|Home | About | Journals | Submit | Contact Us | Français|
While overall hydrophobicity is generally recognized as the main characteristic of transmembrane (TM) α-helices, the only membrane system for which there are detailed quantitative data on how different amino acids contribute to the overall efficiency of membrane insertion is the endoplasmic reticulum (ER) of eukaryotic cells. Here, we provide comparable data for TIM23-mediated membrane protein insertion into the inner mitochondrial membrane of yeast cells. We find that hydrophobicity and the location of polar and aromatic residues are strong determinants of membrane insertion. These results parallel what has been found previously for the ER. However, we see striking differences between the effects elicited by charged residues flanking the TM segments when comparing the mitochondrial inner membrane and the ER, pointing to an unanticipated difference between the two insertion systems.
The overwhelming majority of mitochondrial proteins are encoded on nuclear chromosomes, synthesized in the cytoplasm, and post-translationally imported into mitochondria through translocon complexes in the outer and inner membrane (Bolender et al, 2008). In the case of inner membrane proteins, two different translocons (TIM23, TIM22) can mediate membrane insertion of proteins arriving from the intermembrane space (IMS) side of the inner membrane, while the Oxa1p translocon inserts membrane proteins coming from the matrix side (Neupert and Herrmann, 2007; Becker et al, 2009).
The role of the TIM23 complex is analogous to that of the Sec61 translocon, which mediates translocation and insertion of proteins in the endoplasmic reticulum (ER) membrane (Rapoport, 2007; Skach, 2009). For both translocons, transmembrane (TM) segments are thought to be integrated into the lipid bilayer by lateral exit from the translocon pore (Becker et al, 2009; Skach, 2009). Despite the apparent functional similarities, the TIM23 and the Sec61 translocons are not evolutionarily related (Herrmann, 2003; Dolezal et al, 2006). Thus, it is unknown to what extent the two translocons use similar principles of recognizing TM segments in membrane proteins.
Here, we present an analysis of the sequence characteristics governing the insertion of membrane proteins by the TIM23 complex. To measure the efficiency of TM-segment insertion into the inner membrane, we have taken advantage of the peculiar maturation pathway of Mgm1p, a nuclearly encoded, dynamin-like GTPase important for mitochondrial fusion, morphology, and genome integrity in Saccharomyces cerevisiae (Sesaki et al, 2003; Wong et al, 2003; Sesaki and Jensen, 2004). Mgm1p is sorted into the inner membrane by the TIM23 complex through a unique process called alternative topogenesis (Herlan et al, 2004) (Figure 1A). Mgm1p has a classical N-terminal mitochondria-targeting presequence followed by two hydrophobic segments and a C-terminal globular domain. After cleavage of the presequence by the matrix processing peptidase, the first hydrophobic segment is inserted into the inner membrane in about half of the molecules. This gives rise to the long isoform of the protein (l-Mgm1p). In the remaining Mgm1p molecules, the first hydrophobic segment slips through the TIM23 channel into the matrix whereupon the second hydrophobic segment is cleaved by a rhomboid-like protease, Pcp1p, in the inner membrane (Herlan et al, 2003, 2004; Schäfer et al, 2010), releasing a truncated C-terminal part of Mgm1p (s-Mgm1p) to the IMS. Both isoforms are required for function (Meeusen et al, 2006).
We reasoned that Mgm1p carrying an engineered hydrophobic segment (H-segment) in place of the first TM segment should give rise to the two isoforms in varying amounts, depending on whether the H-segment is integrated into the inner membrane or not. We therefore replaced the first hydrophobic segment of Mgm1p by a set of 19-residue H-segments of varying hydrophobicity (Figure 1B), for which we have already measured membrane insertion in the mammalian and yeast ER (Hessa et al, 2007, 2009). For immunodetection, a triple-haemagglutinin (HA) tag was fused to the C-terminus of Mgm1p in all constructs. C-terminal epitope tagging of Mgm1p with either HA or c-myc does not compromise protein function (Meeusen et al, 2006; Zick et al, 2009) (and our unpublished data), showing that the C-terminal HA tag disrupts neither targeting nor topogenesis. Each Mgm1p construct was expressed in yeast from a low-copy CEN plasmid, and the relative amounts of the two isoforms were quantified from western blots and used to calculate an apparent free energy of membrane insertion, (where R is the gas constant and T is the absolute temperature, T=298 K, and fl and fs denote the fractions of l-Mgm1p and s-Mgm1p molecules, respectively).
