Unearthing the transition rates between photoreceptor conformers
 Robert W. Smith^{1, 2}View ORCID ID profile,
 Britta Helwig^{1, 3},
 Adrie H. Westphal^{3},
 Eran Pel^{1, 3},
 Maximilian Hörner^{4, 5},
 Hannes M. Beyer^{4, 5},
 Sophia L. Samodelov^{4, 6},
 Wilfried Weber^{5},
 Matias D. Zurbriggen^{6},
 Jan Willem Borst^{3} and
 Christian Fleck^{1}Email author
https://doi.org/10.1186/s129180160368y
© The Author(s) 2016
Received: 13 June 2016
Accepted: 7 November 2016
Published: 25 November 2016
Abstract
Background
Obtaining accurate estimates of biological or enzymatic reaction rates is critical in understanding the design principles of a network and how biological processes can be experimentally manipulated on demand. In many cases experimental limitations mean that some enzymatic rates cannot be measured directly, requiring mathematical algorithms to estimate them. Here, we describe a methodology that calculates rates at which lightregulated proteins switch between conformational states. We focus our analysis on the phytochrome family of photoreceptors found in cyanobacteria, plants and many optogenetic tools. Phytochrome proteins change between active (P _{ A }) and inactive (P _{ I }) states at rates that are proportional to photoconversion crosssections and influenced by light quality, light intensity, thermal reactions and dimerisation. This work presents a method that can accurately calculate these photoconversion crosssections in the presence of multiple nonlight regulated reactions.
Results
Our approach to calculating the photoconversion crosssections comprises three steps: i) calculate the thermal reversion reaction rate(s); ii) develop search spaces from which all possible sets of photoconversion crosssections exist, and; iii) estimate extinction coefficients that describe our absorption spectra. We confirm that the presented approach yields accurate results through the use of simulated test cases. Our test cases were further expanded to more realistic scenarios where noise, multiple thermal reactions and dimerisation are considered. Finally, we present the photoconversion crosssections of an Arabidopsis phyB Nterminal fragment commonly used in optogenetic tools.
Conclusions
The calculation of photoconversion crosssections has implications for both photoreceptor and synthetic biologists. Our method allows, for the first time, direct comparisons of photoconversion crosssections and response speeds of photoreceptors in different cellular environments and synthetic tools. Due to the generality of our procedure, as shown by the application to multiple test cases, the photoconversion crosssections and quantum yields of any photoreceptor might now, in principle, be obtained.
Keywords
Background
The aim of computational biology is to obtain a better quantitative understanding of biological systems and the principles upon which evolution has constrained cellular networks. Building on such knowledge, networks can be designed that yield certain optimal responses in both natural and synthetically constructed organisms. However, despite the wealth of experimental techniques and data available, biological parameters can only be estimated relative to standard values, if they can be calculated at all. This has increasingly led to the development of optimisation techniques that aim to find the most accurate or probable set of parameters that describe a particular set of data (see [1, 2]). Here, we will focus on one instance where this is a particularly pertinent problem  namely the rates at which a lightregulated protein (photoreceptor) is able to switch between different conformational states.
One such family of photoreceptors is the phytochromes. Phytochromes have been found throughout the cyanobacterial and land plant kingdoms [3, 4]. Each of these proteins respond to light through a similar chromophorebased mechanism [5]. Upon illumination of red light, a conformational change in the chromophore leads to the activation of phytochromes (P _{ A } = active phytochrome) to regulate downstream networks [6, 7]. This process can be reversed, leading to increased populations of inactive phytochrome (P _{ I }), by illumination with a second light source of a different wavelength or leaving the protein in the dark, where thermal reactions control the conformational change (thermal reversion) [8–12]. The wavelength of light required for phytochrome inactivation changes between phytochrome species, with plant phytochromes requiring farred light whilst bacterial and algal phytochromes can be inactivated by wavelengths across the visible light spectrum [4, 11]. The transition rates between conformers is related to the photoconversion crosssections of the photoreceptor and the light conditions that the photoreceptor is exposed to (i.e. the intensity and spectral composition of the light source) [13, 14]. photoconversion crosssections are, essentially, the ability of a molecule to absorb a photon of a given wavelength resulting in a conformational change [15]. A range of spectroscopic techniques have been used to observe these state changes [9, 11, 16–18]. For phytochromes, absorption spectra show a relatively large spectral overlap implying the presence of mixed populations of P _{ A }, P _{ I } and, in some cases, intermediate conformations (such as LumiR [19–21]), coexisting within the protein sample. By using absorption spectra, one can not only observe levels of P _{ A } or P _{ I } within a phytochrome sample, but the level of absorbance can be related to sample concentration, molecular weight and extinction coefficient [10, 11, 13, 15]. The extinction coefficient is defined as the ratio between the photoconversion crosssection of a phytochrome species and its quantum yield, where the quantum yield is the ratio of absorbed photons that lead to a molecular change with the total number of photons that are absorbed [15].
Therefore, photoconversion crosssections can be inferred directly from absorption spectra if some other parameters are known. This is the case often described by Butler’s model [15]. However, to accurately estimate the photoconversion crosssection and quantum yield of one phytochrome conformational state using this model, an experimental configuration is needed whereby 100% of the population is in a single state. For many phytochromes, no experimental condition can achieve such requirements due to overlapping absorption spectra that are indicative of mixed populations [13, 15]. Previous studies of plant phytochromes from Avena sativa have estimated the photoconversion crosssections and quantum yields by assuming specific percentages of P _{ A } within the phytochrome population under different conditions (∼87% after saturating red light illumination, 100% after farred illumination), allowing Butler’s model to be solved [10, 11]. We shall refer to these estimates as the ‘Mancinelli spectra’ and ‘Mancinelli quantum yields’ following the review by Alberto Mancinelli [13].
