Tag Archives: superfolder

Design guidlines for tandem fluorescent timers

Almost 4 years ago, I wrote a post on tandem fluorescent timers (tFTs). The idea is to have two different fluorescent proteins fused together to the protein of interest. In the paper from 4 years ago, it was superfolder GFP (sfGFP) and mCherry. sfGFP matures very fast (within minutes) and mCherry  matures more slowly (t1/2 ~40min). The ratio beween green to red fluorescent signal indicates the percentage of new vs old proteins, thus acts as a “timer”.  This latests paper on tFTs from the same group of Michael Knop’s lab, found that analyzing tFTs might be more complicated due to some possible problems of this system.

Continue reading

Advertisements

Visualizing translation: insert TRICK pun here

Unlike transcription, it is much harder to image translation at the single molecule level. The reasons are numerous. For starters, transcription sites (TS) are fairly immobile, whereas mRNAs, ribosomes and proteins move freely in the cytoplasm, often very fast. Then there are only a few TS per nucleus, but multiple mRNAs are translating in the cytoplasm. Next, there’s the issue of signal to noise – at the transcription site, the cell often produces multiple RNAs, thus any tagging on the RNA is amplified at the transcription site.  Last, it is fairly easy to detect the transcription product – RNA – at a single-molecule resolution due to multiple tagging on a single molecule (either by FISH or MS2-like systems). However, it is much more difficult to detect a single protein, be it by fluorescent protein tagging, or other ways (e.g. FabLEMs).

The rate of translation is ~5 amino acids  per second, less than 4 minutes to a protein 1000 amino-acids long. This is faster than the folding and maturation rate of most of even the fastest-folding fluorescent proteins. This means that by the time the protein fluoresce, it already left the ribosome. However, attempts were made in the past with some success.

Continue reading

Cells reach out their “hands” to create new limbs

Communication between cells takes many forms. There could be communication by sending out microvesicles with important messages inside, by sending out free molecules (like hormones) or by special structures (e.g. synapses).

Sonic hedgehog (SHH) is a signaling protein that is important for the development of vertebrate limbs. It was thought to be release from a small group of cells at the posterior end of the limb bud, and is recognized by receptors on cells a long distance away.

Not this Sonic Hedgehog… (image taken from http://sonic.wikia.com)

A new paper publish in Nature from Maria Barna’s lab shows that SHH actually remains bound on the external side of the cells that produce it. The cells simply send very long thin protrusions (here named filopodia) that reach all the way to similar filopodia of the receiving cells.

I think that not only the story is very novel and interesting, but the images are very pretty.

Several “technical” issues:

In order to study the SHH signaling in live chick embryos, they designed a custom made live in ovo microscopy system: a temperature controlled plate; on it an egg container chamber, and an objective that is dipping into the yoke.

They show that standard fixation methods (e.g formaldehyde) destroy these filopodia. Also, a volume marker (in this case sfGFP) that just fills the cytoplasm does not give a strong enough signal to detect these filopodia (possibly since they are very thin, and packed with actin filaments  and other proteins, so there’s very little free volume left).

So, they used palmitoylated fluorescent proteins, pmeGFP (green) and pmKate2 (red) that target them to the plasma membrane. This enabled them to visualize these very thin and long filopodia. Here’s a video movie from their paper.

They use a variety of cytoskeletal proteins fused to EGFP or to mKate2 to learn about the structure of these filopodia. Their conclusion is that these structures contain only a specialized form of actin filaments.

They show beautifully that SHH (fused to EGFP) travels to the tip of these filopodia:

They used a split GFP technology to show that SHH is actually found on the outside of the cell membrane.  In split GFP, two fusion proteins are produced, each one is fused to “half” of a GFP protein (its not exactly half but let not go into that now). If the two fused proteins are in close proximity, the two halves associate to produce an GFP that fluoresce irreversibly. The two separate halves do no fluoresce. So one half was fused to SHH and the other was anchored to the extracellular leaflet of the plasma membrane, and when both were expressed, they got green fluorescence.

In total – a very nice and pretty paper.
ResearchBlogging.orgSanders TA, Llagostera E, & Barna M (2013). Specialized filopodia direct long-range transport of SHH during vertebrate tissue patterning. Nature, 497 (7451), 628-32 PMID: 23624372

New and improved – the next generation of GFP?

A new and improved green fluorescent protein, named mNeonGreen, was developed.

