Structural factors governing alkane substrate specificity in the alkane monooxygenase (AlkB) family

Abstract

Alkane monooxygenase (AlkB) is a family of integral membrane non-heme diiron enzymes that catalyzes the initial step in the oxidation of alkanes by numerous microorganisms. It plays a key role in the global carbon cycle, but the scope of this role is unknown. The AlkB family is thought to catalyze the oxidize alkanes from pentane to C₄₄H₉₀; implicit in this assertion is the idea that an individual AlkB only catalyzes the oxidization of a relatively narrow range of substrates. The structural basis of this discrimination is unclear. In this paper, eight AlkBs for which we possess reliable biochemical information about their substrate ranges were selected and AlphaFold3 was used to predict their three-dimensional structure. Several different cavity-determining programs were used to identify and measure the substrate channel in these predicted structures. Correlations between substrate selectivity and structure were made. Combining information available from different sources enables new hypotheses about the factors that control substrate selectivity in this important family of metalloenzymes to be made.

Introduction:

Alkanes enter the environment from a multitude of sources and while these reduced carbon molecules are energy-rich enough to be good cellular fuel sources, the inertness of their strong and non-polar bonds poses a C-H bond activation challenge. Combined with their membrane-disrupting lipophilicity, alkane consumption is limited to microorganisms that possess alkane-oxidizing machinery. These alkane-oxidizing enzymes can be categorized both by whether they use oxygen (either monooxygenases or dioxygenases) or if they are found in anaerobic organisms, and by the chain length of alkanes that they transform. A number of reviews have been published about alkane-oxidizing enzymes that are important in the environment.[1–8]

This article focuses on one of the family of alkane-oxidizing enzymes, the alkane monooxygenase (AlkB) family. The first member of the family was identified in the late 1960’s-early 1970’s by a group of researchers at the University of Michigan from the organism Pseudomonas oleovorans, now known as Pseudomonas putida GPo1.[9–14] This family of enzyme is purported to oxidize the bulk of the liquid alkanes that enter the environment. However, no single AlkB is predicted to oxidize the full range of linear alkanes whose transformation has been attributed to AlkB. Rather, individual members of the AlkB family are thought to have a narrower substrate range. In 2005, researchers found that decreasing the bulk of a single side chain changed the substrate preference of an AlkB from octane to hexadecane.[15] Conversely, modifying the analogous residue in an AlkB that afforded bacterial growth on hexadecane to a bulkier residue shifted the organism’s substrate growth profile towards shorter alkanes.[15] The authors speculated that this critical residue resided inside a substrate channel and pointed into the channel at a distance about 13 Å from the active site.[15]

This speculation, based on hydropathy plots and sequence analysis was remarkably accurate.

Almost 20 years after this provocative paper, the cryo-EM structure of AlkB was finally solved.[16, 17] Two groups published structures of different members of the AlkB family, both coming from the same thermophilic organism Fontimonas thermophila (Ft). Both structures share similar key features. Both structures show the presumed substrate channel and had electron density that was resolved as dodecane in the channel. Even though the overall AlkB protein fold is unique, the AlphaFold predicted structures match the experimentally determined structures quite well (TM scores ~0.98).[16]

In this paper, we build on the success of these two cryo-EM structures and the revolution in macromolecular structure prediction tools to revisit many decades of information about AlkB’s activity in the context of the structural factors that govern it. We also critique these methods and point to factors that limit our ability to predict the substrate range of any AlkB simply from its amino acid sequence. We do so with some confidence that further evolution of tools complemented with more experimental data will improve our predictive power.

Methods

Choice of AlkB Species

The following AlkBs were selected because biochemically reliable information was available about the alkane substrates they preferentially oxidize: P putida AlkB, F. thermophila AlkB, A borkumensis AlkB, M tuberculosis AlkB, D cinnemea AlkB, and AlkB 1 and 2 from P aeruginosa. The amino acid sequences for each of these proteins is given in the Zenodo repository associated with this manuscript. Their assigned UniProt IDs are presented in the Supporting Information (Table S1). Some protein variants were also included: P putida mutant BMO2 evolved by the Arnold group (PpAlkBBMO2)[18], PpAlkBW55V, FtAlkBW62V and DcV91W.