Building on Butler’s model, the Verméglio group has proposed an analytical method to determine the photoconversion crosssections of phytochrome proteins and the ratio of the quantum yields between the P _{ A } and P _{ I } states within a mixed population [22, 23]. By using absorption spectra obtained as a result of three different illumination conditions, one can calculate the photoconversion crosssections by solving a system of linear equations [23]. Quantum yields need to be calculated using other methods. However, these analytical equations do not take into account any effects of thermal reversion which, under certain conditions, can play a significant role in the reactions governing phytochrome transitions [14, 24]. Another set of methods have also recently been developed that can estimate the number of subpopulations within a proteins reaction cycle [16–18]. As a result these methods have enabled comparisons of quantum yields between a phytochrome subfamily but does not provide any information about photoconversion crosssections or transition rates [25]. Thus, neither of these approaches provide the same level of information obtained from the ‘Mancinelli spectra’ that allows researchers to calculate, for any wavelength of light, how the phytochrome populations switch between different states.
An example of a phytochrome protein, whose dynamics have been studied both within its natural context and in optogenetic tools, is phytochrome B (phyB) from Arabidopsis thaliana [14, 26–28]. Generally, phyB works as follows regardless of host cell. Once synthesised in the cytosol, phyB covalently binds tetrapyrrole chromophores (phytochromobilin (P ΦB) in plants, cyanobacterial phycocyanobilin (PCB) in cyanobacteria and synthetic systems) to form a holoprotein [6, 7]. As with other phytochromes, phyB is activated by red light and inactivated by thermal reversion and/or exposure to farred light. Upon activation, phyB is able to interact with other proteins, translocate into the nucleus and control downstream processes [27, 29]. These downstream processes include transcription [26, 30], protein degradation [31] and protein sequestration, potentially through the formation of nuclear bodies [27, 32, 33]. In plants, the interactions of phyB eventually lead to the suppression of seedling elongation and inhibition of the vegetative to flowering transition [34].
So far in optogenetics, the truncated Nterminal version of phyB (consisting of the first 650 amino acids and referred to as phyBN from hereon) has been used [9, 26, 27]. Importantly, phyBN has been shown to be the most stable construct that still exhibits the correct responses and dynamics to light signals compared to the fulllength variant [9, 26, 27, 35] whilst thermal relaxation of phyBN is slower in vitro than fulllength phyB in vivo [8, 12, 28]. Futhermore, compared to the activity of fulllength phyB in planta, which has been shown to act as a dimer, phyBN in synthetic systems is believed to function as a monomer [36, 37]. Formation of phyB dimers in plants means that the networks regulating downstream activity are different between natural and orthogonal synthetic systems [28, 38, 39]. Through an increasing number of systems biology studies, the networks and kinetic rates describing phyB’s role in controlling plant development have been estimated and verified [14, 28, 40]. Since these studies have dealt with the fulllength phyB protein, they have assumed that the ‘Mancinelli spectra’ are the correct photoconversion crosssections that determine phyB conformational changes. Despite the ‘Mancinelli spectra’ being calculated using fulllength plant phyA, this approximation is justified since plant phyA and phyB molecules respond to light in a similar manner. However, as a wider range of phytochromes and their structural variants are being studied for their photophysical properties and dynamics in natural and optogenetic systems, there is a real need to accurately estimate these photoconversion crosssections and the rates at which phytochromes switch between conformational states and how efficient these processes are. With such knowledge, one could potentially look to compare and optimise the transition rates of photoreceptors for use in optogenetic networks via protein engineering.
Here, we build on the analytical methods proposed by the Verméglio group to include further phytochrome reactions, such as thermal reversion and dimerisation, to obtain upper limits for the quantum yield ratio and search spaces for the photoconversion crosssections. Using this information, we then propose an optimisation algorithm that reliably estimates the photoconversion crosssections and quantum yields of any photoreceptor. The generality of our approach is highlighted by simulating different members of the phytochrome family  including a red/green cyanobacteriochrome and a red/farred plant phytochrome. First, our output is compared to those obtained by Butler’s model, showing that our algorithm can find the correct set of photoconversion crosssections and quantum yields when Butler’s model fails. Subsequently we show, for the red/farred photoreceptor reminiscent of fulllength phytochromes found in plants, that our algorithm can find the correct photoconversion crosssections and quantum yields even in the presence of multiple thermal reversion reactions. Furthermore, the results are insensitive to noise. Finally, our method is applied to the phyBN protein used in optogenetics and provides the photoconversion crosssections for this protein for the first time in vitro. Given adequate data, we propose that this procedure can be adopted for any photoreceptor protein and will be of use to those wishing to obtain quantitative understanding of lightregulated processes in natural and synthetic biological systems or comparisons of phytochrome family members and structural variants.
