Structure, biogenesis, and evolution of thylakoid membranes

Abstract

Cyanobacteria and chloroplasts of algae and plants harbor specialized thylakoid membranes (TMs) that convert sunlight into chemical energy. These membranes house PSII and I, the vital protein-pigment complexes that drive oxygenic photosynthesis. In the course of their evolution, TMs have diversified in structure. However, the core machinery for photosynthetic electron transport remained largely unchanged, with adaptations occurring primarily in the light-harvesting antenna systems. Whereas TMs in cyanobacteria are relatively simple, they become more complex in algae and plants. The chloroplasts of vascular plants contain intricate networks of stacked grana and unstacked stroma thylakoids. This review provides an in-depth view of TM architectures in phototrophs and the determinants that shape their forms, as well as presenting recent insights into the spatial organization of their biogenesis and maintenance. Its overall goal is to define the underlying principles that have guided the evolution of these bioenergetic membranes.

Introduction

Oxygenic photosynthesis provides the bioenergetic foundation for almost all life forms on Earth. It is responsible for virtually all carbon fixation—and thus biomass production—in the biosphere. Its oxygen-producing activity has shaped the planet's geosphere and atmosphere by enabling the onset of aerobic respiration and the subsequent emergence of complex eukaryotic organisms. As such, it represents one of the central drivers of biological evolution.

During photosynthesis, sunlight is converted into chemical energy in the form of ATP and NADPH, which fuel cellular metabolism, in particular by means of CO₂ fixation via the Calvin-Benson-Bassham cycle. Energy conversion takes place in specialized membrane systems called thylakoids. Here, the multi-subunit protein/pigment complexes known as PSII and PSI capture light and use its energy for charge separation events that initiate a directed process of electron transport across thylakoids to NADP+ and ultimately generates a proton-motive force for ATP production by ATP synthase. The ensuing charge imbalance is offset by the extraction of electrons from water at PSII, which results in the release of molecular oxygen. The cooperative action of PSII and PSI is also known as the Z-scheme of oxygenic photosynthesis, the origin of which dates back more than 2.4 billion years to the time when atmospheric O₂ reached concentrations of around 10% of their present level (Gumsley et al. 2017; Ossa et al. 2022). As recently reviewed by Oliver et al. (2023), the precise timing and paths of the early evolution of the photosynthetic complexes, in particular PSII, and their functional interconnection remain controversial. According to the latest phylogenetic analyses, the onset of oxygenic photosynthesis can be dated to approximately 3 billion years ago (Fournier et al. 2021).

Regardless of how and when cyanobacteria emerged as the major representatives of phototrophic organisms, evolution subsequently further shaped some of them into endosymbionts of eukaryotic heterotrophs. This development in turn gave rise to a new organelle—the chloroplast—and the 3 major lineages of photoautotrophic eukaryotes: glaucophytes and red and green algae (McCourt et al. 2023). The latter 2 lineages underwent additional secondary and even tertiary endosymbiotic events, which account for the current diversity of algae with complex plastids (Fig. 1). The descendants of green algae later conquered the terrestrial portion of the planet, finally developing into vascular plants (Harholt et al. 2016). During their diversification and evolution, thylakoid membrane (TM) systems have undergone a considerable degree of morphological diversification (Fig. 1; Mullineaux 2005). This ranges from simple photosynthetic plasma membranes (PMs) and internal membrane sheets in cyanobacteria to the elaborate network found in chloroplasts, which is further differentiated into highly stacked grana and unstacked stroma thylakoids (Fig. 1).

Figure 1.

TM systems in various phototrophs. Schematic depictions of thylakoid architectures (green) found in species ranging from cyanobacteria to vascular plants. A)Gloeobacter violaceus. B)Synechocystis sp. PCC 6803. C)Arthrospira sp. PCC 8005. D)Synechococcus elongatus PCC 7942. E)Coleofasciculus chthonoplastes PCC 7420. F)Geminocystis papuanica. G)Anabaena sp. PCC 7120 with heterocyst. H) Cyanelle of Cyanophora paradoxa. I) Chloroplast of Porphyra pulchella. J) Chloroplast of Chlamydomonas reinhardtii. K) Secondary plastid of Emiliana huxleyi. L) Tertiary plastid of Symbiodinium microadriaticum. M) Fully developed Chloroplast of vascular plants on the right, its proplastid precursor on the left with conections to the envelope via membrane protrusions. Biogenesis centers (B), T-zones (J), and grana margins (M) are depicted in blue. Pyrenoids (H to L) are shown as grey circles. The size differences only show tendencies but do not represent a statistical comparison as a schematic illustration.

Despite this complex evolutionary history, both the lipid matrix and the core molecular machinery of thylakoids that mediates photosynthetic electron transport, i.e. PSII, the cytochrome b₆f complex, PSI, and mobile electron carriers like plastoquinone and plastocyanin, have remained almost unchanged with only minor modifications (Allen et al. 2011). The major exception is the nature of the light-harvesting antenna systems, which have developed from membrane-attached phycobilisomes (PBS) into membrane-integrated light-harvesting complexes (LHCs). This transition alleviated steric constraints and increased the flexibility of the TM system, which ultimately led to the distinctive membrane substructure of plant chloroplasts: the granum (Fig. 1; Mullineaux 2005).

This article provides an overview of the variety of TM architectures in phototrophs. We then review our current understanding of their constituent parts and outline how these contribute to the shaping of key morphotypes. Finally, the question of how and where thylakoids are maintained is addressed. The last section includes a description of the key membrane-shaping factors, whose function has recently been clarified at the molecular level. Throughout the text, the evolutionary trajectory from cyanobacteria to vascular plants is considered in order to pinpoint conserved principles of the organization of TMs as bioenergetic membranes.

Architectures of TM systems

Cyanobacteria: the evolutionary starting point for thylakoid development

Cyanobacteria represent a diverse phylum of Gram-negative bacteria. They exhibit an astounding range of shapes and sizes, ranging from unicellular coccoids to multicellular filaments and colonies. The clade has colonized all possible habitats, including marine, freshwater, and terrestrial environments, and many have established intracellular symbioses with other organisms (Whitton 2012; Mareš et al. 2019). Using transmission electron microscopy (TEM), Mareš et al. (2019) have compared the thylakoid morphologies of more than 200 different cyanobacterial species. As exemplified in Fig. 1, A to G, the highly flexible architectural designs of thylakoids range from parietal to radial, coiled and parallel, and even more specialized types. The ecophysiological driving forces of this diversity remain unknown, but cell sizes smaller than 2 microns appear to restrict TM complexity. Phylogenetic analyses have revealed that parietal arrangements, like that of Synechococcus elongatus PCC 7942 (hereafter Synechococcus 7942; Fig. 1D), represent the ancient form of cyanobacterial TM organization. The spatial orientation of the PBS (Table 1) and their height of 36 nm determine the dimensions of the interthylakoidal space (Domínguez-Martín et al. 2022). This parietal form subsequently evolved into more complex architectures that can occasionally revert to the parietal state, as in Synechocystis sp. PCC 6803 (hereafter Synechocystis 6803; Fig. 1B) with its peripheral fascicles (Mareš et al. 2019). Indeed, the concentric multilamellar organization of these cyanobacterial thylakoids has been compared with the “wrap-around” myelin sheaths that envelope the axons in animal nervous systems—and proposed to represent a case of convergent evolution (Morelli et al. 2021), since both are involved in gas absorption/transport processes (CO₂ vs O₂) and delivery of ATP to the centers of their respective structures. Indeed, artificial concentric multilayer systems that are similar to thylakoids or myelin sheaths have recently been built for chemical synthesis (Cybulski et al. 2020).

Table 1.

Proteins and protein complexes involved in positioning, shaping, and maintenance of TMs

Protein or protein complex	Function	Cyanobacteria	Glaucophyta	Rhodophyceae	Chloroplastida	References
Positioning
PBS	TM positioning	✓	✓	✓	–	(Gantt and Conti 1969; Rast et al. 2019; Adir et al. 2020)
PSII-LHCII-supercomplex	Grana formation/TM stack stabilization	–	–	–	✓	(Albanese et al. 2020; Wood and Johnson 2020)
RIQ1/2	LHCII organization/grana formation	–	–	–	✓	(Yokoyama et al. 2016)
STN7/8	LHCII or PSII phosphorylation/grana size and stacking	✓	✓	✓	✓	(Bonardi et al. 2005)
KEA3, VCCN1, MGT10	Grana size and stacking	✓	✓	✓	✓	(Zhang et al. 2022; Dukic et al. 2023)
Shaping
CURT1	TM bending and maintenance/vesicle fusion	✓	✓	✓	✓	(Armbruster et al. 2013; Heinz et al. 2016; Sandoval-Ibáñez et al. 2021)
AncM	TM attachment to the PM	✓	–	–	–	(Ostermeier et al. 2022)
Maintenance
VIPP1	TM biogenesis and maintenance/vesicle fission, fusion and trafficking	✓	✓	✓	✓	(Vothknecht et al. 2011; Zhang et al. 2012; Gupta et al. 2021; Liu et al. 2021)
VIPP2	Sensing and coping with chloroplast membrane stress !only in some members of the Chlorophyceae	–	–	–	✓	(Theis et al. 2020)
CPRabA5e	Vesicular fission and trafficking	✓	✓	✓	✓	(Karim et al. 2014)
FZL	TM biogenesis and maintenance/vesicle fission, fusion and trafficking	✓	✓	✓	✓	(Gao et al. 2006; Findinier et al. 2019; Fauser et al. 2022)
CPSFL1	Vesicle trafficking	–	✓	✓	✓	(García-Cerdán et al. 2020; Hertle et al. 2020)
CPSAR1	TM biogenesis and maintenance/vesicle trafficking	✓	✓	✓	✓	(Garcia et al. 2010)

Protein conservation (✓) is indicated.

