vector (Additional file 4) for 12 h in these three mediums and recorded the variation in the pH values and the expressed enzyme activity (Additional file 1: Table S5). As shown in Fig. 2a, the pH values of the control culture and the test culture began to increase at 6 h post-inoculation. The increase in the pH within the control culture was associated with two processes: (1) the metabolism of the substantial increase in the bacterial population, and (2) the consumption of the medium. However, the change in pH within the test culture was also dependent on a third process—expression of L-LDH, which helped facilitate the digestion of lactose and the pH increase. The increased pH caused by L-LDH was evident in pH set I. The pH of the test culture increased to a higher degree than that of the control culture (test culture: 4.54 ± 0.02 to 5.31 ± 0.075; control culture: 4.54 ± 0.01 to 4.9 ± 0.072). The pH increase in the test culture was also observed in pH sets II and III, but it was not as obvious as that in pH set I.
As shown in Fig. 2b and c, both the β-GAL activity and L-LDH activity, which were caused by the expression of the lacZ gene and ompA-lldD genes, respectively, in the circuit of BL21: pETDuet1-1, were higher in the test group as compared to the control group. Before 4 h, the enzyme activity measurements were unavailable because of the minimal amount of bacteria. After culturing for 4 h, the β-GAL activity of the test group continued to steadily increase in all three pH sets. In addition, 8–10 h post-inoculation, the β-GAL activity of the test group increased to the greatest extent and later flattened in pH sets II and III. Additionally, the L-LDH activity of the test group began to decrease in pH set II and pH set III 10 h post-inoculation. The corresponding pH range of the test group 8–10 h post-inoculation was 6.43 ± 0.10 to 7.23 ± 0.07 in pH set II and 6.58 ± 0.03 to 7.34 ± 0.07 in pH set III, which indicated that the dual-function switch of the circuit was completed for these pH ranges. These results suggested that relatively low pH values promote L-LDH expression in the circuit in order to remove the lactic acid to prevent the increase in pH. The increased pH then makes the circuit begin to switch gradually from L-LDH expression to β-GAL expression, which would continue until the pH is close to neutral.
The tri-stable-switch circuit helped mice to recover the pH drop caused by excess lactose intake
The in vitro experiments confirmed the theoretical feasibility of the tri-stable-switch circuit to alleviate LI by switching between β-GAL expression and L-LDH expression, but whether it could work in vivo remained unclear. We thus divided 84 mice into five groups, including (1) initial set (n = 4), (2) untreated group (n = 20), (3) model group (n = 20), (4) control group (n = 20), and (5) test group (n = 20) in order to investigate how the circuit functioned in vivo. As shown in Fig. 3a, mice in the control and test groups were administrated bacteria (BL21: pETDuet1-0 in the control group and BL21: pETDuet1-1 in the test group; OD600 = 1) in a total volume of 0.3 mL in a 0.9% NS suspension daily during the first week. The bacteria were confirmed to colonize the colons of the mice, which lasted for at least 24 h (Additional file 2). The other groups were given the same volume of normal saline (NS) daily. The pH of the colons of the mice in the initial set was set as the pH value at 0 h for all groups. At the time point of 0 h, mice of the model, control, and test groups were administrated the lactose solution (12 mg of lactose per 20 g of body weight), and mice of the untreated group were administrated the same volume of 0.9% NS. The pH values of the colons of the mice in the remaining four groups were then tested at each time point (Additional file 1: Table S6) and graphed to illustrate the variation in pH (Fig. 3b). The colon pH of the model group and control group decreased to 4.66 ± 0.15 and 4.72 ± 0.25, respectively, from 0 h to 3 h, and then recovered to 4.89 ± 0.24 and 4.94 ± 0.1, respectively, from 3 h to 6 h. However, the colon pH value of the untreated group without lactose intake and the colon pH of the test group with BL21: pETDuet1-1 were relatively stable. Thus, these results indicated that the tri-stable-switch circuit prevents the pH drop in mouse colons caused by an excessive intake of lactose, thereby restoring intestinal homeostasis and relieving LI.
Moreover, we then tested fecal β-GAL activity using another set of the four groups of mice, including (1) untreated group (n = 3), (2) model group (n = 3), (3) control group (n = 3), and (4) test group (n = 3). The operations in the first week and at the time point of 0 h were the same as the operations described above (Fig. 3c). We then tested the β-GAL activity in the feces of mice before the time point of 0 h, and at the time point of 3 h (Additional file 1: Table S7, Fig. 3d). The β-GAL activity of the test group was found kept at a high and stable level (P = 0.27, Student’s t test), suggesting that the colonization of the BL21: pETDuet1-1 has prepared enough β-GAL activity for the following lactose intake. No evident variation in the β-GAL activity of the untreated group was observed as well (P = 0.68, Student’s t test). Nevertheless, the β-GAL activity of the model and control groups significantly decreased at 3 h (model group: P = 0.0072, control group: P = 0.0015, Student’s t test). These results indicated that the colon pH drop might decrease the intestinal β-GAL activity, and the tri-stable-switch circuit could keep intestinal pH stability and high intestinal β-GAL activity.