We first measured ΔGapp for a series of 19-residue Leu/Ala-based H-segments, with the composition GGPG-nL/(19-n)A-GPGG (all constructs are listed in Supplementary Table S1). The secondary-structure breaking GGPG…GPGG flanks were included to ‘insulate' the H-segments from the surrounding protein sequence (Hessa et al, 2005). As anticipated, H-segments with a higher number of Leu residues produced a higher fraction of the membrane-anchored l-Mgm1p isoform relative to s-Mgm1p (Figure 2A). The measured ΔGapp is linear with the number of Leu residues (n), with ΔGapp=0 kcal/mol (50% membrane insertion) reached for, n≈5–6 (Figure 2B). This ‘threshold hydrophobicity' is increased by roughly one Ala → Leu replacement compared with the yeast ER (Hessa et al, 2009) and by two such replacements compared with the mammalian ER (Hessa et al, 2005). These differences may at least in part depend on charged residues outside the GGPG…GPGG flanks (see below).
As a control for proper mitochondrial import, we compared the size of the l-Mgm1p isoforms from whole-cell lysates prepared from yeast transformants carrying either HA-tagged Mgm1p or the HA-tagged Mgm1p(6L/13A) construct with the corresponding in vitro-translated full-length products (including the presequence) (Figure 2C). A clear difference in molecular weight was observed between the in vitro-translated presequence-containing Mgm1p and the Mgm1p isoforms isolated from whole-cell lysates, indicating that the presequence has been cleaved from the Mgm1p constructs and therefore that they are correctly imported into the mitochondria.
To determine the contribution of each of the 20 naturally occurring amino acid to ΔGapp, we prepared a second set of H-segments (Supplementary Table S1). Each test amino acid was placed in the middle of the H-segment, so that it would be embedded in the hydrophobic core of the lipid bilayer when integrated into the membrane. As the maximal sensitivity of the assay is obtained when ΔGapp≈0 kcal/mol (50% membrane insertion), the Ala/Leu composition of the H-segments was adjusted such that, for each test amino acid, the efficiency of insertion was not too far from 50%. Assuming a simple additive model for the contributions of Leu and Ala residues to ΔGapp for the 19-residue H-segments shown in Figure 2B, we have n+1.64 kcal/mol, which yields =−0.20 kcal/mol and =0.09 kcal/mol. Possible contributions to ΔGapp from the Gly-Pro flanking sequences should be small (Hessa et al, 2005), and are ignored here. For the remaining 18 natural amino acids (X), ΔGappX was calculated from measured ΔGapp values as , see Supplementary Table S1.
Figure 3A shows the mitochondrial ΔGappX scale. As expected, the hydrophobic amino acids are at the lower end of the scale, whereas the charged amino acids are found at the higher end. Comparison to the scale obtained for membrane insertion into the mammalian ER shows a good correlation (R2=0.81) (Figure 3B), but the mitochondrial scale spans a somewhat broader range compared with the ER scale (−0.2 to +3.0 kcal/mol versus −0.5 to +1.8 kcal/mol) and the zero point is different, as noted above.
Meier et al (2005) have reported that the presence of Pro residues in TM segments renders TIM23-mediated membrane insertion less efficient. To better understand the role of Pro in TM helix insertion, we analysed a set of H-segments carrying a single or a symmetrically disposed pair of Pro residues in different positions. As seen in Figure 4A and B, Pro residues strongly reduce membrane insertion when located in the middle section of the H-segment and to a lesser but significant extent when near the ends. This suggests that formation of an α-helical structure of the H-segment is an important determinant for TIM23-mediated recognition and insertion of the H-segment into the inner membrane.