It was engineered from a Yellow fluorescent protein (LanYFP) that was isolated from the cephalochordate Branchiostoma lanceolatum. Therefore, LanYFP is genetically unrelated to the commonly used Aequorea victoria GFP.

LanYFP has a high quantum yield (0.95) and extinction coefficient (~150,000 M−1 cm−1) – making it a very bright protein.  LanYFP is a tetramer – not useful for most applications.

Directed evolution to make it a monomer produced the new, monomeric protein, mNeonGreen. The ex/em of mNeonGreen is slightly shifted to the yellow compared to EGFP (509/516 compared to 488/509), making it a better choice to separate from CFP emission.

Though slightly less brighter than its parent protein LanYFP, mNeonGFP is 2-3 times brighter than EGFP and actually brighter than most green & yellow proteins.

Another great advantage of this new protein is that is it fast folding – the authors claim it is <10 min at 37C. This is fairly close to the superfolder GFP.

It is also very photostable (comparable to EGFP), performs well as a fusion construct at N & C termini many tested proteins, performs 4-times better that EGFP in stochastic single-molecule superresolution imaging and is a better FRET partner (both as acceptor and donor) than other proteins.

mNeonGreen fused to histone H2B shows the different stages of the chromosomes during cell division. Source: Shaner et al., (2013) Nature Methods 10:407-409.

mNeonGreen fused to histone H2B shows the different stages of the chromosomes during cell division. Source: Shaner et al., (2013) Nature Methods 10:407-409.

In short, this may very well be the “next generation” of fluorescent proteins. It has all the good qualities, and seems to have none of the bad ones. It performs better than most, if not all fluorescent proteins in every tested parameter.

Only its name is rather plain. I would call it something like wonderGFP or GreenLantern (hey, it even has the Lan from the animal they developed it from).

(Update: see here for details on how to get your hands on this protein).

ResearchBlogging.orgShaner NC, Lambert GG, Chammas A, Ni Y, Cranfill PJ, Baird MA, Sell BR, Allen JR, Day RN, Israelsson M, Davidson MW, & Wang J (2013). A bright monomeric green fluorescent protein derived from Branchiostoma lanceolatum. Nature methods, 10, 407-409 PMID: 23524392

Folding and maturation (part 3) – fluorescent timers

In the previous two parts (1)(2), I described the directed evolution of fast folding fluorescent proteins. But why is it important? Why do we need fast folding GFP? Why do we need to know the maturation time?

For most applications, it usually doesn’t matter. If we express these proteins constitutively, then we should already have enough fluorescent protein in the cells when we get to the experiment. Even in induced systems, we rarely take into account the maturation time of the protein. We follow fluorescent as it appears; usually without taking into account that the immature protein may have been present in the cells minutes or hours before.

However, when studying cellular dynamics, timing is important. The spatiotemporal dynamics of a protein is not an easy task to determine. Fluorescent timers (FT) are fluorescent proteins which change color with time due to a chemical conversion of the chromophore. Early FTs were first developed on the basis of mCherry. These FTs change color from blue to red at rates ranging from 10 min (fast-FT) to 28hr (slow FT). How are FT used? Basically, the ratio between the two colors should indicate the passage of time. For instance, a ration of red/blue=0 will indicate that the FT is a nascent protein that has not converted yet. A ratio of red/blue=1 indicates that there are no new proteins – all have converted. The ratio between 0 and 1 will tell us how much new protein still exists at the specific cellular location.

However, because of their tendency to oligomerize, or because of low brightness, these FTs were not widely used. Another recently developed monomeric green to orange FT is called Kusabira green orange (mK-GO). However, the 10h color transition is not suitable to measure fast dynamics.

A new paper sent to me by one of the blog readers (thanks Andrius) has taken a unique approach to develop FTs which they call tandem FT (tFT).

FT – regular fluorescent timer. tFT – tandem FT. m1, m2 – maturation rates. If m1<<m2 than option 2 is negligable compared to option 1.

Instead of using directed evolution to develop a novel FT, they took two known fluorescent proteins which have significantly different maturation times (sfGFP and mCherry) and fused them together in tandem. The logic of this system is quite beautiful: the sfGFP will mature fast, and the green signal will remain consistent. The mCherry will mature slower (half-time of 40min). Thus, they get a two color timer. If the signal is only green, it means that the protein was recently synthesized. A larger ratio of red/green will indicate passage of time based on mCherry maturation rate. A ratio of 1 will indicate that no new protein is present at that particular location.

