Results

Summary of Results

---

Our project successfully characterised a novel phosphatase, PafA (BBa_K5172000), different from all previous phosphatases on the registry as it exhibits minimal end-product inhibition, making it an ideal candidate for industrial applications. We also applied LigandMPNN to produce very stable variants of the enzyme, to enable our enzyme to be used in harsh conditions such as wastewater treatment. We created a guide for how to redesign enzymes using ligandMPNN (see our AI redesign guide - AI redesign cycle). The few redesigns tested exhibited low activity compared to the wild-type. Going forward, we plan to repeat the redesign cycle, and “fix” (maintain) a larger number of residues, so the high activity of the wild-type PafA can be conserved. You can see a summary of our results below:

• Consurf was used to output the top 70, 50 and 30% conserved residues from PafA. Residues within 5Å of the docked ligand and two Zn²⁺ ions and in each redesign either 70, 50 or 30% of the most conserved residues were designated as fixed. This ensured that LigandMPNN kept the active site and most evolutionarily “important” residues constant.
• Due to time constraints, only six redesigns were chosen to order as gBlocks from Integrated DNA Technologies (IDT). Two from each level of conservation were selected: 70.1 and 70.2; 50.1 and 50.2; 30.1 and 30.2. A phylogenetic tree of redesigned PafAs was used to select the most distinct sequences to test the most diversity in sequences.
• Of the six redesigns of PafA, only 70,1, 70.2, 50.2 and 30.1 were successfully cloned into the backbone of pET21a(+)-BsaI-sfGFP due to errors in G block amplification.
• PafA, PhoA codon optimised for E. coli but derived from Citrobacter (BBa_K3767001), and variants 70.1, 70.2 and 50.2 were easily expressed and the protein was produced in E. coli however after purification, SDS showed that variant 30.1 was degraded. Size exclusion chromatography was used to remove impurities, during which 70.1 was lost.
• Redesigned PafA variants remained thermostable, up to >95°C - As opposed to wild-type PafA and PhoA, which aggregated around 70°C.
• PafA from F. johnsoniae (BBa_K5172000) displayed much higher activity compared to the most active phosphatase on the registry, PhoA (BBa_K3767001).
• Variants displayed some, but minimal, phosphatase activity, indicating that a larger pool of redesigned sequences produced by LigandMPNN would need to be characterised to identify one with activity comparable to the wild-type. More residues need to be identified and fixed in future redesigns to ensure the activity is maintained.
• An ideal working pH range for PafA from F. johnsoniae (BBa_K5172000) was determined to be between pH 7.4 and 9.5. Further increments within this range would be required to identify an optimum pH.
• Due to the low activity of the variants, assaying the optimum pH proved unsuccessful however, it was discovered that all of the variants at high concentrations in temperature assays were deemed to be very thermophilic and worked optimally at higher temperatures.

Redesign of PafA

---

Figure 1: Graphical representation of the AI redesign cycle described in the AI guide.

Using our AI guide (AI redesign cycle), with the cycle summarised in Figure 1, we redesigned the PafA enzyme, generating the sequences in the following spreadsheet: Sequences. Briefly, we used Consurf to identify the top 70%, 50% and 30% evolutionarily conserved residues (Figure 2) and ChimeraX to identify residues within 5Å of the bound ligand (G6P) or two cofactor Zn²⁺ ions. These were the residues that LigandMPNN was instructed not to sample in the design process, i.e. “fixing them”.

Figure 2: Showing the percentage of residues fixed, prioritising fixing the more evolutionarily conserved residues, identified by Consurf.

Due to time constraints, we generated a phylogenetic tree by aligning protein sequences using MUSCLEv5 then the tree was built with iqtree. This was then uploaded to iTOL and displayed as an unrooted tree (Figure 3)

Figure 3: Phylogenetic tree of sequences produced from LigandMPNN generated by aligning protein sequences with MUSCLEv5 then the tree was built with igtree visualised as an unrooted tree on iTOL. Circled sequences were ordered as G-blocks from IDT. The starred sequence shows the wild-type PafA sequence.

