Grana are stacked thylakoid membrane structures in land plants that contain photosystem II (PSII) and light-harvesting complex II proteins (LHCIIs). We isolated two Arabidopsis mutants, reduced induction of non-photochemical quenching (riq1 and riq2), in which stacking of grana was enhanced. The curvature thylakoid 1a curt1a) mutant was previously shown to lack grana structure. In riq1 curt1a, the grana were enlarged with more stacking, and in riq2 curt1a, the thylakoids were abnormally stacked and aggregated. Despite having different phenotypes in thylakoid structure, riq1, riq2, and curt1a showed a similar defect in the level of non-photochemical quenching of chlorophyll fluorescence (NPQ). In riq curt1a double mutants, NPQ induction was more severely affected than in either single mutant. In riq mutants, state transitions were inhibited and the PSII antennae were smaller than in wild-type plants. The riq defects did not affect NPQ induction in the chlorophyll b-less mutant. RIQ1 and RIQ2 are paralogous and encode uncharacterized grana thylakoid proteins, but despite the high level of identity of the sequence, the functions of RIQ1 and RIQ2 were not redundant. RIQ1 is required for RIQ2 accumulation, and the wild-type level of RIQ2 did not complement the NPQ and thylakoid phenotypes in riq1. We propose that RIQ proteins link the grana structure and organization of LHCIIs.
In the chloroplasts of land plants, the photosystem II (PSII) supercomplex consists of two PSII cores and associated light-harvesting complex II proteins (LHCIIs). This supercomplex is enriched in highly stacked thylakoid regions of appressed membranes known as grana (Dekker and Boekema, 2005). Light energy absorbed by LHCIIs is transferred to a special pair of chlorophyll molecules in the PSII reaction center to drive electron transport though downstream protein complexes in the thylakoid membrane. Under very high light intensities, excess absorbed light energy is dissipated as heat. This process can be monitored as the qE (energy-dependent quenching) component of the non-photochemical quenching of chlorophyll fluorescence (NPQ). qE is induced by lumenal acidification, which depends on photosynthetic electron transport. A thylakoid-localized protein, PsbS, senses low lumen pH to induce qE (Li et al., 2000). Lumen acidification also activates violaxantin de-epoxidase, which catalyzes conversion of violaxanthin to zeaxanthin via antheraxanthin (xanthophyll cycle) (Niyogi et al., 1998).
As a result of these events, some LHCIIs detach from PSII, leading to aggregation of the disassociated LHCIIs around the PSII-LHCII supercomplex. This dynamic reorganization results in qE induction at both aggregated LHCIIs and still bound LHCIIs, which are referred as quenching sites Q1 and Q2, respectively (Holzwarth et al., 2009; Johnson et al., 2011, Minagawa, 2013). Besides qE, several additional components of NPQ have been proposed. qZ is a zeaxanthin-dependent component and is formed and relaxed with 10-15 min lifetime (Nilkens et al., 2010). qT is induced as a result of state transitions and qI is induced following PSII photoinhibition. State transitions balance the excitation pressure between PSII and PSI via relocation of LHCIIs (Kouřil et al., 2012). qI is associated with damage to the D1 protein, leading to photoinhibition and reduced photosynthetic capacity (Aro et al., 1993). The impaired PSII reaction center is able to quench fluorescence directly (Horton et al., 1996), but the exact mechanism is still unknown. Although qZ, qT and qI components are slowly induced and relaxed (several minutes to hours), the kinetics of qE induction and relaxation are so fast (several seconds to minutes) that plants can respond to rapid fluctuations in light conditions (Li et al., 2009; Jahns and Holzwarth, 2012). Recently, a qM component has been discussed to reflect the chloroplast movement in the light (Cazzaniga et al., 2013).