We next assessed the positional dependence of the effects on membrane insertion elicited by pairs of aromatic residues (Phe, Trp, Tyr; Figure 4C) and by single-charged residues (Lys, Asp; Figure 4D). Membrane insertion increased when Trp and Tyr were placed near the ends of the H-segment, while the effect of Phe was largely independent of position. The positively charged Lys residue was found not to affect insertion compared with the parent 6L/13A H-segment when placed in the first or last four positions. These results are consistent with our previous findings for the mammalian ER (Hessa et al, 2007). In contrast, the negatively charged Asp residue was only tolerated when placed in the four C-terminal positions, whereas in the mammalian ER Asp is tolerated at both the N- and C-terminal ends of the H-segment.
Finally, we investigated the role of charged and polar flanking sequences on H-segment insertion. On average, three positively charged Arg or Lys flanking residues (RRPR…, …RPRR, KKPK…, …KPKK) reduce ΔGapp by 1.0–1.3 kcal/mol from both the matrix and IMS sides compared with the GGPG… and …GPGG flanks (Figure 5). This corresponds to approximately four Ala → Leu replacements in the hydrophobic part of the H-segment.
In contrast, a negatively charged DDPD… flanking sequence on the matrix side increases ΔGapp by 1.1 kcal/mol, while a …DPDD sequence on the IMS side has little effect compared with the …GPGG flanking sequence, in agreement with the results for a single-Asp residue in Figure 4D. The same qualitative effects are seen with XXPX…,…XPXX flanking sequences where X=Glu, Gln, Asn, and His (see constructs #107–114, Supplementary Table S1); that is, it is not the charge per se but the high polarity of the side chain that matters in these cases. Interestingly, His does not behave as a positively charged residue in this context, which is consistent with the high pH of the mitochondrial matrix (around 8.0 in HeLa cells and rat cardiomyocytes (Llopis et al, 1998)), a value that is well above the pKa for the imidazole side chain.
Positively charged residues thus promote membrane insertion from both sides of the membrane. This is in contrast to the mammalian ER, where positively charged residues only promote insertion if they are present on the cytoplasmic side of the H-segment (Hessa et al, 2007). Negatively charged and highly polar residues reduce membrane insertion only when placed at the matrix side of the H-segment in the mitochondrial system, and only when placed at the lumenal side of the H-segment in the mammalian ER (Hessa et al, 2007).
Given the importance of flanking residues located outside the hydrophobic segment itself and the relatively high threshold hydrophobicity for membrane insertion seen with the Mgm1p constructs, we also determined the threshold hydrophobicity using a second inner membrane protein, CoxVa. CoxVa has a single TM segment that is integrated into the inner membrane via the TIM23 translocon (Glaser et al, 1990; Miller and Cumsky, 1993). We replaced the CoxVa TM segment by GGPG-nL/(19-n)A-GPGG H-segments of varying hydrophobicity and determined the efficiency of membrane insertion by an established protease-accessibility assay (Glaser et al, 1990) (Figure 6A). As seen in Figure 6B, 50% membrane insertion is observed for n≈2–3, compared with n≈5–6 for Mgm1p, corresponding to a difference of ~1 kcal/mol in the threshold hydrophobicity between the CoxVa and Mgm1p constructs. Although minor differences between in vivo and in vitro import assays may explain part of this difference (as has been seen for insertion into the ER (Hessa et al, 2005)), it is likely that sequence context outside the GGPG…GPGG flanks also impact the threshold hydrophobicity.
Positively and negatively charged flanking residues have similar effects on membrane insertion in the context of CoxVa as seen for Mgm1p, that is, positively charged flanks tend to increase insertion while negatively charged flanking residues on the matrix side of the hydrophobic segment strongly reduce insertion (Figure 6C).