Special cases of thylakoid architecture in cyanobacteria include irregular arrangements found in heterocyst-forming cyanobacteria, such as Anabaena sp. PCC 7120 (Fig. 1G), and the absence of internal thylakoids in Gloeobacter, one of the earliest branching lineages in the cyanobacterial phylogenetic tree (Fig. 1A). In Gloeobacter violaceus, the photosynthetic complexes are located in patches of 140 ± 100 nm diameter within the PM (Rexroth et al. 2011; Rahmatpour et al. 2021). This organizational principle is generally assumed to represent an early stage in the evolution of photosynthetic membranes similar to that of anoxygenic green-sulfur bacteria (Guglielmi et al. 1981; Mareš et al. 2019). However, it cannot be ruled out that PM-localized photosynthesis might result from secondary loss of thylakoids in Gloeobacter (Mareš et al. 2019).

EM-based tomography has revealed that cyanobacterial thylakoid sacs occur as a connected network with a continuous lumen (Nevo et al. 2007; Ting et al. 2007; Liberton et al. 2011). Strikingly, these sacs contain perforations, which have been proposed to enable the exchange of metabolites between the center and periphery of the cytoplasm (Nevo et al. 2007). High-resolution cryo-electron tomography (cryo-ET) has uncovered thylakoid convergence zones in Synechocystis 6803, areas at which thylakoids bend toward the PM and presumably form sites of close contact or fusion between TMs and PMs (Liberton et al. 2006; van de Meene et al. 2006). Recent in-situ cryo-EM data has revealed that the converging TMs in these regions actually fuse to form a tube (TCT: thylakoid convergence tube) that is oriented perpendicular to the TM sheets and parallel to the PM (Rast et al. 2019). Occasionally, this membranous tube contacts the PM at sites called thylapses, owing to their synapse-like appearance (Fig. 2, A, B, and D; Rast et al. 2019). There is no evidence that TM and PM fuse at thylapses to form a continuum between periplasm and thylakoid lumen. Instead, the connection is likely to be made by protein material that bridges the 3-nm gap between the 2 membranes (Fig. 2D; Rast et al. 2019). In contrast to Synechocystis 6803, Synechococcus 7942 cells do not contain obvious TM convergence zones (Fig. 1, B and D). Nevertheless, cryo-ET–based analysis of Synechococcus 7942 under conditions favoring the biosynthesis of new TMs revealed transient tight junctions—but no membrane fusions—between PM and TMs that are similar to those seen in Synechocystis 6803 (Huokko et al. 2021; see “Thylakoid shaping proteions”).

Figure 2.

The thylakoid convergence zone (TCZ) in Synechocystis 6803. A) 3D segment of a tomogram volume from Synechocystis 6803. The outer membrane (light blue), PM (dark blue), thylakoids (green), membrane-associated ribosomes (pink), free cytosolic ribosomes (grey), and phycobilisomes (yellow) are depicted (Rast et al. 2019). B) Thylapse region from A, rotated 90°, without outer and PM. C) 3D reconstruction of a fluorescence micrograph (side and top views) showing chlorophyll autofluorescence (red), CurT (light blue), and AncM (dark blue). D) Current working model for initial PSII assembly at a TCZ, based on panels A and C. Thylakoids (green), membrane-associated ribosomes (pink), free cytosolic ribosomes (grey), phycobilisomes (yellow), CurT (light blue), and AncM (dark blue) are indicated, and PSII intermediates are marked.

The remarkable diversity of TM architectures in cyanobacteria underscores their fundamental role in the evolution of increasingly intricate and complex thylakoid architectures. The functional implications of these structural variations across different cyanobacterial lineages remain to be clarified. The potential role of TM substructures such as TCTs and thylapses in Synechocystis 6803 is further discussed in the section on TM biogenesis and maintenance.

Algae: the evolution of chloroplasts

Eukaryotic algae evolved via multiple endosymbiotic events (Adl et al. 2005). A central component of most algae (and some hornworts) is the pyrenoid, a functional analog of the carboxysomes found in cyanobacteria, as both mainly consist of paracrystalline clusters of CO₂-fixing Rubisco (Liu 2022; Ni et al. 2022; He et al. 2023). Unlike carboxysomes, however, the pyrenoid inevitably influences the architecture of thylakoids. As recently summarized by He et al. (2023), they can enter and indeed traverse this intra-organellar, phase-separated subcompartment.

The first endosymbiotic event led to 3 major algal lineages: Glaucophyta, Rhodophyceae, and Chlorophyceae. Glaucophytes are a small and primordial lineage of freshwater algae that contain 2 cyanelles instead of chloroplasts (see Cyanophora paradoxa in Fig. 1H). Their thylakoid systems are concentric with occasional terminations or bifurcations, and pyrenoid-traversing thylakoids have not been reported (Price et al. 2019). Rhodophytes (red algae) are a highly diverse group of uni- and multicellular algae that are predominantly found in marine environments and possess unstacked TMs and PBS as their main light-harvesting antenna (Su et al. 2010). In Porphyridium purpureum UTEX 2757, similar to some cyanobacteria, branching and fusion regions were observed between neighboring TMs building “stairs,” which end in a large perforation on top of them (Li et al. 2021). In some red algae, such as Porphyra sp., thylakoids are arranged in a simple net-like structure with PBS-free thylakoids winding through the pyrenoid (Fig. 1I; Ackland et al. 2006).

Chlorophyceae include green algae, as well as their evolutionary descendants—land plants (Embryophyta). They are a diverse group of photosynthetic organisms and exhibit a wide range of thylakoid architectures. In contrast to glaucophytes and red algae, green algae lost PBSs and developed membrane-integral LHCs based on chlorophyll b as their light-capturing pigment (Green 2019). This innovation had a significant impact on their TM architecture. Due to reduced steric constraints associated with the bulky PBS complexes, thylakoid sheets were able to form multiple tightly stacked layers (Mullineaux 2005; Adir et al. 2020). In situ cryo-ET of the unicellular green alga Chlamydomonas (Chlamydomonas reinhardtii) has shown that the TMs in its chloroplasts exhibit an elaborate architecture with an interconnected network of unstacked and stacked lamellae. However, the highly structured membrane layers characteristic of the grana of vascular plants are absent in Chlamydomonas (Fig. 1J; Engel et al. 2015). Although the TM stacks of some green algal taxa, such as Coleochaetales and Charales, have been described as grana, their nanomorphology differs from that of vascular plants in the number and size of stacks (Gunning and Schwartz 1999; Chen et al. 2018; Mazur et al. 2021). On average, the appressed “grana-like” TM areas are composed of 3 to 4 thylakoids per stack, with a lateral repeat length of about 22 nm and an interthylakoidal space of <4 nm (Fig. 3A). This is in sharp contrast to cyanobacteria, in which an interthylakoidal space of 30 to 50 nm has been reported for Synechocystis 6803 (Fig. 2A; van de Meene et al. 2006).

Figure 3.

TM properties. A) Dimensions of TMs in different phototrophs grown under continuous illumination (van de Meene et al. 2006; Kirchhoff et al. 2011; Engel et al. 2015; Ostermeier et al. 2022). B) Influence of the physicochemical properties of thylakoid lipids on membrane structure. For abbreviations see text. C) Pyrenoid tubules in C. reinhardtii traverse the pyrenoid starch sheath and form a twisted tubuli (reticulated region) network inside the pyrenoid (according to Engel et al. 2015). D) Cross-section of a pyrenoid tubule showing multiple minitubules (brown). E) Model of a grana stack with a stromal lamella wound around it. Grana margins with high levels of CURT1 are depicted in light blue. F) Zoom into the grana margin.

Like cyanobacterial TM sheets, however, algal thylakoid stacks are occasionally perforated by fenestrations (Nevo et al. 2007; Ting et al. 2007), which are likely to facilitate metabolite exchange (Engel et al. 2015). Moreover, the work of Engel et al. (2015) detected frequent invaginations in the inner chloroplast envelope (IM), which are derived from the PM of the former endosymbiont. In rare cases, connections were found between the TM and these invaginations, suggesting that lipids and/or proteins might be delivered to TMs via such “membrane bridges” (Engel et al. 2015). This work also revealed the detailed structures of pyrenoid-traversing thylakoids in Chlamydomonas (Fig. 3C). Several thylakoids fuse to form a pyrenoid tubule that is continuous with the lumen of the thylakoids and forms a complex interconnected network, the reticulated region. In addition, they can envelope up to 8 minitubules that are continuous with the stroma in the interthylakoid space and the pyrenoid matrix (Fig. 3, C and D; Engel et al. 2015). Thus, this tubular network probably coordinates the spatially separated steps of carbon fixation by facilitating the exchange of sugar intermediates and CO₂ between the stroma and the interior of the pyrenoid.