Grana are thylakoid structures that develop primarily in land plants. It is thought that LHCIIs play a role in grana biogenesis because mutants or transgenic plants in which the accumulation or organization of LHCIIs is altered show disturbances of grana structure (Labate et al., 2004; Kim et al., 2009; Cui et al., 2011; Pietrzykowska et al., 2014). The charged stroma-facing surface of LHCIIs interacts via the same surface on other molecules present in a different membrane, likely resulting in cohesion of the two membranes (Standfuss et al., 2005). Kinase mutants defective in phosphorylation of LHCIIs and PSII subunits exhibit reduced stacking of grana (Fristedt et al., 2009), demonstrating that LHCIIs phosphorylation is also critical for the stacking of grana. CURVATURE THYLAKOID 1 (CURT1) has recently been identified as another protein required for grana stacking, and acts independently of the mechanism mediated by LHCIIs. CURT1 proteins are localized to the grana margin and probably determine the diameter of the grana disc by causing membrane curvature. The Arabidopsis curt1 mutants have pseudograna, which consist of reduced numbers of thylakoid discs having enlarged diameters and greatly reduced marginal regions, but the accumulation of PSII subunits or LHCIIs is unaltered in these mutants (Armbruster et al., 2013; Pribil and Leister, 2014). Details of the molecular mechanism of grana formation, the thylakoid architecture, and their functional link with photosynthesis and its regulatory processes are not yet fully understood.
Here, we identified two Arabidopsis mutants exhibiting low levels of NPQ induction and thus them reduced induction of non-photochemical quenching 1, 2 (riq1, riq2). We initially focused on these genes in a proteomics approach to clarify the components of PSI cyclic electron transport that depend on PROTON GRADIENT REGULATION 5 (PGR5), but we later recognized that they were not related to PGR5. Actually, in riq mutants, the grana were stacked more highly than those in the wild type (WT). Our analyses revealed that RIQ proteins contribute to NPQ and grana stacking in ways different from those of CURT1 functions. This study provides new genetic evidence for the functional link between grana structure and organization of LHCIIs, which are shown to be related to qE induction and state transitions.
Arabidopsis riq Mutants Cannot Sustain NPQ under Moderate Light
Arabidopsis RIQ1 and RIQ2 contain 158 and 198 amino acid residues, respectively, and share a conserved domain of unknown function (DUF) 1118, which includes two putative transmembrane (TM) domains. Their N-terminal regions were predicted to be transit peptides (TPs) targeted toward chloroplasts (Figure 1A). No other genes in the Arabidopsis genome encode proteins similar to RIQ. RIQ genes are conserved in land plants and some green eukaryotic algae, including Chlorella variabilis. However, only the RIQ1 ortholog was identified in Micromonas pusilla RCC299, one of the dominant photosynthetic eukaryotes in marine ecosystems (Figure 1B), and Chlamydomonas reinhardtii and Volvox carteri do not contain RIQ-related genes, implying that some green algal lineages have lost them.
We investigated the phenotypes of the Arabidopsis T-DNA insertional mutants of riq1 and riq2. In riq1, T-DNA was inserted into the first exon of RIQ1, whereas in riq2, it was inserted into the 5′UTR of RIQ2 (Figure 1C). RT-PCR analyses did not detect riq transcripts in either mutant, suggesting that both alleles were null (Figure 1D). In riq mutants, including the double mutant riq1 riq2, there was no mutant phenotype for growth (Figure 1E), chlorophyll content or chlorophyll a/b ratio (Supplemental Table 1) under the growth conditions used in this study. The maximum photochemical efficiency of PSII (Fv/Fm), which is often used to estimate the PSII photoinhibition, was the same (0.77) among the genotypes (n = 3).
However, a mutant phenotype was detected in both riq1 and riq2 in an analysis of chlorophyll fluorescence. Steady-state NPQ levels were mildly but significantly lower in riq1 and riq2 than in WT, although they were higher than those in npq4, which is defective in PsbS (Li et al., 2000) (Figure 2A). To characterize this phenotype in more detail, induction of NPQ was measured at various white actinic light (AL) intensities, where photosynthesis was activated. At 1,900 μmol photons m–2 s–1, NPQ induction was nearly unaffected in riq1 and riq2, reaching 2.0 by 5 min in the light (Figure 2B). In WT plants exposed to AL at 250 μmol photons m–2 s–1, NPQ peaked at 1.4 within 5 min. By contrast, in riq1 and riq2, the maximum level (1.1) of NPQ was induced after a 2-min exposure to AL and was followed by a gradual reduction in the light to 1.0 after 5 min (Figure 2C). Under a lower AL intensity of 100 μmol photons m–2 s–1, NPQ induction was also transient in WT, perhaps because of the relaxation of ΔpH by ATP synthase. However, the relaxation of NPQ in the light was faster in riq1 and riq2 than in WT (Figure 2D). To confirm that the NPQ phenotypes were due to the riq defects, RIQ genes were introduced into the mutants under the control of their own promoters. This transformation induced the recovery of NPQ levels at 250 μmol photons m–2 s–1 (Supplemental Figure 1). A similar NPQ phenotype was observed under red AL (Supplemental Figure 2), suggesting that the blue-light-dependent chloroplast movement monitored as qM (Cazzaniga et al., 2013), was not affected.