Methods
Methodology overview
In order for the method to work, absorption spectra need to be obtained from different experimental conditions. Importantly, this data needs to illustrate the switching from active to inactive states of the photoreceptor and vice versa. For a photoreceptor that responds to a single region of the light spectrum (such as the cryptochromes [41]) spectra need to be obtained from two experiments: under constant darkness after the photoreceptor population has been completely activated by light, and under prolonged periods of activating light after complete inactivation in darkness (see Section 1.4 of Additional file 1). For phytochromes, that respond to two regions of the light spectrum (for example see [11]), data would need to be obtained after initial illumination of red light followed by constant treatment with inactivating light conditions or darkness and vice versa, recording the activation of phytochromes in red light after initial exposure to inactivating light.
where A _{ s }(λΩ _{ P },Ω _{ E }) is the simulated absorption at wavelength λ given a set of tuples describing the preparatory, \(\Omega _{P} = \{(t_{P},z_{P}^{\lambda })\}\), and experimental, \(\Omega _{E}= \{(t_{E},z_{E}^{\lambda })\}\), light conditions. Here, the light conditions are described by the time of exposure, t _{ P } and t _{ E }, in s and the spectral and intensity composition of the light source, \(z_{P}^{\lambda }\) and \(z_{E}^{\lambda }\), in μmol m ^{−2} s ^{−1}. The terms c ^{ k } are timedependent changes in the fraction of available chromophore, n _{ c }, within the photoreceptor monomer or dimer structure that is in state k. The parameters \(\boldsymbol \sigma = \{\sigma _{1}^{\lambda },\ldots,\sigma _{k}^{\lambda },\ldots,\sigma _{N}^{\lambda }\}\) is the set of wavelengthdependent photoconversion crosssections (in m^{2}/mol) of a sample population switching between states P _{ I } and P _{ A }, and the Φ={Φ _{1},…,Φ _{ k },…,Φ _{ N }} is the set of associated quantum yields of the lightdependent reactions. Furthermore, the c ^{ k }’s depend on the parameters related to thermal reversion, where α={α _{1},…,α _{ n }} is the percentage of the active photoreceptor (P _{ A }) population that undergoes a thermal reversion rate of β={β _{1},…,β _{ n }}, where n is the number of P _{ A } species in the population (see below), and the preparatory/experimental light conditions (Ω _{ P }, Ω _{ E }). In Section 1.2 of Additional file 1, we show the general forms for c ^{ k } used in our study and how this function is altered between different cases.
The total concentration of photoreceptor in our sample, c _{ tot }, is defined in mol/L, the term 2.303 is required to relate absorption to measured light fluences and l is the light path length through the samplecontaining cuvette in cm [15]. For our study, we have a cuvette containing \(c_{tot} = \frac {0.34~\text {mg/ml}}{70000~\text {Da}} (\sim \frac {\text {mol}}{\mathrm {L}})\) sample that is 1 cm thick and so, due to light path corrections, we use l=1.06 cm as in [15].
 1.
The parameters for thermal reversion kinetics are found, such as the rate of reversion, β, and the percentage of the population, α, that undergoes reversion at this rate (Fig. 1, Box 2). Importantly, these parameters play a role in determining c ^{ k }.
 2.
Appropriate search spaces for the extinction coefficients, \(\epsilon _{k}^{\lambda } = \sigma _{k}^{\lambda }/\Phi _{k}\), and the upper bound of the quantum yield ratio, R ^{max}, are found numerically, given the thermal reversion parameters, α and β (Fig. 1, Box 3).
 3.
photoconversion crosssections, \(\sigma _{k}^{\lambda }\), and quantum yields, Φ _{ k }, are obtained by comparing simulations described by Eq. (1) to measured timedependent absorption spectra (Fig. 1, Box 4).
The resulting photoconversion crosssections and quantum yield values should then be validated against known data. In our examples, we use either the known values (where simulated data is used) or our available datasets (in cases where biological data is used; Fig. 1, Box 5). This section will present the details of each step of the approach and the elucidation of photoconversion crosssections and quantum yields. In the results section, we will show the output from a series of test cases before using data obtained for phyBN.
Step 1: Determination of thermal reversion rates
where n is the number of exponentials required to describe the decay of P _{ A }. Using (3), we obtain estimates for α _{ j } and β _{ j } (see Section 1.8 of Additional file 1 for pseudocode). Recall from (1) that the functions c ^{ k } depend on the preparatory light input Ω _{ P }. Consequently, the rates of thermal reversion will change depending on the starting fraction of P _{ A } in the system. Importantly, under darkness, Eq. (3) shows that the percentage decrease of active photoreceptor is independent of the measured wavelength, λ, used for the analysis since \({E_{t}^{D}}\) is not a function of λ. This can be seen in Additional file 1: Figure S1.
Step 2: Finding search spaces for the transition rates and quantum yield ratio upper bound
The next step of our methodology is to obtain the search spaces for \(\sigma _{k}^{\lambda }\) and their respective Φ _{ k }. In order to do this, a previously published method will be extended to include the effects of thermal reversion, using the parameters obtained from the previous section [22, 23]. This section will outline the previously published method for photoreceptors with two lightregulated states and extend this to include a single thermal reversion reaction. In the Additional file 1 we present equations for any n _{ c }, and cases whereby the photoreceptor undergoes multiple thermal reversion processes or when multiple species exist in the photoreceptor reaction cycle.
where \(\epsilon _{k}^{\lambda }\) was calculated such that \(\hat {\epsilon }_{k}^{\lambda } = 2.303{lc}_{tot}{\epsilon }_{k}^{\lambda }\) are the wavelengthdependent extinction coefficients and X _{ k } are scalar constants. Due to the relationship between extinction coefficients, photoconversion crosssections and quantum yields (\(\epsilon _{k}^{\lambda } = \sigma _{k}^{\lambda }/\Phi _{k}\)), the resulting spectra provide us with limits for the optimisation algorithm developed below to minimise the difference between our data and Eq. (1).
These equations are implicit for X _{1} and X _{2} because the fractions \(c_{\lambda _{w}}\) depend, via Eq. (4), on unknown \(\epsilon _{k}^{\lambda }\) (see Sections 1.2 and 1.3 of Additional file 1). Thus, other than in the simplest cases, these equations require solving numerically. By calculating the values for \(c_{\lambda _{w}}\) for given values Φ, one can obtain spectra for \(\hat {\epsilon }_{k}^{\lambda }\) via Eqs. (4) and (6). Doing this for all possible values of Φ provides us with the search space of \(\epsilon _{k}^{\lambda }\) spectra to be used within the optimisation algorithm.