Many algal groups obtained their plastids via secondary or even tertiary endosymbiosis when a non–plastid-harboring eukaryotic cell engulfed a eukaryotic alga. Therefore, these algae possess chloroplasts with 1 or 2 additional surrounding membranes and PBS-free thylakoids (Green 2019). A recent study of 7 distant taxa of algal phytoplankton, based on focused ion-beam scanning electron microscopy (FIB-SEM) and comparative 3D morphometric analysis, has provided new quantitative insights into the plasticity of chloroplast architecture during acclimation to high light levels (Uwizeye et al. 2021). Moreover, thylakoid architecture in the diatom Phaeodactylum tricornutum has been reconstructed from FIB-SEM data, revealing frequent connections between thylakoid layers, which might facilitate rapid transport of photosynthetic electron carriers (Flori et al. 2017). Most secondary/tertiary plastids also harbor pyrenoids with TMs. However, in contrast to Chlamydomonas, these TMs form stacks of 2 or 3 thylakoid layers rather than complex tubular networks within the pyrenoid (Fig. 1, L and K; Blank 1987; Rast et al. 2015; LaJeunesse 2017). Nevertheless, it appears likely that also these pyrenoidal TMs are involved in carbon exchange with the cytoplasm.

In summary, the diverse endosymbiotic events that characterized the evolution of eukaryotic algae have shaped their TM systems into appressed and non-appressed regions with strikingly heterogeneous morphologies. Moreover, thylakoids and the CO₂-fixing pyrenoid are tightly interwoven and thus should facilitate metabolite exchange. Whether the pyrenoid has influenced overall thylakoid architecture—for example, by spatially organizing TM biogenesis—is addressed in the section on TM biogenesis and maintenance.

Vascular plants: conquering terrestrial habitats

Land plants (Embryophyta) evolved from streptophytes, the sister group of chlorophytes, and conquered terrestrial habitats in the Silurian period some 429.3 million years ago (Libertín et al. 2018; Donoghue et al. 2021). They are divided into nonvascular plants (bryophytes) and the monophyletic group of vascular plants (tracheophytes) (De Vries and Archibald 2018; Rensing 2020). Their chloroplasts characteristically contain an elaborate bipartite thylakoid architecture comprised of stacked grana thylakoids and unstacked stroma thylakoids (Fig. 1M).

Fully developed grana thylakoids are made up of stacks of flattened disks, which are intersected at a right-handed helical angle of about 16° in Lactuca sativa to 20° in A. thaliana by thin stromal lamellae (Figs. 1M and 3E; Mustárdy and Garab 2003; Bussi et al. 2019; Mazur et al. 2021). Each granum is separated from the next by a thin layer of stromal fluid, the partition gap (Fig. 3A), whereas the individual sheets are held together in a stack by a balanced combination of van der Waals forces and repulsive electrostatic and hydration forces (Puthiyaveetil et al. 2017). Grana and stroma thylakoids are interconnected by bifurcations via luminal space, whereas different stroma lamellae are connected by left-handed helices for additional luminal communication between the lamellae (Shimoni et al. 2005; Bussi et al. 2019).

The thickness of the grana and its dimensions were first measured by Daum et al. (2010) and verified by Kirchhoff et al. (2011). Under continuous illumination these are very close to the dimensions of stacked TMs measured in Chlamydomonas (Fig. 3A) (Kirchhoff et al. 2011; Engel et al. 2015). The number of thylakoids per stack varies from less than 10 in bright light–exposed chloroplasts to about 100 in plants like Alocasia macrorhiza that thrive in extreme shade (Goodchild et al. 1972). Under dark conditions, the width of the grana lumen is most probably determined by the size of the oxygen-evolving complex of PSII (Ferreira et al. 2004; Kirchhoff et al. 2011; Engel et al. 2015), which in turn suggests that the width of the thylakoid lumen is the same in algae and higher plants (Engel et al. 2015).

Determinants of TM shape

Lipids: the matrix of TMs

In principle, thylakoids are a specialized system of flattened lipid vesicles densely packed with pigment-protein complexes. The lipid bilayer that serves as a scaffold for the protein complexes in thylakoids is comprised of 4 main glycerolipids (i.e. lipids built on a 3-carbon glycerol backbone): the phospholipid phosphatidylglycerol (PG), and the glycolipids monogalactosyldiacylglycerol (MGDG), digalactosyldiacylglycerol (DGDG), and sulfoquinovosyldiacylglycerol (SQDG), all of which are devoid of phosphorus (Fig. 3B). From cyanobacteria to chloroplasts, the neutral glycolipids MGDG and DGDG dominate the lipid composition of thylakoids, comprising 40% to 55% and 25% to 35% of the total lipids, respectively. The anionic glycerolipids SQDG and PG occur in lower proportions of 5% to 25% and 5% to 15%, respectively (Murata et al. 1990; Sakurai et al. 2006). These glycerolipids are not only major components of the TM, but they also fulfill functions as structural components of TM protein complexes. A recent estimate suggests that, in both cyanobacteria and chloroplasts, approximately 2% of the total thylakoid lipids are allocated to PSI; the corresponding figures for PSII are 2% and 6%, respectively, while less than 1% are assigned to the cytochrome b₆f complex (Yoshihara and Kobayashi 2022).

The fact that cyanobacteria and chloroplasts share virtually the same lipid composition reflects their common origin. However, the evolutionary transition from the former to the latter resulted in not only differences in their glycolipid biosynthetic pathways (see the section on TM biogenesis and maintenance; Fig. 4A) but also their lipid profiles. Cyanobacteria additionally contain small amounts (up to 10% of total lipids) of the neutral glycolipid monoglucosyldiacylglycerol (MGlcDG)—a metabolic precursor of MGDG (Petroutsos et al. 2014). Indeed, MGlcDG is already present in the primitive thylakoid-less cyanobacterium G. violaceus, which also lacks SQDG and thus relies on PG as its main anionic glycerolipid (Selstam and Campbell 1996; Rexroth et al. 2011). The acquisition of SQDG as an anionic glycerolipid has been interpreted as a “sulphur-for-phosphorus” evolutionary strategy, which may have been a key factor in the radiation of early cyanobacteria, as SQDG synthesis provided an anionic surrogate for and flexible source of PG, which in turn ensured the availability of a constant supply of anionic lipids even under phosphorus-deprived conditions (Yu et al. 2002; Yu and Benning 2003; Van Mooy et al. 2006). PG is essential for photosynthetic activity and is therefore an indispensable constituent of photosynthetic membranes (Sato et al. 2000; Babiychuk et al. 2003), while the need for SQDG across cyanobacteria and eukaryotic phototrophs is variable and species dependent (Kobayashi and Wada 2016). As precursors of thylakoid lipids, the glycerolipids phosphatidic acid (PA) and diacylglycerol (DAG) are present in small amounts in cyanobacterial and chloroplast membranes. Evidence from lipid labeling in cyanobacteria (Selao et al. 2014) and from probing of DAG pools in subchloroplastic compartments (Fritz et al. 2007; Muthan et al. 2013) suggests that PA and DAG may occur in TMs. The phospholipid phosphatidylinositol (PI) is a minor component of chloroplast thylakoids (Dorne et al. 1985) and is thus considered to be a eukaryotic acquisition. The mechanisms underlying the presence of PI in the chloroplast remain unknown.

Figure 4.

Biosynthesis of thylakoid glycerolipids in cyanobacteria and chloroplasts. For further explanation, see text. Abbreviations: CW, cell wall; IEM, internal envelope membrane; OEM, outer envelope membrane.

Thylakoid glycerolipids display contrasting biophysical properties that influence membrane organization and dynamics (Fig. 3B). Glycerolipids with a small head-group display a conical shape that induces a negative membrane curvature and favors their self-assembly into inverted tubular micelles, also known as hexagonal II (HexII) phases. Glycerolipids with bulky head-groups have an inverted conical shape, induce a positive curvature, and self-assemble into tubular micelles (hexagonal I, HexI, phases). In contrast, glycerolipids with a similar cross-section along the molecule are cylinder-shaped and form lamellar (Lm) phases. MGDG is a HexII phase-forming neutral galactolipid, whereas DGDG, SQDG, and PG are Lm phase-forming lipids (Jouhet 2013). Non-bilayer lipids such as MGDG can occur in the lamellar phase depending on levels of hydration and temperature, unsaturated fatty-acid profiles, and/or their combination with Lm lipids (Demé et al. 2014). In thylakoid lipid mixtures, the ratio of HexII (MGDG) to Lm (DGDG, PG, SQDG) defines the organization of the lipid phase, as well as phase transitions, such as negative curvature to lamellar (HexII→Lm) phase, or vice versa (Lm→HexII), that occur within a time window of nano- to microseconds (Demé et al. 2014; Guéguen and Maréchal 2022). PA and DAG are also HexII phase-forming lipids that induce negative membrane curvature and packing defects (i.e. membrane regions with low lipid density) and are therefore relevant for membrane remodeling processes, such as fission and fusion (Szule et al. 2002; Noack and Jaillais 2020).

