Role of loops connecting secondary structure elements in the stabilization of proteins isolated from thermophilic organisms

Abstract

It has been recently discovered that the connection of secondary structure elements (ββ-unit, βα- and αβ-units) in proteins follows quite stringent principles regarding the chirality and the orientation of the structural units (Koga et al., Nature 2012;491:222–227). By exploiting these rules, a number of protein scaffolds endowed with a remarkable thermal stability have been designed (Koga et al., Nature 2012;491:222–227). By using structural databases of proteins isolated from either mesophilic or thermophilic organisms, we here investigate the influence of supersecondary associations on the thermal stability of natural proteins. Our results suggest that β-hairpins of proteins from thermophilic organisms are very frequently characterized by shortenings of the loops. Interestingly, this shortening leads to states that display a very strong preference for the most common connectivity of the strands observed in native protein hairpins. The abundance of selective states in these proteins suggests that they may achieve a high stability by adopting a strategy aimed to reduce the possible conformations of the unfolded ensemble. In this scenario, our data indicate that the shortening is effective if it increases the adherence to these rules. We also show that this mechanism may operate in the stabilization of well-known protein folds (thioredoxin and RNase A). These findings suggest that future investigations aimed at defining mechanism of protein stabilization should also consider these effects.

Introduction

One of the fundamental issues in both chemistry and biology is the identification of the structural determinants that dictate protein folding and stability.^1,2 The decoding of the folding code of protein structures would have a major impact on native structure prediction and on de novo design.² This task is particularly difficult to achieve. Unlike synthetic polymers, protein structures combine complexity, fine-tuning, and marginal stability. The replacement of a single atom may have dramatic consequences on the overall architecture of these macromolecules made of thousands of atoms. In general, these structures rely on the delicate balance of a variety of different energetic factors whose contribution is difficult to be dissected.

Despite these difficulties, in recent years, major progresses have been made.^3,4 Among other contributions, the development of Rosetta package has provided a valuable tool for structure prediction and design.² A very recent breakthrough in the field is represented by the discovery of Baker and colleagues that the juxtaposition of basic secondary structure elements (α-helices and β-strands) follows well-defined rules.^5,6 These investigations identified three fundamental rules for the preferences of βℓβ (strand–loop–strand), αℓβ (helix–loop–strand), and βℓα (strand–loop–helix) structural motifs. In particular, it was shown that the chirality of βℓβ and the orientation of βℓα/αℓβ strongly depend on the loop size. In this framework, local/global fold stabilizations/destabilizations may be achieved by ad hoc modifications of the loop connecting the two elements. Intriguingly, the rules have a general validity and are somewhat independent of amino acid sequence. The efficiency of these rules was supported by the ab initio design of five different protein folds.⁵ Notably, the designed scaffolds exhibited unusually high thermal stability, with many of them exceeding melting temperatures of 95°C. This finding was ascribed to the fact that these structures represent a sort of “ideal” models, as they were designed by strictly following the fundamental rules. On the other hand, natural proteins might follow these rules less stringently to achieve protein states that are somewhat structurally frustrated and on the edge of their folding stability.^7,8

The correlation between thermal stability and adherence to the rules prompted us to investigate on its possible involvement in the evolutionary selection of protein sequences in thermophilic organisms. Despite the efforts, the definition of structural determinants of protein thermostability is a highly debated issue for which no consensus view has been reached yet.^9–20 We here compared the impact of these rules on databases of protein structures isolated from either (hyper)thermophilic or mesophilic organisms. By assuming that proteins isolated from (hyper)thermophiles are, in general, more stable than those isolated from mesophiles, we evaluated the correlation between these rules and protein thermostability.