Structural Prediction and Sequence Alignment

To predict the structures of all proteins, the AlphaFold3 webserver was used.[19] This was done even in the case where a cryo-EM structure exists to keep consistency in the results. We have previously shown that the AlphaFold3 predicted structure of FtAlkB aligns well with the cryo-EM structure. The same observation is true of the AlphaFold3 structure of FtAlkBG. In these predictions the ferric iron ions were modelled in, and the structural predictions were satisfactory as ascertained by the pTM scores (Table S1). Regions of disorder were only observed in N- and C-termini, which is to be expected and does not impact the subsequent structural alignment. To allow identification of cavity residues more easily, each predicted structure was aligned with the cryo-EM structure of Fontimonas thermophilia AlkB (PDB Code: 8SBB)[16] using the Structural Alignment Tool[20] on the RCSB Protein Data Bank website (RCSB.org).[21] This alignment used the Template Modeling align algorithm[22] and showed excellent alignments with low root mean square deviation (Table S1) for all studied proteins. T-COFFEE was used for sequence alignment.[23, 24]

An analysis of the residues that line the substrate cavity was done by visualizing all of the aligned structures at once and selecting all residues within 5 Å of the dodecane molecule present in the FtAlkB structure. As the TM alignments created almost a nearly identical superposition of the backbone in the transmembrane region, visualizing the structural overlap was facile. These residues were compared to the sequence alignment, and placed into a table (Table 1).

Table 1.

Selected residues that line the AlkB substrate cavity. The first four enzymes prefer octane while the second set of four enzymes (highlighted with green fill) prefers longer alkanes.

Protein	Pos.	AA	Pos.	AA	Pos.	AA	Pos.	AA	Pos.	AA	Pos.	AA
FtAlkB	62	W	315	L	274	L	137	I	141	L	35	W
AbAlkB	60	W	311	L	270	L	133	I	137	L	33	W
PpAlkB_BMO	55	W	306	L	265	L	128	I	132	L	28	W
PpAlkB	55	W	306	L	265	L	128	I	132	L	28	W
DcAlkB	91	V	351	L	310	L	173	I	177	I	56	I
MtAlkB	69	L	333	L	292	L	151	V	155	V	34	I
PaAlkB1	52	V	306	L	265	L	127	T	132	V	24	L
PaAlkB2	52	V	306	L	265	L	127	T	132	I	24	A

Cavity Analysis

MoloVol v1.1.1[25] was used to quantify the cavity volume via a two-probe approach. For these calculations, a small probe radius of 1.3 Å and a large probe radius of 3 Å with a grid resolution of 0.2 Å was used. Hydrogens were added by ChimeraX[26–28] before analysis. The standard van der Waals radii set for the elements present in the program were used, which correspond to 1.20 Å, 1.77 Å, 1.66 Å, 1.50 Å, 1.89 Å, and 2.44 Å for H, C, N, O, S, and Fe, respectively. The optimal parameters for MoloVol were determined as described (vide infra, results) by trying a range of values based on literature precedent and program guidelines and then checked to ensure that they detected the substrate cavities already structurally identified in the published cryo-EM structures (Table S2–S3). An analysis of the results of these parameter explorations is included in the results and discussion section. The length across the longest cavity dimension was ascertained by fitting a convex hull to the surface produced by MoloVol, where the convex hull functionality in SciPy was used.[29] Visualizations of these convex hulls show a snug fit to the cavity surface (Figure S1).

CAVER 3.0.2 was used for tunnel calculation and clustering.[30] The parameters used in CAVER 3.0.2 for the modelled protein were as follows: minimum probe radius—1.2 Å; shell depth—4.0 Å; shell radius—3.0 Å; and clustering threshold—5.0. The tunnels were clustered by average-link clustering. The starting point was placed at the two active site iron ions. All frames (5000 each) from three 200 ns trajectories were analyzed.