The last remaining bound to be obtained is for R=Φ _{ A }/Φ _{ I }. To highlight how the upper bound for R can be obtained we look at the case where a photoreceptor has a single thermal reversion reaction. For the case of multiple thermal reversion rates the same steps can be followed to calculate R.
where \({S_{i}^{j}} = \sum _{\lambda }\epsilon ^{\lambda }_{i}{z^{\lambda }_{\lambda _{j}}}\). The function R ^{ o } is the same as that calculated using the previously published method in the absence of thermal reversion [23]. Note again that the function of R is, in fact, implicit. For a given set of Φ, one calculates the values of X _{ k } from Eq. (6) to then obtain a value R using Eq. (8). To limit our optimisation search space we wish to find bounds of R given our data. By definition, the lower bound of R is zero. An upper bound for R can be found since the second term of Eq. (8) is always positive since β, Φ _{ I } and \(\epsilon _{A}^{\lambda }\) are all positive. Thus, R ^{ o } is the upper bound for R given specific values of Φ. Therefore, we use the numerical maximum of R, R ^{max}, given all possible combinations of Φ as the upper bound of R.
The above steps therefore provide, in the presence of different numbers of thermal reversion reactions, constraints on the wavelengthdependent photoconversion crosssections and quantum yields. The final step in obtaining photoconversion crosssections and quantum yields is to determine where in the defined search space the minimum difference between simulated spectra and experimental data exists.
Step 3: Simulating timedependent absorption spectra

One obtains a value of Φ _{ I } from R=Φ _{ A }/Φ _{ I }.

These values are then input into Eq. (6) to obtain values of X _{1} and X _{2}.

The obtained values of X _{1} and X _{2} are used in Eqs. (4) and (5) to obtain \(\epsilon _{A}^{\lambda }\) and \(\epsilon _{I}^{\lambda }\).

These spectra, plus the estimates of Φ, are input into Eq. (1) to simulate absorption spectra \(A_{s}\left ({\lambda }\left \left \{\left (t_{D}\right.\rightarrow \infty,z^{\lambda }_{\lambda _{w_{1}}}\right)\right \},\left \{\left (t,z^{\lambda }_{\lambda _{w_{2}}}\right)\right \}\right)\).
In our study all optimisation routines use MATLAB’s fmincon function that is started from 100 random initial points to cover as much of the global parameter space as possible. Furthermore, 95% confidence intervals for Φ _{ k } and thermal reversion parameters are calculated from profile likelihoods following the method outlined in [42].
Software
All computational procedures were performed using MATLAB (Mathworks, Massachusetts, USA). The codes created for our optimisation algorithms can be downloaded from http://gitlab.com/wurssb/Absorption_Spectra_Optimization.
Results
Methodology successfully returns input values for extreme simulated test cases
To test the validity of our algorithm, we have tested the system using simulated data and compared the output to results obtained using Butler’s model [11]. Here, we highlight two extreme test cases in the absence of noise. Notably, we assume that the lightdependent switching of the photoreceptor population occur on the ms to s timescale, comparable to what is known for fulllength plant phytochromes in vivo (see Additional file 1: Figure S3) [13]. For full details of the parameters used in our test cases see Additional file 1: Tables S1 and S2.
To highlight that our algorithm is an accurate generalisation of Butler’s model, we now show the results for a photoreceptor that does not satisfy the conditions required for Butler’s model to produce accurate results. In this test case, we have simulated a photoreceptor that is activated by 660 nm light, but inactivated by 680 nm light with quantum yields of Φ _{ A }=0.1 and Φ _{ I }=0.3. Such a case could occur when analysing point mutations within phytochromes (see [35] for examples). Since this photoreceptor is unable to obtain populations containing 100% of P _{ I } or P _{ A }, Butler’s model fails to predict the correct photoconversion crosssections and quantum yields (blue and green lines in Fig. 2 e, f). However, our algorithm finds the correct photoreceptor parameters and simulations perfectly match the synthetic data (Fig. 2 e–h). Thus, we have shown that, in both cases where Butler’s model can be applied and where it cannot, our optimisation algorithm is able to find the ‘true’ photoconversion crosssections and quantum yields of photoreceptors.
Approach is able to find optimal photoconversion crosssections in the presence of multiple biochemical reactions
As discussed in Section 1.1 in Additional file 1, there is more than one way in which a photoreceptor can undergo multiple thermal reversion reactions. One option is through photoreceptorchromophore interactions crucial for the transformation from active to inactive photoreceptor. The second is through dimerisation, whereby the P _{ A }−P _{ I } heterodimer has a different thermal reversion rate to the P _{ A }−P _{ A } homodimer, as is the case for fulllength phyB found in plants [28]. Notably, if spectra are obtained from a sample that contain a mix of these populations, we still wish to obtain photoconversion crosssections for the monomeric P _{ A } and P _{ I } species. To show that our methodology can handle both cases, we have simulated a red/farred photoreceptor using the ‘Mancinelli’ spectra and quantum yields in the presence of multiple thermal reversion reactions [13].
By changing the c ^{ k } function, we are able to describe absorption spectra that result from mixed populations of dimerised photoreceptors. Again, using an order of magnitude difference between the thermal reversion rates, we find the parameters used to create our test data (Fig. 3 d, e). The resulting simulations using these parameters match the synthetic datasets (Fig. 3 f). Furthermore, we can find the optimal photoconversion crosssections for any given set of thermal reversion parameters and using the wrong form of c ^{ k } has negative effects on results (Additional file 1: Figure S5).Thus, we have shown that even in complex photoreceptor systems our approach is able to accurately find the biochemical parameters of populationlevel state transitions for photoreceptors.
