We calculated the carbonate chemistry parameters from reconstructed historic ocean pH and DIC (Ridgwell, 2005) using the standard software CO2sys, version 1.01 (Lewis and Wallace, 1998). Calculated ocean carbonate concentrations and reconstructed calcium concentrations (Stanley and Hardie, 1998) were used to determine the calcite saturation state. Atmospheric pCO₂ was obtained from previous reconstructions (Berner and Kothavala, 2001). We compared the carbonate chemistry parameters with the fossil record of pelagic calcifying nannoplankton (Bown et al., 2004) to identify the environmental conditions under which intracellular calcification emerged (Figure 1). The primary objective of this analysis is to compare, in qualitative terms, the reconstructed carbonate chemistry of the ocean with the appearance of calcareous nannoplankton. A precise, quantitative reconstruction of the carbonate chemistry throughout the Phanerozoic goes beyond the scope of our study. However, from a visual inspection, the data we compiled using literature sources and CO2sys are consistent with recent reconstructions (Witkowski et al., 2018; Montañez et al., 2016).

To examine the evolutionary history of calcification-related proteins, we identified proteins across lineages and created gene trees of relevant protein families. We downloaded protein databases from the National Center for Biotechnology Information (NCBI) genome browser for taxa represented in the phylogeny of Read et al. (2013), Supplementary Table S1. To account for differences in annotation completeness in each protein database, we used BLAST to identify homologs of Gephyrocapsa huxleyi proteins in each taxon. We searched the G. huxleyi protein database for annotated sequences in each protein family implicated in calcification (Benner et al., 2013) to use as query sequences (Supplementary Table S2). We then used pBLAST to identify all sequences in the protein database that matched these query sequences with probabilities less than e-6. After discarding duplicates, we aligned the BLAST results for each protein family in Geneious 11.1.3, (Kearse et al., 2012), http://www.geneious.com, using the Geneious global alignment with cost matrix Blosum62, gap open penalty 12, and gap extension penalty 3. We created gene trees using Jukes-Cantor genetic distance module and the UPGMA tree building method in Geneious. We then inspected trees for phylogenetic representation and patterns of protein family expansion. Because the BLAST output results for calcium channels included both cation channels and calmodulin, we split these sequences into two subtrees for better visualization of the relationships among calmodulin-related proteins. Supplementary Figures S1 – S7 show the resulting gene trees.

We calculated the energy required for calcium transport into Golgi-derived vesicles as part of the energy invested to maintain calcium homeostasis. In our calculations, we neglect the costs associated to the transport of golgi vesicles because these vesicles are transported by both calcifying and non-calcifying cells. We also neglect the costs required for the production of polysaccharides and for exocytosis, the process for the secretion of polysaccharides by haptophytes, because also these costs are inccurred by non-calcifying cells and, thus, do not constitute and additional energy requirement for calcifying cells. In order to achieve intracellular calcium homeostasis, total calcium transport out of the cytoplasm has to equal the passive influx via calcium channels. The passive, diffusive uptake via the calcium channels is driven by the difference between the calcium concentration at the cell surface (Ca _surf ) and the calcium concentration in the cytoplasm (Ca _cyt ). Thus, knowing the calcium gradient across the plasma membrane, the energy required to achieve calcium homeostasis (E _ATP ) can be calculated by E ATP = P Ca ⋅ A cell ⋅ C a surf − C a cyt ⋅ Ψ A T P C a , (1)

where P _Ca is the apparent membrane permeability per cell surface area (A _cell ) and Ψ A T P C a is the amount of ATP required per calcium ion transported. The consumption of one ATP molecule for transporting two calcium ions through the membrane is a functional property of the reaction loop of the Ca-ATPase (Toyoshima and Inesi, 2004; Winther et al., 2013) and is independent if transport is over the vesicle or plasma membrane. Without net calcium influx into the cell, Ca _surf is equal to the calcium concentration in the bulk medium (Ca _out ). However, fluxes of calcium into intracellular compartments, such as vesicles, create a concentration gradient across the cell membrane with Ca _surf < Ca _out . With the total flux of calcium into the vesicles (Q _vesicle ), and thus into the cell, the integration of the spherical diffusion equation in the cell environment gives the following concentration at the cell surface: C a surf = C a out − Q vesicle 4 ⋅ π ⋅ r cell ⋅ D Ca , (2)