Molecular Dynamics Simulations

The AlphaFold3 predicted structure of PpAlkB was used as input to the H⁺⁺ web server[31] at pH 7.4, in line with typical experimental measurements of AlkB activity. The interior dielectric was modeled as 10 (protein interior), and the external dielectric constant as 80 (water). A concentration of 150 mM salt was used. The metal coordinating residues were confirmed to be deprotonated before proceeding. All assigned protonation states used for titratable residues are provided in the Supporting Information (Table S4). Metal parameters to treat the diferric iron center were adapted from previous work,[32] where the fit parameters yielded metal-ligand bond distances that agreed well with EXAFS data of the FtAlkB protein. The protein was simulated without the substrate modelled in to avoid assuming a priori a substrate cavity or binding location. Orientation of Proteins in Membranes web server[33] (OPM) was used to predict the orientation of PpAlkB in the membrane with a membrane tilt angle of 19°. The aligned protein structure was used as input to CHARMM-GUI,[34] where a 80% phosphatidylethanolamine (POPE) membrane and a 20% phosphatidylglycerol (POPG) membrane was constructed to mimic the major lipid constituents of the P. putida inner membrane.[35, 36] Molecular dynamics simulations were performed with Amber24[37] and utilized the Lipid21[38], ff14SB[39], and TIP3P[40] forcefields for the lipid, protein and water, respectively. The system consisted of 135,659 atoms for PpAlkB WT with dimensions of 103.90 Å x 104.30 Å x 135.10 Å and simulated in triplicate for 200 ns, following equilibration procedures detailed in our previous work[32], given stepwise in Supporting Information Text S1. Supporting plots showing headgroup to headgroup distance and area per lipid support are intact and consistent with experiment POPE/POPG lipid membrane (Figure S2–S3). The RMSD of the protein backbone remains around 1.1-1.4 Å across all MD runs with and without C- and N-termini (Figure S4). These values suggest the protein is not changing rapidly from the AlphaFold structure and the simulations are stable, with values in line with values observed in our previous work.[32]

Results and Discussion:

General approach:

There are different ways to determine the substrate range of an enzyme. In the classical approach, the protein is purified, activity is reconstituted, substrates are added one at a time, and the identity and yield of products determined. This approach is limited because the environment of a purified protein almost certainly does not match the environment of the enzyme inside the cell. This is particularly true for integral membrane enzymes. Judicious choice of detergents can yield membrane proteins that are active; however, evidence indicates that these conditions often do not recapitulate optimal activity. Characterizing substrate preferences in whole cells has many advantages, but also the disadvantage that ancillary processes, e.g. substrate uptake, impact the observed reactivity. In this work, we used information about the substrate range of each enzyme from experiments done both with purified proteins and with cell-free extracts. Where possible, we confirmed that the differences between these two ways of collecting data were not significant. In cases where we have relied on published data that is not our own, we confirmed that the methods used to determine the substrate range were consistent with best practices.

We include in our analysis data from the four AlkBs whose purification has been reported (PpAlkB, FtAlkB, FtAlkBG and AbAlkB), as well as two single point variants of PpAlkB (W55V) and FtAlkB (W62V).[17, 41–51] These four AlkBs represent a minuscule fraction of the AlkBs identified by sequence homology. The substrate range for FtAlkBG has not been reported. Its reactivity with dodecane was used to indicate the purified protein used in cryo-EM studies was active. We have reported the substrate range for a structurally similar protein, DcAlkB, which is included in this paper.[52] In addition to these AlkBs, we include in our analysis the PpAlkB variant, BMO2, generated through a process of directed evolution by the Arnold group, which preferentially oxidizes butane.[18] Three other AlkBs, found in pathogenic organisms, are included in our analysis as well: PaAlkB1; PaAlkB2; MtAlkB. These chosen AlkBs range from having 35 to 74.4% sequence identity with each other, excluding PpBMO2 and other point variants. FtAlkB, PpAlkB, AbAlkB have over 60% sequence identity to each other, while DcAlkB/MtAlkB and PaAlkB1/PaAlkB2 also show higher sequence identity to each other than other AlkBs analyzed (Figure S5).