Performance is robust to noisy spectra
Due to the inclusion of noise, we also highlight here the accuracy of our approach to find the optimal parameters. In Fig. 4 e, we show the profile likelihoods obtained for Φ _{ I } and R=Φ _{ A }/Φ _{ I }. Following the theory in [42], the 95% confidence interval for these parameters occurs when changing either parameter leads to the optimal score increasing by more than χ ^{2}(0.95,1)=3.8415. From Fig. 4 e, it is clear to see that a very minor change in either parameter would result in increasing the score by more than 3.8415 (note the scale of the yaxis), suggesting that our methodology obtained the optimal parameter values with high confidence. In Fig. 4 f, we show the entire search space for our optimisation algorithm. The search space shows a clear minima at [ Φ _{ A },Φ _{ I }]=[ 0.1,0.2] that our algorithm has successfully found.
On the basis of these results we can confirm that our approach provides accurate results for multiple photoreceptors undergoing different thermal reversion reactions in the presence and absence of noise. Furthermore, the obtained parameters have been shown to be the global minimum of the search space. This would suggest that our methodology is wellsuited to dealing with absorption spectra obtained experimentally. In the following sections the transition rates for the phyBN protein fragment commonly used in optogenetic tools will be determined.
The phyBN protein fragment undergoes slow thermal relaxation
The Nterminal region of phyB (phyBN) contains the key protein domains required for light responses and is relatively stable compared to other phyB fragments upon purification [9]. Consequently, this protein is ideal for optogenetic tools although the kinetic rates describing the interprotein reactions have not yet been quantified. Before detailing the photoconversion crosssections that control the speed of lightdependent reaction, in accordance with our methodology, we need to determine the reaction rates of thermal reactions of phyBN.
Optimal thermal reversion parameters with 95% confidence intervals
Single exponential  Double exponential  

α  1  0.265 ± 0.0094 
β _{1}  3.4 ×10^{−5}± 6 ×10^{−10}  0.0011 ± 0.0011 
β _{2}  0  2.7 ×10^{−5}± 9 ×10^{−6} 
Conformational changes of phyBN are slower and blueshifted compared to fulllength phyB
According to the ‘Mancinelli spectra’ for fulllength phytochromes, the speed of transitioning between a population of active and inactive photoreceptor would occur within 520 s under the experimental conditions of this work (see Section 2 of Additional file 1 and Additional file 1: Figure S3) [13]. However, there are two important differences in conditions between the measurement of the ‘Mancinelli spectra’ and our experimental conditions. First, the ‘Mancinelli spectra’ were calculated using absorption spectra obtained from in vivo samples whereas the samples used here are in vitro. Second, in plant systems, phytochromes use P ΦB chromophores whereas, in synthetic systems, PCB is used as an alternative. Such differences have already been shown to produce ∼20 nm blueshifted absorption spectra that transition more slowly between different states (∼mins) than spectra measured in vivo (∼s) [35, 44]. Our data support these observations, whereby saturating levels of active or inactive photoreceptor can be achieved within 5 min given our experimental conditions (Fig. 5 b).
Discussion
Methodology accurately calculates properties of photoreceptors
In this work we have introduced and tested a methodology that calculates the wavelengthdependent photoconversion crosssections and the quantum yield of photoreceptor states (Fig. 1). Based on theoretical analysis (see [11, 15, 23]) of specific photoreceptor conditions, we have generalised the theory to be used for any studied photoreceptor (see Sections 1.4 and 1.5 of Additional file 1). The key step in the algorithm is the construction of search spaces for the k species within a photocycle. These search spaces contain all possible sets of photoconversion crosssections, \(\sigma _{k}^{\lambda }\), given different values of quantum yields, Φ _{ k } (for example, see Figs. 4 and 6 b). The construction of these search spaces involves solving simultaneous equations that depend on the fraction of active photoreceptor (Eqs. (4) and (6)). Thus, given the constraints of Eq. (4), the wavelengthdependency of the photoconversion crosssections can have no other form (i.e. the solution to the optimisation problem definitely exists within this space). Consequently, we have demonstrated that the approach works for different photoreceptors with multiple thermal reversion reactions and dimer forms in the presence or absence of intrinsic noise (Figs. 2, 3 and 4). In Section 1.4 and 1.5 of Additional file 1, we have presented the equations and an outline of how our analysis could be adapted for further photoreceptors that respond only to a single wavelength of light (i.e. for blue light receptors such as cryptochromes and LOVbased proteins that are used in optogenetics [45]) or for photoreceptors that have reaction cycles containing N species.
The approach presented here has high potential to be used to obtain the photoconversion crosssections for these photoreceptors, aiding both plant and synthetic biologists in understanding their respective light responsive networks in greater detail. One such example where our algorithm can be used is for the plant UVB photoreceptor, UVR8 (UV RESISTANCE LOCUS 8), that has recently been shown to exist as a mixed population of active monomers and inactive dimers under different environmental conditions [46]. Our methodology will be instructive in determining photoconversion crosssections for this photoreceptor and the rates of this lightactivated switch. Furthermore, since our proposed method can handle cases of overlapping photoconversion crosssections more robustly than previous methods, one can envisage that it can be used to obtain transition rates for photoreceptors containing point mutations (as in [35]) or to find photoconversion crosssections of photoreceptor substates, such as LumiR in the case of phytochromes [19–21]. In Section 1.5 of Additional file 1 we highlight how this analysis could be achieved, however it should be noted that obtaining analytical expressions of c ^{ k } and R becomes challenging as the number of system components increases. Thus, one could envisage that numerical solves are incorporated into the analysis to approximate values of c ^{ k } given a specific model system.