Despite the high level of sequence identity, both riq1 and riq2 exhibited a similar NPQ reduction (Figure 2), suggesting that the RIQ1 and RIQ2 functions were not redundant. However, it is still possible that their functions partly overlap. To test this possibility, NPQ induction was analyzed in riq1 riq2. The phenotypes of riq1 riq2 were identical to those of the single mutants in terms of both light-intensity dependence (Figure 2A) and the time courses of induction and relaxation (Figures 2C and 2D), suggesting that RIQ1 and RIQ2 played non-redundant roles in contributing to NPQ.
RIQ1 and RIQ2 are Localized in the Grana Core
To localize RIQ proteins, the specific antibodies recognizing the predicted mature forms of each protein were prepared. Both RIQ1 and RIQ2 were found in the membrane fraction isolated from WT, but were absent in the corresponding single mutant (Figure 3A). Subfractionation of chloroplasts into the chloroplast envelope, thylakoid, and stromal fractions revealed the specific localization of the RIQ proteins to the thylakoid membrane (Figure 3B), consistent with information in the Plant Proteome database (PPDB) (Sun et al., 2008). The anti-RIQ1 antibody detected two signals that were absent in riq1 and riq1 riq2 (Figure 3A). The upper, faint signal was unlikely to be due to non-specific reaction of the antibody, because the same signal was detected using independently prepared antibody raised against a RIQ1 oligopeptide (EEFGVLSAATNPET) (Supplemental Figure 3). Although the RIQ1 level was unaffected in riq2, in riq1 the RIQ2 level was reduced to approximately 25% of that in WT (Figure 3A), suggesting that RIQ1 is required for RIQ2 accumulation. Protein accumulation was restored to nearly the WT level by introduction of RIQ1 or RIQ2 into the corresponding mutant (Figure 3A).
The protein compositions of the grana and the stromal lamella differ strikingly in chloroplasts of land plants (Dekker and Boekema, 2005). The grana margin is the edge region of the grana thylakoid and is rich in CURT1 family proteins (Armbruster et al., 2013; Puthiyaveetil et al., 2014). To clarify the localization of RIQ proteins in the thylakoid membrane, these membranes were subfractionated into the grana core, grana margin, and stromal lamella. Both RIQ1 and RIQ2 were detected mainly in the grana core, as was PsbO (a PSII subunit). A trace amount of RIQ2 was also detected in the grana margin and stromal lamella fractions (Figure 3C).
Both RIQ1 and RIQ2 are Required for Full Induction of NPQ
It is possible that RIQ1 is required just to stabilize RIQ2 (Figure 3A) and that the reduction of NPQ observed in riq1 (Figure 2) occurs via the reduced level of RIQ2. To test this possibility, we expressed RIQ2 under the control of the CaMV 35S promoter in riq1. If this possibility were true, then the riq1 NPQ phenotype would be complemented by the accumulation of RIQ2 to WT level even in the absence of RIQ1. We obtained four transgenic lines, two of which accumulated WT level of RIQ2 (Figure 4A). All the lines exhibited NPQ levels similar to those of riq1 (Figure 4B). This result suggests that both RIQ1 and RIQ2 are required for full induction of NPQ.
RIQ1 is required for RIQ2 accumulation, and both proteins are required for the normal NPQ induction (Figures 3A and 4). To test the possibility that RIQ1 and RIQ2 interact in vivo, co-immunoprecipitation was performed using a polyclonal antibody against RIQ2. As a negative control, riq2, which accumulated the WT level of RIQ1, was used. In WT, only the RIQ1 signal with slower mobility was detected in the immunoprecipitate (Figure 3D), suggesting that at leas, the major form of RIQ1 did not interact with RIQ2.
riq Mutants are Not Defective in the Known qE Machinery
Production of qE is induced by lumenal acidification and is dependent on the trans-thylakoid proton gradient (ΔpH) (Horton et al., 1996; Li et al., 2009). The riq NPQ phenotype (Figure 2) may be due to a defect in the qE machinery that senses lumenal acidification and induces NPQ. Alternatively, the defect may affect the mechanism of ΔpH formation or maintenance. The electron transport rate (ETR) reflects the relative rate of electron transport through PSII and was not significantly affected in any genotypes (Supplemental Figure 4A). Another parameter, 1-qL, which represents the state of reduction of the plastoquinone pool, was slightly but not significantly higher in riq mutants than in WT (Supplemental Figure 4B).