While the overall thylakoid lipid composition is similar across cyanobacteria and chloroplasts, the physiological relevance of glycerolipid abundance, molecular diversity (i.e. types of fatty acids attached to the glycerol backbone), dynamics, and intracellular location remain to be explored.

Photosynthetic complexes

Thylakoids may also be seen as protein-bearing platforms that are interconnected by lipid bilayers, since the proportion of proteins in grana thylakoids can reach up to 80% (Kirchhoff et al. 2008a). Thus, the photosynthetic complexes must be considered not only as drivers of energy production and CO₂ fixation but also as determinants that shape TM architectures. A common theme among prokaryotic and eukaryotic phototrophs is the dense clustering and heterogeneous distribution of photosynthetic complexes in TMs. This is mainly reflected by the lateral segregation of PSII and PSI, a mechanism that avoids energy spill-over between the 2 photosystems (Lokstein et al. 2021). This feature also contributes to the patterning and dynamics of TM architecture, while structural changes in the TM are strongly associated with photosynthetic function and adaptation to light conditions (Li et al. 2020).

Cryo-ET analyses of Synechocystis 6803 cells have revealed the presence of linear arrays of PSII—up to 20 units in length—even in morphologically unstructured cyanobacterial TM sheets (Rast et al. 2019; Fig. 2A). Moreover, these units are separated from neighboring PSI complexes, which minimizes spill-over effects (Akhtar et al. 2024). In addition, the organizational heterogeneity of photosynthetic complexes in TMs can differ between cyanobacterial species, which further underlines the structural plasticity of TM architectures (Fig. 1; Zhao et al. 2020, 2022). In the red algae P. purpureum UTEX 2757, PBSs build supercomplexes with PSII and are orderly packed between 2 parallel TMs defining their minimal interthylakoidal space (Li et al. 2021; You et al. 2023).

The evolutionary advent of the membrane-integral LHC antenna system in the chloroplasts of green algae further optimized this organizational principle via close stacking of TM sheets. This in turn resulted in the lateral separation of PSII and PSI complexes into appressed and non-appressed regions, respectively. Based on in situ cryo-EM data, this separation is very strict in Chlamydomonas, which exhibits virtually no intermixing between the 2 photosystems (Wietrzynski et al. 2020). Similarly, in plant chloroplasts PSII is mainly located in grana stacks, whereas PSI and the ATP synthase are located in stroma lamellae (Albertsson 2001; Kirchhoff et al. 2008b). Grana margins are the only TM regions in which PSII and PSI intermix (Rantala et al. 2020). It is not surprising that PSI and the ATP synthase should be excluded from appressed TM areas, since their stromal moieties are too bulky to fit into the narrow interthylakoid space in grana (Daum et al. 2010; Kirchhoff et al. 2011). This gap is defined by PSII-LHC supercomplexes that stick out into the stroma by 3.6 nm (Kirchhoff et al. 2011; Engel et al. 2015).

LHCs most probably originated from a family of small 1-helix Chl-binding proteins in cyanobacteria, which subsequently evolved into LHC proteins with 3 transmembrane helices via gene duplication/fusion events (Engelken et al. 2010). As the main antenna complex in algae and vascular plants, LHCII accounts for about 50% of grana thylakoid proteins, and PSII-LHCII supercomplexes (Table 1) support the formation and stability of the characteristic stacked grana structure (Albanese et al. 2020). The self-association of PSII-LHCII domains in neighboring membrane layers is thought to be mediated by van der Waals interactions and via electrostatic interactions between the positively charged N terminus of LHCII and the negatively charged stromal surface of the adjacent trimer (Li et al. 2020; Wood and Johnson 2020; Manna and Schlau-Cohen 2022). Even in artificial membranes, high LHCII densities lead to the stacking of proteoliposomes, thus mimicking the in vivo thylakoid grana membranes of green algae and vascular plants (Wilson et al. 2022).

Nevertheless, how and to what extent physicochemical forces between 2 adjacent membranes determine the structure of the granum remains a matter of intense discussion (Moazzami Gudarzi et al. 2021; Müh et al. 2021). According to Puthiyaveetil et al. (2017), grana stacking can be explained by the balance between attractive van der Waals forces and repulsive electrostatic and hydrostructural forces alone. Crucially, this conclusion was based on a revised calculation of the relevant stacking forces, which differs notably from those employed in earlier studies (Sculley et al. 1980; Barber 1982; Chow et al. 2005). This led Moazzami Gudarzi et al. (2021) to state that van der Waals forces had been overestimated, owing to a misinterpretation of the dielectric permittivity of membrane proteins and excessively low values for the surface charge density. This in turn led to corrections of the stated values that were closer to those used in previous reports and reopened the question of how the interplay between electrostatic and lipid-mediated interactions facilitates grana stacking (Müh et al. 2021).

Moreover, it seems paradoxical that the grana should stabilize the ultrastructure of chloroplasts and at the same time allow for dynamic flexibility in response to changing environmental conditions (Wood et al. 2019). Short- and long-term adaptation of the photosynthetic apparatus to changes in light intensity is mainly driven by LHCII phosphorylation, which enables rapid changes in grana size and stacking (Kirchhoff et al. 2011; Anderson et al. 2012; Wood et al. 2019). LHCII and PSII core proteins are reversibly phosphorylated by redox-controlled kinases (STN7 and STN8, respectively; Table 1) and their respective phosphatases (Tikkanen and Aro 2012).

Other determinants of grana stacking include the ion composition and internal osmotic pressure of the lumen, both of which are primarily mediated by thylakoid-localized ion transporter/channels (Kirchhoff 2014; Liu et al. 2018). The thylakoid K⁺/H⁺ antiporter KEA3, the voltage-dependent Cl⁻ channel VCCN1, and magnesium transporters, for example, MGT10/MRS4, are among those thought to be involved in shaping thylakoid architecture (Finazzi et al. 2015; Spetea et al. 2017; Sun et al. 2017). Each of these transporters/channels acts independently in regulating the degree and balance of acidification of the lumen—and the resulting osmotic pressure, which is largely responsible for thylakoid swelling at different light intensities (Li et al. 2020). A knockout mutant of VCCN1 shows altered thylakoid ultrastructure with curved grana (Herdean et al. 2016), while knocking down the inner-envelope Mg²⁺ transporter MGT10 in Arabidopsis thaliana (Arabidopsis), as well as knockout of its homologue MRS4 in Chlamydomonas, results in longer grana stacks and more TM layers (Zhang et al. 2022; Dukic et al. 2023).

Clearly, the macromolecular organization of photosynthetic complexes has a significant impact on TM structure in cyanobacteria and chloroplasts. However, the long-standing questions of how and to what degree grana stacking is mediated by physicochemical forces, hydrogen bonds across the stroma gap, pigments, Mg²⁺ ions, or yet undescribed factors such as dynamin-like proteins need further attention.

Thylakoid-shaping proteins

Besides the complexes that constitute the core of the photosynthetic apparatus, a family of 4 small thylakoid-shaping proteins, named CURVATURE THYLAKOID1A-D (CURT1A-D), was initially identified in A. thaliana as crucial determinants of thylakoid architecture by Armbruster et al. (2013). All CURT1 proteins have a mass of 15 kDa and contain 2 transmembrane domains at the C terminus, as well as an amphipathic helix in their N-terminal region. In vitro, CURT1A displays membrane-bending activity and is located at grana margins (Fig. 3, E and F; Table 1), which is consistent with a role in thylakoid shaping (Armbruster et al. 2013). Indeed, CURT1 proteins control the size and number of grana discs by forming oligomeric complexes (Pribil et al. 2018). As a consequence, the complete absence of CURT1 proteins leads to pseudograna comprised of reduced numbers of grana discs with enlarged diameters, whereas overexpression of CURT1A leads to more discs with reduced diameters (Armbruster et al. 2013; Pribil et al. 2014). Moreover, phosphorylation seems to alter the structure of CURT1's N-terminal helix and/or oligomerization behavior, resulting in a modified bending activity (Armbruster et al. 2013). Despite the dramatic morphological changes of the TM system observed upon loss of CURT1A-D, accumulation of PSII subunits and LHCIIs remains largely unaltered (Armbruster et al. 2013; Pribil et al. 2014). Nevertheless, curt1a-d mutants revealed that CURT1-mediated thylakoid plasticity is required for the fine tuning of photosynthesis under challenging light conditions in the greenhouse, as well as under natural conditions in the field (Pribil et al. 2018). Thus, CURT1 proteins seem to play a crucial and dynamic role in the shaping of grana thylakoid regions in chloroplasts.

