Our project was divided into two wet-lab streams: Surface Display of Nanobodies and Construction of fuGFP-Cellulose Binding Domains fusion protein. These streams would come together to allow for the screening of novel nanobodies against GFPs generated through DNA shuffling.
Statistics were used for data visualisation in both wet-lab streams. In cases where multiple replicates were sampled, we constructed generalised linear models using the “glmmTMB” package (Brooks et al. 2017) in R version 4.2.1 (R Core Team 2022). We used the “testResiduals” function in the “DHARMa” package (Hartig 2022) to assign model families and test assumptions, and then performed Wald Chi-square test using the “car” package (Fox & Weisburg 2019) to extract p-values from models. Where we observed a significant result, we used the “emmeans” package (Lenth 2022) to perform Tukey-Kramer post-hoc contrasts.
Summary
The surface display side of our project focussed on the expression and evolution of anti-sfGFP nanobodies to anti-fuGFP nanobodies through DNA shuffling. Expression was done in the TOP10 E. coli using pUS250 as our vector. Two layers of golden gate cloning were used to clone, and the cymR operon was used to induce expression. Once expression of pre-existing nanobodies had succeeded, we successfully created a new variant nanobody in vitro through DNA shuffling, cloned and transformed and expressed the variant nanobody as well.
Achievements:
Anti-sfGFP Nanobodies
Construction and transformation of Neae-Intimin anti-sfGFP nanobody vector into E. coli
Our first stage focused on inserting our Neae-Intimin Nanobody construct into TOP10 E. coli. This was done through two stages of cloning and transformations - firstly the Neae-Intimin, then the nanobodies (see Experiments for more detail). At each stage we harnessed phenotypic blue-white screening and junction PCRs to confirm successfully cloned colonies.
Creation of surface display plasmid
In order to display our collection of anti-sfGFP nanobodies, we created a plasmid which contained Neae-intimin, a bacterial protein which can display other proteins (i.e. our nanobodies) on the surface of E. coli. As shown in Figure 1, our Neae part contained a lacz-alpha gene flanked by Golden Gate cloning sites so that we could clone in our nanobodies after the fact. This means that this first stage of cloning was just taking the Neae-intimin gene and inserting it into our plasmid backbone.
By replacing the amilCP gene in our plasmid backbone with the Neae-intimin part via Golden Gate cloning, we were able to use the blue colour produced by amilCP as a method of screening. Any blue colonies we saw after transformation had the original backbone, and any white colonies had the Neae-intimin insert.
As can be seen in Figure 2, we were able to use amilCP blue white screening to identify colonies that were likely to contain our Neae insert. To confirm this, we used junction PCR on a few of these white colonies. The gel from this colony PCR can be seen in Figure 3.
The gel in Figure 3 shows that almost every colony we sampled amplified with the expected size of ~350 bp. This means that all of these colonies except for colony 4 had the Neae insert. From the PCR results and our blue-white screening with amilCP, we concluded that we had successfully inserted our Neae part into our plasmid backbone pUS250, and we were ready to progress to the next step of our project. We purified the plasmid from colony 2 to use as our pUS250-Neae plasmid for future stages of the project.
Insertion of anti-sfGFP nanobodies into pUS250-Neae
Once we had purified our plasmid from the last step, we were ready to clone in our anti-GFP nanobodies. To do this we used the Esp3I Golden Gate sites that are present on either side of the lacZalpha region in our Neae part. This means that when we cloned in our nanobodies, they replaced the lacz-alpha. We did five Golden Gate clonings and transformations, one for each of the five anti-sfGFP nanobodies.
In combination with the chemical X-Gal, the lacZalpha gene creates a blue colour. This means that we were able to use blue-white screening again to see which colonies had successfully recombinant plasmids. When plated on agar containing X-Gal and IPTG, the blue colonies were our Neae plasmid with no insert, and the white colonies contain the Neae plasmid with a nanobody insert (Figure 5).
Based on the blue-white screening, we concluded that it was very likely that all of the white colonies contained our insert of interest. To further confirm this, we performed a junction PCR on a few white colonies.
The PCR gel (Figure 6) combined with the blue-white screening were strong evidence that the colonies we had picked did indeed contain our anti-sfGFP nanobodies. We had 4 colonies for each nanobody, besides H where we had three, and 7 where we had two (numbered three and four). We decided to proceed with colony number three for each Nanobody.