They then started playing with in in different scenarios in yeast. First, they followed the formation of the spindle pole body (SPB) (the yeast equivalent to centrosomes). It is already known that in yeast, the SPB duplicates during metaphase and the “old” SPB travels to the bud tip during anaphase. They tagged an SPB protein with their tFT. Indeed, they see that the ratio of red/green in the bud is greater than in the mother cell. A mutation known to affect this segregation similarly reduces the ratio in the bud compared to the mother. In contrast, proteins that are known to be retained at the mother, whereas newly synthesized a transported to the bud show the exact opposite – i.e. red/green signal higher in the mother vs. bud cell.

A very neat experiment used a bud-scar protein tagged with tFT which shows several bud scars forming on the same cell over time, with different color blends (the old scar is red, newer in “yellow” and newest in green. This presents a nice “clock” of for the relative age of each scar. It’s too bad that the authors didn’t add the real time line (in minutes) in their image and tried to correlate the actual time with the tFT “time”.

A protein localized to bud scars is fused to tFT. the older the scar, the more red it is. Source: Khmelinski et al. (2012) Nat Biotech 30:708.

They then started to explore other stuff. First, they looked at the segregation of nuclear pores during mitosis. Nuclear pores (NPC) are complex structures that allow transfer of proteins and RNA in and out of the nucleus in a regulated manner. The question they asked is whether “old” NPCs remain in the mother cell and the bud get newly synthesized proteins, or vice versa (or equal distribution, i.e. non discriminated transfer). They individually tagged each of the NPC proteins (plus some control) – a total of 36 proteins- with their tFT and measured whether we see older proteins at the mother or the bud. Surprisingly, the bud gets the older proteins. Their analysis suggests that there is active transport of the “old” NPCs into the bud. How this is achieved and why the bud should receive old and possibly damaged NPCs are good questions for further research.

NPC proteins tagged with tFT show preferential localization of “older” proteins at the bud compared to mother cell. source: Khmelinski (2012), Nat Biotech. 30:708.

Despite of their interesting results, the authors didn’t stop there and when to test another application. They show that a single readout of the red/green ratio can indicate the stability of the protein. Stabilization/de-stabilization by mutations or different inherent stability can easily be distinguished based on the red/green ratio. They further utilize this approach to screen for protein stability regulators, which was quite successful as they identified most known factors in the specific pathway they studied, as well as identified new factors in this pathway.

In conclusion – this paper shows how one can utilize “ordinary” fluorescent proteins as fluorescent timers. Moreover, this method is much easier and much more flexible than trying to evolve FTs by random or directed mutagenesis. In fact, using different pairs with different maturation times would easily enable us to create FTs for any time intervals we wish and in any color we wish.

They show here how to utilize this system to follow cell cycle events. Obviously, this system can be used to study other scheduled, localized events in the cell.

A straightforward application is to follow protein stability. However, this system can also be used to study mRNA stability, by combining these tFTs with the MS2 system.

I’m sure other applications will be developed in the future.
All in all, a good paper with a great idea!

ResearchBlogging.org Khmelinskii A, Keller PJ, Bartosik A, Meurer M, Barry JD, Mardin BR, Kaufmann A, Trautmann S, Wachsmuth M, Pereira G, Huber W, Schiebel E, & Knop M (2012). Tandem fluorescent protein timers for in vivo analysis of protein dynamics. Nature biotechnology, 30 (7), 708-14 PMID: 22729030

Folding and maturation, or how to evolve your own GFP (part 2)

In part 1 I discussed the directed evolution of fast-folding GFPs. These were developed for specific purposes of improving the solubility, stability and folding of the protein. Now, I will discuss the maturation step and how it was measured for a variety of GFPs.

As mentioned in the first part, maturation is the final step in the transformation of GFP from a chain of amino acids to a fluorescent protein – the creation of the chromofore. The chromofore has a very long name (p-hydroxybenzylideneimidazolidone) which in wild type  GFP is formed from amino acids S65, Y66 and G67. In this process, the amide nitrogen of G67 backbone performs a nucleophilic attack on the carbonyl carbon of S65. Oxidation with atmospheric O2 and dehydration reactions create the imidazolinone ring, which is conjugated to the side chain of Y66.

