Principles of Cold Adaptation of Fish Lactate Dehydrogenases Revealed by Computer Simulations of the Catalytic Reaction

Abstract

Cold-adapted enzymes from psychrophilic and psychrotolerant species are characterized by a higher catalytic activity at low temperature than their mesophilic orthologs and are also usually found to be more thermolabile. Computer simulations of the catalytic reactions have been shown to be a very powerful tool for analyzing the structural and energetic origins of these effects. Here, we examine the cold adaptation of lactate dehydrogenases from two Antarctic and sub-Antarctic fish species using this approach and compare our results with those obtained for the orthologous dogfish enzyme. Direct calculations of thermodynamic activation parameters show that the cold-adapted fish enzymes are characterized by a lower activation enthalpy and a more negative entropy term. This appears to be a universal feature of psychrophilic enzymes, and it is found to originate from a higher flexibility of certain parts of the protein surface. We also carry out free energy simulations that address the differences in thermal stability and substrate binding affinity between the two cold-adapted enzymes, which only differ by a single mutation. These calculations capture the effects previously seen in in vitro studies and provide straightforward explanations of these experimental results.

Introduction

How cold-active enzymes from different species have been adapted by evolution to work under low temperature conditions has been a major problem in enzymology for many decades. Early comparative biochemical studies of orthologous enzymes adapted to cold and warm conditions date back at least to the 1960's, where particularly enzymes of cold-adapted fish attracted attention from marine biologists and biochemists (Kaplan 1965; Cowey 1967; Pesce et al. 1967; Low et al. 1973). Already, these early studies revealed that cold-water fish enzymes generally were more efficient catalysts than their counterparts from warm-blooded animals and that “apparently, cold-blooded animals have evolved enzymes with structures capable of functioning at low temperatures” (Kaplan 1965). A most remarkable finding was that of Somero and coworkers (Low et al. 1973), namely, that the adaptation of catalysis to low temperature seemed to be due to a lowering of the activation enthalpy ($\mathrm{\Delta }{H}^{\text{‡}}$) accompanied by a more negative activation entropy ($\mathrm{\Delta }{S}^{\text{‡}}$). Since it is the former that causes the exponential rate decay as the temperature is lowered, this makes perfect sense, almost as if the fish “knew” about the Arrhenius equation. Since then, comparative studies of orthologous enzymes from differently adapted species have shown that the redistribution of the enthalpic and entropic contributions to the activation free energy barrier is basically a universal characteristic of cold-adapted enzymes (Siddiqui and Cavicchioli 2006; Gerday 2013; Åqvist et al. 2017). That is, it generally holds for a range of different types of psychrophilic and psychrotolerant species from all kingdoms of life.

A particularly interesting analysis of the lactate dehydrogenases (A₄-LDH) from a series of closely related fish species was reported by Fields and Somero (1998). This study compared k_cat and K_M values, as well as thermal stability, of the enzyme from cold-water Antarctic and South American notothenioid fish species with that of warm-water fishes and ectothermic animals. Here, a strong inverse linear correlation was found between the k_cat value at 0°C and the average body temperature (<T_body>), again demonstrating the higher catalytic efficiency of the cold-adapted enzymes compared with their warm-active counterparts. As LDH is an enzyme that is presumably under strong evolutionary pressure due to its role in glycolysis, the catalytic rate has evidently become highly optimized with respect to the environmental temperature that the enzyme experiences. Here, the cold-water notothenioid fishes have a very high degree of sequence identity with just a few mutations between them. Also, the dogfish enzyme (average body temperature ∼20 °C) has a relatively high sequence identity of about 71% to the notothenioids, but its k_cat value is 3–4 times lower, while the warm-blooded bovine A₄-LDH (73% identity with the notothenioids) is about 5 times slower at 0 °C (Fields and Somero 1998).

Moreover, it was found that the K_M values for pyruvate were consistently higher (2- to 3-fold) for the Antarctic than the South American species in the temperature range 0–20 °C, thus indicating that substrate binding is weaker for the more cold-adapted enzymes at a common measurement temperature (Fields and Somero 1998), as often seems to be the case (Siddiqui and Cavicchioli 2006; Gerday 2013; Åqvist et al. 2017). The thermal stability as measured by residual enzyme activity versus time after incubation at 50 °C, on the other hand, showed no correlation with the notothenioid habitat temperature. For example, both the Patagonotothen tessellata (<T_body >= 8 °C) and Champsocephalus gunnari (<T_body >= 0 °C) enzymes were among the least stable, although the Harpagifer antarcticus (<T_body >= 0 °C) and Eleginops maclovinus (<T_body >= 8 °C) LDHs belonged to the most stable category. Moreover, the most stable enzyme as measured by the residual activity assay was found to be that from the Antarctic Dissostichus mawsoni species, which may speak against the general notion that cold-adapted enzymes are always characterized by thermal lability (Siddiqui and Cavicchioli 2006; Gerday 2013; Åqvist et al. 2017). However, in comparison to truly mesophilic LDHs, the C. gunnari enzyme has been shown to have 10–15 °C lower melting temperature than that from Deinococcus radiodurans and pig heart (Coquelle et al. 2007; Khrapunov et al. 2017).