In addition, these proteins were recently shown to participate in the light-dependent conversion of etioplasts into chloroplasts in angiosperms during the reorganization of prolamellar bodies (PLBs) and their transformation into thylakoids (Sandoval-Ibáñez et al. 2021). The Chlamydomonas genome harbors 3 CURT1 genes (Armbruster et al. 2013). However, their molecular functions remain to be defined, although a shaping activity in the formation of 180° loops between thylakoid tips at chloroplast lobes has been proposed (Engel et al. 2015). Moreover, CURT1 proteins might be involved in orchestrating the tubular network that links the TM to the pyrenoid matrix in Chlamydomonas (Fig. 3C; Engel et al. 2015). Thus, the membrane-curving activity of CURT proteins might be involved in thylakoid shaping beyond grana formation, suggesting a broader role for it as an “architect” for the TM.

In line with this idea, although they lack grana structures, most cyanobacteria also possess a homolog of plant CURT1 proteins, named CurT. Inactivation of CurT in Synechocystis 6803 results in dramatic alterations of overall membrane organization, including the loss of thylakoid convergence zones and aberrant formation of disorganized, circular TM sheets (Heinz et al. 2016). As a consequence, photosynthetic performance, and in particular PSII assembly, are compromised in curT mutants (Heinz et al. 2016). Imaging data have revealed that CurT is recruited to the TCT network at the periphery of the Synechocystis 6803 cell, with the highest concentrations occurring at the most tightly curved TCT regions (Fig. 2, B and C). The network is made up of TCMs that are oriented perpendicular to the thylakoid sheets and are likely to be formed by CurT's bending activity (Fig. 2). As mentioned above, at thylapse regions TCTs occasionally make contact with the PM via a proteinaceous bridge (Rast et al. 2019; Fig. 2). Recently, a possible component of this membrane-tethering bridge has been identified based on a suppressor screen for the curT mutant (Ostermeier et al. 2022). The 32-kDa protein AncM (anchor of convergence membranes) contains a single TM domain and is coexpressed with CurT in Synechocystis 6803. Loss of AncM results in detachment of the thylapse from the PM and an increase in the stack size of TMs (Ostermeier et al. 2022). AncM is localized to distinct punctate regions close to the PM, where CurT also accumulates; this supports the idea that AncM marks thylapses (Fig. 2, C and D; Table 1). Thus, the interplay between CurT and AncM appears to be a key element of the overall architecture of the TM system in Synechocystis 6803.

Despite this functional relationship, phylogenetic analyses revealed that, unlike the ubiquitous CURT1/CurT, AncM is only found in cyanobacteria (Ostermeier et al. 2022). Apparently, AncM function was not required for, and might even have hampered, the formation of the sophisticated grana-thylakoid system during evolution, or, alternatively, its function was taken over by other factors. Indeed, even some cyanobacteria lack AncM, and a few lineages contain neither AncM nor CurT, such as taxa from the primordial Gloeobacterales that lack an internal TM system, and simple marine Prochlorococcus species (Ostermeier et al. 2022). On the other hand, more highly developed cyanobacteria, like the filamentous, heterocyst-forming Anabaena sp. PCC 7120, possess 2 homologs of CurT. While CurT1 is abundant in vegetative cells, CurT2 is present exclusively in nonphotosynthetic, N-fixing heterocysts. There it accumulates at honeycomb membranes rather than at peripheral thylakoids, suggesting that it is involved in organizing this specialized membrane system that controls metabolite fluxes in heterocysts (Santamaria-Gomez et al. 2018). Moreover, CurT co-purifies with FtsH4, which is involved in the biogenesis of PSII (Krynická et al. 2022) and possibly of PSI (Koník et al. 2024). These findings point to a membrane-structuring and/or subdomain-forming function for CurT.

The fact that TMs are completely disorganized in the Synechocystis curT mutant but still display some curvature suggests that additional factors are involved in cyanobacterial TM bending. One such factor might be C15-C19 hydrocarbons, which have been shown to accumulate in both TMs and PMs (Lea-Smith et al. 2016). Mutants deficient in these hydrocarbons show reduced thylakoid curvature, suggesting that they promote membrane flexibility and facilitate bending (Lea-Smith et al. 2016). Furthermore, the authors of that study hypothesized that CurT proteins augment and orient the hydrocarbon-dependent disturbances and alterations that give rise to membrane curvature in cyanobacteria.

Other players that might be functionally connected with CURT1 in Arabidopsis chloroplasts are the RIQ proteins (REDUCED INDUCTION OF NON-PHOTOCHEMICAL QUENCHING). In contrast to CURT1, absence of RIQ1 and RIQ2 results in increased grana stacking and reduced LHCII accumulation (Yokoyama et al. 2016). Both proteins are present in land plants but have been lost in certain green algae (Table 1). For instance, Micromonas pusilla RCC299 lacks RIQ2, whereas Chlamydomonas and Volvox carteri have lost both RIQ1 and RIQ2 (Yokoyama et al. 2016; Table 1). RIQ1 and RIQ2 are located in the grana core, but their contribution to grana stacking differs from that of CURT1. By establishing a functional link between grana structure and LHCII organization, they optimize light-harvesting efficiency. Precisely how RIQ proteins do this remains to be elucidated. This will require tests that can determine whether the changes observed in grana structure in riq1 and riq2 mutants result primarily from (1) direct regulation of grana stacking and LHCII organization, or (2) alterations in LHCII organization around PSII that enhance photoprotective functions, which affects grana stacking only secondarily (Yokoyama et al. 2016).

TM biogenesis and maintenance

The striking morphological differences in TM architectures (Fig. 1) raise the question of how their diverse evolutionary trajectories evolved. In recent years, a combination of genetic and biochemical approaches has revealed 2 major principles of TM biogenesis: (1) it is highly ordered in both space and time, and (2) it proceeds through strictly determined sequential maturation steps that are often facilitated by evolutionarily conserved assembly factors. Here, the focus is on the first of these steps and addresses lipid and membrane flow. The second aspect is reviewed in detail by Komenda et al. (PSII assembly) and Bock et al. (PSI assembly) in this issue.

Lipid synthesis

Although the thylakoid lipid compositions of cyanobacteria and chloroplasts are remarkably similar their pathways used to synthesize galactolipids differ (Fig. 4A) (Petroutsos et al. 2014; Sato and Awai 2016). In cyanobacteria, MGDG synthesis occurs in 2 steps. First, the neutral glycolipid MGlcDG is synthesized by the glucosyltransferase MGlcDG synthase (MgdA), which transfers a glucose moiety from UDP-glucose to DAG (Awai et al. 2006); then an epimerase (MgdE) isomerizes the glucosyl group into galactose, yielding MGDG (Awai et al. 2014). In chloroplasts, a galactosyltransferase MGDG synthase (MGD1) catalyzes the transfer of galactose from UDP-galactose onto DAG, thus synthesizing MGDG in a single step (Shimojima et al. 1997). In both cyanobacteria and chloroplasts, DGDG synthesis occurs via MGDG galactosylation, catalyzed by non-orthologous DGDG synthases of cyanobacterial (DgdA) (Awai et al. 2007; Sakurai et al. 2007) or chloroplastic type (DGD1) (Kelly et al. 2016), respectively. Why cyanobacteria and eukaryotic phototrophs use different pathways for galactolipid synthesis is not clear. The lack of viable MgdA knockouts in cyanobacteria (Awai et al. 2006), together with the lethality and nonphotoautotrophic growth of the Arabidopsis mgd1 mutant (Kobayashi et al. 2007), highlight the pivotal role of glycolipids in thylakoid biogenesis and function. Neither partial (Maida and Awai 2016) nor complete (Apdila et al. 2020) replacement of the cyanobacterial galactolipid pathway by the corresponding plant-type pathway in Synechococcus 7942 leads to an obvious phenotype, indicating that the lipid products MGDG and DGDG, but not their biosynthetic pathways, are functionally relevant. This functional relevance strongly relies on the physical properties of the 2 lipids, which are probably maintained after replacement of galactose by glucose. For instance, Synechocystis 6803 MgdE mutants, lacking galactolipids but accumulating MGlcDG, contain thylakoids and are photosynthetically active, albeit with reduced light-energy utilization (Awai et al. 2014).

Like galactolipid biosynthesis, PA biosynthesis differs between cyanobacteria and chloroplasts, since it is performed by nonhomologous proteins that rely on different acyl donors (Fig. 4A). In cyanobacteria, but not in chloroplasts, the first acylation step relies on Acyl-P rather than Acyl-ACP (Matsumoto and Awai 2020). In contrast, DAG biosynthesis (Fig. 4A) depends on cyanobacterial and chloroplast PA phosphatases (LPP) that share a prokaryotic origin (Nakamura et al. 2007). Unlike galactolipid synthesis, the pathways for PG and SQDG (Fig. 4B) are shared by cyanobacteria and chloroplasts (Hagio et al. 2002; Sato et al. 2003). Eukaryotic phototrophs harbor additional extra-chloroplast PG synthesis pathways in the mitochondria and endoplasmic reticulum (Kobayashi et al. 2024); however, these cannot compensate for the loss of chloroplast-synthesized PG (Babiychuk et al. 2003).