Testing the Function of Neae-intimin Construct Displaying anti-sfGFP Nanobodies
In order to test the function of our construct we performed a GFP binding assay (see Protocols) where purified GFPs are added to induced cells, then washed off and measured for fluorescence. Preliminary testing showed that four out of the five nanobodies appeared to bind to sfGFP, with H binding markedly stronger than the other nanobodies. This was validated by our modelling results (see our Modelling page).
The next step was to validate the Neae-intimin construct as a negative control and nanobody H as the positive control for testing of shuffled nanobodies later on. We repeated the Nanobody-GFP binding assay with 5 replicates for each data point for just sfGFP. Induced and uninduced cells expressing just Neae with no nanobody and Neae-NBH were tested. Two different concentrations were measured to refine the assay. Figure 9 shows this. Significances were calculated from a tukeyHSD post-hoc contrast. No significant difference was found between uninduced versions of Neae and H. A significant difference was found between the induced nanobody H binding and all other data points. This validated the use of H as a positive control.
DNA Shuffling of anti-sfGFP Nanobodies
Based on the above results, we were confident that our surface display method of testing nanobody binding was successful. The next step of this project was to create variations on these nanobodies using DNA shuffling, and test the difference in their function.
DNA Shuffling
We shuffled together all of the five anti-sfGFP nanobody sequences that we obtained from the literature. The first stage of this was mixing the five sequences together in one reaction, and cutting them with DNAse I. To confirm that this digest was successful, we ran a gel comparing the product of our digest with the undigested nanobody sequence.
The comparison of the undigested fragment mixture and the digested in Figure 10 clearly shows that the DNAse I digest was successful. Based on this, we proceed to the next phase of DNA shuffling, which is the reassembly of the pieces into new fragments. Once we had performed this reassembly, the new reassembled fragments were PCR amplified.
The gel shown in Figure 11 indicates that there was reassembly of the digested fragments. This indicates that some complete fragments, like digested and reassembled nanobodies, were able to be amplified. This supports the idea that the shuffling steps did proceed successfully, as the gel results correspond to what we would expect.
We then cloned the shuffled products into pUS250-Neae backbone and then transformed into TOP10 E. coli. From the resulting control plate on X-Gal IPTG, all colonies appeared white, indicating efficiency of transformation and successful transformants.
Unfortunately, sequencing results showed a different story. All 11 sequences came back as the lacZ-alpha operon. This indicates that the blue-white screening failed in this case (the plates were not at the right concentration of X-Gal IPTG).
A second batch that made it through our screening for nanobodies binding fuGFP (see screening Protocol) were sent off for Sanger sequencing. The sequences showed that all colonies contained the lacZ-alpha operon, meaning that the screening did not adequately isolate only fuGFP binding nanobodies.
After an Esp3I digest to remove background pUS250-Neae, nine colonies were isolated. A spanning PCR showed one colony (colony #7) with the correct length for a shuffled nanobody, named ‘Allocamelus’ by our team (Figure 12).
All nine colonies and the original 5 anti-sfGFP nanobodies were put through a digest with MspI, and ran out on a gel (Figure 13). The gel showed a unique pattern of bands for Allocamelus when compared to the five anti-sfGFP nanobodies. This was a good indication that Allocamelus may be shuffled.
The allocamelus or ass-camel is a mythical creature with the head of a donkey or mule and the body of a camel (Topsell, 1658). This creature features in British heraldry (Smith, 1928). We named our one successfully shuffled nanobody after this strange beast because our nanobodies sequences come originally from camelids (of which the allocamelus is one) and that they are shuffled, much like the camel and donkey are shuffled together to create the allocamelus. Although it probably wasn’t necessary to give this shuffled nanobody such a fanciful name, we decided to have a bit of fun with it.
After sequencing our “allocamelus” nanobody, we realised that it fit better with the name than we ever anticipated! It turns out that this nanobody was a mix of just two of our original nanobodies (NB2 and NB3), just like how allocamelus is a mix between two animals.
Smith, R. F., 1928. An Early Map of Surrey. The British Museum Quarterly, 3(1), 16. doi:10.2307/4420919
Topsell, E., 1658. History of Four-footed Beasts and Serpents.
