Preparation of the isolated GAF1-only and bidomain GAF1–PHY constructs of All2699 protein are described elsewhere (Xu Q.-Z. et al., 2019; Slavov et al., 2020). For solid-state NMR spectroscopy, two All2699 holoproteins in vitro assembled with uniformly ¹³C- and ¹⁵N-labeled PCB [u-[¹³C,¹⁵N]-PCB-All2699(GAF1) and u-[¹³C,¹⁵N]-PCB-All2699(GAF1–PHY)] as lyophilized powder were used (Xu Q.-Z. et al., 2019). No additional illumination was applied during the NMR acquisition. For ultrafast transient absorption spectroscopy, the in vivo assembled holoproteins were prepared in a final buffer containing 50 mM Tris, 150 mM NaCl, and 5% glycerol at pH 8.0.

All solid-state NMR experiments were acquired on a wide-bore Bruker AVANCE-III 400 MHz spectrometer equipped with a 4 mm double-resonance magic-angle spinning (MAS) probe (Rheinstetten, Germany). All spectra were externally referenced to the COO⁻ signal of solid _L-Tyrosine·HCl at 172.1 ppm. The data was preprocessed with Bruker Topspin 4.0.1 and further analyzed with MestReNova 14.1.0 (Mestrelab Research, Santiago de Compostella, Spain).

The time-resolved transient absorption experiments were performed using a home-built pump-probe setup which was described previously elsewhere (Slavov et al., 2015a). In the set-up a Ti:sapphire amplifier system (Clark, MXR-CPA-iSeries) was used to generate the fundamental laser pulses (775 nm, 130 fs, 1 kHz) which are used to generate the pump and probe pulses. The pump pulses are generated using a home-built two stage NOPA (noncollinear optical parametric amplifier) with a prism compressor in between the two NOPA stages for pulse compression. For probing, white-light continuum pulses were generated by focusing the laser fundamental into a CaF₂-crystal (5 mm). A time resolution of ∼100 fs for the experiment was estimated from the pump-probe cross correlation. The experiments were conducted under magic angle conditions (54.7° pump-probe polarization difference) with excitation pulses at 710 nm and 70 nJ/pulse and probe energy 10–20 pJ/nm or lower. Samples were placed in a 1 mm path-length fused-silica cuvette and continuously translated at high speed in two orthogonal directions within a plane perpendicular to the probe beam. A high-power 625 nm LED (Thorlabs, 1 W) provided continuous illumination to maintain the sample in the P_fr state and prevent photoproduct accumulation. The analysis of the transient absorption data was performed using OPTIMUS (www.optimus.optimusfit.org) (Slavov et al., 2015b).

In most Group I and II phytochromes, the ultrafast P_fr → P_r photodynamics occurs on a similar timescale and is significantly faster than the P_r → P_fr reaction. After rapid departure from the Franck-Condon region (∼100 fs), the excited state relaxes within ∼500–700 fs to form the primary photointermediate Lumi-F (Bischoff et al., 2001; Heyne et al., 2002; Müller et al., 2008; Schumann et al., 2008; Kim et al., 2014b; Yang et al., 2016; Stensitzki et al., 2017), which subsequently generates the blue-shifted Meta-F intermediate (Müller et al., 2008; Kim et al., 2014b; Stensitzki et al., 2017). The P_fr ultrafast data (see SI for experimental details) of the GAF1-PHY construct is dominated by four transient absorption signals (Figure 2a). The broad positive signal from 450–650 nm corresponds to excited state absorption (ESA), while the negative feature at 700–750 nm arises from P_fr ground state bleach (GSB) and stimulated emission (SE). After ∼0.5 ps, a new positive signal emerges at 740 nm (faint yellow feature, Figure 2a), assigned to the primary photointermediate. By 5 ps, this intermediate decays, giving rise to a second positive signal at 550–600 nm, which persists beyond the experimental time window. We analyzed the transient absorption data using lifetime density analysis to resolve the details of underlying kinetics (see SI and ref (Slavov et al., 2015b). At ∼100 fs, the combination of negative (700–750 nm) and positive (650–700 nm) amplitude lifetime distributions indicate the departure of the wavepacket from the Franck-Condon region (Figure 2b). This is followed by a spectrally broad positive distribution from 450 to 600 nm and a negative distribution above 650 nm, reflecting the 300-fs decay of ESA and GSB/SE. Concurrently, a strong negative distribution above 730 nm marks the emergence of the primary hot ground-state intermediate (GSI), Lumi-F. This intermediate undergoes a spectral blue shift with a lifetime of 5–10 ps, as indicated by the distribution at 650–710 nm and the positive distribution at 740 nm. The blue shift continues on longer timescales, as illustrated by the positive distribution at 675–740 nm and the negative distribution at 525–650 nm, resulting in a final set of lifetime distributions. These correspond to the long-lived GSI, Meta-F, whose lifetime exceeds the detection window of our experiment. The overall kinetic behavior is highly similar to our previous observations in SynCph2—a knotless phytochrome also composed of GAF and PHY domains—and appears predominantly homogeneous (Fischer et al., 2021).