Results and Discussion

Trends of the general, nonredundant database

Initial analysis of the chirality or orientation of secondary structural elements as a function of loop lengths was performed using a general and nonredundant database composed of protein chains sharing sequence identities lower than 25% (Data25). Following the definitions of Baker and colleagues,⁵ the so-called fundamental rules were checked on three distinct classes of structural units (βℓβ, βℓα, and αℓβ—“Methods” section and Supporting Information Fig. S1 for the notations and definitions). In the following sections, each state, characterized by a defined loop length, was defined as a very high, high, medium, and low preference state when the occurrence of the most abundant orientation/chirality was >95, 80–95, 70–80, and 60–70%, respectively (Table I). When the occurrence of the most abundant conformation was lower than 60%, no preference was assumed. Preferences exhibited by rare loops (population, <5%) were also considered not significant. Preferences exhibited by uncommon loops (occurrence in the range 5–8%) were considered significant only if high or very high preferences were detected.

Table I.

Occurrence of chirality/orientation of the supersecondary elements βℓβ, βℓα, and αℓβ in different data sets of protein structures^a [Color table can be viewed in the online issue, which is available at http://wileyonlinelibrary.com]

Loop size	βℓβ			βℓα			αℓβ
ℓ	L (%)b	R (%)b	Loop (%)c	P (%)d	A (%)d	Loop (%)c	P (%)d	A (%)d	Loop (%)c
Data25
1	75	25	0	47	53	2	97	3	1
2	100	0	40	87	13	9	88	12	7
3	85	15	6	33	67	9	71	29	18
4	67	33	16	62	38	11	66	34	20
5	9	91	15	51	49	13	62	38	15
6	35	65	6	48	52	12	57	43	10
7	57	43	3	52	48	8	57	43	8
8	62	38	3	55	45	7	57	43	6
9	59	41	2	53	47	5	57	43	4
10	59	41	2	51	49	4	62	38	3
DataEc
1	0	0	0	62	38	1	90	10	1
2	100	0	40	85	15	10	85	15	7
3	89	11	6	30	70	10	75	25	19
4	63	37	16	57	43	10	62	38	22
5	6	94	14	52	48	15	59	41	15
6	41	59	6	47	53	11	56	44	10
7	57	43	3	51	49	8	55	45	8
8	55	45	3	48	52	8	54	46	6
9	65	35	3	54	46	5	53	47	4
10	57	43	2	50	50	4	61	39	2
DataHs
1	20	80	0	44	56	1	98	2	1
2	100	0	36	90	10	9	78	22	8
3	86	14	6	29	71	10	75	25	16
4	75	25	18	64	36	9	65	35	18
5	8	92	15	54	46	16	58	42	14
6	40	60	7	39	61	14	48	52	10
7	49	51	3	54	46	8	54	46	8
8	65	35	3	53	47	6	51	49	6
9	61	39	2	61	39	4	61	39	4
10	54	46	2	49	51	4	61	39	3
DataPyr
1	0	0	0	56	44	1	100	0	1
2	100	0	45	87	13	10	84	16	11
3	74	26	8	32	68	10	83	17	23
4	55	45	18	62	38	9	73	27	24
5	5	95	9	49	51	14	66	34	14
6	30	70	5	42	58	15	64	36	8
7	44	56	2	56	44	9	56	44	5
8	37	63	3	50	50	6	57	43	4
9	62	38	2	44	56	5	62	38	2
10	33	67	1	41	59	5	62	38	2
DataSul
1	0	0	0	0	100	1	100	0	1
2	100	0	47	88	12	9	95	5	8
3	85	15	6	23	77	12	70	30	25
4	56	44	16	58	42	13	57	43	23
5	17	83	11	52	48	16	66	34	13
6	43	57	7	27	73	10	61	39	10
7	23	77	4	33	67	6	60	40	5
8	75	25	1	55	45	8	61	39	4
9	50	50	2	62	38	3	81	19	3
10	40	60	2	35	65	5	62	38	2
DataTm
1	0	0	0	64	36	2	100	0	1
2	100	0	50	89	11	10	82	18	8
3	88	12	7	30	70	11	78	22	21
4	55	45	13	57	43	12	67	33	25
5	9	91	10	50	50	14	66	34	16
6	46	54	5	48	52	12	54	46	10
7	59	41	3	49	51	8	53	47	7
8	69	31	3	56	44	6	71	29	4
9	33	67	2	65	35	4	57	43	3
10	44	56	2	34	66	4	65	35	2

^aIn the general database Data25, very high/high preferences are denoted in red, whereas medium/low preferences are denoted in cyan (see the text for definitions).