Some of our results from a cycle of the redesign visualised on ChimeraX can be seen in Figures 4 and 5. In this cycle, we constrained 30% of the residues, based on ConSurf conservation scores and residues within 5 Ångstroms of the ligand and Zn²⁺ ions, identified using ChimeraX. LigandMPNN generated sequences with increased stability which was then input into AlphaFold to predict a new protein structure; as a result the space within the active site may have changed in a way that we didn’t visualise which could explain some of the reductions of activity explored later in the results. Figure 5 shows that the backbone of PafA was maintained in the redesigns, even in the designs with the least amount of fixed residues (30%).

Figure 4: A ChimeraX overlay of the wild-type (cream) and improved versions of PafA. (A) displays 30.1 in blue, (B) displays 50.2 in pink, (C) displays 70.2 in green. The ligand binding site is conserved between the two structures.

Figure 5: An overall structural comparison of the wild-type (cream) and 30.1 (blue) as visualised in ChimeraX

Cloning of gBlocks into a pET-21a(+) backbone

---

A sample of pET-21a containing pafA from F. johsoniae was kindly gifted from the lab of Dr Andrew Hitchcock, Photosynthesis Research Group - University of Sheffield as shown in Figure 6. We have a modified backbone: pET-21a(+)-BsaI-sfGFP. The addition of BsaI restriction sites, our defined overhangs and a default fluorescent marker facilitated cloning by type IIS digestion, and the lack of fluorescent phenotype is indicative of successful ligation (Figure 7). This was used to clone our G-blocks.

Figure 6: A plasmid map of pAH668 (pet21a + PafA)

The PafA variants and PhoA were cloned into the pET21a(+)-BsaI-sfGFP backbone, Figure 7, to allow easier screening of transformations.

Figure 7: The plasmid map of the backbone pET21a(+)-BsaI-sfGFP.

Expression Results

---

Following cloning in DH5ɑ and sequencing, plasmids were miniprepped and transformed into BL21 LEMO. Protein was purified via Immobilised Metal Affinity Chromatography (IMAC), and relevant fractions were analysed via SDS-PAGE electrophoresis. Purification of 30.1 (see Figure 8) yielded a large band at around 24 kDa and no band at the predicted size of ~58 kDa, suggesting that this protein has been degraded. Observed smaller molecular weight bands corroborate this theory. For all other PafA and variant purifications, a high yield was observed at the correct size. For PhoA, yield was lower but still significant in elutions 4-6. Elutions with proteins were pooled and the buffer was exchanged into a glycerol-based buffer.

Figure 8: SDS-PAGE results of PhoA and the PafA variants post-protein purification. The flow-through (FT), binding buffer (BB), wash buffer (WB), and elutions (E1-10) were all run on the gel to identify which elutions contained protein.

Due to the extra banding present, it couldn’t be guaranteed that the protein concentration determined via Biodrop accurately represents the concentration of our desired protein, or if contamination impacted the accuracy. Therefore, further size exclusion chromatography (SEC) was carried out. Then, 10µl of the sample was run on an SDS gel (see Figure 9). This enabled us to pool fractions which specifically had our protein in (and the least amount of impurities) for assays. The protein concentrations of each of the variants post-SEC are shown in Table 1.

Figure 9: SDS-PAGE results of PafA WT, PhoA, and PafA variants post-SEC. 30.1 had the wrong-sized bands, suggesting degradation

Table 1

PafA	1.28 mg/ml	20650nM
30.1	0.352 mg/ml	6060nM
70.2	0.465 mg/ml	7740nM
PhoA	0.213 mg/ml	4420nM
50.2	263 mg/ml	4860nM
70.1	0.046 mg/ml	762nM (Lost in SEC)

Determining the stability of the proteins

---

Purification steps expose the proteins to harsh conditions, such as Imidazole, that could cause them to unfold, reducing the accuracy of upcoming characterisation assays. Circular dichroism (CD) experiments were carried out to ensure the proteins were folded correctly. Circular dichroism spectroscopy can show the stability of the proteins by determining the temperature at which they aggregate out of solution or unfold. For our uses, we primarily used CD to deduce the thermostability of our proteins.