In most papers that measure GFP maturation, the authors do not distinguish folding from maturation. In many cases, the experiments involve complete denaturation of the protein in vitro (e.g. by urea or guanidinium chloride) and then measuring fluorescence following re-folding. In vivo, it would be difficult to assess folding kinetics (because you do not really have an exact time zero for when the protein is being translated).  [In our lab, we are trying to develop a system to visualize translation in vivo in real time by imaging techniques.  If it works, it would be useful for determining exact folding & maturation rates in vivo of a single fluorescent protein].

In any case, it is easier to measure the kinetics of the maturation step, both in vivo and in vitro, simply by removing and adding oxygen. I mentioned in part 1 that this is how the maturation rate of GFP-S65 was calculated in vivo.  A new paper from 2011 from the lab of Funatsu analyzed the maturation rates of several GFPs in vitro. Their idea was simple – using a commercial kit, the performed in vitro transcription-translation reaction of the FPs at anaerobic conditions (they added catalase and glucose oxidase to get 0.1mg/l of oxygen). They then stopped the reaction and added oxygen, then followed fluorescence.

The proteins they assayed were: wild-type GFP, GFP-S65T, GFP-S65T/S147P, EGFP, sfGFP, Emerald, GFPm, GFPmut2, GFPmut3, sgGFP, “cycle 3”(a.k.a GFPuv), frGFP (a.k.a GFPuv3), GFPuv4 and 5 and two yellow FPs: EYFP and Venus..

Their results are interesting. As expected, the S65T mutation improved the maturation rate by 3.2 fold. However, EGFP matured only 20% faster than the S65T alone. The S65T/S147P is a variant that is stable under a range of temperatures. This mutant matured faster than EGFP, at a rate 40% faster than S65T alone. GFPmut2 and GFPmut3 (S65G/S72A) both showed a 7-fold faster maturation than WT GFP (>2-fold the S65T alone). sgGFP (SuperGloGFP, F64L/S65C/I167T, from Qbiogene) showed improved maturation rate compared to S65T (70%  faster).  Emerald, which contains 4 mutations on top of the EGFP mutations showed only a slight increase in maturation rate (similar to S65T/S147P). “cycle 3” was only slightly better than WT GFP, frGFP (which is cycle 3 +EGFP mutations combined) showed a maturation rate which was 10% less than EGFP.  Addition of the I167T mutation to create GFPuv5 increased the maturation rate by ~70% (just like in sgGFP).

Most interesting is the super-folder GFP (sfGFP), which showed a maturation rate of only 2.3-fold over the WT (that is ~70% of that of S65T). Thus, though this protein may be more stable and may fold very fast compared to other variants, the important step of maturation is the slowest among all variants tested (except the WT). Since folding assays measure fluorescence as the output for a mature protein, it means that the folding step (prior to maturation) is much faster than previously appreciated.

GFPm is a weird case.  GFPm, developed by David Tirrell, is a variant with mutations from cycle 3 and GFPmut3. However, Tirrell tried something unique –  to replace the leucines in the protein with 5,5,5- tri-fluoroleucine (tfl). The fluoreinated form proved to be insoluble, and fluorescent of the cells (E. coli) was 500-fold less than the regular protein. They then went on with mutagenesis, developing new variants which were brighter (up to 650-fold over GFPm). Personally, I do not understand how this can be of much use, since using tfl will probably have major effects on every aspect of cell biology we are interested in. Anyway, the maturation rate of GFPm (not fluoreinated) is itself pretty high – almost 3-fold over S65T and is actually the fastest maturing GFP variant tested. Oddly, they do not discuss this result anywhere in the text. Perhaps there is a technical issue they wanted to avoid?

Last but not least, the two YFPs showed maturation rates which are similar to WT GFP (Venus) or ~2.3-fold better (EYFP).

So how does all this information help us?

First, if we know which mutations enhance maturation and which slow it down, we can design faster-maturing proteins.

Second, we can use this data to estimate translation or translocation rates in vivo. However, we should remember that the data obtained in vitro (at 37C) does not neccessarily agree with actual maturation rates in vivo in every cell type. for instance, yeast or fly grow in colder temperatures which may affect maturation. Oxygen levels in tissue culture dish are differnt than in entire animals, and also differnt in differnt organs.  Also, if the GFP is fused to another protein, it may also affect folding as we learned in part 1, as well as maturation. Finally, the in vitro environment in the tube lacks many biomolecules (proteins other than used for translation, small molecules, ions, oxidizing molecules, antioxidants etc) which can affect oxygen availability in the immediate environment of the newly translated GFP.