Not only the profile but also the location of the final steps in TM lipid biosynthesis is important because such sites can define the starting points of bilayer assembly. In cyanobacteria (Fig. 4A), according to membrane fractionation experiments, MGDG is synthesized in thylakoids and the PM, whereas DGDG synthesis is restricted to the PM (Omata and Murata 1986; Selao et al. 2014). In chloroplasts (Fig. 4A), thylakoidal MGDG is synthesized in the inner envelope (Miège et al. 1999; Awai et al. 2001) and DGDG is synthesized in the outer envelope (Froehlich et al. 2001; Joyard et al. 2010), where the characteristic N-terminal domain of DGD1 is essential for galactolipid translocation between envelopes (Kelly et al. 2016). The topological segregation of the enzymes that synthesize MGDG and DGDG has been proposed to control the relative abundance of these major HexII- and Lm-forming lipids—and thus membrane curvature—by regulating the transient accumulation of MGDG (Guéguen and Maréchal 2022). Synthesis of the anionic lipids PA, PG, and SQDG (Fig. 4, A and B) occurs in the IM (Andrews and Mudd 1985; Essigmann et al. 1998; Tietje and Heinz 1998), where they may modulate the local abundance of MGDG, since they influence MGD1 activity (Dubots et al. 2010; Makshakova et al. 2020; Nitenberg et al. 2020). The work of Selao et al. (2014) shows that PG and SQDG pathways are active in the TM and PM from cyanobacteria (Fig. 4B). However, it remains unclear how or even whether these pathways communicate between the 2 membranes.

An exciting question for future research is whether sites of lipid biosynthesis might also serve as assembly hotspots for protein-pigment complexes and facilitate coordinated biogenesis of the 2 major TM components.

Membrane functionalization by lipids

Lipids can perform functions beyond their structural roles. Due to their biophysical properties, lipid species with conical or inverted-conical shapes trigger packing defects (Fig. 3B) that, even under low abundance, can functionalize specific membrane domains. Among these, phosphoinositides, PA, and DAG are lipid species well-known to act in nonphotosynthetic membranes as signaling molecules and key players in membrane dynamics, either by influencing physical membrane properties or by mediating specific protein interactions (Gerth et al. 2017; Noack and Jaillais 2020). If these lipid species were to play such a role also in chloroplasts, this would have enormous implications for our current understanding of plastid biology.

Accumulating evidence not only supports the presence of PI, but also its phosphorylated derivatives, that is, phosphoinositides, in chloroplasts (Schroda 2020). Immunolocalization coupled to fluorescence microscopy using phosphoinositide-specific antibodies has confirmed the presence of phosphatidylinositol 4-phosphate (PI4P) in chloroplasts (Okazaki et al. 2015; Hertle et al. 2020). In A. thaliana, PI4Kα1 is the main contributor to chloroplast levels of PI4P (Okazaki et al. 2015). PI4P itself is a negative regulator of chloroplast division in A. thaliana, probably owing to its interaction with the outer-envelope proteins PLASTID DIVISION 1 and 2 (Miyagishima et al. 2006; Glynn et al. 2008), both of which specifically bind PI4P in vitro (Okazaki et al. 2015). These results restrict the role of phosphoinositides to the outer envelope. However, accumulating evidence supports their involvement in intra-chloroplast dynamics, since increasing numbers of chloroplast proteins have been shown to associate with the inner envelope and/or the TM, which selectively bind PI4P in vitro: The chloroplast-localized, Sec-14-domain protein CPSFL1 (Hertle et al. 2020), VESICLE-INDUCING PROTEIN IN PLASTIDS1 (VIPP1) (Theis et al. 2019), VIPP2 (Theis et al. 2020), WHEAT KINASE START1 (Gou et al. 2015) and DYNAMIN-LIKE PROTEIN 2a (Gene Kang et al. 1998; Kim et al. 2001), although the evidence for the last one is controversial because its chloroplast location could not be confirmed (Arimura et al. 2004). In addition to in vitro evidence, using super-resolution immunolocalization, PI4P localized throughout Arabidopsis chloroplasts and partially colocalized with CPSFL1 (Hertle et al. 2020). Moreover, contrary to the complementation of the Arabidopsis cpsfl1 mutant with the native gene, CPSFL1 constructs harboring point mutations that reduced the interactions with phosphoinositides, rescued seedling lethality but did not fully recover the growth phenotype of the cpsfl1 mutant (Kim et al. 2022). This highlights the functional relevance that intra-plastidic phosphoinositide-mediated interactions may have in chloroplast biology.

PA and DAG can transiently recruit proteins and are key signaling molecules in eukaryotic heterotrophs (Toker 2005; Wang et al. 2006). In eukaryotic phototrophs, the signaling function of PA is well established (Testerink and Munnik 2011; Hou et al. 2016), whereas that of DAG is debated (Ha and Thompson 1991; Dong et al. 2012). The specific binding of CPSFL1 not only to PI4P but also to PA (Hertle et al. 2020) supports a role for PA in intra-chloroplast membrane dynamics (Fig. 4B). Although the evidence is just emerging, there is an ongoing quest in plastid biology for both lipid and protein players key to intra-plastidic interactions. Phosphoinositides, PA, and DAG are exciting candidates for the functionalization of photosynthetic membranes.

Vesicle formation and trafficking

Besides the on-site synthesis of lipids and proteins, the transport of lipids, proteins and pigments is essential for thylakoid biogenesis and maintenance. However, the specific molecular processes involved remain poorly understood. Several non-exclusive mechanisms have been proposed to enable lipid transport between the inner envelope (or the PM in cyanobacteria) and the thylakoids: (I) direct invagination of the inner membrane via nonvesicular lipid phase transition and/or protein-mediated processes, (II) direct membrane contact sites, (III) vesicle trafficking, and (IV) nonvesicular transport mediated by stromal proteins (Hurlock et al. 2014; Awai 2016; Mechela et al. 2019; Guéguen and Maréchal 2022).

Especially in cyanobacteria, the existence of an internal vesicle transport system is still under debate. In Microcoleus sp., vesicle-like structures have been observed that are considerably larger in diameter (150 to 300 nm) than chloroplast vesicles (30 to 70 nm) (Westphal et al. 2001; Nevo et al. 2012; Lindquist et al. 2016; Lindquist and Aronsson 2018). Although cyanobacterial genomes harbor homologs of vesicle-associated proteins, including TVP38 (TLG2-COMPARTMENT VESICLE PROTEIN of 38 kDa) (Jilly et al. 2018) and THF1 (THYLAKOID FORMATION 1), their molecular functions await clarification (Keller and Schneider 2013).

Vesicle formation in chloroplasts was observed by TEM over 60 years ago (Mühlethaler and Frey-Wyssling 1959). Bioinformatic studies suggested that a putative chloroplast vesicle transport machinery might have evolved via transfer to the chloroplast of components of the secretory pathway, which then underwent divergent evolution (Vothknecht and Soll 2005; Lindquist and Aronsson 2018). In line with this, several factors have been proposed to play functional roles during vesicle transport.

THF1 has been found at the chloroplast inner envelope, the stroma, and in TMs (Peltier et al. 2006) and interacts with LHCB proteins (Huang et al. 2013), a proposed cargo for chloroplast vesicles (Tanz et al. 2012; Khan et al. 2013). The chloroplast Sec14-domain protein CPSFL1 is essential for photoautotrophic growth and is required for vesicle formation at the inner envelope in Arabidopsis chloroplasts, probably through specific binding to PA and PI4P lipids (Fig. 4B; Hertle et al. 2020). CPSFL1 is also involved in the regulation of phytoene synthesis and carotenoid transport in the Chlamydomonas chloroplast (García-Cerdán et al. 2020).

Several plastid dynamin-like proteins (DLPs) have been linked to vesicle transport and maintenance of TMs (Backues et al. 2010; Jilly et al. 2018). These DLPs show a complex distribution in chloroplasts, including the inner envelope, stroma, and TM (Peltier et al. 2006). The chloroplast-localized Rab GTPase mutant cpraba5e (Table 1) is characterized by irregularly shaped chloroplasts, suggesting its involvement in vesicle transport and thylakoid biogenesis in Arabidopsis (Karim et al. 2014). The GTPase SECRETION-ASSOCIATED RAS1 (CPSAR1) is a well-studied yeast cytosolic vesicle transport protein that is localized to the chloroplast inner envelope and stroma. Arabidopsis cpsar1 mutants are embryo lethal, owing to a lack of thylakoids. Accordingly, roles in the initial steps of vesicle formation and thylakoid biogenesis have been proposed for CPSAR1 (Garcia et al. 2010). Another small GTPase and dynamin-related membrane-remodeling protein is FUZZY ONIONS (FZO)-like (FZL), which is found in both green algae and vascular plants. In Chlamydomonas, FZL is involved in thylakoid fusion and optimal pyrenoid function (Findinier et al. 2019; Fauser et al. 2022). Loss of FZL in Arabidopsis results in a disorganized thylakoid structure with reduced uniformity of grana lamellae, inflation of grana margins, and accumulation of vesicles, all of which support its role in membrane fusion and TM biogenesis (Gao et al. 2006).