where D _Ca is the diffusion coefficient of calcium in seawater and r _cell is the cell radius. Here, we assume that the transport of calcium into vesicles represents a fixed fraction of total calcium transport (f _v ) as compared to the transport over the plasma membrane (1 − f _v ), i.e.,: Q vesicle = f v ⋅ P Ca ⋅ A cell ⋅ C a surf − C a cyt . (3)

In other words, we assume that while calcium is rushing in through the ion channels, ATPases pump calcium back through the cell membrane into the extracellular medium but at the same time also into the vesicles. The volume inside the vesicle, however, can be regarded as outside of the cytoplasm from a thermodynamical point of view. In order to maintain calcium homeostasis, the flux into the cytoplasm has to be balanced by the fluxes out of the cytoplasm. Inserting the expression for Q _vesicle into Eq. 2 and subtracting Ca _cyt from both sides of Eq. 2 results in an implicit equation for the concentration difference Ca _surf - Ca _cyt , which has the solution: C a surf − C a cyt = C a out − C a cyt 1 + f v ⋅ P Ca ⋅ A cell 4 ⋅ π ⋅ r cell ⋅ D C a (4)

Replacing in Eq. 1 the concentration difference Ca _surf − Ca _cyt with Eq. 4 gives E _ATP as a function of the relative fraction of calcium transport into the calcifying vesicles: E ATP = P Ca ⋅ A cell ⋅ C a out − C a cyt 1 + f v ⋅ P Ca ⋅ A cell 4 ⋅ π ⋅ r cell ⋅ D C a ⋅ Ψ A T P C a (5)

This equation shows that the higher f _v , the lower the absolute requirements for cellular energy. In order to quantify how much the cellular energy consumption can potentially be reduced by intracellular calcification, we calculate the energy requirements to maintain calcium homeostasis as a function of f _v from 0 (non-calcifying cell) to 1 (Supplementary Figure S8). The coccolithophore Gephyrocapsa huxleyi produces approximately 1 coccolith per hour (Paasche, 1962) and one coccolith contains around 22 fmol of CaCO₃ (Young and Ziveri, 2000). This results in a calcification flux (Q _vesicle ) of about 6.11 ⋅ 10^–18 mol s⁻¹ (see: Holtz et al., 2013). Since the permeability of the plasma membrane for calcium ions is not known in coccolithophores, we estimate a permeability that is based on the conductance of calcium channels (Lux and Nagy, 1981) and measured densities of calcium channels in cell membranes of snail axons (Krishtal et al., 1981; Tsien, 1983). All the parameter values used to calculate the extra energy required by intracellular calcification are reported in Supplementary Table S3. We provide the numerical code for calculating the energy required to maintain calcium homeostasis by single-celled calcifying nannoplankton as open-source software (https://github.com/systemsecologygroup/Calcisphere) so that it can be used, modified, and redistributed freely.

The energy for the Ca²⁺ pumping ATPase is provided by the hydrolysis of ATP at a high ATP/ADPxPi ratio in the cytosol with a free energy gain, ΔG, of 15 kcal/mol (Meldolesi and Pozzan, 1998). Cooperative binding of 2 Ca²⁺ per ATPase implies a Ca²⁺/ATP coupling ratio of 2 (Inesi and Tadini-Buoninsegni, 2014). We assume an electroneutral Ca²⁺/2H⁺ co-transport for the Ca²⁺ pump and a physiological pH at both sides of the Ca²⁺ pump. If all of the free energy available from ATP hydrolysis is stored in the concentration gradient of Ca²⁺ between the cytosol and its environment, i.e.,: ΔG = 2RT ⋅ ln (Ca _surf /Ca _cyt ), the calcium gradient can be up to 10⁵. If the Ca-ATPases work near this thermodynamic limit, the cytosolic Ca²⁺ concentration of 0.1 μmol L⁻¹ and the gradient of 10⁵ yields a Ca²⁺ concentration of 10 mM at the cell surface, which is, in fact, exactly the Ca²⁺ concentration in today’s ocean (Langer et al., 2006). Ca²⁺ transport near the thermodynamic limit would have the advantages that: 1) pumping Ca²⁺ from the cytosol to the environment proceeds with a minimum dissipation of energy in ATP consumption, and 2) a steep Ca²⁺ gradient provides fast influx in cell signaling. We therefore postulate that a constant Ca²⁺-gradient close to the thermodynamic limit represents a part of calcium homeostasis. Moreover, the Ca²⁺-gradient across the plasma membrane should be quite independent of the bulk Ca²⁺ concentration in seawater.

