Metabolic modelling of the C3-CAM continuum revealed the establishment of a starch/sugar-malate cycle in CAM evolution

The evolution of Crassulacean acid metabolism (CAM) is thought to be along a C3-CAM continuum including multiple variations of CAM such as CAM cycling and CAM idling. Here, we applied large-scale constraint-based modelling to investigate the metabolism and energetics of plants operating in C3, CAM, CAM cycling and CAM idling. Our modelling results suggested that CAM cycling and CAM idling could be potential evolutionary intermediates in CAM evolution by establishing a starch/sugar-malate cycle. Our model analysis showed that by varying CO2 exchange during the light period, as a proxy of stomatal conductance, there exists a C3-CAM continuum with gradual metabolic changes, supporting the notion that evolution of CAM from C3 could occur solely through incremental changes in metabolic fluxes. Along the C3-CAM continuum, our model predicted changes in metabolic fluxes not only through the starch/sugar-malate cycle that is involved in CAM photosynthetic CO2 fixation but also other metabolic processes including the mitochondrial electron transport chain and the tricarboxylate acid cycle at night. These predictions could guide engineering efforts in introducing CAM into C3 crops for improved water use efficiency.


32
Crassulacean acid metabolism (CAM) photosynthetic CO 2 fixation is an evolutionary 33 descendant of C 3 photosynthesis, which is known to have evolved independently multiple times 34 in at least 35 plant families comprising about 6% of flowering plant species (Winter and Smith, 35 1996a;Silvera et al., 2010). CAM is an adaptation of photosynthetic CO 2 fixation typically 36 associated to limited water availability (Cushman and Borland, 2002). By closing their stomata 37 during the light period and fixing atmospheric and/or respiratory carbon dioxide (CO 2 ) 38 exclusively in the dark period, CAM allows plants to use water more efficiently while fixing 39 carbon for growth. The engineering of CAM into C 3 crops has been suggested as a possible 40 strategy to meet the demands on agriculture for food, feed, fibre, and fuels, without exacerbating 41 the pressures on arable land area due to climate change (Borland et al., 2014). However, as a 42 carbon-concentrating mechanism, CAM is thought to be more metabolically expensive than C 3 43 (Winter and Smith, 1996b), which suggests that transferring a CAM pathway into C 3 crops 44 would incur a crop yield penalty. To investigate the energetics of C 3 and CAM, large-scale 45 metabolic models were applied which showed that engineering CAM into C 3 plants does not 46 impose a significant energetic penalty given the reduction in photorespiration from the carbon-47 Model simulations with flux balance analysis 89 Based on the constraints and objective function stated in the Results section, parsimonious flux 90 balance analysis (pFBA) was performed using scobra (https://github.com/mauriceccy/scobra), an 91 extension of cobrapy (Ebrahim et al., 2013). The scripts for running the simulations in this study 92 can be found in Supplementary File S2. In this study, we primarily reported the results from the 93 pFBA simulations (Supplementary Tables S1, S2, S3). The conclusions made based on the pFBA 94 results for C 3 , CAM, CAM cycling and CAM idling were confirmed using flux variability 95 analysis (Mahadevan and Schilling, 2003) applied on the primary objective (Supplementary 96

99
Predicted metabolic fluxes of C 3 , CAM, CAM cycling and CAM idling 100 In this study, we simulated the metabolism of leaves undergoing C 3 , CAM, CAM cycling 101 and CAM idling using a recently published core metabolic model of Arabidopsis which was used 102 to model C 3 and CAM plants (Shameer et al., 2018). Minor modifications of the model were 103 outlined in the Materials and Methods section. The constraints for simulating the core metabolic 104 functions of mature leaves, namely export of sucrose and amino acids into the phloem and 105 cellular maintenance, were set based on the values in Shameer et al. (2018). All simulations, 106 except CAM idling, were constrained to have a phloem export rate of 0.259 μ mol m −2 s −1 based 107 on the value of C 3 plants in Shameer et al. (2018). The set of constraints for modelling the four 108 different modes of photosynthesis are summarised in Table 1. The primary objective function of 109 minimising photon demand was used throughout this study, which allows us to study the 110 metabolic efficiencies of the different modes of photosynthesis. Parsimonious flux balance 111 analysis (pFBA), i.e. minimisation of absolute sum of fluxes, was applied as a secondary 112 objective to eliminate substrate cycles. Results from pFBA were confirmed using flux variability 113 analysis (Mahadevan and Schilling, 2003) performed on the primary objective. 