These measurements of the A₄-LDH properties from a series of cold-adapted notothenioid fishes thus provide a very useful data set for examining the principles of cold adaptation on a detailed structural level. Computer simulations of the catalytic reactions of differently adapted orthologous enzymes have proven to be very useful for eliciting the structural and energetic origins of adaptive effects observed by in vitro experiments (Åqvist et al. 2017). In particular, simulations that enable reaction free energy profiles to be calculated at different temperatures make it possible to computationally obtain accurate Arrhenius plots, from which thermodynamic activation parameters can be obtained. It then also becomes possible to understand where changes in activation enthalpies and entropies (Low et al. 1973) come from, in terms of the enzyme structure. We have earlier reported both quantum mechanics/molecular mechanics (QM/MM) calculations and molecular dynamics (MD)–based empirical valence bond (EVB) simulations (Åqvist and Warshel 1993; Åqvist et al. 2017) of the similar NADH-dependent reaction catalyzed by the psychrophilic Psychrobacter arcticus hydroxybutyrate dehydrogenase (PaHBDH), where acetoacetate or 3-oxovalerate is reduced to 3-hydroxybutyrate or 3-hydroxyvalerate, respectively (Machado et al. 2020; Koenekoop et al. 2022). The first of these studies used kinetic isotope effects (KIE), QM/MM calculations and X-ray crystallography to dissect the PaHBDH mechanism and its energetics. In the second paper, we used MD/EVB simulations to explore the role of enzyme oligomerization on the thermodynamic activation parameters of the rate-limiting hydride transfer reaction with the 3-oxovalerate substrate. The EVB potential energy surface was then parametrized on the earlier QM/MM calculations and the MD/EVB simulations showed that the activation parameters were basically unaffected by the oligomeric state (monomer, dimer, or tetramer) of PaHBDH.

Since the LDH and HBDH reactions, substrates, and active site geometries are very similar, we can use here essentially the same EVB model to describe also the LDH reaction. The chemical step is rate limiting for PaHBDH with 3-oxovalerate at all temperatures and with acetoacetate at low temperature (Machado et al. 2018). No such data appear to be available for the fish A₄-LDHs, but a detailed kinetic study of pig heart LDH (Zhadin et al. 2008) shows that the limiting rate constants for loop opening/closure are of similar magnitude to that of the pyruvate→lactate hydride transfer step (217/s vs. 637/s at 23.1 °C and 470/s vs. 1337/s at 32.5°C). Moreover, in the pig heart enzyme, a KIE ≈ 1.6 was found for the overall rate and a KIE ≈ 1.9 for the hydride transfer step at 23.1 °C (Zhadin et al. 2008). This indicates that the chemical step is at least partly rate limiting in the porcine LDH at this temperature (see also, Peng et al. 2014). Interestingly, a distinct break in the Arrhenius plot for k_cat was also observed at 35 °C (Khrapunov et al. 2017), which may signal a change of the rate-limiting step (Åqvist et al. 2020; Åqvist 2022). In contrast, the C. gunnari enzyme showed a completely linear Arrhenius plot between 10 °C and 40 °C in that study (Khrapunov et al. 2017). Above 35 °C, there was a drastic drop in the activation energy for the pig heart enzyme which, although still somewhat slower than the C. gunnari LDH, now showed a similar activation energy. One interpretation of these results could be that the rate-limiting step shifts from loop opening/closure to chemistry above the breakpoint in porcine LDH and that chemistry is always rate limiting for the C. gunnari enzyme. However, as noted above, no detailed kinetic data have been reported for the fish enzymes and their dominant rate-limiting step thus remains unknown. On the other hand, the ratio of k_cat between the cold-water notothenioid A₄-LDHs and that of the dogfish is >3 (Fields and Somero 1998), and larger than the ratio between loop opening/closure and chemistry in the pig heart enzyme, so it is inevitable that evolution must also have acted on the chemical step in the fish enzymes, even if it is not cleanly rate limiting.

To address the problem of how the notothenioid A₄-LDHs have adapted to the cold environment, we report here extensive MD/EVB computer simulations of the catalyzed pyruvate→lactate conversion at different temperatures. This allows us to accurately calculate reaction free energy profiles and Arrhenius plots, from which the thermodynamic activation parameters of the chemical step can be extracted (Åqvist et al. 2017; Socan et al. 2019). These calculations are done for the mesophilic dogfish enzyme and for two of the cold-adapted notothenioid species that differ in average body temperature. The computer simulations show, in agreement with data from previous in vitro studies (Fields and Somero 1998), that the cold-adapted fish enzymes are three to four times faster than the mesophilic one and provide a straightforward explanation for this. The two notothenioid enzymes only differ by one mutation from each other and yet show a significant difference in substrate binding affinity and thermal stability. This intriguing problem is addressed by standard thermodynamic cycle free energy perturbation calculations, which turn out to explain the effect.