Assuming that the Ca²⁺-gradient across the plasma membrane is maintained at 10⁵ by the Ca²⁺-ATPase, the intracellular calcium concentration will respond to variations in extracellular calcium concentration as experienced during Earth’s history (Stanley and Hardie, 1998), Figure 1. The intracellular concentration of calcium (Ca _cyt ), however, generates a constraint on the intracellular phosphate concentration (PO _cyt ) because calcium and phosphate precipitate to Ca₃(PO₄)₂ when the ion product exceeds the solubility product K s p C a 3 ( P O 4 ) 2 = PO cyt ⋅ Ca cyt . We can therefore calculate the maximum possible intracellular phosphate concentration for any intracellular calcium concentration as follows: P O cyt = K s p C a 3 ( P O 4 ) 2 C a cyt . (6)

Calcium ions play important roles in the physiology of eukaryotic cells, including control of DNA synthesis, chromosomal configuration, and signal transduction (Marcum et al., 1978; Boynton et al., 1980; Krishtal et al., 1981; Lux and Nagy, 1981; Tsien, 1983; Poenie et al., 1985; Carafoli, 1987; Steinhardt and Alderton, 1988; Hepler, 1994; Sanders et al., 2002; Verret et al., 2010). Calcium ions enter the cell passively via channels (Gussone et al., 2006; Verret et al., 2010) that allow for a diffusive influx driven by a steep concentration gradient across the plasma membrane (Langer et al., 2006). Excess intracellular Ca²⁺ is then actively removed by calcium-binding proteins (e.g., calmodulin) and transport proteins (Ca²⁺-ATPase and Na⁺/Ca²⁺ exchanger) with different calcium affinities and maximum transport rates (Carafoli, 1987). The Ca²⁺ concentration inside the cell has to be tightly regulated (calcium homeostasis) at around 0.1 μmol L⁻¹ (Brownlee et al., 1995) for optimal cell functioning (Carafoli, 1987) and to avoid cell poisoning (Simkiss, 1977; Orrenius et al., 1989; Müller et al., 2015).

The same proteins that maintain intracellular calcium homeostasis in non-calcifying cells (Figure 2A) are implicated in intracellular calcification (Figure 2B). To determine whether an expansion of protein families coincides with the emergence of intracellular calcification, we compared all gene families involved in the precipitation of calcium carbonate (von Dassow et al., 2009) in a large number of eukaryotes (Read et al., 2013). These genes include Ca²⁺ transporting ATPases, HCO 3 − transporters, Ca²⁺ channels, calmodulin, carbonic anhydrase, Ca²⁺/H⁺ exchangers, Na⁺/Ca²⁺ exchangers, and vacuolar H⁺-ATPases. The presence of calcium transport proteins in all eukaryotes clearly indicates that these genes are not unique to intracellular calcification. We found no evidence that diversification of these gene families could explain the appearance of nannoplankton calcification (Supplementary Figures S1 – S7).

