MD approaches have affirmed and expanded upon the evidential importance of the central pore for substrate gating. Pioneering work came from modeling the major BMC-H protein (Yang et al., 2020) of the propanediol metabolosome in Salmonella enterica, PduA, using biased potentials (e.g. umbrella sampling, metadynamics) (Park et al., 2017). Calculations indicated a more favorable passage of 1,2-propanediol compared to the higher free-energy barrier for propionaldehyde (Park et al., 2017), confirming earlier assumptions that some intermediates, like the reactive propionaldehyde, are selectively sequestered within the BMC lumen to prevent toxicity (Sampson and Bobik, 2008) or carbon loss due to volatility (Penrod and Roth, 2006). The higher free-energy barrier does not wholly block substrates, like propionaldehyde, from diffusing but does impede the process kinetically. The mechanism was attributed simply to the higher hydrogen-bonding capacity of the central pore constriction, lined with a serine residue (S40) (Crowley et al., 2010), for 1,2-propanediol over propionaldehyde due to the extra hydroxyl group which acts as an additional hydrogen bond donor. Effectively, the serine-lined pore is a better binding site for 1,2-propanediol than propionaldehyde which promotes the formers passage so long as the binding is not overly strong (Bauer and Nadler, 2006). Such features from the Pdu BMC can reasonably be applied to carboxysomes, due to the high sequence and structural conservation between all BMC shell proteins (Melnicki et al., 2021). Indeed, experimental work in the propanediol metabolosome has further attributed the pore-lining S40 of PduA as critical to influencing permeation (Chowdhury et al., 2015) This result was later confirmed via simulation and experimentation on CcmK2 of the Synechocystis sp. PCC 6803 β-carboxysome with its S39 pore (Faulkner et al., 2020).

The residues lining the central constriction alone are not sufficient to explain substrate permeation across ɑ- and β-carboxysome shells as many BMC-H proteins common encode residues like serine and glycine at the pore constriction. Surface electrostatic density around the central pore has also been observed to be an effective mechanism of attracting/repelling substrates in several studies focusing on BMC-H proteins (Faulkner et al., 2020; Mahinthichaichan et al., 2018). In all cases, the pore-adjacent concave surface exhibits a high extent of positive charge effectively turning the outward facing surface into an electrostatic funnel ( Figure 2B ). This may explain the lower free-energy profile for bicarbonate just outside of the pore within the concavity, essentially attracting the negatively charged bicarbonate anions while conversely impeding the passage of neutral CO₂ and O₂ from reaching the interior [or, in the case of CO₂, escaping once bicarbonate is converted luminally by carbonic anhydrase (Cai et al., 2009)] ( Figure 2B ). These works importantly provided the groundwork to understand not just CO₂ concentration within the shell, but also the inhibition of O₂ diffusion too. MD observations that O₂ diffusion is biophysically impeded is substantiated by the presence of bioinformatically identified glycl-radical enzyme associated microcompartments (GRMs) (Zarzycki et al., 2015). GRMs have been speculated to help extend the range of environments that glycyl-radical enzymes can reasonably act in, as oxygen exposure inactivates these enzymes (Zarzycki et al., 2017; Zhang et al., 2001). This notion is further supported by engineered carboxysome shells that package oxygen-sensitive hydrogenases can impart enhanced activity in an aerobic environment (Li et al., 2020).

The specific residues (corresponding to R11, K36, and the pore S39 in CcmK2) responsible for forming a substrate barrier are largely conserved among CcmK2 proteins (Faulkner et al., 2020) and emphasize the importance of both the pore and concave surface overall in substrate gating ( Figure 2B ). The alignment of β-strands proximal to the pore on the concave surface (L31 to K36 in CcmK2) also expose their backbone amines and contribute to this effect. We note, however, that these specifics will differ between shell proteins. For example, BMC-H even among the same class will differ in surface electrostatics (Schmidt-Dannert et al., 2018) and therefore should not be taken as a one-size-fits-all rule (i.e. concave surface always being positive to the same degree) but instead as another layer of consideration when assessing permeation. While Rubisco and the carboxysome shell may not be able to sufficiently differentiate between CO₂ and O₂ (Poudel et al., 2020), the outer shell can enhance the passage bicarbonate and, in combination with the encapsulated carbonic anhydrase, locally increase the CO₂ concentration around Rubisco. MD simulations have helped explain the molecular basis for carbon concentration in the carboxysomal CCM and will be an essential methodology to predictively modify the shell for augmented substrate specificities moving forward.