Third, we can use such data to design cool experiments. So… stay tuned for part 3, which I will dedicate to biological timers.

ResearchBlogging.orgIizuka R, Yamagishi-Shirasaki M, & Funatsu T (2011). Kinetic study of de novo chromophore maturation of fluorescent proteins. Analytical biochemistry, 414 (2), 173-8 PMID: 21459075
Yoo TH, Link AJ, & Tirrell DA (2007). Evolution of a fluorinated green fluorescent protein. Proceedings of the National Academy of Sciences of the United States of America, 104 (35), 13887-90 PMID: 17717085

Folding and maturation, or how to evolve your own GFP (part 1)

GFP is one of the most widely used proteins in research. Its usefulness has advanced our understanding of biology in huge leaps forward. One of the greatest advantages of GFP is that the chromophore is formed in an autocatalytyic manner, no need for an enzyme or cofactor assistance. All that is required is atmospheric oxygen.

GFP has sort of a barrel shape, with the chromophore inside. The chromophore is formed from residues serine 65 (S65), tyrosine 66 (Y66) and glycine 67 (G67). The process of creating the structure of the protein is called folding, whereas the process of creating the chromophore is termed maturation. Maturation occurs after the protein folds to near native conformation.

However, the natural, wild type, GFP is not the best tool. It is rather dim and not very stable at 37°C  (Since the jellyfish lives at low temperatures). Also, it matures rather slowly. Early on, a mutation was introduced, S65T, that increased the brightness of the GFP and also shifted its excitation peak (from a major peak at 395 and a minor at 475, to a single major peak at 488).  In vivo, it was shown that maturation of GFP-S65T takes 27 minutes (this was measured by expressing GFP in E. coli bacteria under anaerobic conditions, then supplementing the bacteria with O2. Therefore, they only measured the last few steps of the maturation process). EGFP, the most common variant used in research labs, contains a second mutation, F64L. This mutation stabilizes the protein at 37°C and was suggested to increase the maturation rate.

Over the years, GFP has been a subject for further “directed evolution” to achieve required traits like different colors, brightness, pH stability and photostability, oligomerization state, as well as folding and maturation.

Why do we care about folding and maturation?

Well, GFP is often fused to other proteins. In some cases, that protein is misfolding, or folding slowly. This can affect the folding of GFP. This is particularly true when expressing fused proteins in E. coli, where in many cases the ectopically expressed protein is aggregating in inclusion bodies. Since GFP itself has a slight tendency to dimerize, any aggregation of the fusion protein may amplify the dimerization of the GFP. Thus, it would be advantageous to have a better folding GFP, with a lower tendency to dimerize.

As to maturation, having a fast-maturing GFP would be very beneficial for studies of translation, or translation-coupled localization.  Because, if the GFP takes 20-30 minutes to mature after it is being translated, we cannot really say what happened and where was the GFP or any fused protein, in that time.

Our story begins in 1996, when the lab of Willem Stemmer tried to improve GFP by a technique called DNA shuffling. Their goal was to improve whole cell fluorescent in E. coli cells expressing GFP. After three cycles of DNA shuffling, and selection based on fluorescence intensity with UV excitation, they isolated a clone (named simply ‘GFP cycle 3’) which showed 45-fold increased fluorescence compared to the then commercial Clontech pGFP. However, the spectral characteristics were unchanged, and the maturation rate (T1/2 = 95min at 37°C in whole cells) also seemed unchanged).  Mutations found in GFP cycle 3 compared to their starting GFP showed that three hydrophobic amino acids were replaced by hydrophilic amino acids. Thus, the improved fluorescence is probably due to reduction in protein aggregation by the hydrophobic surfaces, and increased solubility of the GFP.

CHO cells expressing wt GFP (A) or GFP cycle 3 (B). Source: Crameri et al. (1996) Nat. Biotech. 14:315.

Ten years later, the superfolder GFP (sfGFP) was engineered by the lab of Geoffrey Waldo for the particular purpose of creating a protein with less tendency to misfold when fused to poorly folded proteins in E. coli.

How was this protein engineered?

They started with a folding reporter GFP  (frGFP) which contained the cycle 3 & EGFP mutations.

They fused frGFP to a poorly folding protein, H subunit ferritin, which is insoluble when expressed in E. coli. After several rounds of mutagenesis and screening for fluorescence, they obtained the highly fluorescent sfGFP which bears six new mutations.  Ferritin-sfGFP was found to be 50-fold more fluorescent than the starting fusion protein. When expressed alone, cells expressing sfGFP were twice as bright as with frGFP. Importantly, the excitation & emission spectra , QY and EC were similar in both proteins.