114 The model predictions of C 3 and CAM were very similar to that in Shameer et al. (2018) 115 given the similarities in the constraints used. Without any constraints on malate decarboxylation 116 enzyme and carbohydrate storage, the model predicted net carbon fixation during the light period 117 6 in the C 3 flux prediction, whereas in CAM carbon was fixed in the dark period with 118 phosphoenolpyruvate carboxykinase (PEPCK) being the main predicted route for malate 119 decarboxylation. Starch was predicted to be the main carbohydrate storage in both C 3 and CAM. 120 These results are consistent with the findings in Shameer et al. (2018) Table S6). In this study, we mostly presented the results from simulations 131 with starch as the carbohydrate storage. Similar conclusions can be made for using sugar as the 132 carbohydrate storage. The core set of metabolic fluxes for C 3 , CAM, CAM cycling and CAM 133 idling with starch as the carbohydrate storage is depicted in Figure 1. 134

CAM cycling
135 Similar to C 3 plants, CAM cycling fixes carbon in the light period. CAM cycling is 136 characterised by its closed stomata in the dark period with refixation of respiratory CO 2 and a 137 small diel organic acid flux (Sipes and Ting, 1985;Cushman, 2001, Winter, 2019. To model 138 CAM cycling, we applied the C 3 constraints with an additional constraint of setting CO 2 and O 2 139 exchange at night to zero to simulate the closure of the stomata (Table 1). This resulted in a flux 140 distribution that resembled a weak version of CAM, with nocturnal malate accumulation and 141 increased light period starch accumulation ( Figure 1C). Phosphoenolpyruvate carboxylase 142 (PEPC) was predicted to be active only at night in CAM cycling for CO 2 refixation, in contrast to 143 C 3 where PEPC was only active during the light period (Supplementary Table S2). Another 144 major difference between CAM cycling and C 3 is malate accumulation. While C 3 was predicted 145 to have a very small amount of malate accumulation during the light period, CAM cycling was 146 predicted to have substantial amount of nocturnal malate accumulation (~20% of the amount of 147 malate accumulation in CAM) (Figure 1; Supplementary Table S2), which is consistent with 7 known behaviour of CAM cycling (Ting, 1985;Cushman, 2001). The nocturnal malate 149 accumulation and respiratory CO 2 refixation via PEPC under the CAM cycling scenario were 150 accompanied by changes in fluxes in other parts of metabolism. Malate decarboxylation during 151 the light period was predicted to be active in CAM cycling but not in C 3 (Figure 1). There was a 152 larger flux through gluconeogenesis to convert malate into starch in the light period, which led to 153 more starch accumulation during the light period in CAM cycling compared to C 3 (Figure 1; 154 Supplementary Table S2). Given that CAM cycling has a higher starch accumulation in the light 155 period, it was predicted to have a larger glycolytic flux in the dark to convert starch into 156 phosphoenolpyruvate (PEP) for CO 2 refixation, compared to C 3 (Figure 1; Supplementary Table  157 S2). The activities of most of the other reactions at night were similar in CAM cycling and in C 3, 158 with CAM cycling having a slightly higher flux through the tricarboxylic acid (TCA) cycle and 159 the mitochondrial electron transport chain (ETC), presumably to produce extra ATP for 160 transporting malate into the vacuole for storage at night. 161

CAM idling