Chlorophyll fluorescence analysis implied that riq mutants were defective in the qE machinery rather than in the formation or maintenance of ΔpH. To test this more directly, we measured the amplitude of the total light–dark difference in the electrochromic shift signal (ECSt), which reflects the total size of the proton motive force (pmf) formed in the light. The pmf was saturated at 249 μmol photons m–2 s–1 in WT. As reported previously (Wang et al., 2015), the pmf was significantly lower in pgr5 than in WT under AL of 249 and 1707 μmol photons m–2 s–1. However, the size of the pmf was not affected in riq mutants (Supplemental Figure 5), suggesting that the NPQ phenotype is not due to a reduced ΔpH.
Even with sufficient ΔpH, a defect in proteins associated with the qE machinery, such as PsbS, CP26, and CP29, impairs qE induction (Li et al., 2000; de Bianchi et al., 2008; de Bianchi et al., 2011). Accumulation of these proteins was not reduced in riq mutants, and subunit levels of the other thylakoid protein complexes were also unaffected (Supplemental Figure 6). To assess the possibility that the xanthophyll cycle involved in qE (Niyogi et al., 1998) is affected in riq mutants, we analyzed the carotenoid conversion induced by high light using high performance liquid chromatography (HPLC). In WT, the de-epoxidation state of xanthophyll carotenoids (DEPS) was low in the dark but was elevated to approximately 30 in the light (Supplemental Figure 7). Under high light, DEPS was not affected in riq1 or riq2, indicating that the NPQ phenotype in riq mutants cannot be explained by reduced activity of the xanthophyll cycle (Supplemental Figure 7). Taken together, these findings show that the NPQ phenotype in riq mutants is not caused by defects in pmf regulation, accumulation of NPQ-related proteins, or xanthophyll cycle activity.
LHCIIs Are Required for RIQ-Related NPQ
In Arabidopsis, the main component of NPQ is qE, reflecting the size of thermal dissipation from LHCIIs (Johnson et al., 2011). In riq mutants, accumulation of LHCIIs (Lhcb1 to 6) was not affected (Supplemental Figure 6), consistent with them having a normal chlorophyll a/b ratio (Supplemental Table 1). To determine whether LHCIIs were involved in RIQ-related NPQ in vivo, we crossed riq mutants with chlorina1-1 (ch1-1), which was defective in chlorophyll b synthesis (Murray and Kohorn, 1991; Espineda et al., 1999). Because of the absence of chlorophyll b, ch1-1 does not accumulate any functional LHCIIs except for Lhcb5, resulting in a drastic reduction in qE (Havaux et al., 2007; Takabayashi et al., 2011). The small amount of NPQ remaining in ch1-1 is considered to be related to Lhcb5 and the PSII core (Havaux et al., 2007). Accumulation of RIQ proteins was not affected in ch1-1 (Figure 5A), suggesting that RIQ proteins accumulate independently of LHCIIs. At 250 μmol photons m–2 s–1, in which the reduced size of NPQ was evident in riq mutants, a similar level of NPQ was induced among ch1-1, riq1 ch1-1 and riq2 ch1-1 (two independent lines) (Figures 5B and 5C). The riq defects did not affect the NPQ induction remaining in ch1-1. This indicates that RIQ proteins are required for efficient thermal dissipation from LHCIIs, although the stability of RIQ proteins is independent from LHCII accumulation.