^bRelative occurrence, expressed as percentage, of the R- and L-states for a hairpin with a given loop length.

^cPercentage of motifs with a given length of the loop. The sum is lower than 100 since motifs with loops larger than 10 residues are considered in the statistics but are not reported in the table. The total number of βℓβ loops is 12458, 2222, 6615, 677, 333, and 595 for Data25, DataEc, DataHs, DataPyr, DataSul and DataTm, respectively. The total number of βℓβ loops is 9245, 2610, 3218, 751, 374, and 853 for Data25, DataEc, DataHs, DataPyr, DataSul and DataTm, respectively. The total number of αℓβ loops is 11547, 3165, 3846, 964, 467, and 1098 for Data25, DataEc, DataHs, DataPyr, DataSul and DataTm, respectively.

^dRelative occurrence, expressed as percentage, of the A- and P-states for αℓβ and βℓα motifs with a given loop length.

βℓβ motifs

The search for βℓβ motifs in the database leads to the identification of 12,458 hairpins isolated from 4498 structures. The fundamental rule for this motif assesses that for short loops (two to three residues) L-hairpin is preferred (βℓβ-ℓ2 and βℓβ-ℓ3, see “Methods” section for notations). A preference for the R-chirality is shown by the hairpin βℓβ-ℓ5. In line with these findings (Table I), we observe that βℓβ-ℓ2 exclusively adopts a L-chirality (100%). A high preference for this configuration is also shown by βℓβ-ℓ3 (85%). The increase of the loop length reduces the preference for the L-state. Hairpins with ℓ = 5 show a strong preference for the R-chirality. These findings are in line with those reported previously.⁵ A further increase of the loop length smoothes chirality preferences. It is also worth noting that loops longer than six residues are not very common in protein structures. In conclusion, significant preferences are observed for ℓ = 2 (L, very high), ℓ = 3 (L, high), ℓ = 4 (L, low), ℓ = 5 (R, high), and ℓ = 6 (R, low). It is important to note that the vast majority of βℓβ motifs found in globular proteins (83%) are characterized by loop sizes (ℓ2–ℓ6) that have a preference for one chirality. This finding indicates that the fundamental rules identified by Baker and colleagues for βℓβ hairpins play a major role in the build-up of natural globular proteins containing this motif.

βℓα motifs

The 9245 βℓα motifs, identified in 3456 structures of the database, were classified according to the relative orientation of the strand and the helix (parallel P or antiparallel A) as a function of the loop length. As previously observed, preferences were observed for βℓα motifs with ℓ = 2 (P, high), ℓ = 3 (A, low), and ℓ = 4 (P, low). For this motif, states with no preference for a specific orientation are highly populated (71%). This suggests that the rule about the strand/helix orientation is less widespread in natural proteins compared to that related to β-hairpins.

αℓβ motifs

As for βℓα, the 11,547 αℓβ motifs, identified in 3762 structures of the database, were classified according to the relative orientation of the strand/helix (P or A) as a function of the loop size. As previously observed, short loops prefer the P-orientation, whereas longer ones do not show any preference. In particular, preferences were observed for αℓβ motifs with ℓ = 2 (P, high), ℓ = 3 (P, medium), ℓ = 4 (P, low), and ℓ = 5 (P, low). For this motif, the occurrence of loop sizes with a significant preference for a specific orientation represents the majority of the αℓβ motifs (60%).