162 CAM idling is characterised by the lack of diel gaseous exchange and a small continued 163 diel fluctuation in the organic acids level because of internally recycled CO 2 (Sipes andTing, 164 1985, Winter, 2019). It is usually an adaptation in water-stressed plants, which results in the 165 closure of stomata for the whole 24-hour cycle. To model this, the CO 2 and O 2 exchange during 166 the light and the dark periods were constrained to carry zero flux ( Table 1). Given that there is no 167 CO 2 exchange, we assumed that there is no net carbon fixation, hence phloem export was 168 constrained to zero for CAM idling. 169 The primary metabolic demand for plants in CAM idling is cellular maintenance. The 170 model predicted a starch-malate cycle where starch accumulated in the light period is 171 metabolised in the dark period mainly through glycolysis and the oxidative pentose phosphate 172 pathway (OPPP) to produce ATP and NADPH for maintenance processes ( Figure 1D). While the 173 majority of PEP was used as precursor for carbon refixation by PEPC, a significant proportion of 174 PEP was predicted to be metabolised further through the TCA cycle to feed the mitochondrial 175 ETC for ATP synthesis ( Figure 1D). Given that it is a closed system with respect to carbon, CO 2 176 produced in the OPPP and the TCA cycle is refixed by PEPC, which ultimately leads to the 177 accumulation of malate in the dark. In the light period, PEP from malate decarboxylation was 8 recycled to produce starch via gluconeogenesis, while the CO 2 produced from malate 179 decarboxylation was refixed via the Calvin-Benson cycle similar to the scenario for CAM 180 ( Figure 1). With no net carbon import or export, the amount of carbon stored in starch in the light 181 period was predicted to be equalled to the amount of carbon storage in malate at night. The 182 starch-malate cycle was primarily driven by the energy from the light reactions of 183 photosynthesis, and it acted as a carbon neutral way of storing and transferring energy from the 184 light period to the dark period. Similar results were obtained when sucrose or fructan was used as 185 the sole carbohydrate storage instead of starch (Supplementary Table S6 The metabolic flux predictions of C 3 , CAM, CAM cycling and CAM idling were 191 compared to see how CAM cycling and CAM idling fit into the evolution of CAM from C 3 . 192 Table 2 summarises the predicted fluxes related to energetics and metabolic accumulation in the 193 four simulations. CAM idling was predicted to use the fewest photons, which was expected 194 given that it does not have the metabolic demand for exporting sucrose and amino acids into the 195 phloem. For the same metabolic demand, CAM requires more photons than C 3, as expected. It is 196 interesting to see that the photon demand for CAM cycling falls between C 3 and CAM. A similar 197 trend was observed for other fluxes related to energy metabolism including the ATP and 198 NADPH production by the photosynthetic light reactions and the ATP production by the 199 mitochondrial ATP synthase ( Table 2)  Predicting the metabolic transitions during C 3 -CAM evolution 219 The behaviour of diel CO 2 exchange is the main diagnostic indicator between C 3 and 220 CAM (Silvera et al., 2010). To model the potential metabolic transitions that could happen 221 during the evolution of CAM from C 3 , we varied the CO 2 uptake rate during the light period 222 from 13.12 μ mol m −2 s −1 (the predicted value for C 3 ) to 0 μ mol m −2 s −1 (which had the same 223 effect as gradually increasing nocturnal CO 2 uptake given the overall carbon balance). This 224 simulates the decrease in gaseous exchange during the light period by stomatal closure, hence a 225 similar constraint was set for light period oxygen exchange. As the stomata closes in the light 226 period, i.e. light period CO 2 uptake decreases, it was assumed that the proportion of ribulose-1,5-227 bisphosphate carboxylase/oxygenase (RuBisCO) flux going through the carboxylase reaction 228 increases linearly from 75% (carboxylase to oxygenase ratio of 3:1) to 83.74% (carboxylase to 229 oxygenase ratio of 5.15:1) to account for the reduction of photorespiration. All other constraints 230 remained the same as the C 3 and CAM simulations. This analysis simulates the closing of 231 stomata which decreases atmospheric CO 2 intake during the light period. The full results from 232 this simulation can be found in Supplementary Table S3. 