Results

Calculations of Activation Free Energies

We focus here on comparison of the A₄-LDH enzyme from the three species Squalus acanthias (dogfish), P. tessellata (black southern cod), and C. gunnari (mackerel icefish), where the former is the most mesophilic species with an average body temperature of 20 °C (Fields and Somero 1998). Our EVB model of the A₄-LDH catalyzed hydride transfer reaction with pyruvate was basically copied from earlier MD/EVB simulations of the HBDH reaction (Koenekoop et al. 2022). Besides the two distinct substrates, the main difference between LDH and HBDH is the proton donor to the carbonyl oxygen, which in LDH is a protonated histidine sidechain (His193—dogfish numbering is used throughout) while in HBDH, it is a tyrosine. Our earlier QM/MM calculations placed the (R)-3-hydroxyvalerate product about 3.9 kcal/mol above the 3-oxovalerate reactant in HBDH (Machado et al. 2020). The corresponding equilibrium on the pig heart LDH enzyme (pyruvate $\rightleftharpoons$ lactate) has been shown experimentally to be isoenergetic (Zhadin et al. 2008). This is likely to be at least partly caused by the pK_a difference between histidine and tyrosine, where the former is more favorable for proton transfer due to its lower pK_a value. Hence, if one recalibrates the EVB model for HBDH with a reaction free energy of $\mathrm{\Delta }{G}_0=0$, the barrier drops to $\mathrm{\Delta }{G}^{\text{‡}}=13.9$  kcal/mol at 0 °C, which is very close to that observed for the dogfish LDH, $\mathrm{\Delta }{G}^{\text{‡}}=13.6$ kcal/mol at 0 °C (Fields and Somero 1998). To examine the relative efficiency of the three LDH enzymes here, we thus use the values $\mathrm{\Delta }{G}^{\text{‡}}=13.6$ and $\mathrm{\Delta }{G}_0=0$ for dogfish LDH to calibrate our EVB model. Here, it may also be noted that to the extent that there is a temperature dependence of the pK_a of the His193 proton donor (Yancey and Somero 1978), this will be reflected as part of the temperature dependence of $\mathrm{\Delta }{G}_0$ in the free energy calculations.

Based on the earlier QM/MM simulations of the HBDH reaction (Machado et al. 2020), the chemical step in LDH was also modelled as a concerted process. The MD/EVB simulations yield a transition state for hydride and proton transfer to the pyruvate substrate (fig. 1A) with similar geometry to that of HBDH (Machado et al. 2020; Koenekoop et al. 2022), and a typical reaction free energy profile from the 100 replicate dogfish LDH simulations at 0 °C is shown in figure 1B. Averaging over all the replicas at 0 °C gives activation and reaction free energies of $\mathrm{\Delta }{G}^{\text{‡}}=13.62\pm 0.14$ and $\mathrm{\Delta }{G}_0=0.06\pm 0.26$ kcal/mol in accordance with our target values for the dogfish enzyme [errors are given as ±1 standard errors of the mean (s.e.m.)]. Remarkably, applying the EVB model to the C. gunnari and P. tessellata enzymes yields small but significant reductions of the activation barrier for the two cold-adapted enzymes, with $\mathrm{\Delta }{G}^{\text{‡}}=12.96\pm 0.16$ and $\mathrm{\Delta }{G}^{\text{‡}}=12.85\pm 0.18$ kcal/mol for C. gunnari and P. tessellata, respectively. The corresponding experimentally derived values here are in both cases $\mathrm{\Delta }{G}^{\text{‡}}=13.0$ kcal/mol at 0 °C (Fields and Somero 1998). Hence, the MD/EVB simulations are able to capture the small effect (∼0.6 kcal/mol) on the activation barrier caused by the 82–83 mutations between the mesophilic and cold-adapted enzymes. Of course, this is only made possible by the large number of independent MD/EVB simulations carried out which allows the error bars of the free energy calculations to be pushed to very small values.

Fig. 1.

(A) View of the concerted transition state in the dogfish enzyme, with hydride transfer from NADH to pyruvate and simultaneous protonation of the substrate carbonyl by His193 (dogfish numbering). (B) Reaction free energy profile for hydride transfer from one of the 100 replicate simulations of dogfish LDH at 0 °C. $\mathrm{\Delta }\epsilon$ is the generalized reaction coordinate (Åqvist and Warshel 1993; Koenekoop et al. 2022).

Temperature Dependence of the Reactions

The MD/EVB simulations for the three enzymes were repeated at five different temperatures (0, 10, 20, 30, and 40 °C) in order to calculate Arrhenius plots (fig. 2) and extract the thermodynamic activation parameters (table 1). An experimental Arrhenius plot for the C. gunnari enzyme has been reported by Coquelle et al. (2007) and taking the temperature range ≤ 40 °C from there (since the melting temperature is about 45 °C) yields values of $\mathrm{\Delta }{H}^{\text{‡}}=4.7$ and $T\mathrm{\Delta }{S}^{\text{‡}}=-8.0$ kcal/mol (with T = 273 K as reference). Similar values of $\mathrm{\Delta }{H}^{\text{‡}}=5.9$ and $T\mathrm{\Delta }{S}^{\text{‡}}=-7.1$ kcal/mol are also obtained from the work of Khrapunov et al. (2017). Moreover, an activation energy of 4.7 kcal/mol has been reported for the notothenioid Chaenocephalus aceratus enzyme which differs by only one mutation from C. gunnari (Fields and Houseman 2004). As can be seen from table 1, our predicted values for C. gunnari in the 0–40 °C temperature region are in remarkably good agreement with the experimental values ($\mathrm{\Delta }{H}_{\mathrm{calc}}^{\text{‡}}=4.8$ and $T\mathrm{\Delta }{S}_{\mathrm{calc}}^{\text{‡}}=-8.4$ kcal/mol). Notably, the corresponding calculated activation parameters for the P. tessellata enzyme are very similar to those for C. gunnari and both of these LDHs show a reduction of $\mathrm{\Delta }{H}^{\text{‡}}$ and a more negative $T\mathrm{\Delta }{S}^{\text{‡}}$ compared with the more mesophilic dogfish enzyme. Although there seems to be no such experimental data available for dogfish, the bluefin tuna enzyme with a similar reaction rate and <T_body> was reported to have activation parameters $\mathrm{\Delta }{H}^{\text{‡}}=8.8$ and $T\mathrm{\Delta }{S}^{\text{‡}}=-3.7$ kcal/mol at 5 °C (Low et al. 1973). These values are thus also very close to those we predict here for the dogfish LDH (table 1).

Fig. 2.