The proclivity of BMC shell pores to bind anionic species is not limited to solely bicarbonate. Other anions, like sulfate, have been found in crystal structures of a broad range of BMC-H proteins including CcmK1, CsoS1A, and EutM to list a few (Tanaka et al., 2008; Tsai et al., 2007; Takenoya et al., 2010). MD have also revealed structural aspects of ion coordination. For instance, chloride ions have been observed to occupy pore-adjacent positions (Faulkner et al., 2020; Mahinthichaichan et al., 2018) and coordinate with either backbone amides or basic residues, such as arginine. Similar results were recently found for the metabolosome BMC-H PduA, where chloride ions were found to coordinate with the backbone amide of the pore-lining S40 (Trettel et al., 2023). This study also found that chloride itself also acts competitively with 1,2-propanediol, the intended substrate, for pore access thereby hindering permeation rates. Altogether, simulation data from both carboxysome and metabolosome models both agree on the ability of anions to coordinate with and occupy shell protein pores via non-specific backbone interactions (Faulkner et al., 2020; Trettel et al., 2023). While only currently reported for metabolosome shells, this suggests that ion coordination may be a widespread phenomenon which can also regulate permeation events in carboxysomes. The role of other physiologically relevant anions, such as inorganic phosphate which can regulate Rubisco activity (Marcus and Gurevitz, 2000), has yet to be explored in this context.

Initial attempts at modeling BMC assembly were inspired by carboxysomes where evidence has been found for both concomitant and core-first assembly pathways (Perlmutter et al., 2016) as observed in both ɑ- and β- lineages (Kerfeld and Melnicki, 2016) ( Figure 3 ). The principle differentiating factor was the relative strength of attraction cargo had for other cargo, where weaker interactions led to ‘one-step’ or concomitant assembly (observed in ɑ-carboxysomes) ( Figures 3A, B ) while stronger interactions led to ‘two-step’ or core-first (observed in β-carboxysomes) ( Figures 3A, C ). Specifically, for ɑ-carboxysomes, modeling (Mahalik et al., 2016) and atomic-force microscopy (Sutter et al., 2016; Garcia-Alles et al., 2017) have both suggested that shell facets form by nucleation, which can further provide an area to locally concentrate cargo (Oltrogge et al., 2020) and nucleate ɑ-carboxysome formation, since cargo-cargo interactions are predicted to not be strong enough drivers on their own (Perlmutter et al., 2016) ( Figure 3B ). β-carboxysome cargo (Rubisco and CcmM M35) in two-step assembly modes coalesce strongly enough on their own without the need of a shell-templated trigger ( Figure 3C ). Interestingly, these simulations predicted that cargo would become organized into concentric layers, observed prior in both ɑ- (Iancu et al., 2007; Shively et al., 1973; Schmid et al., 2006) and β-carboxysomes (Kaneko et al., 2006; Iancu et al., 2005). Paracrystalline order was not a prerequisite for forming complete particles in these simulations and in fact would inhibit budding ( Figure 3A ). These observations have held up to additional recent higher-resolution experimental scrutiny, where Rubisco in both ɑ- and β-carboxysomes is now understood to assemble into concentric layers (Evans et al., 2023; Faulkner et al., 2017; Ni et al., 2022) when the internal concentration is sufficiently high (Metskas et al., 2022). While just the first of many follow-up studies, Perlmutter et al.’s above work demonstrated the utility of computational modeling to understand carboxysome assembly. However, the system employed at the time, albeit elegant, only investigated one shell geometry (T = 3), one BMC-H and BMC-P, and one cargo. This initial model has been greatly expanded to include considerations like the impact of cargo packaging on BMC size (Mohajerani and Hagan, 2018), the role of scaffolds (Mohajerani et al., 2021), and even multiple cargos (Tsidilkovski et al., 2022) on microcompartment size, assembly pathway, and packaging efficiency ( Figure 4A ).