In the ultrafast dynamics of the GAF1-only construct, only three transient absorption features are observed, as the positive signal above 730 nm seen the GAF1–PHY data is not clearly detectable (Figure 2c). Notably, the decay of the ESA below 600 nm is markedly prolonged into the picosecond range and includes an additional, well-defined band at 425 nm that is absent in the GAF1–PHY data. The lifetime density analysis (Figure 2d) reveals a complex pattern of lifetime distributions for both ESA and GSB signals. While the ∼100 fs Franck-Condon relaxation is preserved, the relatively narrow lifetime distribution reflecting the decay of the ESA, and the GSB/SE signals in the GAF1–PHY data is replaced by an extended and very structured distribution spanning from ∼500 fs to ∼20 ps (compare Figures 2b,c). These positive- and negative distributions can be assigned to a multiphasic P_fr excited-state decay associated with elevated recovery of the ground state and reduced formation of the primary GSI. The extension of the strong negative distribution to longer wavelengths suggests some formation of the red-shifted primary GSI, similar to that observed in the GAF1–PHY data (Figure 2a) and in SynCph2 (Fischer et al., 2021). The later phases of the excited-state decay, represented by the positive distributions in the 2–20 ps lifetime range, coincide with the early blue spectral shift evolution of the primary photoproduct. This is evident by the dip in the lifetime distribution at 3–10 ps (>730 nm). This dip is the result of the overlap between the negative distribution (SE decay) and the positive distribution (photoproduct blue shift). This data suggests that the slowest components of excited-state decay persist beyond the early evolution of the primary photoproduct. As a result, it is difficult to determine their exact contribution to product formation. Nevertheless, both excited-state decay and photoproduct formation appear substantially stretched in time and heterogenous compared to the P_fr dynamics of the GAF1–PHY construct and SynCph2. A detailed kinetic model will require structural information to account for this heterogeneity.

Guided by the qualitative insights from the LDMs (Figures 2b,d), we constructed kinetic models for the GAF1-only and GAF1–PHY P_fr states, enabling extraction of rate constants and spectral signatures of the underlying species (Figure 3).

The LDM of the GAF1–PHY P_fr dynamics revealed a single excited-state decay component, consistent with a homogenous chromophore population. This justified the construction of a global analysis model based on a single ground-state configuration (Fischer et al., 2021). To capture the ultrafast relaxation, the model included a Franck–Condon (FC) state, a short-lived relaxed excited state minimum, and three consecutive ground-state intermediates (GSIs) (Figure 3a). Constraining the model by the experimentally determined quantum yield of P_fr photoconversion (16%, see Supplementary Figure S1 and SI for details on quantum yield determination) reduced parameter space and yielded physically meaningful rate constants and species-associated difference spectra (SADS) (Figures 3a,b).

Strikingly, this homogenous model closely mirrors that of the related knotless phytochrome Cph2, with nearly identical excited-state lifetimes (0.34 ps vs. 0.33 ps) and similar intermediate spectral features (Fischer et al., 2021). Unlike Cph2, however, we could not resolve a distinct Lumi-F_hot state, likely due to the shorter lifetime of Lumi-F (4.6 ps). We assign GSI-1 to Lumi-F based on its red-shifted absorption (>730 nm) (Figure 3b). Lumi-F proceeds via a single intermediate (GSI-2, 640 nm) to reach Meta-F (GSI-3, 630 nm), suggesting a more direct ground-state evolution in GAF1–PHY.

