In vivo characterisation of fluorescent proteins in budding yeast

Fluorescent proteins (FPs) are widely used in many organisms, but are commonly characterised in vitro. However, the in vitro properties may poorly reflect in vivo performance. Therefore, we characterised 27 FPs in vivo using Saccharomyces cerevisiae as model organism. We linked the FPs via a T2A peptide to a control FP, producing equimolar expression of the 2 FPs from 1 plasmid. Using this strategy, we characterised the FPs for brightness, photostability, photochromicity and pH-sensitivity, achieving a comprehensive in vivo characterisation. Many FPs showed different in vivo properties compared to existing in vitro data. Additionally, various FPs were photochromic, which affects readouts due to complex bleaching kinetics. Finally, we codon optimized the best performing FPs for optimal expression in yeast, and found that codon-optimization alters FP characteristics. These FPs improve experimental signal readout, opening new experimental possibilities. Our results may guide future studies in yeast that employ fluorescent proteins.

yFPs tdTomato, mScarletI, and mYPET (YPET A206K, F208S, E231L, N234D) were codon-optimized and synthesised (Baseclear B.V., Leiden, The Netherlands), generating ytdTomato, ymScarletI and ymYPET. These constructs were digested with NheI and Kpn2I and ligated using T4 ligase into either T2A-mTq2 or T2A-mCherry in which the FP N-terminally of T2A was removed by digestion with the same enzymes. This generated ytdTomato-T2A-mTq2, ymScarletI-T2A-mTq2, ymNeongreen-T2A-mTq2, ymNeongreen-T2A-mCherry and ymYPET-T2A-mTq2 in pDRF1. Msn2-ymNeongreen and ymTq2Δ9 pUC19 plasmids were codon-optimized and synthesised (Baseclear). A PCR was performed using these constructs according to table S1. Next, the products were digested using NheI and Kpn2I and ligated using T4 ligase into T2A-mTq2 and T2A-mCherry pDRF1 plasmids in which the FP N-terminally of T2A was removed by digestion with the same enzymes, which generated ymTq2-T2A-mCherry, ymNeongreen-T2A-mTq2 and ymNeongreen-T2A-mCherry. pDRF1 plasmids containing the single yFPs were generated by performing a PCR according to table S1 on Msn2-ymNeongreen, ymTq2Δ9, ymYPET, ytdTomato, ymScarletI in pUC19 plasmids which added a stopcodon at the C-termini. Subsequently, the PCR products were digested with NheI and NotI and ligated with T4 ligase in an empty pDRF1 vector digested with NheI and NotI which generated ymYPET, ymTq2, ymScarletI, ytdTomato and ymNeongreen in pDRF1. CytERM-ymVenus was created by a mutagenesis PCR according to table S1. Afterwards, pDRF1 containing ymVenus-T2A-mTq2 was constructed by performing a PCR on CytERM-ymVenus according to table S1. Next, the product was digested using NheI and Kpn2I and ligated into a T2A-mTq2 pDRF1 vector in which the FP N-terminally of T2A was removed by digestion with the same enzymes. pFA6a-yFP-CaURA3 plasmids containing the yFPs were generated by performing a PCR according to table S1. Next, the products were digested using PacI and AscI (New England Biolabs) and ligated with T4 ligase into the plasmid pFA6a-link-yomCherry-CaURA3 also digested with PacI and AscI to replace yomCherry with the yFP, which generated pFA6a-yFP-CaURA3 plasmids. pFA6a-link-ymNeongreen-SpHis5 was generated by performing a PCR on Msn2-ymNeongreen pUC19 according to table S1. Next, the product was digested using PacI and AscI and ligated into pFA6a-link-yomKate2-SpHis5 also digested with PacI and AscI (New England Biolabs), replacing yomKate2 with ymNeongreen.
CytERM constructs CytERM-dTomato (addgene plasmid #98834) and CytERM-mTq2 (addgene plasmid #98833) were digested using NheI and NotI and ligated into an empty pDRF1 vector digested with the same enzymes which generated CytERM-dTomato and CytERM-mTq2 in pDRF1. CytERM-yeVenus, CytERM-ymNeongreen, CytERM-ytdTomato, CytERM-ymScarletI, CytERM-ymTq2 and CytERM-ymYPET pDRF1 were created by performing a PCR according to table S1. Afterwards, products were digested using XmaI (New England Biolabs) and NotI and ligated with T4 ligase into a CytERM pDRF1 plasmid in which the FP C-terminally of CytERM was removed by XmaI and NotI which generated the CytERM-yFPs.  TCTCCGGATTTGTACAATTCATCCATA  CCAT  mTq2-T2A-mCherry pDRF1  mTq2 C1  CTGCTAGCGCTACCGG  TATCCGGACTTGTACAGCTCGTCCA  yotagRFPT-T2A-mTq2 pDRF1  pFA6a-link-yoTagRFP-T-CaURA3 (Addgene plasmid  #44877) ATGCTAGCCACCATGGTATCTAAA GGTGAAGAGTTG  Figure S1. pH sensitivity of all characterised FPs. Yeast cells were incubated for 2 hours in a citric-acid/Na2HPO4 buffer with 2 mM DNP and fluorescence was measured using a fluorescent plate reader. Per FP, at least 3 technical replicates were measured. Afterwards, fluorescence was normalized to the pH giving the highest fluorescence and a hill fit was performed to determine the hill coefficient and pKa value. Dots represent the mean of at least 3 replicates, line indicates fitted pH response curve, error bars indicate SD. Figure S2. Day-to-day variation of yFPs depicted by the coefficient of variation (CV) of the mean brightness of 3 days.