As with all methodologies, accuracy of the results is one of two factors that need to be taken into account. Whilst we have shown that our approach produces accurate results, we should also assess computational performance. Currently, running the entire process in a sequential manner on a single iMac core using MATLAB R2013a can take at least a couple of hours when 100 optimisation runs are performed, depending on the complexity of the photoreceptor system being studied. However, within the current framework there are many steps that could be parallelised. Notably, the most timeconsuming step is generating the transition rate search spaces, which becomes particularly inefficient for complex problems (such as phytochrome dimers). Upon parallelisation, this process could be sped up since the same calculations are repeatedly performed for different values of the quantum yields. Furthermore, our optimisation procedure is currently performed 100 times starting from random initial places with the quantum yield parameter space. This could similarly be parallelised such that all optimisation runs take place simultaneously. One could envisage that these steps would decrease the computational speed from hours to minutes.
The phyBN protein fragment used in optogenetic tools responds slowly to changing light conditions
To showcase how our approach can be applied to a wellstudied photoreceptor, we performed our analysis on the phyB Nterminal protein fragment that is used for optogenetic tools [26, 47]. Our results highlight three important insights. First, as previously observed, the data show that the phyBN fragment has very slow thermal reversion kinetics and lightreversibility (Fig. 5 a, b) compared to other phytochrome protein variants (for comparison, see the predicted in vivo lightreversibility of fulllength phytochromes predicted from the photoconversion crosssections in Additional file 1: Figure S3) [13, 28, 35, 43]. Second, the calculated photoconversion crosssections have a lower amplitude for phyBN in vitro than fulllength phytochromes in vivo and have a wavelengthdependency that is relatively blueshifted (Fig. 6 a). This matches with previous observations of measured P _{ A } accumulation and blueshifted absorption spectra of phyBN [9, 35]. Finally, in correlation with the slower transition between P _{ A } and P _{ I }, the efficiency of the phyBN protein to absorb photons (as determined by the quantum yields Φ _{ A } and Φ _{ I }) are decreased compared to estimates for fulllength phytochromes [13, 25]. This is likely due to the different chromophores used by phytochromes under in vivo and in vitro conditions [44]. In summation, these results suggest that responses of phyBN populations in vitro to light are inefficient and slow when compared to the fulllength counterpart in vivo.
This result has implications for the development of optogenetic tools based on the lightswitching of phyBN [26, 27, 47]. These studies have shown that common readouts of phyBbased optogenetic tools, such as gene expression, can take hours to accumulate to high levels after illumination of red light. Notably, naturally occurring processes such as nuclear transport are ratelimiting and can take 20 min before the maximal system response is observed [27]. Since the phyBN population is fully activated by red light within 5 min of our experimental set up (using 11 μmol m ^{−2} s ^{−1} LEDs), our results suggest that accelerating population level switches between photoreceptor states could also, consequently, increase the speed of optogenetic responses. For example, mutated forms of phyBN that have faster activation kinetics (such as the 90–624 amino acid fragment [35]) could be applied in optogenetic tools.
Conclusion
Here, we have built upon previous theoretical work to present a generalised approach for determining photoconversion crosssections and, consequently, transition rates between conformational states of photoreceptors. We have shown the utility of our approach through the use of simulated data for multiple photoreceptors of differing complexity before performing our analysis on the phyB Nterminal fragment used in optogenetic tools. We envisage that the methodology can be used for any photoreceptor protein fragment, allowing the research field to directly compare rates at which photoreceptors and their mutated derivatives switch between conformational states in more detail. The resulting analysis will aid future optogenetic efforts where speed, efficiency and robustness of photoreceptor dynamics are imperative for the application of tools in biobased industries.
Declarations
Acknowledgements
The authors would like to thank Profs. Wolfgang Gaertner and Silvia Braslavsky for discussions about detailed phytochrome photodynamics and for feedback on the study.
Funding
RWS is funded by FP7 Marie Curie Initial Training Network grant agreement number 316723. CF is supported by HFSP Research grant RGP0025/2013.
Availability of data and materials
All data obtained for this study and the computational algorithm can be found in the Additional file 1.
Authors’ contributions
RS and CF designed the study. RS, MDZ, JWB, and CF wrote the paper. RS, BH, AHW, EP, JWB and CF planned and carried out testing of phyBN fragment. SLS, HB, MH, WW and MDZ produced purified phyBN protein. All authors read and approved the final manuscript.
Competing interests
The authors declare that they have no competing interests.
Consent for publication
Not applicable.
Ethics approval and consent to participate
Not applicable.
Further materials
All MATLAB scripts and data used in this study can be downloaded from http://gitlab.com/wurssb/Absorption_Spectra_Optimization.
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.