In contrast, the GAF1-only construct exhibited strikingly heterogeneous P_fr ultrafast dynamics. The transient absorption data and the corresponding LDMs (Figures 2c,d) revealed multiple decay components with substantially different timescales, ranging from sub-ps to tens of picoseconds. These differences could not be reconciled with a single homogeneous population and instead pointed toward the presence of multiple distinct ground-state configurations, giving rise to spectro-kinetically distinguishable excited-state behaviors. To account for this, we constructed a kinetic model incorporating three parallel P_fr ground state populations (P_fr-I, P_fr-II, and P_fr-III), each contributing nearly equally to the observed transient kinetics (Figure 3d). It is worth noting that the NMR analysis (vide infra) also independently revealed three subpopulations with an approximately 1:1:1 distribution. To ensure physically reasonable results and account for the presence of a shunt pathway, we limited the total photoproduct quantum yield to 10% (Trunk et al., 2025) (see SI for details). The Franck-Condon spectra (FC-I, FC-II, and FC-III) are essentially identical and the excited states (ES-I, ES-II, and ES-III) of the corresponding P_fr populations share similar ESA features but differ markedly in decay kinetics (0.55 ps, 3.5 ps, and 16.4 ps, respectively) and spectral bleach patterns (Figures 3d,e), suggesting structurally encoded relaxation differences. Consequently, we developed a model in which all ESs funnel into a shared primary intermediate (GSI-1, analogous to GAF1–PHY and Cph2 Lumi-F), which then either relaxes back to the initial P_fr pool or proceeds forward along a common pathway towards Meta-F (Figure 3d).

We then attempted to determine the photoconversion efficiencies of each subpopulation. For this purpose, we tested a fully heterogenous model with completely independent product formation pathways for each subpopulation (Supplementary Figure S5). While the model fits the data, it is severely overparametrized, which hinders the reliable recovery of some of the kinetic rates and the branching efficiencies. To address this limitation, we implemented a simplified model in which each ES either relaxes back to its original P_fr state or proceeds towards a shared GSI-1 (Supplementary Figure S4). The purpose of this model is not to provide mechanistic insight but solely to extract the branching efficiencies for each subpopulation while avoiding harsh increase in model complexity and the resulting overparameterization. Remarkably, P_fr-I, accounting for just 32% of the initial P_fr population, produces ∼95% of the final photoproduct, whereas P_fr-II contributes only ∼5% and P_fr-III is virtually non-productive. The ∼30% branching toward photoproduct formation in P_fr-I (Figure 3d) does not imply ‘enhanced’ or ‘penalized’ photochemistry overall, but rather reflects the intrinsic productivity of one subpopulation within a heterogeneous ensemble (overall population-weighted quantum yield of ∼10%). Interestingly, the decay dynamics of P_fr-I resemble those of the GAF1–PHY construct, suggesting that this subpopulation may have structural features stabilized by the PHY domain. Notably, the productive population shows a strongly red-shifted SE, consistent with a potential transient deprotonation of ring C or D (Yang et al., 2022), although explaining why P_fr-I is comparatively more productive will require additional structural and theoretical data beyond the scope of this work. Together, these results indicate that, in the absence of the PHY domain, the chromophore binding pocket permits conformational diversity, leading to heterogeneous and overall inefficient photoconversion, whereas the GAF1–PHY architecture constrains the intrinsic bilin flexibility to enforce structural and kinetic uniformity and ensure a defined, well-regulated photosensory output.

To reveal P_fr structural features in the two All2699 constructs, we applied solid-state NMR spectroscopy using uniformly ¹³C- and ¹⁵N-labeled PCB chromophore (u-[¹³C,¹⁵N]-PCB). Two-dimensional (2D) homonuclear ¹³C–¹³C dipolar correlation (DARR) (Figures 4a,b, see SI for sample preparation and experimental details) spectra were recorded with an optimized 50 ms mixing time to enable detection of both nearest-neighbor and medium-range correlations (∼3.3–3.7 Å). Due to this specific labeling pattern, the DARR data exclusively probes the intramolecular structure and heterogeneity of the chromophore itself in different protein environments. The detection range and assignment strategy are based on earlier studies on these proteins in their corresponding P_r states as frozen solution (Xu Q.-Z. et al., 2019) and lyophilized powder (Kim et al., 2020).