Herlan et al (2004) have shown that mutational impairment of the mitochondrial import-motor components Tim44p and Pam18p/Tim14p reduces the formation of s-Mgm1p, implying that membrane integration of Mgm1p is increased in the absence of motor activity. We therefore tested the effect of the import motor on the balance between s-Mgm1p and l-Mgm1p in constructs with different H-segments. In case of wild-type Mgm1p, we observed a strong reduction of the amount of s-Mgm1p in the temperature-sensitive mutant pam16-3 of the motor subunit Pam16p (Frazier et al, 2004) grown at 30°C, the highest temperature at which cells can still grow (Figure 7A). The relative levels of s-Mgm1p were reduced in pam16-3 cells also for Mgm1p carrying H-segments of varying hydrophobicity, but to a smaller degree than for wild-type Mgm1p (Figure 7B). Similarly, for constructs with H-segments carrying charged or polar flanking residues, we saw a reduction in relative s-Mgm1p levels (Figure 7B); the effects of the pam16-3 mutation were especially large when the charged or polar residues are at the matrix-facing, N-terminal end of the H-segment. It thus appears that a fully functional import motor increases the threshold for H-segment membrane insertion in the context of Mgm1p, possibly by pulling on the nascent chain.
In summary, membrane integration of H-segments into the mitochondrial inner membrane depends on overall hydrophobicity and aromatic residues in a similar way as we have found previously for the same set of H-segments in the yeast and mammalian ER membranes (Hessa et al, 2007, 2009). Since these characteristics correlate with physicochemical properties of lipid bilayers, the data suggest that protein–lipid interactions contribute in an important way to TM helix integration both in the mitochondrial inner membrane and in the ER membrane.
While the presence of Arg or Lys residues near both the N- and C-terminal ends of the H-segment promotes insertion, the presence of negatively charged or highly polar residues at the matrix-facing, N-terminal end of the H-segment causes a severe defect in membrane integration. This effect is surprisingly strong: in the Mgm1p context, a markedly hydrophobic H-segment such as 11L/8A that is fully integrated into the inner membrane with neutral GGPG…GPGG flanks only integrates to about 20% with negatively charged DDPD… flanking residues on the matrix side (Supplementary Table S1, entries #9 and #106; see also the CoxVa constructs #118 and #127). A summary comparison of TM helix recognition and integration in the ER and mitochondrial systems is given in Table I.
What could be the basis for the different effects of positively and negatively charged flanking residues on membrane integration? An obvious possibility is that the electrical potential across the inner membrane, which is negative on the matrix side, could explain the strong reduction in membrane insertion caused by negatively charged residues at the matrix-facing end of a TM segment and the increased insertion efficiency caused by positively charged flanking residues. The reduction in membrane insertion seen when highly polar but uncharged residues replace Gly in the matrix-facing flanking sequences cannot be explained by the membrane potential, however; an interesting possibility is that it may reflect a transbilayer asymmetry in the translocon channel itself.
The positive-inside rule, which states that positively charged residues are more abundant in loops that connect TM segments on the cytoplasmic side (or, in the case of mitochondria, the matrix side), applies to membrane proteins in almost all membranes (Wallin and von Heijne, 1998). However, it has been noted that nuclear-encoded mitochondrial membrane proteins do not always follow the positive-inside rule (Gavel and von Heijne, 1992). Indeed, a statistical analysis of TM segments in mitochondrial membrane proteins that are imported either by the ‘conservative sorting' pathway (i.e. they are first translocated into the matrix and then inserted into the inner membrane from the matrix side (Hartl et al, 1987)) or by the ‘stop-transfer' pathway (i.e. they insert laterally from the TIM23 translocon into the inner membrane (van Loon et al, 1986)) shows that, in the latter group, positively charged Lys and Arg residues are abundant on both sides of the TM segments, whereas negatively charged residues seem to be under-represented on the matrix side (Figure 8). This is perfectly consistent with the effects on membrane integration of the Mgm1p and CoxVa constructs uncovered above.