Calculated Arrhenius plots of $\mathrm{\Delta }{G}^{\text{‡}}$ versus T from reaction simulations at five different temperatures of the LDHs from (A) dogfish, (B) C. gunnari, and (C) P. tessellata. Thermodynamic activation parameters are obtained from linear regressions, and the s.e.m. from 100 replicate simulations for the average free energy barriers is in all cases ≤0.20 kcal/mol.

Table 1.

Thermodynamic Activation Parameters (kcal/mol) for the Hydride Transfer Reaction in the Three Fish Enzymes.

	$\mathrm{\Delta }{G}_{\mathrm{calc}}^{\text{‡}}(0^{\circ }\mathrm{C})$	$\mathrm{\Delta }{H}^{\text{‡}}$	$T\mathrm{\Delta }{S}^{\text{‡}}(0^{\circ }\mathrm{C})$ a	$\mathrm{\Delta }{G}_{\mathrm{expt}}^{\text{‡}}(0^{\circ }\mathrm{C})$	$\mathrm{\Delta }{G}_{\mathrm{calc}}^{\text{‡}}({40}^{\circ }\mathrm{C})$
Dogfish	13.62	8.4	−5.2	13.6	14.39
C. gunnari	12.96	4.8	−8.4	13.0	14.23
P. tessellata	12.85	4.8	−8.0	13.0	14.05

^aThe values of $\mathrm{\Delta }{H}^{\text{‡}}$ and $\mathrm{\Delta }{S}^{\text{‡}}$ are obtained from the Arrhenius plots in figure 2. Experimental $\mathrm{\Delta }{G}^{\text{‡}}$ values are from Fields and Somero (1998).

The above type of behavior with a lower activation enthalpy and a more negative entropy term is, in fact, typical for cold-adapted enzymes and appears to be a universal feature (Low et al. 1973; Siddiqui and Cavicchioli 2006; Gerday 2013; Åqvist et al. 2017). Hence, the computer simulations not only correctly predict the cold-water fish enzymes to be faster than that from dogfish at low temperature but also predict their cold adaptation signature in terms of activation enthalpies and entropies. That it is indeed this redistribution of activation enthalpy and entropy penalties that causes cold adaptation can also be seen from the fact that the difference in the predicted free energy barriers at 40 °C for the three enzymes becomes much smaller (table 1).

Differences in Protein Stability and Substrate Binding Affinity between the Notothenioid Enzymes

Even though the P. tessellata and C. gunnari LDHs only differ by a single amino acid, they have been found to have different thermal stability and substrate binding affinity (Fields and Somero 1998). That is, the V316L (P. tessellata to C. gunnari) mutation increases the thermal stability but decreases the pyruvate binding affinity. This residue is a leucine also in the dogfish enzyme and is located about 20 Å from the active site on the opposite side of the protein and is not near any subunit interfaces. The mutated position (fig. 3A) is on the interior side of the helix αH and makes hydrophobic contact with Leu289 on an adjacent β-strand (βL) as well as with two nearby loops (Coquelle et al. 2007; Abad-Zapatero et al. 1987). It thus appears logical that mutation from a shorter to a longer sidechain (V316L) should make the protein more stable as the interior hydrophobic packing appears to be improved. Why the mutation should affect pyruvate binding affinity is less clear, in particular since it is the smaller valine sidechain in P. tessellata that yields the highest affinity with an approximately 2-fold lower K_M value (Fields and Somero 1998).

Fig. 3.

(A) View of the region around the V316L mutation in the αA helix. In the C. gunnari enzyme, the interior hydrophobic packing is enhanced compared with P. tessellata due to the larger leucine sidechain. (B) Thermodynamic cycle describing the effect of the V316L mutation on the pyruvate binding free energy and on the thermal stability of the enzyme·NADH complex. The Ala-X-Ala tripeptide serves as a model for the unfolded state where the X sidechain is fully solvent exposed (Jespers et al. 2019).

To examine the effect of the V316L mutation on stability and binding, we carried out standard free energy perturbation (FEP) calculations where one sidechain was mutated into the other with and without the pyruvate molecule bound to the active site, as well as in an Ala-X-Ala tripeptide (Boukharta et al. 2014; Jespers et al. 2019). The latter is used as a standard reference for a solvent-exposed residue in the unfolded state. Both the relative binding and folding free energies are then estimated from a thermodynamic cycle (fig. 3B) involving the (V316L) mutation in the three relevant states: LDH·NADH·pyruvate, LDH·NADH and tripeptide (unfolded). The temperature was 20 °C in these simulations as in the experimental assays. These free energy calculations turn out to give results in qualitative agreement with the experimental data of Fields and Somero (1998), with a stability free energy change of $\mathrm{\Delta }\mathrm{\Delta }{G}_{\mathrm{fold}}=-2.68\pm 0.29$ kcal/mol for the substrate-free enzymes and $\mathrm{\Delta }\mathrm{\Delta }{G}_{\mathrm{bind}}=+1.01\pm 0.61$ kcal/mol. Hence, the V316L mutation is predicted to stabilize the C. gunnari enzyme which is consistent with its ∼8-fold higher residual activity measured at 20 °C after 5 min of incubation at 50 °C (Fields and Somero 1998). This thus appears to be a rather trivial effect caused by better hydrophobic packing in the protein interior due to the larger sidechain.