Organization of LHCIIs is Affected in riq Mutants
Induction of qE is related to partial disassociation of LHCIIs, which forms two quenching sites (Holzwarth et al., 2009; Johnson et al., 2011). To analyze the impact of the riq defects on the PSII-LHCIIs structure in vivo, the antenna size per PSII reaction center was measured by using a flash fluorescence induction method. In this assay, a saturating flash was applied to a solution containing isolated thylakoids treated with 3-(3,4-dichlorophenyl)-1,1-dimethylurea (DCMU), an inhibitor to block electron transfer from PSII to the plastoquinone pool, to monitor an increase in chlorophyll fluorescence from LHCIIs at high time resolution. The time required to reach the maximum level is proportional to the size of the antennae connected to the PSII reaction center (Nedbal et al., 1999). riq mutants had slower fluorescence induction kinetics than WT (Figure 6A), suggesting that they had smaller PSII antennae per reaction center. Consistent with the NPQ phenotype, riq1 riq2 had a phenotype similar to that of the single mutants (Figure 6A).
To test the possibility that riq mutants were defective in the reorganization process of LHCIIs, we analyzed chlorophyll fluorescence at 77 K to test their capacity for state transitions. At the temperature, chlorophyll fluorescence is emitted from both photosystems, and its level represents the size of antennae attached to each photosystem. After dark adaptation, leaves were exposed to weak red light to induce state 2, where phosphorylated LHCIIs dissociated from PSII and associated to PSI, and far-red light to induce state 1, where dephosphorylated LHCIIs reassociated to PSII (Minagawa, 2013). This artificial induction of state 1 or state2 was followed by being immediately soaked in liquid nitrogen for fluorescence measurement. In the spectra normalized at the emission peak from PSI (736 nm), WT leaves showed a clear difference in fluorescence between state 1 and state 2 at around 685 to 695 nm, corresponding to the fluorescence emitted from the PSII antennae (Figure 6B). This difference reflects the change in antenna size based on state transitions (Fleischmann et al., 1999). Arabidopsis stn7 was fixed in state 1 (Bellafiore et al., 2005) and showed no fluorescence change between the two light conditions (Figure 6C). Although riq mutants maintained their state transition activity, the amplitudes of the fluorescence changes were smaller than those in WT (Figures 6E to 6G), indicating that state transitions were greatly impaired in riq mutants. Immunodetection using a specific antibody against phosphorylated proteins revealed that the phosphorylation levels of LHCIIs isolated from leaves in which state 2 was induced did not differ among WT and riq mutants, but it did differ in stn7 (Supplemental Figure 8). In state 1, LHCIIs were dephosphorylated in riq mutants, as in WT. This result suggests that the inhibition of state transitions observed in riq mutants was not induced via alteration of phosphorylation or dephosphorylation of PSII subunits or LHCIIs.
Grana Stacking is Enhanced in riq Mutants
The stacking level of the grana thylakoid is closely related to the mobility of proteins localized to the region (Dekker and Boekema, 2005; Pribil and Leister, 2014). We used transmission electron microscopy to compare thylakoid architecture among WT and riq mutants. Consistent with the observed normal plant growth (Figure 1E), the thylakoid structure was not seriously disturbed in riq mutants (Figures 7A to 7D; Supplemental Figure 9). However, the grana were more stacked than those in WT; this observation was supported statistically by an analysis of the number of thylakoid stacks in the grana region (Figure 7E). Consistent with the NPQ phenotype (Figure 2) and the reduction in antenna size (Figure 6A), riq1 riq2 had a phenotype similar to that of the single mutants (Figure 7D; Supplemental Figure 9), suggesting that function of RIQ1 and RIQ2 was also not redundant in thylakoid stacking.
RIQ and CURT1 Proteins are Independently Required to Sustain NPQ Levels RIQ proteins may be associated with the organization of LHCIIs via optimization of the level of grana stacking. A similar link between grana structure and NPQ has been reported in curt1a. Because CURT1 proteins contribute to grana formation by inducing membrane curvature (Armbruster et al., 2013), defects in them result in a thylakoid phenotype—a lack of normal grana stacking—that contrasts with that in riq mutants. Armbruster et al. (2013) also reported a reduction in NPQ in curt1a, but the reason for this phenotype remains unclear. To study the possible link between thylakoid structure and the function of LHCIIs, we included curt1a in our analyses. In this study, we used a new curt1a allele obtained from the Arabidopsis Biological Resource Center (Supplemental Figure 10A). CURT1A transcripts were not detected in the mutant (Supplemental Figure 10B), indicating that curt1a is a knockout allele. Introduction of the CURT1A-HA gene complemented the curt1a mutant phenotypes regarding the protein level and NPQ induction (Supplemental Figures 10C and 10D). Figure 8A shows the time-course of NPQ induction at 250 μmol photons m–2 s–1. As in riq mutants, NPQ was transiently induced in curt1a to the WT level (1.2) within 2 min of exposure to light. However, no further increase in NPQ was induced, resulting in a slight decline in NPQ levels during an additional 3 min in the light. This curt1a phenotype in NPQ was similar to that observed in riq mutants (Figure 8A). To analyze the genetic interaction between riq and curt1a mutants, we created double mutants. In riq1 curt1a and riq2 curt1a, the NPQ level was further decreased compared with that in the single mutants (Figure 8A), suggesting that RIQ and CURT1 affect NPQ in different pathways. Consistently, the curt1a defect did not affect the RIQ levels and the riq defects did not affect the CURT1A level (Supplemental Figure 6).