The sequence of MGM1 gene was obtained from Comprehensive Yeast Genome Database (CYGD, Munich information center for protein sequences). This coding region has 21 extra amino acids (MSNSTSLRAIPRVANYNTLVR) at the N-terminus compared with the gene sequence given in UniProt (P32266). The MGM1 gene was initially amplified by PCR using genomic DNA isolated from yeast strain W303-1a (MAT a, ade2, can1, his3, leu2, trp1, ura3) and two primers 5′GTGGTTTGTTACGCATGCAAGCTTGATATCGAAATGAGTAATTCTACTTC3′ and 5′TGGTCTAGAGGTGTAACCACTTGAGTTCTTAGGTAAATTTTTGGAGACGC3′, and subcloned into pJK90 (Kim et al, 2003) as described (Kim et al, 2006). Then, using two primers 5′CTGGACACCCCTTTTCTG3′ and 5′CAATGGCGTACCCCATAC3′, and pJK90-MGM1 as a template, MGM1HA gene was amplified by PCR. This PCR product was subcloned into pHP84HA (Kim et al, 2000) by homologous recombination (Oldenburg et al, 1997), and this plasmid was named pHP84MGM1HA.
To facilitate subcloning by homologous recombination, a SmaI site was generated by an overlap PCR (Spee et al, 1993) in the first hydrophobic segment of MGM1 (291GTGGGTGGCCCCGGGATGGCT309) in pHP84MGM1HA, and this plasmid was named pJK110. The 19 amino acid long hydrophobic segments (H-segments) were amplified by PCR using various pGEM1-based constructs (Hessa et al, 2007) as templates and two primers, 5′ATCTAGTTTGTCCTTAGTAAAAGAGCTAGCTTCTTCTTGTTGCCCTGGTACC3′ and 5′GTTATCACCAGGAGATCGATATCACATTTTCCTAAACTTAAACTTTCCGAGACTAGTG3′ (underlined sequences are the sequences complementing upstream and downstream sequences of SmaI restriction enzyme site in pJK110 for homologous recombination, and the rest are the sequences complementing Escherichia coli Lep-H-segment sequence). For cloning of constructs #96–97 (KPKK-1L/18A-GPGG and GGPG-1L/18A-KPKK; Supplementary Table S1), available pGEM-based constructs (Lundin et al, 2008) were used as PCR templates with the two primers, 5′ATCTAGTTTGTCCTTAGTAAAAGAGCTAGCTTCTTCGCGCCGAAGACTAGT3′ and 5′GTTATCACCAGGAGATCGATATCACATTTTCCTAAATGACGGGATCGGTAC3′; for these, the five-residue flanks from the Lep vector are APKTS…VPIPS (c.f. Figure 1B). All plasmids carrying an H-segment were constructed by homologous recombination (Oldenburg et al, 1997) using a PCR amplified E. coli Lep-H-segment and a SmaI-digested pJK110 with Δmgm1a or α strains (MATa or α his3Δ 1; leu2Δ0; ura3Δ0; mgm1kanMX4) (Westermann and Neupert, 2000). Plasmids were isolated from yeast transformants, the correct sequences were confirmed by DNA sequencing. Correct constructs were then re-transformed into W303-1a (MAT a, ade2, can1, his3, leu2, trp1, ura3), selected on –Leu plates, and these yeast transformants were subjected to further analysis.
For in vitro expression of the full-length p-Mgm1p and p-Mgm1p-6L/13A with a presequence (residues 1–902 plus triple-HA tag), the DNA fragments were amplified by PCR using a pHP84MGM1HA or pHP84MGM1(6L/13A)HA as a template and the 3′ end primer 5′TTAGAGAGCGTAATCTGGAAC3′ and 5′ end primer 5′ATGAGTAATTCTACTTC3′. The PCR products were subcloned into a TOPO blunt end vector (Invitrogen) according to the manufacturer's protocol. Correct insertion into the vector was confirmed by restriction enzyme digestion analysis.
p-Mgm1HAp and p-Mgm1p-6L/13AHAp were expressed in vitro following the manufacturer's protocol for the T7 reticulolysate coupled transcription and translation system (Promega). Whole-cell lysate samples from yeast transformants expressing HA-tagged Mgm1p and Mgm1p(6L/13A) (see below) were subjected to TCA precipitation and analysed by SDS–PAGE and western blotting against an anti-HA antibody.