Figures 10A and 11A show the CD results plotting ellipticity against wavelength for PafA and PhoA respectively. At 222nm, and 208nm (figures 10A and 11A), dips are present, indicating that both phosphatases are consistent primarily of alpha helices. As the temperatures increase, plots flatten out as the protein aggregates from the solution or unfolds. Both PafA and PhoA can be seen to fully aggregate out of solution by 70°C, seen in figures 10C and D and 11C and D respectively.

Figure 10A shows a repeat at 50℃ to check the circular dichroism spectrophotometer was appropriately heating the sample, by heating the repeat in a heat block to 50°C and remeasuring. This caused the two plots for 50°C on Figure 10A to initially not align. However, when normalised as shown in Figure 10B, they looked the same, meaning that even though the spectrophotometer was functioning correctly, the evaporation from heating the same sample for a longer period had caused a change in its concentration, resulting in the unnormalised plots differing in appearance. Figure 11A shows the CD results plotting ellipticity against wavelength for PhoA. The results show a repeat at 25°C, “25 E2” was due to us testing elution 2 of PhoA from the Nickel Column Purification Step, however as shown on the graph the plot is relatively flattened which suggested to us that no protein was present in this sample. Therefore we started the CD process again with another elution and the plot looked closer to a typical CD plot.

Figure 10: (A) CD results for PafA WT, (B) & (C) a comparison of ellipticity to temperature at each wavelength for alpha helices on PafA WT.Figure 11: (A) CD results for PhoA, (B) & (C) a comparison of ellipticity to temperature at each wavelength for alpha helices on PhoA. (D) Percentage of maximum ellipticity (%) against temperature (℃) for normalised PhoA and PafA.

Figure 11: (A) CD results for PhoA, (B) & (C) a comparison of ellipticity to temperature at each wavelength for alpha helices on PhoA. (D) Percentage of maximum ellipticity (%) against temperature (°C) for normalised PhoA and PafA.

The variants of PafA were then put through the CD process, as shown in Figure 12, the plots of ellipticity against wavelength do not flatten out as observed on the corresponding PafA and PhoA graphs. This demonstrates that the variants don’t unfold or aggregate out of solution at the limit of the CD spectrometer’s heating capabilities, 95°C.

Figure 12: CD results for PafA variants 30.1, 50.2, 70.1, and 70.2.

Figure 13 confirms these results as when the ellipticity at 208 and 222 nm is plotted against temperature the sigmoidal shape is not present which confirms that the variants are not aggregating or unfolding, furthermore, the ranges in ellipticity values are much smaller with more consistent values.

Figure 13: CD results for PafA variants at wavelengths 208nm and 222nm (corresponding to alpha helix structure)

Characterising the activity of PafA, PhoA and the redesigns

---

para-Nitrophenyl Phosphate (pNPP) was used to characterise the activity of F. johnsoniae’s PafA and its redesigned variants. pNPP is converted, via phosphatase activity, to inorganic phosphate (Pi) and pNP, a yellow-tinged product that can be identified via measuring absorbance at 405 nm. Our team compared the activity of PafA to a codon-optimised version PhoA from Citrobacter. PhoA (BBa_K3767001) was optimised by Queens Canada 2021 to have 40x the catalytic activity of the previous best phosphatase on the registry (BBa K1216001) making it previously the most active phosphatase on the iGEM registry. Since a plate reader (Tecan Spark) was used to accelerate these assays, upper and lower limits of measurable absorbance needed to be determined, as well as identifying the relationship between pNP concentration and absorbance at 405 nm. These were determined using a 96-well plate with concentrations of pNP ranging from 20mM to 1.19nM. Five readings were taken of each well, and each well was duplicated; the five readings were then averaged, and the absorbance at each of these concentrations was plotted. Figure 14 shows that the linear relationship seems to be maintained up until an absorbance between 2.5 and 3, so the upper bound of the machine was assumed to be an absorbance of 2.5.

Figure 14: As the concentration of pNP in the reaction buffer increases, the absorbance of the solution also increases, in a positive linear relationship up until an absorbance of ~2.5.