Sequencing results for Allocamelus confirmed that it was indeed shuffled. From base pairs one to 259 Allocamelus matched Nanobody 3, and from 292 bp to 393 bp Allocamelus matched the sequence for Nanobody 2 (with the juncture having occurred somewhere in between in a 33 bp conserved region) providing definitive evidence that the shuffling did succeed (Figure 14).
Comparison | p-value |
---|---|
Allocamelus vs NB3 | 0.0001 |
Allocamelus vs NBH | < 0.0001 |
Allocamelus vs Neae | 0.1833 |
NB3 vs NBH | < 0.0001 |
NB3 vs Neae | 0.0387 |
NBH vs Neae | < 0.0001 |
In order to test the function of the Allocamelus nanobody (nanobodyAC), we performed another GFP binding assay with replicates (n = 3). The results from this assay show that nanobodyAC does not bind to sfGFP or fuGFP. Statistical analysis through a tukeyHSD post-hoc contrast in R showed significant difference in the binding to sfGFP between Allocamelus and NBH, our positive control, and also Allocamelus and NB3, the nanobody it shares most of its sequence with, but not significant difference in binding to sfGFP when compared to Neaa-intimin, our negative control. As the original nanobodies that allocamelus is made up of do bind sfGFP, this is a major conclusion in that we have succeeded in changing the function of a nanobody utilising DNA shuffling.
Future Directions
Given another month or two, we would make more use of the modelling data we have been given late on in this project. The next step would be to use modelling to design nanobodies that may be closer to strong binding to fuGFP, order these as gBlocks and go through the shuffling process again. This would hopefully generate a diversity of shuffled nanobodies, that, with fixed blue-white screening, could be put through the column and tested for binding to fuGFP. This would increase the likelihood of obtaining a nanobody that does bind fuGFP.
Summary
We developed fusion proteins of fuGFP bound to cellulose-binding domains (CBD) which immobilise fuGFP onto a cellulose matrix and enable cost-effective high-throughput screening of our DNA-shuffled nanobodies with novel affinity for fuGFP.
We started our fusion protein design by searching the parts registry for existing work on cellulose-binding domains and based our designs on sfGFP-CBD proteins used by Imperial College’s 2014 Aqualose team. Our results showed fuGFP-CBD can be expressed in TOP10.
Although we found that expression in TOP10 is significantly limited in comparison to BL21(DE3), especially for three out of our four CBD designs. From our testing, the function of fuGFP-CBDs was further improved with the engineered addition of a flexible linker.
Most importantly, we found that fuGFP-linker-CBDcipA binds tightly to cellulose and can be eluted using glucose to effectively purify the protein. Ultimately, our results strongly suggest that CBD fusion proteins can be applied in novel nanobody screening or even adapted as a more affordable tag for protein purification using cellulose.
Achievements:
fuGFP-CBD expression in TOP10 and cellulose binding
We cloned each fuGFP-CBD sequence into pUS250v3 and transformed them into TOP10 E. coli for expression. When transformed E. coli were induced with cumate on LB agar plates, we observed green fluorescence under UV which was most clear for cells expressing fuGFP-CBDcipA (Figure 16). Interestingly, while some fluorescence was observed this was markedly less compared to control cells expressing fuGFP alone which may reflect lower expression levels of fuGFP-CBD fusion proteins in TOP10 compared to fuGFP.
We tested the binding of fuGFP-CBD proteins to cellulose using paper filter disks and by measuring differences in fluorescence. TOP10 E. coli expressing each of our fusion proteins were lysed by bead beating to obtain cell lysates containing fuGFP-CBDs. We incubated each cell lysates with a piece of paper filter disk in a spin column and centrifuged the column to collect the flow through. We also washed each filter disk by adding TE buffer to the spin column and collecting the wash fraction by centrifugation. The fluorescence readings in Figure 17 indicate that fuGFP-CBDcipA cell lysate was the most fluorescent and also significantly increased the fluorescence of the filter disk in comparison to a negative empty vector control. This suggests that out of our 4 candidate fusion protein designs, fuGFP-CBDcipA is most easily expressed by TOP10 E. coli and is functional in binding cellulose.
However, cell lysates containing fuGFP-CBD -cenA, cex, and -clos did not have noticeable fluorescence compared to cell lysate from an empty vector control and did not cause the filter disk to gain noticeable fluorescence. A possible reason is that the proximity of fuGFP to the CBD domain affects the folding of the whole fusion protein.