The public release of AlphaFold, including AlphaFold3, which allows for metal ions to be included in folded structures, has changed the landscape of structural biology.[19, 53] Since AlphaFold was trained on data from the PDB, it was initially unclear how well it would predict structures of AlkB and other novel proteins without structural analogues. We showed for AlkB that the AlphaFold structure matched well to the experimentally determined structure, providing confidence that using it to fold the structure of other AlkBs will generate reasonably accurate models.[16] These models, however, represent a single snapshot of a low temperature, potentially cryo-reduced, model of the protein. They do not capture the motions likely needed for substrate and product translocation.[54] To sample conformational motions, we ran classical molecular dynamics (MD) simulations at 310 K for one of our selected AlkBs (Pp) to explore protein internal architecture and identify cavities as a function of time and conformational sampling.

The cryo-EM structures of AlkB both show a substrate channel that enables straight chain alkanes to enter the enzyme from the periplasm, where they arrive after either active or passive import into cells.[55–57] In this work, we articulate and test the hypothesis that the substrate range a particular AlkB will oxidize depends on the shape and volume of the substrate channel. The substrate channel from FtAlkB, along with the channel lining residues, is shown in Figure 2.

Figure 2.

Labeled tunnel residues in FtAlkB from the FtAlkB cryo-EM structure. PDB code 8SBB was used to make this figure, and the surface of dodecane bound in the structure (ligand ID D12), is given as transparent pink.

Several tools have been developed for identifying cavities and channels in porous materials and proteins and measuring their volume and length.[25, 30] While algorithms vary, many use a rolling probe approach, in which two spheres are rolled over the structure. The smaller sphere interrogates the cavities and a larger sphere identifies the exterior of the object. The user must select the inner and outer sphere radii. Typically, one would select an inner sphere radius that approximates at least the radius of a hydrogen atom or a water molecule and an outer sphere radius that is larger than the anticipated pore openings. The primary substrates for AlkB are linear alkanes, whose smallest diameter is similar to that of water.

Results and discussion for each specific AlkB:

We first predicted the structures of the AlkBs using AlphaFold3 and aligned them to the FtAlkB reference. High template modeling scores and confidence metrics as reported by AlphaFold support the predictions (Table S1), as evidenced by the superior structural alignment (Figure 3).

Figure 3.

Overlaid cryo-EM and AlphaFold structures with corresponding color scheme used. FtAlkB = Fontimonas thermophila AlkB, AbAlkB = Alcanivorax borkumensis AlkB, PpAlkB = Pseudomonas putida GPo1 AlkB, BMO2 = Pseudomonas putida GPo1 Butane monooxygenase variant 2 AlkB from[18], MtAlkB = Mycobacterium tuberculosis AlkB, DcAlkB = Dietzia cinnamea, PaAlkB1 = Pseudomonas aeruginosa AlkB1, PaAlkB2 = Pseudomonas aeruginosa AlkB2. The disordered termini are omitted for visual clarity.

The first AlkB identified was isolated from the soil organism Pseudomonas oleovorans (later known as Pseudomonas putida GPo1).[10, 14] The substrate specificity of this protein has been studied in purified and whole cells.[58, 59] It prefers to hydroxylate octane but can oxidize alkanes as short as pentane and as long as dodecane, although with decreased efficiency.[15, 16, 58] When the residue first identified by van Beilen et al. is changed from W to V (W55V), the substrate preference shifts to longer alkanes.[15, 16] The active sites of PpAlkB and PpAlkB (W55V), as determined by AlphaFold3, are shown in Figure 4a as well as the substrate cavities identified by MoloVol. Cavity dimensions are provided in Table 2. The ability of the W55V variant to oxidize longer alkanes is supported by the finding of larger volume for the substrate cavity in this enzyme.

Figure 4.