The most significant difference between calcifying and non-calcifying cells is that calcium carbonate precipitates inside Golgi-derived vesicles. The vesicles fuse with the plasma membrane and the calcium carbonate is extruded outside the cell where it remains attached to the extracellular organic matrix. The same vesicular pathway is used in non-calcifying cells to produce new membrane material and to transport proteins to the plasma membrane (Figure 3). All ion transporters and channels that help establish calcium homeostasis in non-calcifying cells, therefore, reach the plasma membrane via Golgi-derived transport vesicles (Pfeffer and Rothmann, 1987). However, the Ca²⁺ transporting ATPase initially occurs in its auto-inhibited form and becomes activated by the presence of calmodulin (Tidow et al., 2012) at the plasma membrane. We thus argue that the emergence of intracellular calcification was triggered by the activation of calcium pumps during the passage of transport vesicles from the Golgi apparatus to the plasma membrane, following the accumulation of calcium in the transport vesicles. This explanation is consistent with the simple structure of sub-micron scaled calcite crystals of proto-calcispheres found in previous studies and attributed to calcareous dinoflagellates (Janofske, 1992; Preto et al., 2013b; Dal Corso et al., 2021). The activation of the calcium transport inside the Golgi apparatus would later lead to the formation of more elaborated coccolith-like platelets due to the specific endomembranous structure of the Golgi apparatus (Figure 3). This is also in line with the finding that inhibition of the cytoskeleton and microtubule formation induces malformations in modern coccolith (Langer et al., 2010).

Such a change in timing of activation would not be detected in gene sequence comparison, as they require protein functional assays. We conclude, however, that all components necessary for the emergence of intracellular calcification were already present in non-calcifying ancestor cells and that no major modifications of cellular physiology and structure were required to induce intracellular calcification. An alternative and speculative hypothesis to the activation of calcium pumps could be that those pumps were already active and that the precipitation of calcium carbonate in the vesicles was prevented by inhibitors, which were subsequently lost thus leading to the emergence of intracellular calcification. But also in this alternative case, the emergence of intracellular calcification would not require major innovations in cellular structure, which is our main argument here.

Since ion transport requires energy, a previous study (Monteiro et al., 2016) assumed that intracellular calcification is energetically costly, thus implying the presence of associated benefits to overcome natural selection. But because non-calcifying cells also export calcium from their cytoplasm, the energetic costs of calcification have to be quantified by comparing the costs required for ion transport into vesicles in calcifying cells versus those required by ion transport across the plasma membrane (to maintain calcium homeostasis) in both calcifying and non-calcifying cells. We therefore assessed the energetic costs of intracellular calcification by calculating the total energy required by cellular calcium transport to maintain calcium homeostasis. Our calculations indicate that the precipitation of CaCO₃ inside the vesicle removes calcium from the solution and generates a sink that depletes calcium around the cell, Eq. 2. Consequently, the gradient across the plasma membrane that drives the diffusive calcium influx through the channel is reduced with respect to non-calcifying cells, Eq. 4, thus fewer ions have to be transported out of the cytoplasm in order to maintain calcium homeostasis, Eq. 5. Since the energy required for active calcium transport out of a calcifying cell is negligibly lower than that required in non-calcifying cells (Supplementary Figure S8), we conclude that extra energy is not needed to transport calcium for intracellular calcification.

However, our energy calculations also show that the reduction in energy requirements is very small because the calcification flux is slow compared to extracellular calcium diffusion so that no significant diffusive boundary layer develops at the cell surface. If all calcium ions were transported into the vesicles and no calcium were exported through the plasma membrane, f _v  = 1.0 in Eq. 5, the energy requirements would only be 0.16 percent less than if no calcium were transported into vesicles and all ions were exported through the plasma membrane, f _v  = 0.0 in Eq. 5.

Besides calcium ions, also carbonate ions are needed in the vesicle to sustain calcification. The transport of carbon could thus be another process requiring energy. However, the transport of hydrogen ions that is coupled to calcium transport creates an alkaline medium inside the calcifying vesicle that shifts the equilibrium of the carbonate system towards carbonate ions and lowers the CO₂ concentration (Holtz et al., 2013; Holtz et al., 2015). CO₂ can, therefore, passively diffuse through the vesicle membrane to replenish the carbonate ions consumed by calcification, thus sustaining calcification at moderate rates (see Figure 3 and Table 4 in Holtz et al., 2013) without additional metabolic costs. Additional carbon transport into the vesicles may only be required to sustain higher calcification rates, as observed in some coccolithophore species (e.g., G. huxleyi) that appeared later in the geological records. Data in support of these arguments can be found in Paasche (1962) and in Young and Ziveri (2000). Since the bicarbonate concentration in the calcifying vesicle is higher than in the cytoplasm, bicarbonate would then need to be transported actively under consumption of metabolic energy (Holtz et al., 2015). However, the bicarbonate gradient across the vesicle membrane is much smaller than the calcium gradient and CO₂ can freely diffuse across the membrane, thereby functioning as a major source of carbon to the calcifying vesicle (Holtz et al., 2015) and, thus, preventing additional energetic requirements for bicarbonate transport.