In the context of microcompartment size, dynamical simulations reveal that shells packaged with cargo, generally, tend to be larger than non-packaged shells (Mohajerani and Hagan, 2018). Further, these simulations showed that BMC size also correlates with assembly pathways where core-first assembly modes, dominated kinetically by relatively stronger cargo-cargo interactions, led to larger particles and up to 5-fold more packaged cargo than concomitant modes (Mohajerani and Hagan, 2018). These results similarly apply to multi-component systems, where assembly pathway is primarily delineated by the sum of the cargo interaction strengths and the strength of self-cargo interactions ( Figure 4B ) can dictate packaging stoichiometry (Tsidilkovski et al., 2022). This has been likewise observed in real BMC systems where β-carboxysomes, which follow a core-first assembly pathway, tend to be larger than their ɑ- counterparts (Whitehead et al., 2014). Empty, synthetic structural models reported thus far are also always far smaller than native BMCs (<40 nm diameter) (Sutter et al., 2017; Ni et al., 2023; Tan et al., 2021; Sutter et al., 2019a; Kalnins et al., 2020; Greber et al., 2019; Sutter et al., 2019b).

Dynamical simulations have also revealed that shell components, while not the predominant factor, can also influence final morphology. For instance, simply increasing the ratio of shell proteins to cargo can lead to overnucleation (Mohajerani and Hagan, 2018) and thus smaller particles ( Figure 4B ). This has also been found in simulations which assume shell proteins demonstrate no spontaneous curvature of their own [motivated by atomic force microscopy studies on shell subunits (Sutter et al., 2016; Garcia-Alles et al., 2017)] and can essentially trap a growing cargo droplet out of equilibrium (Rotskoff and Geissler, 2018). This, however, may depend on the system of study as shell proteins have been observed to form sheets, nanotubes, and empty icosahedra among other morphologies, sometimes within the same sample (Ferlez et al., 2023), without the need of cargo templating to induce curvature (Ferlez et al., 2023; Trettel and Winkler, 2023; Uddin et al., 2018; Hagen AR. et al., 2018; Noël et al., 2015). The presence of excess pentamers or stronger pentamer-hexamer interactions can likewise lead to more pentamer insertion and thus overnucleation into smaller particles (Mohajerani and Hagan, 2018) ( Figure 4B ). This latter point is interesting since many BMC operons encode for more than one BMC-P (Sutter et al., 2021). BMC-P proteins appear to play different roles in different contexts, where in some metabolosomes they can directly influence BMC morphology (Mills et al., 2022) and in others they are completely dispensable and can be added exogenously to “cap” the icosahedron (Sutter et al., 2019b; Hagen A. et al., 2018; Kirst et al., 2022). Observations from simulations further emphasize the importance of studying the effects of BMC-P and how they can influence morphology, packaging efficiency and permeability.