233 Given that the metabolic demands remained constant throughout the analysis, a decrease 234 in CO 2 uptake in the light period led to a shift from C 3 to CAM photosynthesis with an increase 235 in flux through the starch-malate cycle including starch degradation, glycolysis, PEPC, and 236 malate accumulation at night, and malate decarboxylation and starch accumulation during the 237 10 light period (Figure 2A,B; Supplementary Table S3). Note that dark period CO 2 uptake increased 238 as light period CO 2 uptake decreased due to the carbon balance of the model in exporting a fixed 239 amount of sucrose and amino acids into the phloem. CAM cycling occurs at the point when dark 240 period CO 2 uptake is zero. 241 Despite the constrained decrease in RuBisCO oxygenase contribution as light period 242 CO 2 uptake decreased, the amount of energy (in terms of photons) required to sustain the same 243 metabolic demand increased by about 7% from C 3 to CAM ( Figure 2C) as extra energy is needed 244 to run the starch-malate cycle. This is correlated with the increase in flux through the 245 photosynthetic light reactions. Besides plastidial ATP synthesis, there was also an increase in 246 ATP synthesis by the mitochondrial ETC in the light period as the simulation shifted from C 3 to 247 CAM ( Figure 2D). The contribution of mitochondrial ATP synthesis increased from 18.2% in C 3 248 to 35.6% in CAM ( Figure 2E), which is likely to be related to the increase in NADH produced 249 during malate decarboxylation. In our simulations, the RuBisCO carboxylase flux was predicted 250 to be remain relatively constant while the total RuBisCO flux (carboxylase + oxygenase) 251 decreased from C 3 to CAM due to the decrease in RuBisCO oxygenase activity ( Figure 2F). 252 There were two major factors affecting RuBisCO carboxylase flux, i) refixation of 253 photorespiratory CO 2 , and ii) starch accumulation to support energy demand in the dark period. 254 In this case, the two factors counteract each other throughout the simulation where 255 photorespiration decreases and the energy demand for running the starch-malate cycle (mostly 256 for pumping malate into the vacuole) increases from C 3 to CAM. For the simulations with 257 sucrose or fructan as the sole carbohydrate storage, the model predicted an increase in RuBisCO 258 carboxylase flux from C 3 to CAM as the energy required for running the sugar-malate cycle is 259 higher than the starch-malate cycle (due to the cost of pumping sugars into the vacuole for 260 storage). 261 During the night, other than the increase in glycolytic flux as part of the starch-malate 262 cycle from C 3 to CAM, the model predicted an 87% increase in flux through the TCA cycle and 263 an 83% increase in flux through the mitochondrial ETC ( Figure 2G). This increase in 264 mitochondrial ATP synthesis was mostly used to support the ATP-dependent tonoplast proton 265 pump for the increasing nocturnal vacuolar malate accumulation. The cytosolic OPPP flux was 266 predicted to decrease by 30% in the night from C 3 to CAM (Figure 2h). This could be explained 267 by the increase in the TCA cycle flux which contributed to the production of NADPH in the 268 mitochondrion by the NADP-isocitrate dehydrogenase. This lessened the demand for the 269 production of cytosolic NADPH required to be shuttled into the mitochondrion for maintenance and are closed at night (Lüttge, 2004;Silvera et al., 2010, Winter, 2019. With these constraints, 277 our model predicted the known features of CAM cycling including the refixation of respiratory 278 CO 2 in the dark period, and a small amount of nocturnal malate accumulation (Cushman, 2001, 279 Winter, 2019. To support these metabolic behaviours, our model predicted the establishment of 280 a starch-malate cycle in CAM cycling, which included increased flux through malate 281 decarboxylation, gluconeogenesis and starch synthesis and accumulation during the light period, 282 and starch degradation and glycolysis during the dark period, when compared to C 3 plants. The 283 main metabolic advantage of CAM cycling over C 3 is its higher carbon conversion efficiency 284 when photosynthesis is limited by stomatal conductance in the light period, i.e. carbon limited. 285 Given the same metabolic outputs, CAM cycling was predicted to require 20% less external CO 2 286 compared to C 3 due to the refixation of nocturnal respiratory CO 2 . This comes with a minor cost 287 of 4.8% more photons and 1.6% more RuBisCO activity required, assuming that there is no 288 reduction in photorespiration, which could be affected by limiting stomatal conductance and 289 internal CO 2 generation from malate decarboxylation. Given an environment that limits stomatal 290 conductance in the light period, e.g. high temperature and drought, the evolution of CAM 291 cycling, together with the establishment of the starch/sugar-malate cycle, was predicted to be 292 advantageous in maximising carbon conversion efficiency. The metabolic activities of all 293 reactions in the starch-malate cycle in CAM cycling were predicted to be at an intermediate level 294 between C 3 and CAM. The same applies to other supporting reactions such as the TCA cycle in 295 the dark and the mitochondrial ETC during the light and dark periods. These findings suggest 296 12 that CAM cycling is likely to be a possible evolutionary step along the path to the evolution of 297