Yeast transformants carrying various MGM1HA H-segments were grown overnight in 5 ml of –Leu medium at 30°C and harvested by centrifugation at 3000 g. Whole-cell lysate preparation, SDS–PAGE, and western blotting using an antibody against the HA tag were done as described (Kim et al, 2006). Western blots were developed on a Fuji LAS-3000 system and the bands were quantitated using the Image Gauge V 3.45 software.
The temperature-sensitive mutant strain pam16-3 and the PAM16 isogenic wild-type strain (Frazier et al, 2004) were kind gifts from Dr Nikolaus Pfanner, Freiburg. MGM1HA carrying various H-segments were transformed into the pam16-3 and the isogenic wild-type strains. Yeast transformants were grown at 30°C as described above. Whole-cell lysates were subjected to TCA precipitation before loading on to the SDS gels. Western blotting and quantitation were done as described above.
The COXVa-coding sequence was amplified by PCR using the primers coxVa(EcoRI)F (5′CCGAATTCTGCACCAAACACAAGATCAACTAAGAACGC3′ and coxVa(BamHI)R (5′CCGGATCCCCTATTGATTTCCTTTCAAAATTTCTGGCC3′). The PCR fragment was treated with EcoRI and BamHI and inserted between EcoRI/BamHI-digested pGEM4z vector (Promega). COXVa H-segment constructs were made by overlap PCR using the primer pair SP6 5′ATTTAGGTGACACTATAG3′ and: CoxVa(P2)R 5′TAAGCGCAGACCAGTCGGCCATTCGCCTTCTTGCTTATCTCCAGGACCACCAATAAAACTGGAATC3′; CoxVa(Mgm1p)R 5′AGCCATCCCGCCACCCACATATATGGGTAACCTAATAATTTTAGATATAATAATAAAACTGGAATC3′; CoxVa(8L/11A)R5′CAGTAGGAGAAGCAATAACAGTGCGGCGGCAGCAGCTGCTCCAGGACCACCAATAAAACTGGAATC3′; CoxVa(5L/14A)R 5′AAGAGCTGCAAGTGCTGCTAAAGCTAATGCTGCAGCTGCTCCAGGACCACCAATAAAACTGGAATC3′ and the primer pair T7 5′TAATACGACTCACTATAGGG3′ and: CoxVa(P2) 5′TGGCCGACTGGTCTGCGCTTAAGTCGAATTGGCGGCATCGGACCTGGTGGGACCATGAATAAGGAG3′; CoxVa(Mgm1p) 5′TATGTGGGTGGCGGGATGGCTGCTGCAGGGAGTTATATAGCTTATAAGATGACCATGAATAAGGAG3′; CoxVa(8L/11A) 5′CTGTTATTGCTTCTCCTACTGTTAGCTGCCGCTGCTGCAGGACCTGGTGGGACCATGAATAAGGAG3′; CoxVa(5L/14A) 5′TTAGCAGCACTTGCAGCTCTTGCACTAGCTGCAGCAGCTGGACCTGGTGGGACCATGAATAAGGAG3′. The resulting fragments were cloned into the pGEM4z vector (Promega) by EcoRI/BamHI digestion. The constructs CoxVa(1L/18A), CoxVa(2L/17A), CoxVa(3L/16A), and CoxVa(4L/15A) were obtained by site-directed mutagenesis using construct CoxVa(5L/14A) as template. The constructs CoxVa(KKPK-3L/16A-GPGG), CoxVa(GGPG-3L/16A-KPKK), CoxVa(DDPD-3L/16A-GPGG), and CoxVa(GGPG-3L/16A-DPDD) were obtained by site-directed mutagenesis using construct CoxVa(3L/16A) as template. Constructs CoxVa(DDPD-5L/14A-GPGG), CoxVa(DDPD-8L/11A-GPGG), and CoxVa(KKPK-1L/18A-GPGG) were obtained by site-directed mutagenesis using constructs CoxVa(5L/14A), CoxVa(8L/11A) and CoxVa(1L/18A) as template.