The above studies ascribed the connection of homogenous cargo to end morphology. However, the models used may be more applicable to metabolosomes, where cargo directly interacts with the shell (Fan et al., 2010; Aussignargues et al., 2015), than carboxysomes, where scaffolds act as an intermediary connecting the shell and cargo domains (Oltrogge et al., 2020; Wang et al., 2019). Accordingly, Mohajerani et al. have also conducted a study, motivated by ɑ-carboxysomes specifically, on the role of a CsoS2-inspired scaffold proteins in BMC assembly (Mohajerani et al., 2021) ( Figure 4A ). Scaffolds proteins, as a type of cargo themselves, can potentially affect shell size and assembly pathway in a much more programmable manner than typical cargo due to their modular nature (Chaijarasphong et al., 2016) ( Figure 1C ). Simulations parametrized these physical aspects by defining the length of the CsoS2-inspired scaffold (L), the length of the cargo binding domain [L_c, with longer L_c meaning more cargo binding sites, akin to more CsoS2 N-terminal domains (NTDs)] and the fractional length of the cargo binding domain (f_sc = L_c/L) as shown in Figure 4A . Importantly, simulations reveal that there is a critical interplay between the total length of the scaffold and its valency with cargo. By fixing the overall scaffold length (L) and increasing L_c (and therefore f_sc), simulations showed that cargo packaging likewise increases. Moreover, increasing f_sc, analogous to the number of cargo binding sites, transitioned systems from a one-step to a two-step assembly pathway ( Figure 4B ) where two-step pathways are again associated with more cargo packaging (Mohajerani et al., 2021). Similarly to cargo packaging alone (Mohajerani and Hagan, 2018), as the scaffold is itself a type of cargo, physically longer scaffolds also generally result in larger shells to a point as they increase volume requirements (Mohajerani et al., 2021). These simulations are supported by work in the model H. neapolitanus α-carboxysome that demonstrates a requirement for a minimal threshold of NTDs in CsoS2 to be met to achieve carboxysome formation (Oltrogge et al., 2020). Further, more recent work by Oltrogge and colleagues likewise agree that increasing CsoS2 length by increasing the number of middle region (MR) repeats leads to larger α-carboxysomes (Oltrogge et al., 2023). However, they ascribe this phenomenon to the MR repeats of CsoS2 stabilizing the low-curvature regions (i.e. the facets) of the carboxysome shell, enabling their extension, while Mohajerani et al. argue for the need to meet increased volume requirements. We note that these arguments are not mutually exclusive.

MD simulations of shell permeation to date have focused on a small subset of model BMC-H. While impactful, future permeation studies may wish to sample a greater diversity of BMC-H to develop a deeper understanding of the natural biophysical diversity shell proteins can accommodate. For instance, sampling a wider array of carboxysomal BMC-H may highlight subtle differences that influence bicarbonate, O₂, and 3-PGA diffusion. Similar methodologies can and should be applied towards describing permeation in mixed heterohexamer systems, like those reported for CcmK3/K4 (Sommer et al., 2019; Garcia-Alles et al., 2019) or purely synthetic systems with the potential for asymmetric pore designs (Česle et al., 2023) that may further regulate substrate diffusion in ways that homo-hexameric BMC-H cannot. The various classes of BMC-T should also be considered to better grasp their hypothesized connection to substrate gating (Klein et al., 2009; Tanaka et al., 2010). Simulation scale also needs to be accounted for, and future studies may wish to engage with physiologically relevant systems with multiple components like the small synthetically-derived BMC shells (Sutter et al., 2017; Ni et al., 2023; Tan et al., 2021; Sutter et al., 2019a; Kalnins et al., 2020; Greber et al., 2019; Sutter et al., 2019b), as a proxy for larger native-like systems. Investigating more complex shells will progress our understanding of how chemical gradients, a physiologically critical component, behave and evolve within a BMC context. For instance, differences in density and packing of Rubisco within carboxysomes (Kaneko et al., 2006; Ni et al., 2022) may result in CO₂/O₂ gradients proportional to the enzymes’ proximity to the shell. Detailed permeation studies could discern the packaging attributes within BMCs that would result in more efficient catalytic properties in engineered architectures. Simulating whole-BMC shell models can also limit pore-centric bias and explore if flux exists in non-porous areas such as the hexamer-hexamer interfaces or corner junctions where three hexamers meet. Similarly, permeability studies can be expanded to study the diffusion of a wider swath of metabolites and cofactors through BMC shell structures. Current research suggests that these cofactors, like NAD(P)H, are maintained as private pools that are internally recycled and do not appreciably diffuse through the shell barrier (Huseby and Roth, 2013; Cheng et al., 2012). Regardless, novel BMCs may be sought to transform metabolites far larger than those found in current model systems. MD simulations of permeation, therefore, will continue to facilitate rapid in silico prototyping of permeation through protein shells for altered substrate specificities or enhanced carbon concentration within the carboxysome lumen.

