cgreGFP Robustness and Physiochemical Stability

This repository contains code for the data analysis for Bachelor's Thesis "Mutational Robustness and Physical Stability of Fluorescenct Proteins" conducted at Institute of Science and Technology in Evolutionary Genomics Laboratory under supervision of Prof. Dr. Fyodor Kondrashov and Louisa Gonzalez Somermeyer. The work is based on Gonzalez Somermeyer et al. (see GitHub).

Table of contents

1. Introduction
   • Hypothesis
   • Testing
2. Methods
   • Protein purification and concentration
   • Urea sensitivity assays
   • Thermosensitivity assays
3. Resluts and Discussion
   • Mutational robustness
   • Urea sensitivity assays
   • Thermosensitivity assays
4. Conclusion
5. Code availability
   • Structure
6. Usage
   • Software requirements
   • Installation
7. References

Introduction

One of the main goals of biology is to describe how genotype determines the phenotype and the evolutionary success, the fitness, of an organism [3, 4, 13]. In 1932, Sewall Wright introduced the concept of a fitness landscape, also called adaptive landscape [13, 14]. Fitness landscape relates the genotype to the fitness and can be visualized in a simplified way as a 3D representation, where the high-dimensional space of all possible genotypes is reduced to a 2D plane and the fitness is plotted on the z-axis. The topography of the fitness landscape displays the interactions between different mutations and their joint effect on the fitness [2, 12, 13]. Evolution can be interpreted as a `walk` on the fitness surface and the adaptation as a `climb` on the hill [2, 13]. A smooth landscape with a single peak implies a lack of epistasis, where the effect of each mutation is fixed and additive [12]. In a rugged fitness landscape, where e.g. multiple fitness peaks are separated by valleys of low fitness, the effect of each mutation depends on the other present mutations, implicating epistatic interactions [12, 13].

Fig. 1: Conceptual representation of a fitness landscape. One hundred different random hills were generated using mathematical function $c \cdot exp(−((x − a)^2 + (y + b)^2 ))$, as parameters a, b and c were chosen randomly and the hills visualized using custom Python script .

A protein fitness landscape relates the changes in the amino acid sequence to the general function of the studied protein [5]. In this study, green fluorescent proteins (GFPs) were investigated. GFPs make up a structurally homologous protein family with weak sequence identities, first discovered in coelenterates in 1962 with the unique structure of a β-barrel of eleven strands surrounding a central helix (see Fig. 2) [7, 8, 9, 10]. The helical segments on the barrel ends shield the chromophore from the solvent [10]. GFP has found many applications in biotechnology due to its valuable properties, such as posttranslational autocatalytic chromophore formation without requirement of substractes, its ability to serve as a fusion partner, and low sensitivity to enironmental factors [10, 11, 15]. Therefore, it is also a good model for the adaptive landscapes as its fitness can be directly quantified by measuring its fluorescence [4, 5, 11].

Four extant GFPs (avGFP (Aequorea victoria, Hydrozoa), amacGFP (Aequorea macrodactyla, Hydrozoa), cgreGFP (Clytia gregaria, Hydrozoa), and ppluGPF2 (Pontellina plumata, Copepoda)) sharing different degrees of sequence identity (17% to 82%) have been explored by deep mutational scanning and used for building a predictive neural network [4, 11]. For each protein a library of random mutants was generated and labelled with a random combination of nucleotides (barcodes) using error-prone PCR. Each library was sorted according to the intensity of green fluorescence and sequenced by HTS using barcodes. Finally, statistical analysis of all four fitness peaks was performed and the data was used to train a neural network to predict functional genotypes many mutations away from the wild type sequence.

The neural network showed higher accuracy for predictions using the sharp cgreGFP fitness peak, as the deleterious effect of specific mutations as well as negative epistatic interactions were easier to detect. It has been suggested, that these observed differences in mutational robustness could be explained by the difference in thermodynamic stability [1, 6, 11]. cgreGFP is a dimeric 27 kDa protein with an absorbance peak at 485 nm and a fluorescence maximum at 500 nm (see Fig. 2) [7, 8]. It has been observed to be the most kinetically unstable protein out of 4 previously mentioned GFPs, furthermore, reinforcing the idea that the fitness peaks could also be characterized by the thermodynamic stability of the proteins [4].

Fig. 2: Visualization of the β-barrel structure of 27 kDa cgreGFP generated by custom Python script in PyMOL.

Hypothesis

The mutational robustness of a protein can be correlated to its physiochemical stability.