The radiolabelled precursor proteins were synthesized in a cell-free translation system with rabbit reticulocyte lysate (Promega) in the presence of [35S] methionine. Isolated mitochondria from the yeast wild-type strain W3031-1a were incubated with radiolabelled precursor proteins in import buffer (250 mM sucrose, 10 mM MOPS-KOH, pH 7.2, 80 mM KCl, 5 mM MgCl2, 2.5 mm potassium phosphate, 2 mM ATP, 2 mM methionine, 2 mM NADH, and 1% (w/v) bovine serum albumin) at 25°C. The import reaction was stopped by addition of 10 μg/ml valinomycin. Mitoplasts were prepared by diluting mitochondria 10-fold in EM buffer (10 mM MOPS-KOH, pH 7.2, 1 mM EDTA). Protease treatment was performed by incubating the mitochondria (or mitoplasts) with 100 μg/ml proteinase K for 20 min on ice, which was inactivated by subsequent addition of 1 mM PMSF. Mitochondria (or mitoplasts) were reisolated by centrifugation, and proteins were analysed by SDS–PAGE and radioimaging with a Typhoon Variable Mode Imager 9200 (GE Healthcare) or Pharos FX Plus Molecular Imager (Bio-Rad).
Membrane integration of each H-segment was quantified from western blots by measuring relative amounts of l-Mgm1p and s-Mgm1p isoforms. The relative amounts of the two isoforms were used to calculate an apparent equilibrium constant between the membrane integrated and non-integrated forms: Kapp=fl/fs, where fl is the fraction of membrane integrated and fs is the fraction of membrane non-integrated isoforms. Finally, the membrane insertion efficiency was expressed as an apparent free energy difference, ΔGapp=−RT lnKapp. The ‘biological' ΔGappX scale for the 20 individual amino acids (X) was derived from a simple additive model, , where ΔGapp is the measured insertion free energy of the construct in question (see Supplementary Table S1). At least three independent experiments were carried out for each H-segment to measure ΔGapp.
Two sets of (mostly single spanning) TIM23-dependent mitochondrial inner membrane proteins were analysed, one containing eight proteins known to be first imported into the matrix and then inserted by the ‘conservative sorting' pathway into the inner membrane from the matrix side, and the other containing 21 proteins that are inserted laterally by the ‘stop-transfer' pathway from the TIM23 translocon into the inner membrane; see Supplementary Table S2 for sequence identifiers and references.
To analyse the charge distribution in the regions flanking the TM segments, the 10 N- and C-terminal residues flanking each TM segment (as predicted by SCAMPI (Bernsel et al, 2008)) were extracted. The mean frequencies of positively charged (R+K) and negatively charged residues (D+E) were calculated for the N- and C-terminal segments, along with the standard error of the mean.
This work was supported by grants from the Swedish Cancer Foundation, the Swedish Research Council, the Swedish Foundation for Strategic Research, and the European Research Council (ERC-2008-AdG 232648) to GvH, from the Research Settlement Fund for the new faculty of SNU and the National Research Foundation of Korea (NRF 0409-20100093) to HK, and from the Deutsche Forschungsgemeinschaft SFB 594 project B8 and the Cluster of Excellence ‘Macromolecular Complexes' at the Goethe University Frankfurt DFG Project EXC 115 to ASR. KY and TE acknowledge support of this work by Grants-in-Aid for Scientific Research from the Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT). SCB is a recipient of a graduate student fellowship (SFRH/BD/36107/2007) from Fundação para a Ciência e a Tecnologia, Portugal.
The authors declare that they have no conflict of interest.