The tri-stable-switch circuit helped the murine gut microbiota recover from the effects of excessive lactose intake
In order to understand the effects of the engineered bacteria on the murine gut microbiota, we conducted a time-series experiment using a high-frequency sampling of mice fecal samples (Additional file 1: Table S8). As shown in Fig. 4a, four groups of mice (i.e., untreated group, model group, control group, and test group) were subjected to different interventions. The experiment lasted for 21 days and was divided into the four phases: normal care (Phase I), lactose challenge (Phase II), bacterial treatment (Phase III), and restoration (Phase IV). For Phase I, during which the four groups received normal care, the objective was to stabilize the physical signs and the gut microbiota of the mice in the four groups. For Phase II, during which lactose was fed to the model, control, and test groups, the objective was to investigate the influence of excess lactose on the gut microbiota. Phase III, in which BL21: pETDuet1-1 was fed to the test group while empty-vector-containing BL21: pETDuet1-0 was fed to mice in the control group, was used to determine whether BL21: pETDuet1-1 can alleviate LI. In Phase IV, we intended to observe whether the bacteria caused any side effects in the host mice.
The dynamics of the mice gut microbiota differed among the four groups over the 21-day trial. From days 3 to 11, most of the gut microbiota samples from the Untreated group trended toward the right end of the principle coordinate 1 (PCo1) axis, whose degree was more considerable than those of the mouse groups administrated with lactose (Fig. 4b, c). In other words, excessive lactose intake inhibited the shift in the microbiota toward the right end of the PCo1 axis during this period, which began to be obvious at day 7. Nevertheless, among lactose-affected gut microbiotas, only those of the test group arrived at the same degree as those of the untreated group (Figs. 4b, c) after a time lag. Thus, it appeared as though the engineered bacteria were able to weaken some restrictive effects of the lactose.
We then constructed an unweighted-glasso network based on amplicon sequence variants (ASVs). To figure out the exact taxa affected by the inhibitory effects, we calculated the mean abundances of the top 50 most abundant ASVs in the network using samples from three data subsets including “normal condition,” “lactose feeding,” and “treatment” (Fig. 4d, Additional file 1: Table S8). We used samples of the untreated group at days 7 and 11 for the network of “normal condition” because they were most prominent characteristics in the normal stage (Fig. 4c). We used the samples from the model and control at days 5, 7, and 11 and test groups at days 5 and 7 for the network of “lactose feeding” because these samples were under the effects of lactose. We used the samples from the test group at days 11 and 13 for network of “treatment” because these samples have been treated with engineered bacteria. The networks showed that the microbial patterns would be largely affected by lactose intake and then recovered to the pattern that was similar to the original normal pattern after bacterial treatment (Fig. 4d).
Moreover, we found that 35 out of 50 most abundant ASVs in the networks were classified as class Bacteroidia (Additional file 1: Table S9), and genus Bacteroides was the most common genus in the altered microbiota and whose altered abundance was highly accordant with the microbiota variation against PCo1 (Fig. 4c, e). As compared to the “normal condition” network, 29 out of 50 most abundant ASVs were differentially abundant in the “lactose feeding” network, but had an abundance similar to the “treatment” network. For instance, an ASV that is most likely Bacteroides acidifaciens (confidence: 0.90) had an abundance of 0.285 ± 0.036 in the “normal condition” network and a similar abundance of 0.282 ± 0.037 in the “treatment” network, but it had a decreased abundance of 0.199 ± 0.016 in the “lactose feeding” network (P = 0.027, Wilcox test). Other ASVs that were also most likely Bacteroides acidifaciens were found to have similar distributions (Additional file 1: Table S9). Additionally, an ASV that is most likely Lactobacillus murinus (confidence 0.92) had increased abundance in the “lactose feeding” network as compared to that in “normal condition” network (P = 0.001, Wilcox test), while there was no significant difference in its abundance found between the “normal condition” network and the “treatment” network. These results indicated that excessive intake of lactose might inhibit the growth of the genus Bacteroides in the gut microbiota of mice during Phase II, while the administration of BL21: pETDuet1-1 removed this inhibition, such that the murine gut microbiota proceeded to the variation in Phase III, similar to the gut microbiota of the untreated mice in Phase II.