PpAlkB cavities in PpAlkB WT and PpAlkB single and triple point mutants. Panel A: Overlay of PpAlkB cavity (dark blue) and PpAlkBW55V cavity (light blue). The key amino acid residue (W or V) is also shown in stick form. Panel B: The AF3 structure of the triple point mutant of PpAlkB (BMO2), has three residues that differ from PpAlkB. These are depicted as van der Waal spheres and labeled. The substrate channel as identified by MoloVol is shown as a pink surface

Table 2.

Substrate cavities identified by MoloVol. All cavities were identified using an inner probe radius of 1.3 Å and an outer probe radius of 3 Å in MoloVol. Cavity length was estimated by fitting a convex hull to the resulting volume element and taking the longest dimension. Proteins highlighted in orange have been shown to preferentially oxidize alkanes with a chain length of 8 carbons or less while those highlighted in green preferentially oxidize longer alkanes (generally greater than 10 carbons).

AlkB	Cavity volume occupied (Å3)	Cavity volume accessible (Å3)	Cavity length (Å)
PpAlkB	324.7	47.4	16.5
PpAlkB W55V	430.5	65.9	19.5
PpAlkB BMO2	333.7	55.1	15.9
AbAlkB	314.8	44.9	16.4
FtAlkB	327.3	50.7	16.2
FtAlkBW62V	437.5	73.0	17.9
PaAlkB1	425.4 + 133.9 (559.3)	67.0 + 10.4 (77.4)	27.0
PaAlkB2	727.9	111.3	30.0
MtAlkB	273.3+105.9+76.7 (455.9)	29.0+9.9+3.6 (42.5)	28.0
FtAlkBG	416.3	55.5	18
DcAlkB	431.9	59.7	18.6
DcAlkB V91W	258.9	27.5	14.6

The Arnold group reported a study in which evolutionary pressure was put on PpAlkB to evolve it to oxidize shorter-chain alkanes.[18] The most successful variant, BMO2, had three point mutations (V129M, L132V, and I233V). In this work, which includes discussion of multiple AlkBs, this variant is identified as PpAlkBBMO2. The AlphaFold3 structure of PpAlkBBMO2 with its substrate cavity shown as identified by MoloVol shown in Figure 4b. The dimensions of this cavity are provided in Table 2. While no significant difference in cavity volume is observed between PpAlkB and the PpAlkBBMO2 variant, the mutation of residues within the substrate channel (V129 and L132) is likely the cause of the substrate preference shift to shorter alkanes.

Alcanivorax borkumensis (Ab) was the most abundant organism found in the Gulf of Oman after the first Iraq war and is frequently associated with marine oil spills.[60] An obligate alkane-oxidizing organism, it harbors an AlkB that has been widely associated with the oxidation of medium chain alkanes in marine environments.[48, 61–74] The substrate specificity of this protein has been studied as a purified protein and in whole cells.[63] It prefers to hydroxylate octane but can oxidize alkanes as short as pentane and as long as dodecane, although again with decreased efficiency. The AlphaFold3 structure of AbAlkB is shown in Figure 5a as well as the substrate cavity identified by MoloVol and its dimensions (Figure 5a, Table 2).

Figure 5.

AlphaFold3 structures of Ab and PaAlkB. Panel A: AlphaFold structure of AbAlkB with substrate cavity shown as a surface. The tryptophan residues (W33 and W60) associated with substrate preference are represented with gray licorice. Panel B: Overlay of AlphaFold3 figures of PaAlkB1 (red) and PaAlkB 2 (orange) with their corresponding substrate cavities in the same color shown as surfaces. The residues in the key positions (L24 PaAlkB1/A24PaAlkB2V52 PaAlkB1/PaAlkB2) that vary between long and short alkane are given with red sticks.