Specific structural components of the shell, such as the extended C-termini on many BMC-H, should also be addressed. While typically ignored due to missing crystallographic data, these termini can now be predicted and integrated into computational models thanks to emerging computational tools. These outward-facing, flexible/disordered (Faulkner et al., 2020) termini have been implicated in functions such as assembly (Trettel et al., 2022; Klein et al., 2009) like in viruses (Xue et al., 2014), but some data suggests they also reach into the concavity of adjacent subunits (Trettel et al., 2022) which may impact permeation or fine-tune assembly in environmentally responsive ways.

The collective knowledge on carboxysome systems continually expands and reinvents our understandings of these complex systems. Incorporating simulations to complement emerging experimental insights will lead to more meaningful outputs to inform design choices. For instance, future modeling may wish to explore evidence-informed shell-cargo interaction sites that form from predominantly (i) the edge-edge interaction surface of two adjoining shell proteins (Ni et al., 2023) and, in some cases, (ii) interactions with specific interior-oriented domains such as the N-terminus of the PduB BMC-T (Trettel et al., 2022; Lehman et al., 2017; Kennedy et al., 2022). This is further underpinned by the multitude of different shell proteins BMCs can encode and their synthetic interchangeability (Cai et al., 2015b; Slininger Lee et al., 2017) which certainly influence shell-shell (including curvature) and shell-cargo/scaffold interactions. For example, many BMCs encode for BMC-T proteins where every other edge may be better attuned for specific shell interactions on adjacent subunits (Trettel et al., 2022; Waltmann et al., 2023) and influence factors like shell curvature and/or shell-cargo interactions. Many others bioinformatically identified BMC loci entirely lack these factors for unknown reasons (Sutter et al., 2021).

The luminal organization of Rubisco is also now known to differ between related carboxysomes and may be tied to overall carboxysome activity, For instance, Halothiobacillus α-carboxysomes exhibit ~2-fold higher activity than Cyanobium α-carboxysomes and have different modes of Rubisco organization (Ni et al., 2022). Future simulations may have an opportunity to explain how these subtle structural differences arise (i.e. Rubisco surface charge difference, CsoS2 binding affinity, internal Rubisco concentration), ascribe functional consequences, and reveal how to program desired internal conformations.

Shell-focused assembly simulations can also help better define and explain the mechanisms behind the varied supramolecular structures BMC shell proteins can form in vivo and in vitro, such as nanotubes, for designer protein scaffolds (Young et al., 2017). Recent work suggests that BMC-H curvature trends can be inferred by their crystal structural arrangements (Garcia-Alles et al., 2023) and that these trends can be modulated rationally with computationally-informed amino acid substitutions (Li et al., 2021). However, factors like buffer/environmental composition (Faulkner et al., 2019), shell protein class and stoichiometry, and protein disorder undoubtedly also factor into supramolecular, and native-like, structures in unknown ways. In particular, the disordered termini many BMC-H proteins carry, predominantly on their outward facing C-terminus (Sutter et al., 2017; Trettel et al., 2022) have been speculated to fine-tune both shell-shell and shell-cargo interactions (Fan et al., 2012). Further, currently described simulation systems may already be attuned to ascribe the role of the multiple pentamers BMCs can encode for by tuning their relative stoichiometry and interaction strengths. Computational studies will undoubtedly continue to address these considerations and many more for custom carbon-fixing scaffolds.

Assembly-focused simulations teach us that assembly pathway is chiefly governed by cargo interaction strengths ( Figure 3A ) while final morphology is determined by both cargo and shell contributions ( Figure 4B ). In terms of assembly pathway, stronger cargo-cargo (including scaffold) interactions or higher cargo stoichiometries are typically associated with two-step assembly pathways that lead to larger shells with more cargo (Mohajerani and Hagan, 2018; Mohajerani et al., 2021; Tsidilkovski et al., 2022) ( Figure 4B ). Conversely, weaker cargo-cargo interactions or higher shell stoichiometries are associated with one-step assembly pathways and smaller shells (Mohajerani and Hagan, 2018; Mohajerani et al., 2021; Tsidilkovski et al., 2022) ( Figure 4B ). These findings carry direct carboxysome design implications related to assembly kinetics that manifest physically in the forms of (i) expression system design and (ii) scaffold design.