Testing

In order to test the hypothesis, we examined 16 fluorescing cgreGFP variants at different distances from wild-type (WT) cgreGFP which had been previously predicted to be functional by a neural net trained on the cgreGFP mutant library. Two different sequences were predicted at each distance of 6 to 48 mutations with increments of 6 mutations. For each protein, mutational stability was estimated by generating mutant libraries and sorting the cells based on the intensity of green fluorescence. The structural stability was estimated by urea sensitivity and thermosensitivity assays. Finally, the thermodynamic and mutational robustness were compared to each other.

Methods

16 cgreGFP variants with differing distances from the WT cgreGFP were predicted to fluoresce by the neural network. The variants had from 6 to 48 amino acid changes (in- cremented by 6 mutations, 2 predictions at each step) from the WT sequence. Random mutant libraries were constructed for each variant by error-prone PCR and sorted using fluorescence-activated cell sorting (FACS). The generated data was filtered using custom Python scripts and provided by the host group to estimate the mutational robustness of each variant (see data).

Protein Purification and Concentration

Coding sequences for neural network-generated cgreGFP variants with N-terminal His-tags and flanked by BsaI restriction sites were cloned into T7 expression vectors with KanR via modular cloning strategy, MoClo. DH5α E. coli cells were transformed with MoClo reaction mixture via heat shock and grown on LB agar supplemented with 50 µg/mL kanamycin. From each plate two colonies were screened for possible screened for correct construct insertion by polymerase chain reaction (PCR) as well as sequencing. One colony was chosen for further investigation. Plasmid was extracted from liquid cultures of one selected colony and chemically competent BL21-DE3 (New England Biolabs) cells were transformed and grown on ten 12 × 12 cm LB agar 50 µg/mL kanamycin and 20 µM IPTG plates. The cells were sonicated and proteins were isolated using nickel-sepharose protein purification resin (Cytiva). The concentrations of the isolated proteins were determined using Nanopore and Bradford assay using Pierce Coomassie Protein Assay Kit (ThermoFisher) (see Bradford assay data analysis). The quality was further assessed using SDS-PAGE and Western Blotting.

Urea sensitivity assays

Absorbance and fluorescence spectra on Biotek SynergyH1 plate readers were measured for 16 cgreGFP variants. For absorbance, 200 µL reaction mixtures of 17.6 µM purified protein in 1 × PBS and either 0 or 9 M urea were prepared. The absorbance was then measured in 5 nm steps from 300 nm to 700 nm for approx. 60 hr at 42 ℃ in 96-well clear and flat-bottomed plates. For fluorescence, 200 µL samples contained 0.15 µM purified protein in 1 × PBS and either 0 or 9 M urea. The samples were excited by the wavelength of 420 nm and the emission was measured in 5 nm steps from 450 nm to 700 nm for for approx. 60 hr at 42 ℃ in 96-well black-walled and flat-bottomed plates. See data and its analysis.

Thermosensitivity assays

The fluorescence of 16 purified cgreGFPs was measured in different urea concentrations (from 0 to 8 M, in 1 M intervals) during heating from 20 ℃ to 99 ℃ at an increase rate of approx. 2 ℃ /min on a Roche LightCycler 480 monitoring the SYBR-Green channel (excita- tion at 465 nm, collection at 510 nm). For this purpose, 20 µL samples of 0.1 mg/mL purified protein concentration in 1 × PBS were prepared in white 96-well plates and four technical replicates were made for each measurement. The melting temperatures were determined as the peaks of the melting curves averaged over all replicates. See data and its analysis.

Results and discussion

Mutational robustness

Mutant libraries containing thousands of variants of the 16 cgreGFPs along with the WT cgreGFP were sorted using FACS based on green fluo- rescence intensity (provided dataset). In order to visualize the mutational robustness of the variants, the median measured brightness of mutant variants was plotted as a function of their divergence from the original sequence (number of mutations). The steepness of the descent of the curve, thus the sharpness of the fitness peak, directly correlates with mutational robustness, as more fragile proteins lose their functionality quicker as distance from the fitness peak increases. Using custom python script, different functions were fitted to the plot of the number of mutations against the mean brightness. Sigmoid function fitted best, so through its inverse function the number of the mutations needed to reach half of the initial measured brightness was taken as a measure for the mutational robustness of the protein. See Jupyter notebook.

Urea sensitivity assays

Absorbance and fluorescence spectra were measured in 1 × PBS (see Fig. 6) and 9 M urea (see Fig. 7) for 16 cgreGFP variants. As time progressed, neither absorbance nor fluorescence changed in 1 × PBS. As the proteins were kept in a buffer under stable reaction conditions, no change in the intensity of absorbance or fluorescence was expected. On the contrary, in 9 M urea absorbance as well as fluorescence of the proteins decreased as time progressed, with fluorescence seeming to be more affected. This could be attributed to the fact, that the fluorescence relies more on the correct folding of the protein as the chromophore has to be properly shielded from the environment.