Pseudomonas aeruginosa is an opportunistic pathogenic organism found in environmental, plant, and animal sources.[75] Strains of this organism, such as the wound-isolated PAO1 and the environmental RR1, contain two AlkB homologs (referred to as AlkB1 and AlkB2) which are structurally similar to the AlkB found in P. putida GPo1.[76–78] The substrate specificity of these AlkBs have been studied using analysis of cell growth utilizing n-alkanes as sole sources of carbons and energy.[79] AlkB1 and AlkB2 from P. aeruginosa PAO1 are predicted to preferentially oxidize dodecane through hexadecane based on their observed growth utilizing these n-alkanes.[79] A strain of P. putida GPo12 without the OCT plasmid and an alkb-deletion mutant of P. fluorescens both expressing the PaalkB1 and PaalkB2 genes have also been able to restore growth on alkanes ranging from dodecane to hexadecane.[80] Introduction of n-alkanes ranging from decane to C₂₂H₄₆/C₂₄H₅₀ induced alkB1 and alkb2 gene expression in the RR1 strain of P. aeruginosa, with the highest induction resulting from exposure to dodecane to octadecane.[78] The AlphaFold3 structures of both AlkBs and their substrate cavities as calculated by MoloVol are also shown in Figure 5b. The cavities in PaAlkB1 are separated into two adjacent cavities by a small bottleneck caused by the position of leucine residues (Figure S6). The dimensions of both cavities are provided in Table 2. We presume that dynamic fluctuations in sidechain conformations and uncertainty in predicted structure enable this to be a continuous channel by which alkanes can reach the enzyme active site. This is the first AlkB discussed in this paper with the ability to oxidize longer alkanes, and structural differences can be seen in the larger cavity volume and lack of bulky residues near the end of the cavity (Figure 5b). These findings underscore a common theme that volumetric data often must be examined on a case-by-case basis.

Mycobacterium tuberculosis has a functional AlkB whose role, if any, in pathogenicity is unclear.[81–83] It has been shown to preferentially oxidize alkanes from dodecane to hexadecane.[15, 79] The AlphaFold3 structure of MtAlkB is provided in Figure 6. MoloVol identifies three adjacent cavities, which are also shown in Figure 6. The dimensions of these cavities are provided in Table 2.

Figure 6.

AlphaFold3 structure of MtAlkB with substrate tunnel shown as a purple, blue, and pink surface. The substrate tunnel was not detected as a single cavity with a small probe radii of 1.3 angstroms, likely due to sidechain positioning for this given structure

Dietzia cinnamea is a bacterium that can cause infections in humans, particularly those that are immunocompromised.[84, 85] It contains an AlkB to which the proximate electron transfer partner, a rubredoxin, is covalently bound, a motif that is fairly common among AlkBs.[52] It is 74% similar to FtAlkBG, whose cryo-EM structure was recently solved.[17] We showed that it preferentially catalyzed the oxidation of nonane-undecane and that a mutation in the residue analogous to PpW55 (V91W), generated a variant with a marked shift in substrate preference away from longer alkanes and towards shorter alkanes.[52] The AlphaFold3 structures of DcAlkB and DcAlkBV91W are provided in Figure 7. The cavity volumes are provided in Table 2 and show a significant change in volume, easily seen in Figure 7, and consistent with the observed changes in substrate specificity.[52]

Figure 7.

AlphaFold3 structure of DcAlkB (yellow) and DcAlkBV91W (gray) with substrate cavity and position 91 illustrated.

While static structures provide a lens to compare across species, they do not capture sidechain rotamers or other conformational changes that could impact channel shape and size. Thus, we ran molecular dynamics simulations of three trajectories of 200 ns each to verify that the static structure substrate channel we observe in PpAlkB is consistent with MD calculations. Indeed, substrate channels are observed (yellow, Figure 8), albeit at a lower frequency in general in the simulations (Table S5). We compared the positions of bulky tryptophan residues in snapshots with and without tunnels. While some changes in the position of W residues appear to create a more open vs a more closed conformation, rigorously determining this will require free energy calculations and advanced sampling approaches (Figure S8). The absence of a detected substrate channel in some of the MD simulations is probably because, in the absence of water or substrate, interactions between the hydrophobic residues that line the channel keep the channel diameter below the detection limit. Both W55 and W28 are near the end of the substrate channel for all aligned snapshots of PpAlkBs (Figure S7). We also detect additional cavities that we hypothesize may be for molecular oxygen or water, depending on whether they open into the membrane or bulk solution respectively. Work is ongoing by our group to better understand the factors that orchestrate the arrival of substrate and protons and electrons and that facilitate the release of products. Other chemists have begun to explore substrate translocation in FtAlkBG (PDB code: 8F8T), and have examined substrate translocation of dodecane through the analogous tunnel in yellow (Figure 8) in their MD trajectories.[54] Our aim is to predict and analyze the size of substrate that would be preferred from an apo structure across different species, ideally without expensive classical simulations. Interestingly, FtAlkBG has isoleucine as opposed to W in these positions and the translocation out of the substrate cavity occurred rather easily when it was simulated in FtAlkBG. These results show the potential of using classical simulations, predicted structures, and cavity analysis to understand alkane size selectivity in AlkB