Overall, the data retrieved from the general database of protein structures fully confirm the previous findings. We evaluated the possibility that exceptions to the fundamental rules could (at least in part) depend on the accuracy of structure determinations. In particular, we plotted and compared the values of the resolution, R-factor, and R-free values for structures containing highly preferred or uncommon states. For preferred states, we selected βℓβ-ℓ2 (Type L), βℓβ-ℓ3 (L), βℓβ-ℓ5 (R), βℓα-ℓ2 (P), and αℓβ-ℓ2 (P). For uncommon states, we selected the less populated chirality/orientation for the same states (βℓβ-ℓ2 (R), βℓβ-ℓ3 (R), βℓβ-ℓ5 (L), βℓα-ℓ2 (A), and αℓβ-ℓ2 (A)). Resolution and crystallographic R-factor and R-free of the structures containing uncommon states are similar to those containing preferred ones (data not shown). This suggests that the observed relative occurrence of preferred/nonpreferred states is not related to the overall quality/accuracy of the three-dimensional structures.

Trends detected on databases of structures isolated from mesophilic or thermophilic organisms

The focus of the present investigation is to elucidate the role that fundamental rules play in the stabilization of proteins isolated from thermophilic organisms. To this end, we evaluated the occurrence and the impact of these rules on structures of proteins isolated from either thermophilic or mesophilic species (for details, see “Methods” section). As representative of mesophiles, protein structures from Escherichia coli (DataEc), and Homo sapiens (DataHs) were considered. For thermophiles, proteins from Thermotoga maritima (DataTm), the genus Pyrococcus (DataPyr), and the genus Sulfolobus (DataSul) were considered. These organisms were selected on the basis of their optimal growth temperature and on the relatively high occurrence of structures in the Protein Data Bank (PDB). It is worth mentioning that Sulfolobus and Pyrococcus species, both belonging to the archaea domain, present an optimal growth temperature of 75–80 and 100°C, respectively. T. maritima is the only bacterium with an optimal growth temperature as high as 80°C.

The analyzed preferences for specific chirality/orientation of the structural units generally follow the trends observed in the general database Data25 (Table I). It is worth noting that minor differences may also be owing to the limited occurrence of certain states in the specific databases. On this basis, the fundamental rules seem to be followed in the same way by the proteins isolated from organisms living in different conditions. However, some differences are observed when the occurrence of states with a specific loop size is considered. The most frequent state in both the general and the specific databases is βℓβ-ℓ2, a state that shows an exclusive preference for L-chirality. The analysis of individual databases unveils that this state is significantly over-represented in proteins of thermophilic organisms. Indeed, the percentage of βℓβ-ℓ2 state among all hairpins is in the range of 45–50% for DataTm, DataPyr, and DataSul, whereas it is 36–40% for DataEc, DataHs, and Data25. When structures of either thermophilic (DataTer = DataTm + DataPyr + DataSul) or mesophilic (DataMes = DataEc + DataHs) organisms are considered collectively, these differences persist. In particular, the percentage of βℓβ-ℓ2 in DataTer is 47%, whereas it drops to 37% for DataMes. The percentage of βℓβ-ℓ2 occurrence in ensembles generated by random selections (ten trials) of half of the DataTer structures is in the range of 46–49%. If the same procedure is applied to structures isolated from DataMes, this percentage remains close to 36–38%. The increase in the abundance of a very selective state in proteins isolated from thermophilic organisms may suggest that these proteins occasionally use a strategy aimed to reduce the possible conformations of the unfolded ensemble to stabilize the native state (negative design). Indeed, for states where the impact of the rule is prominent, the number of possible association modes of the secondary structure elements in the unfolded ensemble is reduced. As a result, a free energy contribution to the folding stability of thermophilic proteins could arise from the reduction in the entropic difference between the folded and the unfolded states. Notably, the entropic contribution to the free energy is particularly strong at high temperatures. Indeed, the shortening of the loop size from states with limited or no preferences to ℓ = 2 (very high preference for L-chirality) reduces the accessible states by excluding, also in the unfolded state, hairpins with R-chirality. For the αℓβ state, in the database of the proteins isolated from Pyrococcus, the states that present high or medium preferences (ℓ = 2 and 3) account for the 29% of the total αℓβ motifs. This value is significantly larger than that observed for DataHs (18%) or DataEc (21%). In DataSul (25%) and DataTm (23%), this value is intermediate between that observed in mesophiles and DataPyr. This may suggest that also in this case a sort of negative design may occasionally take place. Among the various databases, no significant difference is observed in βℓα motif, for which, as mentioned above, fundamental rules are less used in natural proteins.