The equations of the graphs in Figure 15 are similar (y=2.0997x+0.1107 versus y=2.0774x+0.193), however, due to the high distribution of readings under absorbance of 0.2, the lower bound needed to be determined. Below an absorbance of 0.118, readings for the two repeats become further apart, and below an absorbance of 0.112, the linear relationship collapses. Therefore a lower bound of 0.112 was used.

Figure 15: The two repeats carried out for the relationship between pNP concentration and absorbance.

Figure 16: Determining the lower bounds of the machine. Two repeats of [pNP] against absorbance at 405nm. Points below an absorbance of 0.12 were visualised on the graph to identify if the linear relationship is maintained at lower absorbance levels.

Figure 17 takes a mean of the two repeats, and determines the equation of the line to be y=2.0814x+0.113. The equations of the line from figures 15 when rounded to 2sf are the same as this, therefore the equation of the line is taken as y=2.1x+0.11, accurate to 2sf, where y is absorbance and x is the concentration of pNP (mM).

Determining the activity of PafA compared to PhoA

Absorbance measurements, as shown in Figures 18 and 19, were converted to the concentration of pNP using their determined linear relationship. Gradients of the linear section of the line were determined using a Matlab script that identified the gradient between every successive 5 points, and outputted the maximum gradient, capturing the maximum velocity of the reaction without being thrown off by artefacts near the beginning of each time series. By automating this process, we saved time and increased accuracy against the traditional technique of plotting all the concentrations of pNP against time graphs and determining gradients by hand- a method that is limited by determining qualitatively where the linear section ends and the graphs begin to level out. The gradients were normalised by enzyme concentration and were plotted against pNPP (substrate concentration) and GraphPad Prism was used to fit the data to a Michaelis-Menten curve (Figure 20).

Figure 18: As the concentration of pNPP substrate increases, PafA WT shows a greater rate of catalysis and is able to react more substrate before plateauing.

Figure 19: As the concentration of pNPP substrate increases, PhoA shows a greater rate of catalysis and is able to react more substrate before plateauing

Figure 20: Michaellis-Menten fit on GraphPad Prism for [pNPP] (mM) against velocity of reaction divided by enzyme concentration (v/[E]) (min^-1). A. PafA at an enzyme concentration of 1nM, and B. PhoA at an enzyme concentration of 1nM.

Figures (20) and Prism showed that PafA has a V_max of 13210 min^-1 ± 720 (95% Confidence interval (CI)), and Km is 0.02236 mM± 0.00524 (95% CI). PhoA has a V_max of 6774 min^-1 ± 405 (95% CI) and Km of 0.01549 mM ± 0.014422(95% CI), showing that PafA was more active than PhoA (BBa_K3767001), making PafA (BBa_K5172000) the most active phosphatase currently on the registry.

The activity of the Redesigned PafA Variants

Only our PafA variants 70.2, 50.2 and 30.1 were able to be characterised, as 50.1 and 30.2 were never successfully cloned and 70.1 was lost during SEC. The variants had a lot less activity than the wild-type (see Figure 21), suggesting more residues would need to be fixed in the next cycle of redesign, or a much larger number of redesigns would need to be ordered and characterised, neither of which was possible within the iGEM timeline.

At 1nm and 10nm, there’s no activity when compared to the control of no enzyme (see Figures 21 and 22 showing 70.2 compared to the control).

Figure 21: A control of pNPP without enzyme, as the pNPP concentration increases, absorbance.

Figure 22: Increasing the concentration of pNPP with (10nM) 70.2 increases the absorbance of the solution due to an increase in the enzyme’s activity. However, the absorbance at each concentration does not increase with time.

At 100nM of the enzyme variants, as shown in Figures 23, 24 and 25, the activity is observable, however much lower than that of our WT enzymes PafA and PhoA (see Figures 18 and 19).

Figure 23: As pNPP concentration increases, this increases the absorbance of each solution with the (100nM) 50.2 variant. Each solution’s absorbance i.e. enzyme activity increases exponentially.

Figure 24: As pNPP concentration increases, this increases the absorbance of each solution with the (100nM) 30.1 variant. Each solution’s absorbance i.e. enzyme activity increases exponentially.