Cargo and shell constructs can be designed in both a continuous synthetic operon or discontinuously into different plasmids with different modes of induction for testing (Lee and Tullman-Ercek, 2017). Single-vector/operon designs have been successful using a variety of induction approaches (Bonacci et al., 2012; Graf et al., 2018; Flamholz et al., 2020; Jiang et al., 2023). Notably, similar strategies also result in morphologically and functionally sound carboxysomes when genomically integrated and expressed in plants (Chen et al., 2023; Long et al., 2018). Double-vector systems, which independently express shell and cargo components, have also been described (Jiang et al., 2023; Jakobson et al., 2016; Wagner et al., 2017) although they do need to be tuned and timed appropriately (Lee and Tullman-Ercek, 2017; Jakobson et al., 2016; Nichols et al., 2019) likely due to kinetic effects of aggregation described by simulations. In one case, researchers redesigned a carboxysome for hydrogen production by serially inducing hydrogenase cargo followed by a β-carboxysome shell (Li et al., 2020). BMCs with concomitant assembly pathways, like those commonly employed for heterologous α-carboxysomes formation, may benefit from single vector designs which promote co-expression of both shell and cargo components under native-like controls (i.e. ribosomal binding sites). Similarly, two-step pathways may be promoted by a well-tuned cargo-preaggregation step proceeded by shell expression. Researchers should consider the kinetics of interactions and expression to prevent off-target assemblies.

Scaffold choice and design is also an emerging route for modification. In ɑ-carboxysomes, modifying CsoS2 and the ratios of CsoS2A and CsoS2B (analogous to f_sc in simulations) or the number of NTD/MR repeats (Oltrogge et al., 2023) are approachable routes to alter morphology and Rubisco packaging for CCM augmentation. Similarly, in β-carboxysomes, modifications of CcmM and CcmN may also be sufficient routes for modification. However, both classes of scaffolds act as specific adaptors between the Rubisco cargo and the shell domains and therefore cannot coalesce a more diverse range of cargo by themselves without extensive modification. Heterologous encapsulation and assembly methods may wish to rely on carboxysome-inspired fusions (Gonzalez-Esquer et al., 2015) or metabolosome EPs which trigger both shell-cargo (Fan et al., 2010) and cargo-cargo (Lawrence et al., 2014). Cargo fused with metabolosome EPs may act more akin to the assembly models produced in several assembly simulation works to date (Perlmutter et al., 2016; Mohajerani and Hagan, 2018; Tsidilkovski et al., 2022).

One bottleneck with biomass productivity lies in the connection between photosynthetic efficiency and carbon fixation. To alleviate these bottlenecks, some groups have installed various CCM components into plant chloroplasts including Nicotiana benthamiana (Lin et al., 2014), Rhodosprillum rubrum (Long et al., 2018), and Nicotiana tabacum (Chen et al., 2023). These studies have been able to generate carboxysomes nearly structurally and catalytically equivalent to native carboxysomes and support photosynthesis (Chen et al., 2023). Further additional factors like the incorporation of bicarbonate transporters, removal of the stromal carbonic anhydrase, and including Rubisco activates (Chen et al., 2022) may be needed to significantly enhance growth under ambient CO₂ conditions. A deeper fundamental understanding of carboxysome assembly offered by computational simulations may assist in full implementation of cyanobacterial CCMs into C3 plants.

Simulations have played a critical role in exploring the physical phenomena that underpin carboxysome assembly. However, many conclusions remain explored at low resolution, partially due to the technics used. More investment in multi-resolution calculations is required for incorporating high accuracy detailed mechanisms at commensurate computational investment. Such methodologies may only be possible after exercising high fidelity energy landscape reconstruction based on accelerated MD or AI assisted methodologies for the fast interconversion between low resolution models (e.g. supra coarse-grained) and fully atomic detailed structures. Only this approach would be able to lead to a more fine-tuned and robust rational carboxysome manipulation.