Model of photochromicity 1 Model
Bleaching of Fluorescent Proteins (FPs) is often modelled by a system of linear differential equations [Dickson et al., 1997, Dean et al., 2011, Morisaki and McNally, 2014, Sinnecker et al., 2005. These models incorporate different states with transition rates that describe blinking (fast switching between a fluorescent state and a non-fluorescent state), bleaching and irreversible bleaching behaviour. Since we are only interested in photochromicity and bleaching, we did not model blinking behaviour since this happens on a very short timescale of milliseconds. We assumed that every wavelength of light induces different switching rates, and that not all of the FPs are in the fluorescent state initially. We think that this led to the smallest possible model that describes bleaching, irreversible bleaching, and photochromicity. We model the FPs as having three states, a natural (nat) state, a dark (dark ) state, and an irreversible dark state (irrdark ). In the natural state, the FPs are fluorescent: light is emitted after exposure to light in their defined absorption range. Both under light exposure and spontaneously, FPs can transition from the natural state to the dark state. In the dark state, the FPs are not fluorescent, but can return to the natural state both under the influence of light as spontaneously. For four different wavelengths, we include different rate constants for these transitions, as well as for the spontaneous transitions. FPs can also transition spontaneously from the dark state to an irreversible dark state. In this irreversible dark state no fluorescence is possible and during the course of the experiment, the FPs will not turn back to a fluorescent state anymore.

Equations
The transition rates between the states are defined as v nat→dark = nat · k nd−spont , In these rate equations P RFP denotes the light power in the red-wavelength range (which is around 570 nanometers), and similar for green (≈ 490 nm), cyan (≈ 438 nm) and yellow (≈ 500 nm) light. The dynamics for the different FP-states are now determined by the following differential equations We assume there is no significant exposure to light before the experiment starts, and that the transition from dark to the irreversible dark state is negligible. We thus get an equilibrium between the natural and dark state, depending on the spontaneous rate constants. If we also rescale the state concentrations such that nat(0) = 1, we get

Numerical solution and fitting procedure
To keep our approach as generic as possible, we decided to solve the differential equations numerically. In this way, the solutions can, in principle, also be found if the experimental protocol involves changing light powers. At constant light powers, the equations are analytically solvable, as we will show in Section 1.3.
Since the FPs can only be fluorescent when they are in the natural state, we fit the predicted dynamics of the natural state (nat(t)) to the fluorescence measurements. We fit all rate constants, so that we have 11 fitting parameters. In many experiments, however, the FPs will not be exposed to all wavelengths. The parameters, corresponding to the absent wavelengths, will then not be fitted.
Our set of experiments all involved a periodic exposure protocol, with short phases of light and longer phases of darkness. To make the fitting procedure computationally feasible, we approximated these experiments by calculating the average light power per wavelength.

Analytical solution
If the light powers at the different wavelengths are known, we can calculate overall rate constants k nd = k nd−spont + k nd−RFP · P RFP + k nd−GFP · P GFP + k nd−CFP · P CFP + k nd−YFP · P YFP , k dn = k dn−spont + k dn−RFP · P RFP + k dn−GFP · P GFP + k dn−CFP · P CFP + k dn−YFP · P YFP , We are left with a set of linear differential equations, which can be solved by a sum of two exponentials. The integration constants can be calculated by demanding that nat(0) = 1, and dark (0) = dark 0 . We further simplify the solutions by introducing new variables: We are left with

Biological interpretation of the parameters
A sum of two exponential decays Note that both α − and α + are positive, since Therefore, the solution is a sum of two contributions that decay exponentially in time. The exponential decay rates will differ by If this difference is large, then the fluorescence dynamics will be bi-exponential, while if this difference is small, the decay could be fitted by a single exponential already. We see that bi-exponential behaviour will thus occur for example when k nd is large compared to k di . This is often the case. The contribution that decays fastest, i.e. with the larger decay rate: α + , corresponds to the FPs settling in a quasi-equilibrium between the dark state and the natural state. We call this a quasi-equilibrium because on the longer timescale, i.e. with rate α − , fluorescence will decay by the transition of FPs from the dark to the irreversible dark state.
Approximating the change in fluorescence due to fast decay The equilibration of natural and dark state can lead to both an increase or a decrease in fluorescence, depending on the initial conditions. The amplitude of this rapidly decaying contribution is given by the coefficient of e − 1 2 tα + . Typically, the rate of the transition from dark to irreversible dark (k di ) will be smaller than the other rates, so that we can make the approximation ω ≈ (k di + k dn + k nd ). The change in fluorescence due to the fast decay is then approximated by If positive, this last quantity can be used as a measure of FP photochromicity. If light exposure decreases the ratio k nd k dn as compared to k nd−spont k dn−spont , then fluorescence increases after exposure to light. This phenomenon is called photochromicity and is shown by several of the FPs that we have tested.
We enlist these measures of photochromicity for all FPs in Table S1.
If negative, we do not speak about photochromicity, but the change in fluorescence will lead to a first fast decay of fluorescence, before the longer timescale, due to the lower rate constant k di , will determine the dynamics.
Photochromicity can be induced by light of different wavelengths in various extents. Because we have measured fluorescence dynamics of the FPs under exposure to these different wavelengths, we can find out the photochromicity per wavelength if we reintroduce the wavelength-specific rate constants. This becomes a very complicated expression. However, if we linearly approximate the contributions of all wavelengths photochromicities (by calculating derivatives of the change in fluorescence with respect to the light powers P FP at the point in which all powers are zero), then we find The coefficient before P FP is similar to the photochromicity measure and we take this as the photochromicity measure for light of this specific wavelength.