Also, the loss of binding affinity for pyruvate caused by the V316L mutation is qualitatively captured by the free energy calculations, where a binding free energy change of about 0.5 kcal/mol can be estimated from the experimental data at 20 °C (Fields and Somero 1998). The reason for this binding free energy change due to a rather remote mutation from the pyruvate binding site is less trivial to explain. However, as noted above, Leu316 makes hydrophobic contact with the loop harboring the catalytic residue His193 and the histidine residue interacts directly with pyruvate via a hydrogen bond. So, it appears that the effect of the mutation indirectly affects the interaction between His193 and the substrate. This is also supported by a comparison between the experimental (but incomplete) C. gunnari apoenzyme structure (Coquelle et al. 2007) and the dogfish inhibitor complex (Abad-Zapatero et al. 1987), which indicates that there is a small movement of the βG-βH loop holding the catalytic His193 upon pyruvate binding. In both of these structures, Leu316 in helix αH maintains direct contact with the βG-βH loop suggesting that there may be some resistance to the loop movement, which could be relieved by the smaller P. tessellata valine sidechain in the mutated position.

Origin of the Cold Adaptation of the Notothenioid Enzymes

The strongest signature of cold adaptation of the two notothenioid enzymes is clearly their high catalytic activity at low temperature and the characteristic redistribution of the enthalpy and entropy components of the activation free energy. We have earlier shown that the lower $\mathrm{\Delta }{H}^{\text{‡}}$ and more negative $T\mathrm{\Delta }{S}^{\text{‡}}$ of psychrophilic enzymes are generally caused by the cold-adapted ones being softer, or less stiff, than their mesophilic orthologs, particularly the protein surface (Isaksen et al. 2016; Åqvist et al. 2017; Socan et al. 2019). A key indicator of this phenomenon is thus the surface flexibility as measure by root mean square positional fluctuations (RMSF). Such flexibility is difficult to reliably assess from crystallographic temperature factors since surface loops are often involved in crystallographic contacts but can readily be obtained from MD simulations.

As might be expected, a comparison of the calculated RMSF profiles at 0 °C along the protein sequence immediately shows that the mobilities of the two notothenioid enzymes are virtually identical (fig. 4A). Here, the peaks of the backbone mobility pattern usually correspond to loop regions connecting the less mobile ordered secondary structure elements (Åqvist et al. 1985). The dogfish enzyme, on the other hand, can be seen to be substantially more rigid than the other two LDHs precisely in some of these loop regions. The most prominent flexibility differences are found in the following regions (fig. 4B): the αC helix (55–68), the βD-αD active site loop (98–110), the α2F-βG loop (175–185), the long βH-α1G loop region (200–230), and the α1G- α2G loop (241–252), where the dogfish sequence numbering is used. It should be noted here that the poorly structured βH-α1G region was modelled with an incorrect amino acid sequence in the dogfish crystal structure from 1987 (Abad-Zapatero et al. 1987). Although that structure did improve an earlier determined amino acid sequence (Eventoff et al. 1977), based on fitting to the electron density, it was not until 1995 that the correct sequence of dogfish LDH was obtained from cDNA (Stock and Powers 1995). This corrected sequence (fig. 5) also only has two insertions with respect to the notothenioid enzymes, one in the very N-terminal end of the chain (Leu3) and one in the αC-βC loop (Ala75).

Fig. 4.

(A) Calculated average backbone RMSFs per residue along the amino acid sequence of the three LDH enzymes. (B) View of the locations of the 82–83 mutations (cyan spheres) between the dogfish enzyme and the two notothenioid LDHs. The regions of increased mobility in the latter are indicated in red. (C) Calculated average positional RMSFs per residue (heavy atoms) for the active site residues and substrates (NADH, pyr). The RMSF calculations are all done at 0 °C.

Fig. 5.

Alignment of the dogfish (upper) and P. tessellata (lower) A₄-LDH amino acid sequences (UniProt entries P00341 and O93546, respectively). The C. gunnari sequence has a leucine at position 316 (as has the dogfish), which in that case is the only mutation with respect to P. tessellata (indicated in blue).

All of the five sequence segments with increased mobility in the two notothenioid enzymes correspond to surface regions of the protein (fig. 4B), as has also been observed in other cold-adapted enzymes (Isaksen et al. 2016; Åqvist et al. 2017; Socan et al. 2019). With as many as 82–83 mutations compared with the dogfish, it is not surprising that these are more or less scattered all over the structure. It is, however, noteworthy that two of these regions, namely, the βD-αD active site loop and the α1G- α2G loop, are totally free of mutations and thus completely conserved. This is likely due to the fact that these loops make up the substrate entrance surface and are involved in closure of the βD-αD loop upon substrate binding (Coquelle et al. 2007; Abad-Zapatero et al. 1987). The fact that they, nevertheless, are more mobile in the notothenioid structures appears to be due to (packing) effects of “secondary” mutations located in adjacent secondary structure elements (I94V, S237G, and I326V; fig. 5). Also, the increased mobility of the αC helix, which only carries one mutation (M62V), is affected by the nearby mutations L50M and S79G in two adjacent β-strands (βB and βC). The two other loops of high mobility, α2F-βG and βH-α1G, have 4 and 11 mutations within their sequences, respectively, which apparently causes their higher flexibility (fig. 4B). Overall, the average backbone RSMF values for the P. tessellata and C. gunnari LDHs (0.99 and 1.00 Å) are significantly higher than that of the dogfish structure (0.88 Å), which illustrates the generally increased flexibility of the cold-adapted enzymes. Moreover, a comparison of the mobility of active site residues within 5 Å of the substrate carbonyl carbon shows that the dogfish residues are invariably more rigid than in the two notothenioid enzymes (fig. 4C).