Potential effects of rules on specific systems: the cases of thioredoxins and ribonucleases

We evaluated the potential impact of the fundamental rules on two proteins (thioredoxin [Trx] and pancreatic ribonuclease) that have been the subject of extensive investigations aimed at elucidating structure–stability relationships.

Trx belongs to the Trx fold class of oxidoreductases.²¹ Structures of this protein isolated from different organisms have been reported. It has been recently suggested that the exceptional stability of the Trxs isolated from Sulfolobus solfataricus and S. tokodaii may be owing to the shortening of the loop connecting helix 1 and strand 2. We performed a comparative analysis of the loop size of Trx structures isolated from thermophilic (S. solfataricus [TrxA1 and TrxA2], S. tokodaii, Thermus thermophilus) or mesophilic (E. coli, Spinach chloroplast) sources. As observed earlier, the loop connecting helix 1 to strand 2 in the first αℓβ motif undergoes a shortening from 6/7 residues in mesophilic organisms to two residues in thermophilic organisms (Fig. 1).^22,23 We analyzed this shortening in terms of the preferences here reported. αℓβ-ℓ6 and αℓβ-ℓ7 do not show any favorite orientation, whereas the αℓβ-ℓ2 state exhibits a strong preference for the P-orientation (Table I), the one observed in the final folded structures. This suggests that, in thermophilic Trxs, stabilization may occur by affecting the conformational freedom of the unfolded states. It should be mentioned that there is an additional shortening of the loop connecting helix 2 and strand 3 in two out of four Trxs isolated from thermophiles (TrxA2 of S. solfataricus and Trx from S. tokodaii). In this case, the reduction of the loop size (from 5 to 4) does not lead to significant alteration of the relative occurrence of the two orientations (Table I).

Figure 1.

Multiple sequence alignment of Trx isolated from different organisms: Spinach chloroplast, E. coli, S. solfataricus, S. tokodaii, and T. thermophilus. α-Helices and β-sheets are denoted in red and green, respectively. The residue numbering refers to SsTrxA1 sequence. The boxes identify the loop regions discussed in the text. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

We extended our investigations to pancreatic-like ribonucleases.²⁴ The prototype of this family, namely RNase A, displays a thermal stability of about 62°C. Onconase is a particularly thermostable member of the family (T_m = 89°C) which shares a sequence identity with RNase A of 30%.²⁵ Onconase is characterized by smaller loops being its sequence 20 residues shorter than RNase A although secondary structure elements are essentially preserved. Therefore, the system Onconase/RNase A represents an attractive model to study the impact of loop shortening and chirality/orientation preferences on protein stability. Some of the loop deletions occurring in onconase do not affect the pattern of preferences. On the other hand, the shortening of the loop of the final hairpin occurring in onconase (from four to two residues) may have an impact on the protein stabilization (Fig. 2). In both onconase and RNase A, the final hairpin shows a L-chirality which is required by the specific fold. However, in onconase the two-residue loop motif (βℓβ-ℓ2) exclusively prefers the L-chirality, whereas in RNase A the four-residue loop motif (βℓβ-ℓ4) shows only a low preference for this chirality. A smaller stabilizing effect is likely produced by a second deletion occurring in the second to last hairpin. For this R-hairpin, the loop size is reduced from 10 (no preference) to 6 (low preference for the R-chirality). Although other effects certainly play a role in the stabilization of onconase,²⁶ our data suggest that loop adaptation to the fundamental rules discovered by Baker and coworkers may be an important cofactor.

Figure 2.