The peak signal intensity over time in 9 M was plotted in order to visualize the urea sensitivity of the proteins. Steeper curves indicate that the protein variant is more sensitive. Several measures have been considered to quantify the urea sensitivity of the proteins. The sigmoid function was taken again and its inverse enabled to calculate the precise time, at which the absorbance/fluorescence is halved. Pearson’s correlation coefficient was then used to calculate the relatedness between the mutational robustness and urea sensitvity of the proteins (see Table 1).

Table 1: The Pearson’s correlation coefficients and corresponding p-values between different urea sensitivity measures and the mutational robustness of 16 cgreGFP protein variants and the WT cgreGFP. Different methods were considered to correctly estimate the urea sensitivity. The slopes were determined for the linear part of the graphs by linear regression models. As another solution, the first time-point was taken at which fluorescence/absorbance dropped to less than the half of its initial value. Finally, sigmoid function was fitted to the curve and the specific time-point was predicted using the pretrained parameters and the inverse sigmoid function. See data analysis.

	Abs. corr. coef.	Abs p-value	Fl. corr. coef.	Fl p-value
Slopes	0.37	0.16	0.45	0.08
Time at ≤ half the signal	0.54	0.03	0.29	0.28
Sigmoid function	0.37	0.16	0.29	0.28

Thermosensitivity Assays

The melting temperature of the proteins was taken as the peak of the derivative of the melting curve. As the urea concentration increases, melting temperatures decreases. Therefore, the destabilizing effects of heat and chaotropic agent on protein stability are cumulative.

The correlation between mutational robustness and thermostability decreases and the p-values increase with increasing urea concentration. Therefore, the melting temperatures measured in 1 × PBS have the highest correlation with the mutational robustness. This could be explained by the possibility that the simultaneous exposure to heat and the chaotropic agent disrupts the protein stability in many different, nonlinear ways.

Conclusion

Some correlation between the mutational robustness and the thermodynamic stability of the investigated cgreGFPs has been found. However, this correlation appeared to be weak and unreliable, based on the p-values. Possibly some additional assays, such as differential scanning fluorimetry (DSF), circular dichroism (CD), and differential scanning calorimetry (DSC), could be run in order to better estimate the thermosensitivity of the 15 von 27 proteins [4]. Otherwise, some other denaturing agents, e.g. guanidine hydrochloride, or assays measuring sensitivity to pH changes could also be used to quantify the thermodynamic stability of the proteins and relate it to the measured mutational robustness.

Code availability

All functions used for data processing and visualization are stored in utils.py files. The main part of data analysis and visualization can be found in corresponding jupyter notebooks.

Structure

The folder Abs_Fluor_analysis contains 6 subfolders:

• Bradford_protein_concentrations - data and code for determination of the isolated cgreGFPs` concentrations
• Intro_figures - code for generation of Fig. 1
• Mut_robustness - code for analysis of provided processed FACS-data
• Pymol - code for visualization of cgreGFP, Fig. 2
• Thermostability - data and code from thermosensitivty assays
• Urea - data and code from urea sensitivity assays

The project has the following structure:

GFP_Robustness
│   README.md
│   requirements.txt  
│
└───Abs_Fluor_analysis
│   │
│   └───Bradford_protein_concentrations
│       │   220620_cgre_bradford.xlsx
│       │   bradford_analysis.ipynb
│   │
│   └───Intro_figures
│       │   random_fitness_landscape.ipynb
│       │   random_fitness_landscape.png
│   │
│   └───Mut_robustness
│       │   Mut_robustness.ipynb
│       │   amacGFP_cgreGFP_ppluGFP2_final_aminoacid_genotypes_to_brightness.csv
│       │   utils.py
│       └───data
│       │      │   ...
│   │
│   └───Pymol
│       │   cgre.txt
│       │   cgreGFP.png
│       │   cgre_GFP.pse
│   │
│   └───Thermostability
│       │   thermostability.ipynb
│       │   utils.py
│       └───data
│       │      │   ...
│   │
│   └───Urea
│       │   corr_results.csv
│       │   urea_sensitivity.ipynb
│       │   utils.py
│       └───data
│       │      │   ...

Usage

All code was run on Python 3.9.7 on Ubuntu 21.04 and Ubuntu 22.04.