Addition of a flexible linker to improve fuGFP-CBD functionalisation
To investigate the low levels of fluorescence and cellulose-binding in our initial fusion protein designs, we added a flexible 15 amino acid glycine-serine linker between fuGFP and CBD domains. The new fuGFP-linker-CBD sequences were again cloned into pUS250v3 and transformed into TOP10 E. coli for testing.
We observed that the cell lysates obtained containing fuGFP-linker-CBD proteins were now all brighter in fluorescence compared to a negative empty vector control. When using the same filter paper disk binding test, we observed that fuGFP linked to CBDcipA again causes the greatest gain in fluorescence by the filter disk (Figure 17). However, fuGFP linked to other CBDs (cenA, cex, clos) still did not result in any significance fluorescence from the filter disk.
Using SDS-PAGE we analysed proteins in each sample from our filter disk binding test. We found that after incubating the filter disk with fuGFP-linker-CBDcipA two significant bands can be observed at around 46 kDa which corresponds to the expected size of the fuGFP-CBDcipA fusion protein (Figure 18). A band in a similar position is observed when the filter disk is incubated with other fuGFP-CBD fusion proteins but is much fainter in comparison to fuGFP-CBDcipA. The presence of these 2 separate bands may result from post-translational modifications or protein degradation. To confirm the results, two bands of interest for fuGFP-linker-CBDcipA were excised and analysed with mass spectrometry which found 29% coverage of the expected amino acid sequence and highly suggests the fusion proteins are bound to the filter disk (Figure 19). Altogether, these results strongly suggest that fuGFP-linker-CBDcipA is the most functional and is able to bind cellulose and cause fluorescence.
fuGFP-linker-CBDcipA binds cellulose and is eluted with glucose
Next we tested whether fuGFP-CBD proteins bound to cellulose can be eluted under controlled conditions. In order to increase our protein yield, we used the BamHI and XhoI sites on each fuGFP-linker-CBD insert to clone the fusion protein sequences from pUS250v3 into pET28c(+) for expression in BL21(DE3). When induced with IPTG, all transformed cells fluoresced bright green (Figure 20).
We chose to focus on fuGFP-linker-CBDcipA since previous experiments suggested that it was the most functional. Cell lysate containing fuGFP-linker-CBDcipA was obtained from transformed BL21(DE3) cells induced with IPTG. Using a method replicating column chromatography, the cell lysate was incubated with microcrystalline cellulose in a test tube and the unbound fraction of cell lysate was removed by centrifugation. The microcrystalline cellulose was then washed with NT buffer which was also removed and collected by centrifugation. Results show that fuGFP-linker-CBDcipA lysate significantly increases the fluorescence of cellulose and is not washed off by NT buffer (Figure 21).
Next, we screened elution conditions using distilled water, cellobiose, maltose, glucose, and glycerol. After binding fuGFP-linker-CBDcipA to microcrystalline cellulose and washing the cellulose, we made 3 successive additions of the eluents and removed each elution fraction by centrifugation each time. We found that with our method, elution occurs after around 2-3 additions of eluent and that fuGFP-linker-CBDcipA is most strongly eluted with 1 M glucose or maltose, and weakly eluted with distilled water (Figure 22).
Due to its non-fluorescent nature and ease of availability, we chose to further examine elution using glucose. We found that glucose elution fractions were significantly fluorescent, and this corresponds with a decrease in the fluorescence of fuGFP-bound cellulose by approximately half after 4 additions (Figure 23). This suggests that glucose is an effective elution condition and could be potentially useful in purifying fuGFP-CBDs from cellulose.
To investigate the purity of proteins obtained using glucose elution we performed an SDS-PAGE. We found a single significant band at around 46 kDA corresponding to fuGFP-linker-CBDcipA in the elution after 3 additions of glucose (Figure 24). The identity of this band was confirmed to be fuGFP-linker-CBDcipA via mass spectrometry which showed 35% amino acid sequence coverage. Furthermore, this band is also observed in the induced lysate, but is much weaker in the supernatant and absent from the wash fractions. Moreover, the band of interest is heavily concentrated in the cellulose sample after washing but is decreased by around 50% after 4 rounds of glucose elution. Densitometric analysis of the band of interest according to ImageLab indicates fuGFP-linker-CBDcipA makes up 21% of the total protein in induced cell lysate compared to 63% of the proteins in the cellulose sample after incubation and washing with NT buffer, and upon elution, fuGFP-linker-CBDcipA comprises 98% of the elution sample proteins.