Authors’ Affiliations
References
 Banga JR. Optimization in computational systems biology. BMC Syst Biol. 2008; 2:47.View ArticlePubMedPubMed CentralGoogle Scholar
 Simon D. Evolutionary optimization algorithms. New Jersey: Wiley; 2013.Google Scholar
 Duanmu D, Bachy C, Sudek S, Wong CH, Jiménez V, Rockwell NC, et al. Marine algae and land plants share conserved phytochrome signaling systems. Proc Natl Acad Sci USA. 2014; 111:15827–32.View ArticlePubMedPubMed CentralGoogle Scholar
 Rockwell NC, Duanmu D, Martin SS, Bachy C, Price DC, Bhattacharya D, et al. Eukaryotic algal phytochromes span the visible spectrum. Proc Natl Acad Sci USA. 2014; 111:3871–6.View ArticlePubMedPubMed CentralGoogle Scholar
 Rockwell NC, Shang L, Martin SS, Lagarias JC. Distinct classes of red/farred photochemistry within the phytochrome superfamily. Proc Natl Acad Sci USA. 2009; 106:6123–7.View ArticlePubMedPubMed CentralGoogle Scholar
 Rüdiger W, Thümmler F, Cmiel E, Schneider S.Chromophore structure of the physioloically active form (Pfr) of phytochrome. Proc Natl Acad Sci USA. 1983; 80:6244–8.View ArticlePubMedPubMed CentralGoogle Scholar
 Li L, Lagarias JC. Phytochrome assembly. J Biol Chem. 1992; 267:19204–10.PubMedGoogle Scholar
 Schäfer E, Schmidt W. Temperature dependence of phytochrome dark reactions. Planta. 1974; 116:257–66.View ArticlePubMedGoogle Scholar
 Hill C, Gärtner W, Towner P, Braslavsky SE, Schaffner K. Expression of phytochrome apoprotein from Avena sativa in Escherichia coli and formation of photoactive chromoproteins by assembly with phycocyanobilin. Eur J Biochem. 1994; 223(1):69–77.View ArticlePubMedGoogle Scholar
 Vierstra RD, Quail PH. Photochemistry of 124 kilodalton Avena phytochrome in vitro. Plant Physiol. 1983; 72:264–7.View ArticlePubMedPubMed CentralGoogle Scholar
 Kelly JM, Lagarias JC. Photochemistry of 124kilodalton Avena phytochrome under constant illumination in vitro. Biochemistry. 1985; 24:6003–10.View ArticleGoogle Scholar
 Medzihradszky M, Bindics J, Ádám É, Viczián A, Klement É, Lorrain S, et al. Phosphorylation of phytochrome B inhibits lightinduced signaling via accelerated dark reversion in Arabidopsis. Plant Cell. 2013; 25:535–44.View ArticlePubMedPubMed CentralGoogle Scholar
 Mancinelli AL. The physiology of phytochrome action In: Kendrick RE, Kronenberg GMH, editors. Photomorphogenesis in plants. Dordrecht: Kluwer Academic Publishers: 1994. p. 211–69.Google Scholar
 Rausenberger J, Hussong A, Kircher S, Kirchenbauer D, Timmer J, Nagy F, et al. An integrative model for phytochrome B mediated photomorphogenesis: from protein dynamics to physiology. PLoS ONE. 2010; 5(5):e10721.View ArticlePubMedPubMed CentralGoogle Scholar
 Johns HE. Photochemical reactions in nucleic acids. Methods in Enzymology. 1969; 16:253–316.View ArticleGoogle Scholar
 van Stokkum IHM, Larsen DS, van Grondelle R. Global and target analysis of timeresolved spectra. Biochimica et Biophysica Acta. 2004; 1657:82–104.View ArticlePubMedGoogle Scholar
 Kim PW, Rockwell NC, Freer LH, Chang CW, Martin SS, Lagarias JC, et al. Unraveling the primary isomerization dynamics in cyanobacterial phytochrome Cph1 with multipulse manipulations. J Phys Chem. 2013; 4:2605–9.Google Scholar
 Kim PW, Rockwell NC, Martin SS, Lagarias JC, Larsen DS. Dynamic inhomogeneity in the photodynamics of cyanobacterial phytochrome Cph1. Biochemistry. 2014; 53:2818–26.View ArticlePubMedPubMed CentralGoogle Scholar
 van Thor JJ, Ronayne KL, Towrie M. Formation of the early photoproduct LumiR of cyanobacterial phytochrome Cph1 observed by ultrafast midinfrared spectroscopy. J Am Chem Soc. 2007; 129:126–32.View ArticlePubMedGoogle Scholar
 Dasgupta J, Frontiera RR, Taylor KC, Lagarias JC, Mathies RA. Ultrafast excitedstate isomerization in phytochrome revealed by femtosecond stimulated Raman spectroscopy. Proc Natl Acad Sci USA. 2009; 106:1784–9.View ArticlePubMedPubMed CentralGoogle Scholar
 Sineshchekov VA. Photobiophysics and photobiochemistry of the heterogenous phytochrome system. Biochimica et Biophysica Acta. 1995; 1228:125–64.View ArticleGoogle Scholar
 Jaubert M, Lavergne J, Fardoux J, Hannibal L, Vuillet L, Adriano JM, et al. A singular bacteriophytochrome acquired by lateral gene transfer. J Biol Chem. 2007; 282(10):7320–8.View ArticlePubMedGoogle Scholar
 Giraud E, Lavergne J, Verméglio A. Characterization of bacteriophytochromes from photosynthetic bacteria: histidine kinase signaling triggered by light and redox sensing. Methods Enzymol. 2010; 471:135–59.View ArticlePubMedGoogle Scholar
 Burgie ES, Vierstra RD. Phytochromes: An atomic perspective on photoactivation and signaling. Plant Cell. 2014; 26:4568–83. tpc.114.131623.View ArticlePubMedPubMed CentralGoogle Scholar
 Gottlieb SM, Kim PW, Chang CW, Hanke SJ, Hayer RJ, Rockwell NC, et al. Conservation and diversity in the primary forward photodynamics of red/green cyanobacteriochromes. Biochemistry. 2015; 54:1028–42.View ArticlePubMedGoogle Scholar