Figure 25: As pNPP concentration increases, this increases the absorbance of each solution with the (100nM) 70.2 variant. Each solution’s absorbance i.e. enzyme activity increases.

At 100nm, variants displayed a low level of activity (figures 23-25). It’s important to note that, when observing Figures 23-25, 30.1 is seemingly as active as 70.2 and 50.2. This is despite 30.1 being predominantly degraded when run on SDS-PAGE ( Figure 9). All variants have very low activity, and the activity of 30.1 being similar to other assayed variants could either suggest that the other variants have been inactivated following the gels, prior to assaying. This seems unlikely as the WT PafA and PhoA were purified and stored for the same lengths of time, and variants are proven to be more stable. It is also possible that the components that 30.1 was degraded into maintained some level of activity. There’s limited knowledge on which aspects of the PafA enzyme are responsible for activity, so this is possible, however unlikely. Further redesign cycles with a greater number of residues fixed could help to maintain phosphatase activity in the redesigns.

Determining ideal pH for PafA and redesigns

---

PafA’s activity at different pHs was determined using sodium acetate buffer (pH 4-6.5) and Tris buffer (pH 6.5- 9.5) adjusted to a pH range. An overlapping pH for the buffers of 6.5 was included in Figure 27 It can be seen that the usage of different buffers had no impact on the activity of PafA. pNPP undergoes low-level spontaneous hydrolysis to form pNP (Duarte et al., 2014). Whilst hydrolysis of pNPP has a high activation energy, high temperatures and pHs can enable this activation energy to be overcome more easily (Duarte et al., 2014). Therefore it was important to ensure the data of pH was blanked, with substrate so as to account for any non-enzymatic hydrolysis that occurs before enzyme addition. The ideal pH range appears to be within the range of 7.4-9 as shown in Figure 26 and 27, however, more increments would be needed within this range to identify the optimum pH.

Figure 26: Results from a pH assay of PafA WT, showing that PafA has the greatest activity at pH 8-9. Only at a pH of ~9 does the concentration of pNPP affect the enzyme activity rate.

Unfortunately, it wasn’t possible to identify the pH range for the redesigned PafA variants as their activity was too low to determine an optimum. We attempted to get results for this, however we realised that the data collected lacked a trend across the ranges and lacked a peak in activity, as seen for PafA in figure 26. This is likely due to activity being so low that the plate reader wasn’t able to accurately differentiate the different activity levels at the different pHs, therefore we have chosen not to report upon this misleading result.

Determining the ideal Temperature for PafA and redesigns

---

Figure 30 Shows the Absorbance at 405nm after 10 minutes of pNPP hydrolysis of PafA at increasing temperatures, from this we can see that after 60°C the activity of the enzyme drops significantly and at 90°C the enzyme is inactive with no change in hydrolysis compared to the control. As shown in Figure 31 we can see the relative activities of the variants and PafA, as the activity of the enzymes is much lower than PafA we decided to display this data as a percentage of maximum absorbance as from this we can see the optimum temperature regardless of the rate of the enzymatic activity and be able to compare results. All of the variants had an optimum temperature of approximately 80°C after negligible activity at lower temperatures.

Figure 30: Results of a temperature assay for PafA WT. Absorbance remains relatively constant up to 70°C and then rapidly decreases before reaching an absorbance of 0 at 90°C.

Figure 31: Temperature assay results for PafA WT and its variants at 10-90°C, showing that variants 30.1, 50.2 and 70.2 have the highest maximum absorbance percentage at ~80°C but PafA has the highest maximum absorbance percentage at ~37°C.

Due to this result, it was not clear to what degree this change in absorbance was due to temperature dependent hydrolysis of pNPP to pNP, leaving little to no pNPP for the enzyme to hydrolyse. To check this, we pre-heated pNPP to 80°C and 90°C (alongside a non-heated control), and allowed it to cool before conducting the same assay, these values were blanked with PNPP solutions that had not been preheated at the corresponding temperature. As you can see in Figure 32 there is no significant difference in the absorbance values depending on if the pNPP has been pre-exposed to high temperatures allowing us to determine that the variants did indeed have an optimum temperature of around 80°C.