Overall, these results show that fuGFP-linker-CBDcipA binds to cellulose and can be selectively eluted using glucose. Although in our experiment, we tested only a single glucose concentration with 4 additions and found that a good amount of fusion protein remained bound to the cellulose. Therefore, future work could determine the effectiveness of higher concentrations or additional rounds of elution using glucose to increase the yield of eluted protein. Importantly, we have shown a single band of pure fuGFP-CBD fusion protein can be obtained which indicates that our method can be scaled-up in the future and applied for purifying CBD-tagged fusion proteins using cellulose.
Summary
In response to feedback through our Human Practices work, we endeavoured to expand the utility of the fusion proteins we had created by generating new fluorescent proteins that are also attached to CBDs, generating a suite of markers that can be immobilised to and eluted from cellulose. This was done through a degenerate primer PCR amplification. We have tested the new fluorescent proteins - fuGFPb, fuGFPy, fuGFPa, fuBFP - and the fluorescent protein-CBD complexes. The associated parts are BBa_K4488016, BBa_K4488017, BBa_K4488018, BBa_K4488019, BBa_K4488020, BBa_K4488021, BBa_K4488025, BBa_K4488026.
Achievements:
Generation of Novel fuGFP variants
Five new fluorescent free-use proteins were generated using our degenerate primer protocol. All five were characterised in the plasmid pUS252, transformed into TOP10 E. coli. Spectral fluorescence data was measured, as seen in Figure 26 below. Based on this data, four out of five new fluorescent free-use proteins were selected to be uploaded as new parts (both basic and composite parts, attached to our cellulose-binding domains). fuBFPt (t for threonine), was not included, as it is almost identical in spectra to fuBFP, however fluoresces at a much lower intensity.
fuGFPb
fuGFPb is a blue-shifted variant of fuGFP. Unlike fuGFP, which is excited in the long-wave UV range, fuGFPb is excited by blue light. This makes fuGFPb compatible with plate readers that do not extend into the UV range for excitation, and adds utility to the free-use protein line as analogues for sfGFP, which does fluoresce under blue light. fuGFPb has a maximum excitation wavelength of around 485 nm.
fuGFPy
fuGFPy is a yellow-shifted variant of fuGFP that appears slightly more yellow in its fluorescence. Similar to fuGFPb, fuGFPy is excited by blue light, however its maximum excitation wavelength is around 488 nm.
fuGFPa
fuGFPa (a for alanine, referring to its mutant amino acid), is a blue-shifted variant of fuGFP. It appears slightly more yellow/orange than fuGFPy on a plate, although the reasons for this are currently unclear. Both its excitation and emission spectra are more blue shifted than fuGFPb and fuGFPy. fuGFPa has a weaker fluorescence profile than fuGFPy and fuGFPb. Its maximum excitation wavelength is around 482 nm.
fuBFP
fuBFP (free-use blue fluorescent protein) is one of two BFPs generated. The other, fuBFPt (t for threonine, referring to its mutant amino acid) fluoresces only very faintly compared to fuBFP. fuBFP fluoresces in the long wave UV range, similar to fuGFP. Its maximum excitation wavelength is around 377 nm (Figure 27).
fuGFP-CBD Variants
One potential application of improved fuGFP with different colour variants is that they can be fused to CBDs and immobilised onto cellulose. The different colour variants could provide a basis for the design of a lateral flow assay that detects multiple analytes. Using the degenerate primer MVS70 (with redundancy at three loci of the fluorophore) we amplified plasmids of pET28-fuGFP-linker-CBDs to generate variant fluorescent proteins which we transformed into BL21(DE3) using the heat-shock method. We induced protein expression in selected colonies of BL21(DE3) and obtained lysates containing variant fuGFP-CBDs. We then tested the function of these variants by binding to microcrystalline and eluting with glucose.
Using the plate reader to measure the fluorescence of samples of eluted protein from cellulose using glucose. We identified the variants of fuGFP-CBDs that we created and characterised their excitation and emission spectra (Figure 28). Subtle differences exist, fuGFP-linker-CBDcipA excites maximally at around 395 nm and has maximum emission at around 510 nm. In comparison fuGFPb-linker-CBDcipA excites maximally at ~485 nm and emits maximally at ~510 nm. fuGFPy-linker-CBDcipA excites maximally at around 490 nm while emitting maximally at around 510 nm. fuGFPa-linker-CBDcipA excites maximally at 480 nm and emits maximally at ~510 nm. Finally, fuBFP-linker-CBDcipA excites maximally at around 380 nm and emits maximally at around 450 nm.