 Müller K, Engesser R, Metzger S, Schulz S, Kämpf MM, Busacker M, et al. A red/farred lightresponsive bistable toggle switch to control gene expression in mammalian cells. Nucleic Acids Res. 2013; 41(7):e77.View ArticlePubMedPubMed CentralGoogle Scholar
 Beyer HM, Juillot S, Herbst K, Samodelov SL, Müller K, Schamel WW, et al. Red lightregulated reversible nuclear localization of proteins in mammalian cells and zebrafish: In press; 2015.Google Scholar
 Klose C, Venezia F, Hussong A, Kircher S, Schäfer E, Fleck C. Systematic analysis of how phytochrome B dimerization determines its specificity. Nat Plants. 2015; 1:15090.View ArticlePubMedGoogle Scholar
 Pfeiffer A, Nagel MK, Popp C, Wüst F, Bindics J, Viczián A, et al. Interaction with plant transcription factors can mediate nuclear import of phytochrome B. Proc Natl Acad Sci USA. 2012; 109(15):5892–7.View ArticlePubMedPubMed CentralGoogle Scholar
 Tepperman JM, Hudson ME, Khanna R, Zhu T, Chang SH, Wang X, et al. Expression profiling of phyB mutant demonstrates substantial contribution of other phytochromes to redlightregulated gene expression during seedling deetiolation. Plant J. 2004; 38:725–39.View ArticlePubMedGoogle Scholar
 AlSady B, Kikis EA, Monte E, Quail PH. Mechanistic duality of transcription factor function in phytochrome signaling. Proc Natl Acad Sci USA. 2008; 105:2232–7.View ArticlePubMedPubMed CentralGoogle Scholar
 Chen M, Schwab R, Chory J. Characterization of the requirements for localization of phytochrome B to nuclear bodies. Proc Natl Acad Sci USA. 2003; 100:14493–8.View ArticlePubMedPubMed CentralGoogle Scholar
 Park E, Park J, Kim J, Nagatani A, Lagarias JC, Choi G, Phytochrome B. inhibits binding of phytochromeinteracting factors to their target promoters. Plant J. 2012; 72:537–46.View ArticlePubMedPubMed CentralGoogle Scholar
 Franklin KA, Quail PH. Phytochrome functions in Arabidopsis development. J Exp Bot. 2010; 61(1):11–24.View ArticlePubMedGoogle Scholar
 Burgie ES, Bussell AN, Walker JM, Dubiel K, Vierstra RD. Crystal structure of the photosensing module from a red/farred lightabsorbing plant phytochrome. Proc Natl Acad Sci USA. 2014; 111(28):10179–84.View ArticlePubMedPubMed CentralGoogle Scholar
 Brockmann J, Rieble S, KazarinovaFukshansky N, Seyfried M, Schäfer E. Phytochrome behaves as a dimer in vivo. Plant Cell Environ. 1987; 10:105–11.Google Scholar
 Wagner D, Quail PH. Mutational analysis of phytochrome B identifies a small COOHterminaldomain region critical for regulatory activity. Proc Natl Acad Sci USA. 1995; 92:8596–600.View ArticlePubMedPubMed CentralGoogle Scholar
 Liu P, Sharrock RA. Directed dimerization: an in vivo expression system for functional studies of type II phytochromes. Plant J. 2013; 75:915–26.View ArticlePubMedGoogle Scholar
 Takala H, Björling A, Linna M, Westenhoff S, Ihalainen JA. Lightinduced changes in the dimerization interface of bacteriophytochromes. J Biol Chem. 2015; 290:16383–92.View ArticlePubMedPubMed CentralGoogle Scholar
 Johansson H, Jones HJ, Foreman J, Hemsted JR, Stewart K, Grima R, et al. Arabidopsis cell expansion is controlled by a photothermal switch. Nat Commun. 2014; 5:4848.View ArticlePubMedPubMed CentralGoogle Scholar
 Mitsui H, Maeda T, Yamaguchi C, Tsuji Y, Watari R, Kobo Y, et al. Overexpression in yeast, photocycle, and in vitro structural change of an avian putative magnetoreceptor cryptochrome4. Biochemistry. 2015; 54:1908–17.View ArticlePubMedGoogle Scholar
 Raue A, Kreutz C, Maiwald T, Bachmann J, Schilling M, Klingmüller U, et al. Structural and practical identifiability analysis of partially observed dynamical models by exploiting the profile likelihood. Bioinformatics. 2009; 25:1923–9.View ArticlePubMedGoogle Scholar
 Hennig L, Schäfer E. Both subunits of the dimeric plant photoreceptor phytochrome require chromophore for stability of the farred lightabsorbing form. J Biol Chem. 2001; 276:7913–8.View ArticlePubMedGoogle Scholar
 Gärtner W, Hill C, Worm K, Braslavsky SE, Schaffner K. Influence of expression system on chromophore binding and preservation of spectral properties in recombinant phytochrome A. Eur J Biochem. 1996; 236:978–83.View ArticlePubMedGoogle Scholar
 Beyer HM, Naumann S, Weber W, Radziwill G. Optogenetic control of signaling in mammalian cells. Biotechnol J. 2014; 10:273–8.View ArticlePubMedGoogle Scholar
 Findlay KM, Jenkins GI. Regulation of UVR8 photoreceptor dimer/monomer photoequilibrium in Arabidopsis plants grown under photoperiodic conditions. Plant Cell Environ. 2016. [doi:10.1111/pce.12724.]
 Müller K, Siegel D, Jahnke FR, Gerrer K, Wend S, Decker EL, et al. A red lightcontrolled synthetic gene expression switch for plant systems. Mol BioSyst. 2014; 10(7):1679–88.View ArticlePubMedGoogle Scholar