Because of its low metabolic requirements, intracellular calcification does not require secondary evolutionary benefits to emerge. Yet, a number of hypotheses in relation to potential benefits of intracellular calcification have been proposed to explain both emergence and success of coccolithophores. These hypotheses include the potential enhancement of photosynthesis via the removal of carbonate ions (Nielsen, 1966; Paasche, 2002), protection against photoinhibition, advantages in hydrodynamic control, and protection against viruses, bacteria, and grazing (Monteiro et al., 2016). However, enhanced calcification does not enhance photosynthesis (Herfort et al., 2004; Trimborn et al., 2007) and non-calcifying coccolithophore cells can grow even faster than calcifying cells (Rost and Riebesell, 2004). Non-calcifying strains appear also not affected by photo-inhibition at high irradiances (Nanninga and Tyrrell, 1996). Additionally, coccolithophores use non-calcifying haploid stages to escape viral infections of calcified diploid cells (Frada et al., 2008), also questioning the relevance of calcification for viral protection. All these cases cast doubts on the assumption that calcification presents relevant physiological advantages.

Previous experiments (Müller et al., 2015) showed that calcifying cells can grow faster than non-calcifying cells when exposed to elevated calcium concentrations, indicating that intracellular calcification may protect against calcium poisoning (Simkiss, 1977). Our calculations show that the Ca²⁺-ATPase can transport ions against a gradient of up to 10⁵ (see Supplementary Material S1). In non-calcifying cells, an increase of extracellular Ca²⁺ concentrations yields an increase of intracellular Ca²⁺ (Figure 2C). Since intracellular phosphate concentrations are usually around 4–5 mM (Auesukaree et al., 2004), cells operate at calcium and phosphate concentrations just below the solubility product of Ca₃(PO₄)₂ (Chow, 2001). Increasing intracellular calcium concentrations, therefore, lead to precipitation with phosphate due to the low solubility of Ca₃(PO₄)₂ (Pytkowicz and Kester, 1967; Chow, 2001). The decrease of cellular phosphate impedes cell growth in non-calcifying cells (Figure 2C). In calcifying cells, calcium transport into intracellular vesicles can still occur under elevated extracellular calcium concentrations because the calcium concentration inside the vesicle (ranging from 0.5 to 10 mM Holtz et al., 2013; Holtz et al., 2015) required for the precipitation of calcium carbonate is lower than outside the cell. The calcium gradient at the vesicle membrane therefore permits Ca²⁺-ATPase activity and allows maintenance of calcium homeostasis even under elevated extracellular calcium concentrations (Figure 2D). Cell division of calcifying cells remains unaffected (Müller et al., 2015).

Another benefit proposed for intracellular calcification is protection against zooplankton grazing (Jaya et al., 2016; Monteiro et al., 2016). However, coccolithophores are grazed by copepods without any significant sign of saturation of ingestion rates at increasing coccolithophore densities (Harris, 1994). This indicates that calcification and thus mechanical stability does not necessarily protect against copepod grazing. Calcification can, however, have an indirect negative effect on zooplankton grazing. Culture experiments (Harvey et al., 2015) showed that ingestion of calcifying cells reduces the growth rate of the grazers (Figure 2F), whereas ingestion of non-calcifying cells leads to an incrase in zooplankton growth rate (Figure 2E). A reduction of coccolithophore grazing mortality would thus provide calcifying cells with a competitive advantage against non-calcifying cells (Monteiro et al., 2016). Though not relevant for the emergence of pelagic nannoplankton calcification, the potential chemical protection against grazing may have contributed to its ecological success.