Furthermore, we tested the ability of these GFP-linker-CBDcipA variants to bind to microcrystalline cellulose and their elution under glucose. Results in Figure 29 show fuGFP-linker-CBDcipA and its variants are functional and bind to cellulose resulting in increased fluorescence detected by the plate reader and can each be eluted with 1 M glucose.
SDS-PAGE of samples eluted from cellulose using glucose shows distinct single bands for each colour variant fusion protein at ~46 kDA corresponding to the expected sizes (Figure 30). These results show that the colour variants of fuGFP-linker-CBDcipA bind to cellulose and can be eluted with glucose to obtain fractions with relatively high purity.
Free-use GFP (fuGFP) was designed by Mark Somerville at the University of Sydney with the aim of creating an unpatented GFP variant that would facilitate any kind of synthetic biology research by any organisation. The part (BBa_K3814004) was added to the iGEM parts registry by the 2021 University of Sydney Team.
Our team has contributed to the characterisation of this part by calculating a theoretical absorbance spectrum (Figure 31) for a 1 mg/ml concentration solution of fuGFP from experimental data, enabling future teams to use spectroscopy to quantify the concentration of this part.
The absorbances were measured in the photospectrometer between 300-700 nm and the average was graphed (Figure 31). We found that fuGFP exhibits peak absorbance at 395 nm with a secondary peak at 475 nm. In comparison, sfGFP displays only a single peak at 490 nm.
We obtained purified fuGFP and sfGFP stock solutions from Mark Somerville and made 1:5 and 1:8 dilutions of both proteins. In addition to these concentrations, we made up a 1:9 dilution in 0.2 M NaOH. This deprotonates the chromophore of the fluoroproteins such that they take on a homologous conformation between sfGFP and fuGFP, resulting in identical absorption spectra at these wavelengths (Ward et al. 1981). The extinction coefficient of GFPs with this denatured chromophore at 446 nm is 44,000 M−1 cm−1 (Bomati et al. 2015).
The Beer-Lambert law was used to back calculate the concentrations of the original solutions, which were determined to be 1.5 mg/ml and 1.4 mg/ml for fuGFP and sfGFP respectively. From these concentrations, the extinction coefficient of fuGFP was calculated to be 27618 M−1 cm−1. To normalise the absorption values for a 1 mg/ml solution, we divided the absorption values by their respective concentrations and graphed the resulting spectra (Figure 31).
Excitation and emission spectra of 200 µL from the same dilutions and stock solutions were measured using the plate reader. Excitation values were measured from 340-520 nm and emission values were measured from 480-560 nm for both sfGFP and fuGFP and normalised against their peak RFU values (Figure 32).
References
Bomati, E.K., Haley, J.E., Noel, J.P., and Deheyn, D.D., 2015. Spectral and structural comparison between bright and dim green fluorescent proteins in Amphioxus. Scientific Reports, 4(1).
Brooks, M. E., Kristensen, K., van Benthem, K. J., Magnusson, A., Berg, C. W., Nielsen, A., Skaug, H.J., Maechler, M., Bolker, B.M., 2017. “glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Generalized Linear Mixed Modeling.” The R Journal, 9(2), pp.378-400.
Fox, J., Weisberg, S., 2019. An R Companion to Applied Regression, Third edition. Sage, Thousand Oaks, CA. https://socialsciences.mcmaster.ca/jfox/Books/Companion/
Hartig, F., 2022. DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models_. R package version 0.4.6, https://CRAN.R-project.org/package=DHARMa
Lenth, R., 2022. emmeans: Estimated Marginal Means, aka Least-Squares Means. R package version 1.7.4-1. https://CRAN.R-project.org/package=emmeans
R Core Team, 2022. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. https://www.R-project.org/
Ward, W., DeLuca, M. and McElroy, W., 1981. PROPERTIES OF THE COELENTERATE GREEN-FLUORESCENT PROTEINS - Bioluminescence and Chemiluminescence. 1st ed. Saint Louis: Elsevier Science, pp.235–242.