Figure 32: Temperature assay results for PafA WT and its variants with pNPP preheated to 80 and 90°C compared to no preheating (represented by 0).

Determining if minimal end-product inhibition is maintained in redesigns

---

A concern we had during the redesign is that it is currently unknown which structural region(s) of the PafA enzyme are mechanistically responsible for the minimal end-product inhibition. Therefore, during the redesign process, it wasn’t possible to fix important residues needed to maintain this property. By fixing evolutionarily conserved residues (at 70%, 50% and 30% of the top conserved residues being fixed) the hope was that this property would be maintained since minimal product inhibition seems to be evolutionarily conserved across PafA variants. To ensure this was the case, we attempted to assay the end product inhibition of PafA and the designs. However, as identified in earlier assays, non-enzymatic pNPP hydrolysis increases at non-ideal storage conditions, e.g. high pH’s. The phosphate salt used (H₂NaPO₄) acts as a weak acid, lowering the pH of the buffer. As we hadn’t observed end product inhibition at as high as 20mM pNPP - whose hydrolysis would release Pi - we predicted that high concentrations of Pi would be required to observe the minimal end product inhibition quality. From Figure 33 it can be seen that increasing concentrations of H2 NaPO4 caused the pNPP to spontaneously hydrolyse, forming pNP.

Figure 33: Results of blanking the Pi solution. As the Pi solution concentration increased, the V_max also increased in a positive linear relationship.

Furthermore, likely, the change in pH due to the addition of the weak acid meant that the buffer (previously at pH7.4) was no longer within the ideal working range for the enzymes, so whilst the results of Figure 34 suggest PafA is inhibited by higher concentrations of inorganic phosphate, this may also be due to the decrease in pH making the conditions non-ideal.

Figure 34: Absorbance at 405nm against time (minutes) for PafA at a range of H₂NaPO₄ concentrations.

PhoA (Figure 35) seems to show that PhoA was inactive, this may be attributed to us postulating that PhoA does not remain active with different divalent cations to its preferred Mg²⁺ due to been assayed with Zn²⁺.

Figure 35: Absorbance at 405nm against time (minutes) for PhoA at a range of H₂NaPO₄ concentrations.

Conclusion

---

Our project successfully characterises the novel phosphatase, PafA demonstrating its significantly higher activity compared to the previously best-performing phosphatase on the iGEM registry PhoA (BBa_K3767001).

We were able to employ AI tools including LigandMPNN to redesign PafA to increase its suitability for industrial applications. Circular Dichroism (CD) analysis confirmed that the redesigned variants possessed superior thermostability compared to wild-type PafA, remaining stable up to 95°C, compared to PafA unfolding at 70°C. However, the activity of the redesigned variants was notably lower compared to the wild-type PafA. This suggests that during the redesign process, the active site of the enzyme was disturbed. To remedy this, we could try to fix a larger number of residues around the active site to achieve variants with both high activity and thermostability. The limited number of variants tested was another factor that may have contributed to the lower activity. A larger pool of redesigned variants would provide a greater diversity of sequences and increase the likelihood of identifying highly active variants.

Alternatively, it’s possible that there exists an inherent tradeoff between the rate of catalysis and the rigidity of an enzyme. Many psychrophilic enzymes have evolved to be more flexible than their homologues — sacrificing thermostability for a higher Kcat. It’s possible that, by dramatically increasing the thermostability (and therefore rigidity) of our PafA variants, we’ve done the reverse and have sacrificed Kcat.

In conclusion, this research provides valuable insights into the potential of AI-driven enzyme redesign for enhancing the properties of phosphatases. While further optimisation is required to achieve PafA variants with both high activity and thermostability, the results demonstrate the promise of this approach for developing enzymes suitable for industrial applications. For more information about our proposed implementation into a column device and the results we obtained regarding this see, Column

References

---

• Duarte, F. et al. (2014) ‘Resolving apparent conflicts between theoretical and experimental models of phosphate Monoester hydrolysis’, Journal of the American Chemical Society, 137(3), pp. 1081–1093. doi:10.1021/ja5082712