VIPP1, a key player for TM integrity

VIPP1, also known as IM30 (Kroll et al. 2001; Westphal et al. 2001), was originally proposed to be involved in vesicle formation, although this role remains controversial. In any case, VIPP1 is essential for the biogenesis and maintenance of TMs in all organisms that are capable of oxygenic photosynthesis (Kroll et al. 2001; Westphal et al. 2001; Nordhues et al. 2012; Zhang et al. 2012). Recent cryo-EM structures revealed that VIPP1 and its bacterial ancestor Phage shock protein A (PspA) are members of the ancient ESCRT-III (ENDOSOMAL SORTING COMPLEX REQUIRED FOR TRANSPORT) family of membrane remodeling proteins (Gupta et al. 2021; Junglas et al. 2021; Liu et al. 2021). There are 8 different ESCRT-III proteins in humans (Pfitzner et al. 2021), while cyanobacteria and chloroplasts contain only a single VIPP1 type, of which paralogs like VIPP2 (Table 1) exist in some algae (Theis et al. 2020). Despite low sequence conservation, VIPP1/PspA and ESCRT-III proteins share a common core structure made up of a coiled-coil hairpin motif formed by helices α1 and α2. In the so-called open conformation, helices α2 and α3 merge (Fig. 5A). Flexibility in this open conformation is mediated by 3 hinge regions, located at the C terminus of helix α2 (shoulder), between helices α3 and α4 (elbow), and between helices α4 and α5 (wrist), respectively (Nguyen et al. 2020; Gupta et al. 2021; Liu et al. 2021).

Figure 5.

Structural features of VIPP1. A) Structure of Synechocystis 6803 VIPP1 (PDB 7O3W). The designation of the α-helices follows the nomenclature of ESCRT-III proteins. Sequences beyond α5 are not resolved. Hydrophobic residues in helix α0/AHa are shown in yellow, basic ones in blue, acidic ones in red. A pairwise structural alignment between a VIPP1 subunit of chain E in the Synechocystis VIPP1 ring and an AlphaFold structure prediction of VPL2 are shown in the inset (RSMD = 1.74 Å, TM score = 0.35). The alignment was done on RCSB.org ( Berman et al. 2000). B) Filament of 4 interwoven VIPP1 subunits showing the interaction of α5 of subunit “i” with the closed end of the α1-α2 hairpin of subunit “i + 3“. C)Synechocystis VIPP1 ring with C14 rotational symmetry and 6 layers. All images were created with Mol* (Sehnal et al. 2021).

In the open conformation, VIPP1/PspA and ESCRT-III proteins polymerize into filaments, which share the common feature that helix α5 of subunit “i” passes behind 3 to 4 neighboring subunits and binds the closed end of the helix α1-α2 hairpin of subunit “i + 3/4” (Fig. 5B). In vitro, VIPP1 filaments form rings that stack into basket-like structures (Fig. 5C) (Gupta et al. 2021; Liu et al. 2021), as well as helical rods that can engulf membranes (Theis et al. 2019; Naskar et al. 2023). Furthermore, they can form spirals and planar sheets on supported lipid bilayers enriched in negatively charged lipids (Naskar et al. 2023). In contrast to other ESCRT-III proteins, VIPP1/PspA appears not to form co-polymers. Although Chlamydomonas VIPP1 was found to interact with its paralog VIPP2, which forms helical rods shaped like those of VIPP1, it appears more likely that they form hetero-oligomers (Theis et al. 2020). Interestingly, VIPP1 PROXIMITY LABELING 2 (VPL2) was recently discovered in the immediate vicinity of VIPP1 and VIPP2 in C. reinhardtii (Kreis et al. 2023; Yilmazer et al. 2023). VPL2 has the same hairpin motif that is characteristic of ESCRT-III family members but lacks helix α0 and conserved structural motifs beyond helix α2 (Fig. 5A). VPL2 might intercalate into VIPP1/2 filaments to alter the structure of rings and rods and/or to functionalize them for interactions with other proteins.

In vivo, ESCRT-III polymerization is initiated by targeting factors (e.g. ESCRT-0) that recruit assembly factors (e.g. ESCRT-I and II) (Vietri et al. 2020). ESCRT-III polymers are disassembled by the hexameric AAA ATPase Vps4 (Monroe et al. 2017). There is also evidence for the existence of a homolog of an ESCRT-II protein called VIPP1 ASSOCIATED PROTEIN 1 in Arabidopsis and VPL3 in Chlamydomonas (Yilmazer et al. 2023). Assembly and disassembly of VIPP1 oligomers is facilitated by the stromal HSP70 chaperone system, and a specific J-domain co-chaperone (CDJ2) for VIPP1 binding is present in chloroplasts (Liu et al. 2005, 2007). Moreover, in contrast to ESCRT-III polymers, VIPP1/PspA bind and hydrolyze ATP and GTP, which in Synechocystis VIPP1 is mediated by a nucleotide-binding pocket formed by 3 subunits in 2 layers of a basket (Ohnishi et al. 2018; Gupta et al. 2021). Whether this activity plays a role in VIPP1/PspA (dis)assembly dynamics remains to be elucidated.

ESCRT-III proteins bind to lipids with anionic head groups. Binding is mediated via basic amino-acid residues in helix α1 or helices α2/3 and is sometimes supported by helix α0 serving as a membrane anchor (Buchkovich et al. 2013; Nguyen et al. 2020). In contrast, membrane binding by VIPP1/PspA depends only on helix α0 (or amphipathic helix a, AHa) and forms columns in the lumen of baskets and helical rods (Fig. 5C) (Gupta et al. 2021; Naskar et al. 2023). Two membrane properties mediate membrane binding of VIPP1/PspA: stored-curvature elastic (SCE) stress and anionic lipids (McDonald et al. 2015, 2017; Theis et al. 2019; Naskar et al. 2023). SCE stress generates hydrophobic cavities in the membrane (known as lipid-packing defects) into which the hydrophobic face of the AHa inserts. Anionic lipids interact with basic amino-acid residues in the polar face of the AHa (Fig. 5A). The importance of the hydrophobic face in the AHa of VIPP1/PspA is reflected by the almost complete abolition of membrane binding when valine 11 is mutated to glutamate (Jovanovic et al. 2014; McDonald et al. 2017; Gupta et al. 2021). The importance of interactions of basic residues in the polar face of the AHa with anionic lipids is underlined by the fact that the number of membrane interactions of VIPP1/PspA AHas increases with increasing anionic lipid content (McDonald et al. 2017). Chlamydomonas VIPP1 rods cannot engulf liposomes containing 5% PG as anionic lipid but are able to engulf liposomes containing 5% PI4P (Theis et al. 2019), presumably because the latter has 3 negative charges, while PG has only 1 (Fig. 3B).

VIPP1-coated membrane tubules that connect thylakoids to the chloroplast envelope have been observed in Chlamydomonas, suggesting a role for VIPP1 in mediating lipid exchange between the 2 membranes (Gupta et al. 2021) (Fig. 3E). The reduced numbers of such connections in vipp1 knockdown plants might explain why their thylakoids were located in a limited area of the chloroplasts and why thylakoids were completely absent in vipp1 knockout plants (Zhang et al. 2012). A role for VIPP1 in connecting envelopes with thylakoids is compatible with in vitro observations of VIPP1 rings that link liposomes and with VIPP1's ability to mediate liposome fusion (Hennig et al. 2015; Liu et al. 2021).

Four lines of evidence support a role of VIPP1 in the biogenesis of TM protein complexes. First, Arabidopsis, Chlamydomonas, and cyanobacterial vipp1 knockdown lines display reduced levels of at least one of the major TM protein complexes (Kroll et al. 2001; Fuhrmann et al. 2009; Nordhues et al. 2012; Zhang et al. 2014). Secondly, VIPP1 enhances protein export via the twin-arginine transport pathway (DeLisa et al. 2004; Lo and Theg 2012). Thirdly, recombinant VIPP1 stimulates the formation of a D1 insertion intermediate and was found to form a complex with cpSecY, Alb3, and cpFtsY in an in vitro reconstituted system (Walter et al. 2015). Finally, VIPP1 has been found in association with biogenesis factors in several TM complexes (Kreis et al. 2023). One can only speculate about the underlying mechanisms. Like eisosomes (Malinsky et al. 2013), VIPP1 rings might bind to membranes and induce membrane invaginations to create lipid nanodomains enriched in specific lipid types. These molecules in turn might be required for the functions of integrases, transporters, and channels, and/or might serve as structural lipids during the biogenesis and repair of TM complexes (Theis and Schroda 2016).