Figure 8.

Tunnel analysis performed in Caver for PpAlkB MD simulations. 5000 frames from each 200 ns trajectory were analyzed, with one structure from each trajectory given in transparent cartoon. All identified tunnels are shown and colored as yellow, red, blue, or green. The red, blue, and green tunnels have hypothesized functions based on their opening to the surface or membrane. The membrane and disordered termini are not shown for visual clarity.

Discussion of cavity programs.

Tools for calculating tunnels and cavities in proteins vary in their approaches.[86] Caver[30], Fpocket[87], Voidoo[87], 3V[88], and MoloVol[25] are among the most commonly used. We chose MoloVol for its ease of use and speed, and its focus on cavity detection and volumes. Caver focuses on the length of tunnels and channels and gives more statistics about radii. The two programs use different algorithms and give different information and both were employed. As with all cavity analysis, the choice of atomic radii and how finely space is discretized is important. Additionally, cavity analysis is sensitive to protein conformational changes, such as side chain placement. Describing what the surface of a protein really is, is often difficult to discern. Thus, these analyses must be performed with care. More complexities arise when analyzing AlkB, as there are no experimental structures of an apo form of AlkB, or one in which products are trapped. Having additional structural data would greatly enhance our understanding of alkane selectivity. It seems likely that in an apo state the substrate channel in the hydrophobic transmembrane region would simply have the hydrophobic sidechains more tighter packed and would not be filled with water, (consistent with MD simulations) but in the absence of experimental data, we cannot be certain.

Discussion of the significance of these findings

For almost 50 years, researchers have probed AlkB, trying to understand the structural determinants of its reactivity, or engineer it to catalyze new reactions. The nature of our understanding changed markedly with the publication of two cryo-EM structures of AlkB and the emergence of AlphaFold3, whose predicted structures overlay well the published structures.[16, 17] Querying AlphaFold3 structures with experimental data also provides confidence in their validity. For example, PpAlkBBMO2 is a PpAlkB variant that emerged from a series of directed evolutionary screens designed to evolve an AlkB that was better adapted for catalyzing the oxidation of short chain alkanes. Three point mutations occurred in the most evolved enzyme. AlphaFold3 structures indicate that two of the three mutations occur in the substrate channel, very near the residue first identified by van Beilen as being important in determining substrate selectivity.[15] No significant difference in cavity volume is detected by MoloVol. This suggests that conformational sampling or changes may be important to consider, and while point mutations can be generated with AlphaFold, its inability to predict structures of point mutations that are destabilizing or have alternate folds is a known limitation.[88, 89]

AlkBs that oxidize longer alkanes consistently have a less bulky amino acid in the pivotal position, first identified by van Beilen.[15] (See Table 1). Some positions that line the substrate channel are completely conserved among all the AlkBs we have examined, but the FtAllkBW62 and FtAlkBW35 positions are more variable. We identify here that FtAlkBW35 also is a smaller residue in some of the AlkB species that act on longer alkanes.