As a direct proof of the connection between the increased flexibility of the two notothenioid enzymes and the thermodynamic activation parameters, we carried out a set of additional MD/EVB simulations of the C. gunnari enzyme with restraints applied to regions that differ significantly in mobility compared with the dogfish enzyme. In these calculations, weak positional restraints (with force constant 1 kcal/mol/Ų) were applied to the 29 α-carbons that showed an RMSF increase of more than 0.4 Å compared with the dogfish LDH. With this ΔRMSF cutoff, some Cαs belonging to the first four marked regions of figure 4A are thus restrained (αC, βD-αD, α2F-βG, and βH-α1G). The resulting Arrhenius plot for this partly restrained C. gunnari enzyme (fig. 6A) now gives activation parameters of $\mathrm{\Delta }{H}^{\text{‡}}=7.4$ and $T\mathrm{\Delta }{S}^{{}^{\text{‡}}}=-5.7$ kcal/mol, which have clearly approached those of the dogfish enzyme (table 1). The corresponding plot of backbone RMSF values also shows that the fluctuation pattern has become much more similar to that of the dogfish, as intended (fig. 6B).

Fig. 6.

(A) Calculated Arrhenius plot of $\mathrm{\Delta }{G}^{\text{‡}}$ versus T from reaction simulations at five different temperatures of the partly restrained C. gunnari LDH. Thermodynamic activation parameters are obtained from linear regression, and the s.e.m. from 100 replicate simulations for the average free energy barriers is ≤0.19 kcal/mol. (B) Calculated average backbone RMSFs per residue along the amino acid sequence of the dogfish, C. gunnari, and restrained C. gunnari LDHs.

Discussion

The present computational analysis of cold adaptation of the A₄-LDH enzyme in three different fish species reveals several recurring features that have been observed in psychrophilic enzymes. First, the reaction rate at low temperature (0 °C in our case) is predicted to be significantly higher for the cold-adapted enzymes than for their more mesophilic counterparts, in our case a factor of three to four times higher for P. tessellata and C. gunnari compared with the dogfish LDH. Second, the increased rate is primarily caused by a redistribution of the enthalpy and entropy components of the activation free energy. Here, the two cold-adapted enzymes (P. tessellata and C. gunnari) show a ∼4 kcal/mol lower activation enthalpy and a ∼3 kcal/mol more negative $T\mathrm{\Delta }{S}^{\text{‡}}$ term at 0 °C, which results in a ∼0.7 kcal/mol lower value of $\mathrm{\Delta }{G}^{\text{‡}}$ than the dogfish enzyme. Third, as found in several earlier cases (Isaksen et al. 2016; Åqvist 2017; Åqvist et al. 2017; Socan et al. 2019), the enthalpy–entropy redistribution involved in cold adaptation originates from differences in rigidity, or flexibility, of the protein structures. Here, we find strong signals both in terms of the overall backbone mobility and the mobility of the active site residues, which show that the dogfish enzyme is considerably more rigid than the other two. Although there are simply too many mutations (82–83) between the two notothenioid and the dogfish enzyme to dissect their individual effects, the combined effect of the sequence changes is clear. That is, the lower activation enthalpy originates from a softer, more flexible, enzyme surface which reduces the protein resistance to the structural perturbations caused by the chemical reaction (Isaksen et al. 2016: Åqvist et al. 2017). This comes at the expense of an increased entropy penalty that partly compensates the enthalpy effect.

Observations of an increased structural flexibility of cold-adapted enzymes have been reported in many MD studies (see, e.g., Åqvist et al. 2017 and references therein). For example, among the dehydrogenases, structural models of cytosolic malate dehydrogenase from a series of marine molluscs have recently been examined by MD simulations (Dong et al. 2018; Liao et al. 2019), where significant correlations between adaptation temperature and protein backbone mobility were also found. Hence, the general notion that enzyme cold adaptation is intimately connected to protein flexibility must now be considered well established. Most of these MD studies, however, only report correlations between structural flexibility and adaptation temperature without addressing the causal relationships with catalytic rates. These relationships can only be established by simulations that actually calculate activation free energy barriers and catalytic rates, as pointed out already by Bjelic et al (2008). Particularly, the origin of the universally observed enthalpy–entropy shift of the thermodynamic activation parameters is one of the major questions, since it is this shift that gives increased reaction rates at low temperature. This question was answered by the approach of calculating Arrhenius plots with restraints applied to control the protein flexibility (Isaksen et al. 2016). Here, again, we have employed such a restraining approach to the C. gunnari enzyme and the results unambiguously show that a damping of the protein surface flexibility makes the enzyme more mesophilic in terms of the thermodynamic activation parameters.

The subtle single mutation V316L between the P. tessellata and C. gunnari LDHs was considered particularly interesting since it affects both enzyme stability and substrate binding. It turns out that our free energy simulations at least qualitatively agree with the experimental results for the mutation, which causes the C. gunnari enzyme to become more thermostable but binds pyruvate less strongly (Fields and Somero 1998). A lower thermal stability, together with the activation enthalpy–entropy redistribution, seems to be the strongest characteristics of cold-adapted enzymes (Low et al. 1973; Siddiqui and Cavicchioli 2006; Gerday 2013; Åqvist et al. 2017). This would thus suggest that the P. tessellata LDH may actually be more “psychrophilic” than the C. gunnari enzyme, despite its slightly higher average body temperature. The value of k_cat/K_M obtained from Fields and Somero (1998) is also 2-fold higher for P. tessellata than C. gunnari at 0 °C, which might support the notion that the former enzyme may be better optimized for low temperature.