Sequence alignment of RNase A and onconase. α-Helices and β-sheets are denoted in red and green, respectively. The residue numbering refers to RNase A sequence. The boxes identify the loop regions discussed in the text. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Concluding Remarks

The identification of the structural determinants of protein thermostability is an intricate problem that has not had any conclusive answer yet. It is commonly believed that different stabilization mechanisms operate for different proteins. Among other factors, it has been suggested that loop shortening is a widespread mode to achieve stabilization in proteins from thermophilic organisms. As loop size is highly correlated with chirality/orientation preferences according to the rules recently unveiled, simple loop shortenings¹⁵ may not be enough for stabilizing proteins. In other words, ad hoc variations that increase the preference for the chirality/orientation of secondary structure elements required by the protein fold may be crucial. Data here presented show that, in natural proteins, the optimization of these preferences by adapting the loop size is sometimes used in thermophilic proteins. This seems particularly relevant for βℓβ motifs. It is important to note, however, that exceptions to the rules are not rare in natural thermostable proteins. This observation suggests that in these cases a certain level of “frustration” is likely essential for proteins to carry out their biological functions. Although we are aware of the potential limitations of the approach used (size of the databases and heterogeneity of databases isolated from different organisms), we believe that future investigations aimed at defining mechanism of protein stabilization should also consider the aspects discussed here. A full definition of the impact of this type of stabilization would strongly benefit from the extensive use, in future studies, of mutagenesis investigations coupled with thermodynamic characterizations of specific protein classes.

Methods

Notation

Individual states are characterized by a specific structural motif (βℓβ, βℓα, and αℓβ) and a given loop size. In the text, each state is denoted by the three letters identifying the motif (βℓβ, βℓα, and αℓβ) followed by the notation ℓx, where x is the loop length (e.g., βℓβ-ℓ2).

Databases

Our statistical analysis was based on a number of different structural databases. The initial investigations, performed for comparative purposes with the previous data,⁵ were conducted on a significantly large database (Data25). This database is composed of an ensemble of structures selected from the PDB²⁷ by using the server PISCES (http://dunbrack.fccc.edu/PISCES.php).²⁸ Data25 includes X-ray structures of 6690 single-polypeptide chains, which satisfy the following criteria: resolution, ≤2.5 Å; R-factor, ≤ 0.3; sequence lengths, 40–10,000; and percentage of sequence identity, ≤25%. From this initial ensemble, 5349 structures containing at least one of the following motifs βℓβ, βℓα, and αℓβ were selected.

To gain insights into the role played by the fundamental rules in the thermostabilization of natural proteins, the preferred junction modes of secondary structural elements were also evaluated in data sets of protein structures isolated either from mesophilic or from thermophilic organisms. In particular, as examples of proteins from mesophilic organisms, two distinct databases composed by proteins isolated from E. coli (DataEc) or H. sapiens (DataHs) were considered. In the case of thermophiles, three distinct data sets containing structures of proteins belonging to T. maritima (DataTm) species and to Pyrococcus (DataPyr) and Sulfolobus (DataSul) genera were generated. The structures of each data set were selected by using PISCES and ad hoc manipulation. The criteria for structure selection were equal to those applied to generate Data25 with the exception of protein sequence similarity that was set to ≤90%. As a result, DataEc, DataHs, DataTm, DataPyr, and DataSul contain 1160, 3149, 318, 312, and 177 polypeptide chains, respectively.

Algorithm

An in-house program was written to evaluate either the chirality of βℓβ or the orientation of αℓβ and βℓα motifs. As shown earlier,⁵ βℓβ hairpins can show two distinct chiralities: left-handed (L) and right-handed (R) (Supporting Information Fig. S1). Similarly, αℓβ and βℓα motifs may be characterized by a parallel (P) or antiparallel (A) relative orientation of the helix and the strand in the unit (Supporting Information Fig. S1). Following the approach used by Baker and coworkers, the occurrences of these states were evaluated as a function of the loop size. The overall algorithm and the selection of the motifs were designed by following the methods used by Baker and coworkers.⁵

Glossary

DataEc, DataHs, DataTm

databases containing protein chains isolated from Escherichia coli, Homo sapiens, and Thermotoga maritima, respectively

DataPyr and DataSul

databases containing protein chains isolated from organisms belonging to the genera Pyrococcus and Sulfolobus, respectively

Data25

database containing protein chains with sequence identities lower than 25%

PDB

Protein Data Bank

Trx

thioredoxin

Supplementary material

Additional Supporting Information may be found in the online version of this article.