There is strong evidence for a role of VIPP1 in protecting chloroplast membranes from stress. Thus, Chlamydomonas vipp1 knockdown mutants and Synechocystis lines expressing VIPP1 with negatively charged residues on the hydrophobic side of their respective AHas (Fig. 5A) show aberrant thylakoid conversion zones under ambient conditions, and swollen thylakoids after exposure to high light levels (Nordhues et al. 2012; Gupta et al. 2021). Moreover, in Arabidopsis vipp1 knockdown mutants, chloroplasts swell under hypotonic stress conditions (Zhang et al. 2012). In contrast, overexpression of VIPP1 enhances the recovery of photosynthetic capacity after heat stress (Zhang et al. 2016a) and complements a chloroplast-swelling phenotype in the Arabidopsis nyc1 stay-green mutant (Zhang et al. 2016b). VIPP1 has recently been shown to recognize membranes that are experiencing SCE stress and to form planar sheets of VIPP1 filaments at such sites within minutes (Junglas et al. 2020; Naskar et al. 2023). These sheets most likely prevent proton leakage. Moreover, damaged membranes could be repaired via buds forming at central rings within these sheets involving membrane squeezing and fusion processes (Naskar et al. 2023). In Chlamydomonas, membrane stress sensed by VIPP1/2 appears to trigger a retrograde signal resulting among others in the expression of genes encoding VIPP1/2, small heat shock proteins, the protein disaggregase CLPB3, and the protease DEG1C (Perlaza et al. 2019; Theis et al. 2020). This response suggests that membrane damage often is caused by the accumulation of misassembled or aggregated proteins in membranes that must be removed (McDonald et al. 2015; Schroda and Devitry 2023).

Biogenic regions of TMs

Thylakoid biogenesis involves the coordinated assembly of lipids, protein complexes, and pigments. Perhaps the most intensively discussed issue is whether the synthesis and assembly of photosynthetic complexes takes place in biogenic regions that are dedicated to the production of thylakoids, and if so, where such areas are located within cyanobacteria and chloroplasts. With the identification and subcellular localization of several assembly factors and assembly intermediates, cytological landmarks for the biogenesis process have now been pinpointed. Indeed, accumulating evidence strongly supports the idea that specialized biogenic substructures exist, where biosynthetic pathways for the required proteins, pigments, and probably lipids converge and are coordinated and accelerated by substrate-channeling effects (Nickelsen and Rengstl 2013). One advantage of biogenic microcompartments lies in the locally increased concentrations of cofactors, for example, metals, which would shift chemical equilibria toward assembly and could exclude potentially toxic intermediates from photosynthetically active TM areas (Rast et al. 2015).

In line with this, in the cyanobacterium G. violaceus (which is devoid of thylakoids), segregation of biogenic and photosynthetic patches in its PM has been observed (Rexroth et al. 2011). In Synechocystis 6803, TCTs (see Fig. 2) have been postulated to serve as regions for PSII biogenesis, based on several observations. First, TCTs exhibit low Chl autofluorescence and might therefore be nonphotosynthetic membranous subcompartments that would allow for the segregation of photosynthetic from biogenic processes (Ostermeier et al. 2022). Secondly, TCTs have been shown to be free of PBSs (a marker for mature PSII complexes), and ribosomes are found in their immediate vicinity; both observations are compatible with co-translational protein insertion events (Rast et al. 2019). Furthermore, the PSII-related assembly factor PratA has been localized to TCTs (Stengel et al. 2012). Finally, the inactivation of factors that shape TCTs, such as CurT and AncM, affects PSII biogenesis, but interestingly, PSI is not compromised under these conditions (Heinz et al. 2016; Ostermeier et al. 2022).

On the other hand, analyses of the spatial distribution of mRNAs encoding photosynthetic complexes, in particular PSII, strongly suggest that the synthesis of PSII subunits is localized to the cytoplasmic surface of the innermost thylakoid sheet in Synechocystis 6803 (Mahbub et al. 2020). The fact that this thylakoid area is also decorated with numerous membrane-associated ribosomes supports this assumption (Rast et al. 2019). Whether the sites of membrane protein synthesis also represent sites of photosynthetic complex assembly in Synechocystis 6803 remains to be seen. However, a recent study in Synechococcus 7942 revealed that the synthesis of the PSII subunits D1 and D2 occurs in spatially separated punctate loci, which implies that freshly made subunits must diffuse along the membrane to find their assembly partners (Mullineaux and Mahbub 2023).

In chloroplasts, biogenic functions have been assigned to distinct TM subcompartments. The best characterized case is the T-zone concept in Chlamydomonas (Sun et al. 2019). This predicts that the synthesis and assembly of photosynthetic complexes occurs in specialized T-(translation) patches close to the pyrenoid in the basal region of the chloroplast (Fig. 1J). Synthesis and assembly of PSII subunits was localized to T-zones under conditions that favored the synthesis of new TMs (Uniacke and Zerges 2007; Schottkowski et al. 2012). More recently, chlorophyll synthesis and the biogenesis of PSI have both been associated with T-zones, based on the localization of PSI subunits/assembly factors and light-dependent POR, respectively (Sun et al. 2019). These findings confirm that T-zones represent hubs upon which anabolic pathways converge to orchestrate the biogenesis of photosynthetic complexes, including the incorporation of the necessary organic and probably also inorganic cofactors. Interestingly, cryo-EM images of C. reinhardtii cells reveal invaginations of the inner envelope membrane near the T-zone (Engel et al. 2015). Moreover, in the vicinity of T-zone regions, TOC and TIC components of the chloroplast import machinery appear to accumulate upon mild illumination (Schottkowski et al. 2012). This suggests an assembly path that integrates the synthesis of all photosystem subunits encoded by the nuclear genome in the cytoplasm with protein import into the chloroplast, chloroplast genome-encoded subunit synthesis, and chlorophyll assembly in Chlamydomonas (White and Hoober 1994; Sun et al. 2019).

In plant chloroplasts, grana margins appear to function as biogenic regions, which correspond to TCTs and T-zones in cyanobacteria and algae, respectively (Figs. 1, 2, C and D; Bussi et al. 2019). In general, stromal TMs, as well as the margins of grana thylakoids, are accessible to ribosomes. It is therefore assumed that the assembly of photosynthetic complexes occurs there. Indeed, membrane fractionation experiments have revealed that PSII intermediates and assembly factors are enriched in stromal thylakoids, which supports the idea that these represent biogenic and/or PSII repair regions (Suorsa et al. 2014; Koochak et al. 2019). In contrast to the case in green algae and bryophytes, plastids in vascular plants undergo an ontogenetic development that begins with undifferentiated proplastids (Fig. 1M). The work of Charuvi et al. (2012) showed that the 3D structure of small thylakoid networks connected to the envelope found in the shoot apical meristem of Arabidopsis already exhibit the basic structural features of mature thylakoid networks. Depending on the developmental programs and the environmental conditions, these can become specialized for specific sub-types of plastids with distinct functions, for example, chloroplasts, chromoplasts, or etioplasts (for review see Choi et al. 2021). Moreover, especially chloroplasts in C4 plants differ considerably in a tissue-specific manner (mesophyll and bundle sheath), meaning that the development of thylakoids in chloroplasts varies in different tissues/developmental statuses, and some plastids are even devoid of thylakoids (Majeran and van Wijk 2009; Kirchhoff et al. 2013). These transitions are accompanied by radical alterations of the membrane systems, as best exemplified by the above-mentioned case of the CURT1-dependent transformation of PLBs into TMs in etioplasts (see the section on determinants of TM shape). Future work will shed further light on the underlying molecular mechanisms that mediate these dramatic adaptations of thylakoid architectures.

Authors contributions

All authors wrote the article. M.O., A.G.-H. and M.S. designed the figures. M.O. and V.J.C.H. designed the table.

Funding

Work of the authors was financially supported by the Deutsche Forschungsgemeinschaft in the context of the earlier Research Unit FOR2092 (JN and MS) and the Collaborative Research Centre TRR175 [projects A06 (JN) and C02 (MS)] as well as the DFG project Ni390/13-1 (JN). We thank P. Hardy for critical reading of the manuscript.

Data availability

There is no new research data in this review.

Dive Curated Terms

The following phenotypic, genotypic, and functional terms are of significance to the work described in this paper:

• FZL Gramene: AT1G03160

• FZL Araport: AT1G03160

• CLPB3 Gramene: At5g15450

• CLPB3 Araport: At5g15450

• POR Gramene: AT4G39920

• POR Araport: AT4G39920

• STN7 Gramene: AT1G68830

• STN7 Araport: AT1G68830

• STN8 Gramene: AT5G01920

• STN8 Araport: AT5G01920

• DGD1 Gramene: AT3G11670

• DGD1 Araport: AT3G11670

• MGD1 Gramene: AT4G31780

• MGD1 Araport: AT4G31780

• HSP70 Gramene: AT3G12580

• HSP70 Araport: AT3G12580

• KEA3 Gramene: AT4G04850

• KEA3 Araport: AT4G04850

• LPP CHEBI: CHEBI:138807