The detailed kinetics of the LDH-catalyzed reaction has unfortunately been not been measured for either of the three fish enzymes studied herein. For the mesophilic pig heart enzyme (Zhadin et al. 2008) and the thermophilic variant from Bacillus stearothermophilus (Clarke et al. 1986), the rate-limiting step at room temperature is the closing/opening of the βD-αD active site loop, which inserts the key sidechain of Arg106 into the binding site where it interacts with the pyruvate substrate (Coquelle et al. 2007; Abad-Zapatero et al. 1987). Hence, in the B. stearothermophilus enzyme, no H/D kinetic isotope effect was observed (KIE = 1.0) and no NAD⁺ burst was detected, indicating a conformational rate-limiting step prior to the chemical one. In the pig heart enzyme, loop opening associated with product release was found to be the slowest step at room temperature (Zhadin et al. 2008). There is indication of a change of a rate-limiting step in the porcine enzyme above 35 °C, where the activation energy suddenly drops and becomes similar to that of C. gunnari LDH (Khrapunov et al. 2017). Bearing in mind that mesophilic and thermophilic enzymes are generally stiffer than psychrophilic ones, it appears that they would be more susceptible to rate-limiting conformational changes. Such conformational changes would naturally become more facile at high temperature, which might explain the break in the Arrhenius plot observed for the pig heart enzyme (Khrapunov et al. 2017).

The generally increased flexibility of cold-adapted enzymes would seem to be an obvious way of not only reducing the activation enthalpy of the chemical step(s) but also increasing the rate of conformational changes. In analogy with HBDH, it thus appears possible that the chemical step has become rate limiting in our fish enzymes at low temperature, which would explain the excellent agreement between our calculated thermodynamic activation parameters for hydride transfer and those derived from experimental k_cat values (Fields and Somero 1998; Fields and Houseman 2004; Coquelle et al. 2007; Khrapunov et al. 2017). The fact that the active site loop is found to have a higher mobility in our two psychrophilic enzymes could also be an indication of that the loop closing/opening has become more facile in these cases. In this respect, it would thus be of considerable interest to characterize the detailed kinetics also of the cold-adapted fish enzymes.

In summary, the present work shows how one can effectively combine computer simulations of reaction free energy profiles at different temperatures with standard free energy calculations that address the effects of amino acid mutations on protein stability and substrate binding. When comparing closely related, but differently adapted, enzymes, this approach thus allows for quantification of what is thought to be the three main features of thermal adaptation of enzymes. That is, the effects of mutations on activation enthalpy and entropy modulation, thermal stability, and substrate binding affinity can all be addressed.

Methods

System Preparation

The structural model for the dogfish (S. acanthias) A₄-LDH was based on a refined crystal structure in complex with NAD⁺ and the pyruvate analog oxamic acid, PDB entry 1LDM (Abad-Zapatero et al. 1987). Released in 1989, this structure exhibits both a sequence region of poor resolution between residues 207 and 211 where the original sequence WNALKE was replaced by NVASIK, as well as a number of misidentified aspartic acid and asparagine residues (Abad-Zapatero et al. 1987). Comparing the amino acid sequence of this crystal structure with the dogfish LDH sequence deposited in UniProt (entry P00341), a number of mismatches can easily be identified. To ensure a structural model in full agreement with the UniProt sequence and a reliable 3D protein conformation, the template structure 1LDM was aligned to the apo A₄-LDH crystal structure of the mackerel icefish C. gunnari, PDB entry 2V65 (Coquelle et al. 2007), and the holo A₄-LDH complex of pig heart, PDB entry 5YTA (Ho 2018). Amino acid mutations to the dogfish structure were implemented with PyMOL, and the sidechain conformations were based on the aligned template structures. In order to compensate for the poor resolution loop region, the structure of residues 207 through 229 was replaced by that of the pig heart LDH, along with the required amino acid mutations to match the dogfish sequence.

Black southern cod (P. tessellata) and C. gunnari A₄-LDH models were generated using Modeller (Sali and Blundell 1993) based on their sequences deposited as UniProt, entries O93546 and O93541, respectively. The dogfish crystal structure was used as template, and active site residues in close contact with the substrate and cofactor were kept fixed. For both models, the loop sequence of resides 207 through 229 was replaced as before, this time using the C. gunnari apo-LDH structure as template. Note that this experimental structure (2V65) is incomplete and in the significantly different apo-conformation and can thus not be used in its entirety for simulations of the holoenzyme. Further structure preparation was done with Schrödinger's Maestro (Schrödinger 2021) to verify asparagine and glutamine flips and histidine protonation states, along with determination of protonation states of ionizable residues based on their pK_a values, as predicted by PROPKA (Olsson et al. 2011) at pH 7.0. Oxamic acid was modified to pyruvate by substituting the amine group for a methyl group, and the nicotinamide ring of NAD⁺ was protonated to NADH. Crystallographic water molecules beyond a distance of 20 Å from the substrates were removed. The N-terminal tail (residues 1–18) of the sequence was also removed, as it protrudes out from the monomeric structure and is involved in tetrameric subunit interactions. In analogy with HBDH, where it was shown that the thermodynamic activation parameters were insensitive to the oligomeric state (Koenekoop et al. 2022), all calculations herein were done for the LDH monomers.