Measuring the folding rate was done by first denaturing the proteins with urea. Then washing the urea and measuring the time it takes the GFP variant to regain fluorescence.  This method actually measures folding and maturation and this should be taken into account. In any case, though both proteins showed 95% yield within 4 minutes, sfGFP showed a 3.5-fold faster initial rate. sfGFP also tolerated higher urea levels, and showed better fluorescence with a bunch of different fused proteins with different characteristics (expression level, solubility etc…).

fusion of frGFP or sfGFP to 18 differnt proteins shows sfGFP much brighter in all cases. 0.125s, 1s: exposure time. below the images of the bacteria are Western blot images of the fusion proteins. Source: Pedelacq et al. (2006). Nat. Biotech. 24(1):79

They found that the most contributing mutation to the folding  kinetics and stability is the S30R mutation. The change from an oxygen group (that forms a hydrogen bond with E17) to a positively charged side chain mediates an electrostatically charged network of the β-barrel which seems to create a global stability to the structure. A second mutation, Y39N, slightly changes the angle of the backbone, and allows a hydrogen bond with D36, further stabilizing the structure. The other four mutations slightly contribute to the stability.

Although sfGFP seems to fold fast (though no comparison to EGFP was made in this paper), the assays here did not really differentiate folding from maturation.

A year later, the group of David Liu used sfGFP  in an effort to create a new GFP protein that is highly soluble. In order to do that, they replaced multiple neutral amino acids on the outer side of the protein with either positively or negatively charged amino acids to produce super-charged GFP variants GFP(+36) and GFP(-30) [compared to the net charge of sfGFP which is (-7) or GFP which is (-8)]. This scGFP was highly soluble, and also highly resistant to denaturing, even by boiling to protein at 100°C. The spectral characteristics of the scGFPs remained similar to those of GFP and sfGFP indicating that the structure of the outside “barrel” does not affect the chromophore features inside the barrel.

In 2008, Fisher & DeLisa produced yet another variant of GFP called superfast GFP. By exploiting the cell’s protein secretion quality control mechanisms, they screened for new super folding variants. The secretion mechanism exports unfolded proteins to the periplasm of E. coli. In the periplasm, GFP is apparently non-fluorescent. If the protein folds fast enough, the quality control mechanism prevents its export and the protein remains in the cytoplasm.  Their starting GFP was GFPmut2, a variant that was previously optimized for FACS analysis, harboring the mutations S65A/V68L/S72A. Following several round of selection, they isolated a clone which they named superfast GFP. Based on their analysis (using guanidinium chloride (GdnCl) to unfold the proteins and then diluting the GdnCl to refold the proteins, the T1/2 for superfast GFP refolding (resumed fluorescence) was 11 minutes, compared to 33 min (GFPmut2), 20min (sfGFP) and 73min (frGFP).

Graph showing folding of GFP proteins over time. black: square – frGFP; circle – sfGFP; triangle – GFPmut2. open circle – superfast GFP; triangle, square – two other mutants. Source: Fisher AC & DeLisa MP. (2008). PLoS One. 3(6):e2351.

So, far we discussed fast folding GFP proteins. However, the methods to measure folding of these proteins do not discriminate folding from maturation.  Furthermore, the in vitro results do not necessarily represent the in vivo folding rates (i.e. 11 min in vitro does not mean 11 min in vivo).

In the next post I will discuss the issue of maturation.

ResearchBlogging.orgCrameri A, Whitehorn EA, Tate E, & Stemmer WP (1996). Improved green fluorescent protein by molecular evolution using DNA shuffling. Nature biotechnology, 14 (3), 315-9 PMID: 9630892
Pédelacq JD, Cabantous S, Tran T, Terwilliger TC, & Waldo GS (2006). Engineering and characterization of a superfolder green fluorescent protein. Nature biotechnology, 24 (1), 79-88 PMID: 16369541
Lawrence MS, Phillips KJ, & Liu DR (2007). Supercharging proteins can impart unusual resilience. Journal of the American Chemical Society, 129 (33), 10110-2 PMID: 17665911
Fisher AC, & DeLisa MP (2008). Laboratory evolution of fast-folding green fluorescent protein using secretory pathway quality control. PloS one, 3 (6) PMID: 18545653