There is less information about AlkBs that oxidize even longer alkanes (> tetradecane) at the same time there are several reasons why heightened attention is currently being given to these AlkBs. The first reason for increased interest in long-chain degrading AlkBs is the discovery less than five years ago of the “cryptic carbon cycle”.[90] Researchers have shown, in two different and complementary ways, that cyanobacteria produce between 300-800 million tons of alkanes, most of it pentadecane. Pentadecane does not build up in the water column, presumably because it is degraded almost as rapidly as it is produced. Which enzymes are degrading pentadecane is not known, but high levels of alkB genes have been found in many of the impacted areas.[90, 91] The second reason is an association between polyethylene (PET) degradation and AlkB. PET is made up of very long chain of hydrocarbons and is resistant to degradation (both biological or abiological). Precisely what role AlkB plays in instances with enhanced microbial degradation of PET is not clear, but presumably AlkB can functionalize long alkanes that are severed from the primary polymer.

AlkM is a name that has been given to a group of AlkB-like proteins that facilitate bacterial growth on long-chain alkanes (~ctadecane and longer), and whose genetic organization is different than that of the canonical PpAlkB. Initially the name was given to distinguish them from AlkB[92] but the name now seems misplaced because many AlkBs are regulated different from PpAlkB and are found in different gene neighborhoods.[52, 93, 94] AlkM proteins are indistinguishable from AlkB proteins on the basis of both sequence and structure (as predicted by AlphaFold). Largely found in the Acinobacter family, there are few structural clues as to why enzymes placed in this category are capable of catalyzing the oxidation of alkanes that are solids at room temperature.[95]

An important frontier then in AlkB research is to understand what distinguishes AlkBs that can oxidize solid alkanes from those that oxidize liquid alkanes. The evidence linking a particular enzyme to the oxidation of longer alkanes is more circumstantial than the evidence we required to select the cases described here. No purified AlkBs have been shown to oxidize longer alkanes and the oxidation product, while presumed to be the corresponding aliphatic alcohol, has not been confirmed. There are three AlkBs: the AlkB from Dietzia sp. DQ12-45-1b (named AlkW1)[96], and two AlkBs from Rhodococcus sp strain 91[97], strongly correlated with the ability to oxidize alkanes longer than C14. The Alpha-folded structures of these AlkBs have a cavity that points from the active site into the cellular membrane through a gap in the membrane spanning α-helices (Figure S9). Additional experiments are required to determine whether this observation is significant or not, and is a direction for future work.

Conclusions.

Rapid advances in biomolecular structural tools have given chemists an advantage in approaching biological questions because of the paradigmatic way that chemists seek to connect structure and function. However, as we begin to use these tools, we see limitations that arise from the static single conformation of the structures that are generated and the concomitant need for more efforts to connect predicted structures and experimentally determined functions. This investigation into the structural determinants of AlkB substrate selectivity supports the earlier observation that there is a key amino acid residue (or key amino acid residues) that is (are) important in substrate discrimination. More experimental work is needed to categorize the substrate selectivity of additional AlkBs, especially those that preferentially oxidize alkanes with 14 or more carbons. AlkB remains a difficult protein to work with because of its integral membrane nature, its requirement for two iron ions in the active site, and because the substrates are inherently lipophilic. In vivo, AlkB is assisted in metabolizing alkanes by a host of cellular machinery, including an alkane importer AlkL.[55–57] Developing appropriate analytical tools to definitively assess the substrate specificity of a larger set of AlkBs and obtaining structures of AlkBs with products bound and in apo forms will also assist the effort to generate a comprehensive picture of how the structure of AlkB facilitates its function so that it will soon be possible to predict the substrate range of any given AlkB from its amino acid sequence alone.

Supplementary Material

Structural alignment metrics and UniprotIDs of corresponding AlkB sequences analyzed, MoloVol sensitivity tests, molecular dynamics simulation details and analysis. Additionally, a Zenodo repository with AlphaFold structures, forcefield parameters, raw data from MD analysis of tunnels, and other computational data is available at 10.5281/zenodo.17933473

Figure 1.

Picture from original paper by J van Beilen et al. that identified the substrate channel and critical residue in AlkB.[15] Reproduced by permission.