MD Simulations

The Q software package (Marelius et al. 1998; Bauer et al. 2018) was used to perform all the MD simulations using the OPLS-AA/M force field (Robertson et al. 2015). Interaction parameters to describe pyruvate, lactate, NADH, and NAD⁺ were generated with Schrödinger's ffld_server (Banks et al. 2005), with the exception of the torsional parameters connecting the amide to the pyridine ring. These were determined by performing a relaxed surface scan in the gas phase at the B3LYP-D3BJ (def2-SVP) level of theory using the ORCA program package (Neese et al. 2020). The monomeric enzyme was fully solvated in a spherical water droplet with diameter 80 Å with origin at the protein center of mass. Water molecules at the sphere boundary were subjected to radial and polarization restraints according to the SCAAS model (King and Warshel 1989; Marelius et al. 1998). All nonbonded interactions involving the reacting fragments (pyruvate, the nicotinamide ring, and ribose moiety of NADH and the sidechain of His193) were calculated explicitly. The local reaction field method (Lee and Warshel 1992) was used for other long-range electrostatic interactions beyond a direct cutoff of 10 Å. The MD simulations employed a 1-fs time step, and a flat-bottom (>2.5 Å) harmonic restraint (force constant of 10 kcal mo/l/Å) was applied to the distance between the donor–acceptor atom pairs in the hydride and proton transfer reactions (C···C and O···N).

EVB Model

The MD/EVB simulations were carried out as described previously for the HBDH enzyme (Koenekoop et al. 2022). Here, hydride transfer from NADH and proton transfer from His193 to the pyruvate substrate was described by a two-state EVB potential, with the two states corresponding to reactants and products, both represented by a standard force field (Robertson et al. 2015). As earlier, a more physical exponential repulsion U_rep = C_ij exp(−a_ijr_ij) is used between atoms involved in bond breaking and formation instead of the regular Lennard–Jones potential (Åqvist 2022; Koenekoop et al. 2022). This involves interactions between the transferring hydride/proton and their respective heavy atoms (a_ij = 4.0 Å⁻¹ and C_ij = 250 kcal/mol). Bonds in the reacting fragments were represented by Morse potentials as described earlier (Koenekoop et al. 2022). There are two key EVB parameters (Åqvist and Warshel 1993; Koenekoop et al. 2022) that represent the gas phase energy difference (Δα = 85.27 kcal/mol) between reactants and products and the off-diagonal coupling element (H₁₂ = 105.29 kcal/mol) between the two VB states. These were calibrated using the experimental free energy barrier for dogfish LDH at 0 °C ($\mathrm{\Delta }{G}^{{}^{\text{‡}}}=13.64$ kcal/mol) and a pyruvate–lactate equilibrium free energy of $\mathrm{\Delta }{G}_0=0.06$ kcal/mol (Fields and Somero 1998; Zhadin et al. 2008).

Reaction Free Energy Profiles

The FEP umbrella sampling approach was used as described earlier to obtain reaction free energy profiles (Åqvist et al. 2017; Koenekoop et al. 2022), using 51 evenly spaced sampling windows between reactants and products. The free energy calculations were carried out at five different temperatures between 273 and 313 K, and 100 replicate MD/EVB simulations with different initial conditions were generated at each temperature. The initial heating and equilibration protocol was identical to that used for HBDH (Koenekoop et al. 2022). Each average free energy profile at each temperature corresponds to a total of 51 ns of data collection, which yields s.e.m. for the calculated activation free energy barriers of ≤ 0.20 kcal/mol in all cases. Activation enthalpies and entropies were obtained by linear regression from the corresponding Arrhenius plots of $\mathrm{\Delta }{G}^{\text{‡}}$ versus T. The same simulation protocol was applied to each of the three LDH enzymes, as well as to the partly restrained C. gunnari enzyme. In the latter case, weak positional harmonic restraints with a force constant of 1.0 kcal/mol/Ų were applied to Cαs of residues 56, 57, 60–67, 101, 102, 106, 182, 201, 205–207, 211–216, 218–221, and 223.

Free Energy Calculations on the V316L Mutation

Amino acid mutations were carried out with the QresFEP single topology FEP protocol (Boukharta et al. 2014; Jespers et al. 2019), where the wild-type P. tessellata Val316 sidechain was stepwise annihilated to alanine. In a separate simulation, starting from the C. gunnari residue Leu316, the same procedure with mutation to alanine was repeated. These transformations were also duplicated in the absence of the pyruvate substrate to complete the thermodynamic binding cycle (fig. 3B). The initial structures used in these calculations were the P. tessellata and C. gunnari models, generated as described above, where the pyruvate substrate was simply removed in the LDH_V/L316·NADH simulations. As noted above, the C. gunnari apo structure is incomplete and also devoid of NADH (Coquelle et al. 2007), wherefore it could not be used here. In addition, data from reference calculations on the same transformations in an Ala-X-Ala tripeptide (Jespers et al. 2019) to account for the unfolded state were used to construct the protein stability thermodynamic cycle (fig. 3B). These MD simulations were performed with a 50 Å diameter sphere centered on the Cβ atom of the mutated residue and followed the standard QresFEP protocol (Jespers et al. 2019). All ionizable residues within 3 Å of the sphere boundary and beyond were neutralized, although those within the simulation sphere were treated in the protonation state predicted by PROPKA (as above). Each FEP simulation was repeated 10 times with randomly assigned initial velocities. The simulation systems were first heated from 0.1 to 273 K although successively releasing solute heavy atom restraints, after which followed further heating to 293 K and unrestrained equilibration for 100 ps. The production phase then employed 126 discrete sampling windows for the Leu→Ala transformation and 105 windows for Val→Ala, adding up to a total of 23.1 ns simulation time per leg of the thermodynamic cycle and thus 69.3 ns for the entire FEP cycle.

Data Availability

The source code for the Q program is available at https://github.com/qusers/Q6. Input files and other data are available from the authors upon reasonable request.