Software Requirements

• Python 3.9.7
• Ubuntu 21.04 or Ubuntu 22.04
• Bash

Installation

1. You need git to be installed. Open terminal (Crtl+Alt+t) and run following commands:

git clone https://github.com/AnnaToi01/GFP_Robustness.git
cd GFP_Robustness

2. Create virtual environment

   • Via venv

     • Create virtual environment

       python -m venv venv

     • Activate it

       source venv/bin/activate

   • Via virtualenv

     • Install virtualenv if it is not installed.

       pip install virtualenv

     • Create virtual environment

       virtualenv venv --python=3.9

     • Activate it

       source ./venv/bin/activate

   • Via conda

     • Install Anaconda if not already installed (see Instructions).
     • Create virtual environment

       conda create --name <env_name> python=3.9

     • Activate it

       conda activate <env_name>

3. Install necessary libraries

pip install -r requirements.txt

After all required libraries installation you can launch jupyter notebooks and work with chosen datasets.

References

1. Shimon Bershtein et al. “Robustness-epistasis link shapes the fitness landscape of a randomly drifting protein”. In: Nature 444.7121 (2006), pp. 929–932. issn: 1476-4687. doi: 10.1038/nature05385.
2. Luca Ferretti et al. “Evolutionary constraints in fitness landscapes”. In: Heredity 121.5 (2018), pp. 466–481. issn: 1365-2540. doi: 10.1038/s41437-018-0110-1.
3. Inês Fragata et al. “Evolution in the light of fitness landscape theory”. In: Trends in Ecology & Evolution 34.1 (Jan. 2019), pp. 69–82. issn: 0169-5347. doi: 10.1016/j. tree.2018.10.009.
4. Louisa Gonzalez Somermeyer et al. “Heterogeneity of the GFP fitness landscape and data-driven protein design”. In: eLife 11 (2022). Ed. by Daniel J Kliebenstein and Naama Barkai, e75842. issn: 2050-084X. doi: 10.7554/eLife.75842.
5. Emily C. Hartman and Danielle Tullman-Ercek. “Learning from protein fitness land- scapes: a review of mutability, epistasis, and evolution”. In: Current Opinion in Systems Biology 14 (2019), pp. 25–31. issn: 2452-3100. doi: 10.1016/j.coisb.2019.02.006.
6. Ryo Kurahashi, Satoshi Sano, and Kazufumi Takano. “Protein Evolution is Potentially Governed by Protein Stability: Directed Evolution of an Esterase from the Hyper- thermophilic Archaeon Sulfolobus tokodaii”. In: Journal of Molecular Evolution 86.5 (2018), pp. 283–292. issn: 1432-1432. doi: 10.1007/s00239-018-9843-y.
7. Natalia P. Malikova et al. “Green-Fluorescent Protein from the Bioluminescent Jellyfish Clytia gregaria Is an Obligate Dimer and Does Not Form a Stable Complex with the Ca2+-Discharged Photoprotein Clytin”. In: Biochemistry 50.20 (May 2011), pp. 4232–4241. issn: 0006-2960. doi: 10.1021/bi101671p.
8. Svetlana V. Markova et al. “Green-fluorescent protein from the bioluminescent jelly- fish Clytia gregaria: cDNA cloning, expression, and characterization of novel recombi- nant protein”. In: Photochem. Photobiol. Sci. 9 (6 2010), pp. 757–765. doi: 10.1039/ C0PP00023J.
9. S. James Remington. “Green fluorescent protein: a perspective.” eng. In: Protein sci- ence : a publication of the Protein Society 20 (9 2011), pp. 1509–19.
10. Karen S. Sarkisyan et al. “Local fitness landscape of the green fluorescent protein”. In: Nature 533.7603 (2016), pp. 397–401. issn: 1476-4687. doi: 10.1038/nature17995.
11. Roger Y. Tsien. “THE GREEN FLUORESCENT PROTEIN”. In: Annu. Rev. Biochem. 67.1 (June 1998), pp. 509–544. issn: 0066-4154. doi: 10.1146/annurev.biochem.67. 1.509.
12. Jeremy Van Cleve and Daniel B. Weissman. “Measuring ruggedness in fitness land- scapes”. In: Proceedings of the National Academy of Sciences 112.24 (June 2015), pp. 7345–7346. doi: 10.1073/pnas.1507916112.
13. J. Arjan G. M. de Visser and Joachim Krug. “Empirical fitness landscapes and the predictability of evolution”. In: Nature Reviews Genetics 15.7 (2014), pp. 480–490. issn: 1471-0064. doi: 10.1038/nrg3744.
14. S. Wright. “The Roles of Mutation, Inbreeding, crossbreeding and Selection in Evolu- tion”. In: Proceedings of the XI International Congress of Genetics 8 (1932), pp. 209–222.
15. Marc Zimmer. “Green Fluorescent Protein (GFP): Applications, Structure, and Re- lated Photophysical Behavior”. In: Chem. Rev. 102.3 (Mar. 2002), pp. 759–782. issn: 0009-2665. doi: 10.1021/cr010142r.