To investigate the phenotype of NPQ induction in riq and curt1a mutants in more detail, NPQ induction and relaxation were monitored during 5-min light (250 μmol photons m–2 s–1) and 1-min dark cycles. In WT, a high level of NPQ was induced during the first min in the second light period (Figure 8B), whereas riq or curt1a mutants showed a much lower level of NPQ. The same trend was observed in the third light period. This phenotype was more evident in riq1 curt1a and riq2 curt1a than in the single mutants (Figure 8B), but was different using a longer dark interruption (10 min). As was the case with the initial light period (0 to 5 min), a high level of NPQ was transiently induced in all genotypes, including the double mutants, during the first 1 min in the second light period (15 to 20 min) after the long dark period (Figure 8C). In both riq and curt1a mutants, the ability to induce high levels of NPQ (more than 1.5) transiently was recovered during the 10-min dark adaptation.
Synergistic Effects of the riq and curt1a Mutations on Thylakoid Structure
The riq and curt1a mutants exhibited opposite mutant thylakoid structure phenotypes (Figure 7; Armbruster et al., 2013), although the NPQ induction phenotypes were similar (Figure 8A). To study the link between thylakoid structure and the organization of LHCIIs, we used electron microscopy to observe chloroplasts of riq curt1a double mutants. Consistent with previous reports (Armbruster et al., 2013; Pribil et al., 2014), fewer stacked grana structures, with significantly increased diameters, were formed in curt1a (Figures 9A to 9D; Supplemental Figures 11A to 11D). In riq1 curt1a, the grana were similarly elongated but their stacking was enhanced, as in riq single mutants (Figures 9E and 9F; Supplemental Figure 11E). This finding was supported by a quantitative analysis of grana stacking (Figure 9I). Notably, the mutant phenotype for thylakoid structure in riq2 curt1a was synergistic rather than additive. At the distal parts of the lens-shaped structure in riq2 curt1a, the thylakoids were abnormally stacked and aggregated (black arrowheads in Figures 9G and 9H; Supplemental Figure 11F). Probably because of the disturbance of the thylakoid structure, the chloroplast envelope was unusually wavy (white arrowheads in Figures 9G and 9H; Supplemental Figure 11F). Holes were detected in the stroma, reflecting dents in the envelope (asterisks in Figure 9G; Supplemental Figure 11F).
Consistent with the phenotype observed in NPQ induction (Figure 2) and stacking of the grana in the riq single mutant (Figure 7), the analysis of the double mutants with curt1a suggested that RIQ1 and RIQ2 had non-overlapping functions. The function of RIQ1 was not only required to accumulate RIQ2 but also essential to fully induce NPQ (Figures 3A and 4). However, it is still possible that the phenotype observed in the thylakoid structure was caused solely by the reduced RIQ2 level. If this were the case, the different phenotypes observed in riq1 curt1a and the riq2 curt1a would be explained by the different levels of RIQ2 independently of RIQ1. To test this possibility, the thylakoid architecture of the RIQ2-complemented riq1 plants, which accumulated WT level of RIQ2 in the absence of RIQ1, was analyzed (Figure 4A). Two transgenic lines exhibited the same level of grana stacking as did the non-transgenic riq1 mutant (Figure 10). Therefore, RIQ1 is also required for the optimization of grana stacking, independently of its requirement for the accumulation of RIQ2, as was observed in NPQ induction (Figure 4B).