CAM. 298
As opposed to CAM cycling, CAM idling is thought of as a form of very strong CAM 299 (Lüttge, 2004, Winter, 2019. In CAM idling, stomata remain closed throughout the day and 300 night with small, sustained diel fluctuations in organic acids (Cushman, 2001;Silvera et al., 301 2010, Winter, 2019. By constraining our model with closed stomata in both the light and dark 302 periods, the model predicted the operation of the starch/sugar-malate cycle as the most energy 303 efficient way to sustain cellular activities. From an evolutionary perspective, if a plant often 304 experiences conditions that require the closure of stomata throughout day and night, such as long 305 periods of severe drought, the evolution of CAM idling would be advantageous for the plant to 306 stay alive. While the evolution of CAM through CAM cycling seems more likely given its 307 similarities to C 3 , it is not impossible that some lineages could establish the starch/sugar-malate 308 cycle through CAM idling. 309 310 Stomatal conductance as a determinant along the C 3 -CAM continuum 311 It has been proposed that CAM evolution occurs along a continuum from C 3 to CAM 312 (Silvera et al., 2010;Bräutigam et al., 2017). Our model analysis showed that by varying the CO 2 313 exchange in the light period, as a proxy for stomatal conductance, there existed a C 3 -CAM 314 continuum with gradual metabolic changes along the continuum (Figure 2). The key metabolic 315 changes included the processes in the starch/sugar-malate cycle, the TCA cycle at night, and the 316 chloroplastic and mitochondrial ETCs. The fact that a gradual continuum was predicted to be the 317 most energetically favourable way to adapt to a change in stomatal conductance suggests that the 318 fitness landscape between C 3 and CAM is a smooth one. Given our results, it is not surprising to 319 see many facultative CAM plants which can easily switch between C 3 and CAM. Based on our 320 model predictions, it is hypothesised that we could find plants anywhere on the C 3 -CAM 321 continuum. A prime example is CAM cycling which falls within the C 3 -CAM continuum at the 322 point when nocturnal CO 2 exchange is zero. Given the flexibility shown in facultative CAM 323 plants and our results on the C 3 -CAM continuum, it could be possible to find existing plants or 324 engineer new plants that can switch not only between C 3 and CAM but also at different points on 325 the continuum depending on the environmental conditions. 326 13 327

328
Using a core metabolic model of Arabidopsis, we were able to model the metabolic 329 behaviours of CAM, CAM cycling and CAM idling by changing a few simple constraints on 330 gaseous exchange and phloem export. Our results showed that CAM cycling and CAM idling 331 could potentially be evolutionary intermediates on the path to CAM evolution by establishing an 332 intermediate flux through the starch/sugar-malate cycle. By varying the light period CO 2 333 exchange as a proxy for stomatal conductance, the model predicted a continuum from C 3 to 334 CAM with gradual metabolic changes. Besides the insights gained in CAM evolution, the results 335 from this study are informative to guide engineering efforts aiming to introduce CAM into C 3 336 crops by identifying the metabolic changes required to convert C 3 to CAM. In additional to the 337 starch/sugar-malate cycle involved in CAM photosynthesis, our model showed that the fluxes of 338 other metabolic processes, including the TCA cycle and the mitochondrial ETC, need to be 339 altered from C 3 to optimise CAM. 340 341 Tables   Table 1: Sets of constraints for modelling C 3 , CAM, CAM cycling and CAM idling. Phloem export rate was set based on the predicted value of C 3 plants in Shameer et al. (2018). RuBisCO carboxylase:oxygenase ratio was set to 3:1 when stomata is opened, and 5.15:1 when stomata is closed based on Shameer et al. (2018).   The photorespiratory pathway is shown in chloroplast for simplicity, which in reality spans multiple compartments. Flux from 3-phosphoglycerate to PEP was taken as the flux for glycolysis and gluconeogenesis. Flux for succinate dehydrogenase was taken as the TCA cycle flux. RuBisCO carboxylase flux was taken as the flux through the Calvin-Benson cycle.