Eukaryotic Translational Coupling in UAAUG Stop-Start Codons for the Bicistronic RNA Translation of the Non-Long Terminal Repeat Retrotransposon SART1

Western blotting.

Approximately 10⁶ Sf9 cells were infected in a 24-well plate with recombinant AcNPV at a multiplicity of infection of 10 PFU per cell. At 72 h postinfection, cells were scraped, pelleted at 1,000 × g for 5 min, and extracted. Cell extracts were separated on sodium dodecyl sulfate-polyacrylamide gels and electroblotted onto Fluoro Trans W polyvinylidene difluoride membranes (Nippon Genetics). Blots were blocked with 5% nonfat dry milk (Bio-Rad) in TBS-T (50 mM Tris-HCl, 0.3 M NaCl, 0.05% Tween 20, pH 8.0). Membranes were incubated for 16 h at room temperature with mouse monoclonal anti-HA antibody (Roche) diluted 1:10,000 or mouse monoclonal anti-His antibody (Amersham) diluted 1:3,000 in TBS-T. Bound antibodies were detected using an ECL Plus Western blotting detection system (Amersham) with peroxidase-conjugated anti-mouse immunoglobulin G (Amersham) diluted 1:25,000 in TBS-T. Between steps, membranes were washed according to the manufacturer's instructions.

SART1 ORF2 is translated separately from ORF1, not by frameshifting.

The primary question is whether SART1 ORF2p is translated by ribosomal frameshifting or is translated from the first AUG codon in ORF2. We expressed SART1 in an AcNPV expression system. A GST-His₆ tag was fused at the N terminus of ORF1p, and an HA tag was fused at the C terminus of ORF2p (Fig. ​(Fig.2A).2A). The sequence of the overlapping region of ORF1 and ORF2 in respective constructs is the same as in the native SART1. We made three constructs differing at the C-terminal truncation of ORF2p. S1-3666HA has ORF2 truncated just after the endonuclease domain (Fig. ​(Fig.2A,2A, ​,1).1). S1-4304HA has ORF2 truncated just before the reverse transcriptase domain (Fig. ​(Fig.2A,2A, ​,2).2). S1-6110HA has almost full-length ORF2 (Fig. ​(Fig.2A,2A, ​,3).3). In addition, we made a construct in which ORF1 and truncated ORF2 were fused (Fig. ​(Fig.2A,2A, ​,4).4). RNA including both ORF1 and ORF2 was presumed to be transcribed from the AcNPV polyhedrin promoter, based on the observation of our previous report (29). Because the SART1 5′ UTR was replaced by the GST-His tag sequence in the AcNPV-expressed SART1, the length of SART1 RNA expressed by AcNPV was 127 nt shorter than the native SART1.

Open in a separate window

FIG. 2.

ORF2 translation assay by Western blotting. (A) Structures of AcNPV-expressed constructs. ZF, zinc finger; EN, endonuclease; RT, reverse transcriptase; GST, glutathione S-transferase; HA, hemagglutinin influenza virus epitope tag. (B) AcNPV-expressed ORF1p visualized by Western blotting with anti-His antibody. Numbers of lanes indicate constructs shown in panel A. (C) AcNPV-expressed ORF2p visualized by Western blotting with anti-HA antibody.

Open in a separate window

FIG. 3.

Summary of ORF2 translation assay. Red filled circles indicate actual start codons, a red open circle indicates a non-AUG start site, and black open circles indicate in-frame AUG codons which were not used for translation initiation. The distances of AUG codons from the actual stop codon are shown below the ORF structures. Mutations are summarized at the middle, and the detailed sequences are at the bottom. Mutagenized nucleotides are in lowercase. Putative start codons are underlined, and actual start codons are red. Putative stop codons are boxed, and actual stop codons are in purple boxes. Results of the ORF2 translation assay are summarized at the right. ORF2p expression is indicated as the efficiency relative to the wild type (a). Minuses indicate no expression. MW, molecular mass.

Open in a separate window

FIG. 4.

Results of ORF2 translation assay. (A) AcNPV-expressed ORF1p visualized by Western blotting with anti-His antibody. Letters of lanes indicate constructs shown in Fig. ​Fig.3.3. The molecular weight of ORF1p of each construct is shown below. (B) AcNPV-expressed ORF2p visualized by Western blotting with anti-HA antibody. The arrow indicates the ORF2p bands.

GST-His-tagged ORF1p was expressed in three constructs at the expected size (106 kDa), and the GST-His-tagged ORF1-ORF2 fusion protein (133 kDa) was expressed in S1-F3666HA (Fig. ​(Fig.2B).2B). We observed only a single strong band in each lane. Faint bands in lanes 2 and 4 were probably degraded products. The HA-tagged ORF2p bands were observed in both S1-3666HA and S1-4304HA but not in S1-6110HA (Fig. ​(Fig.2C).2C). Even when we increased the exposure time, we could not detect ORF2p bands in S1-6110HA (data not shown). It is possible that no band for the full-length ORF2p (S1-6110HA) (Fig. ​(Fig.2C,2C, lane 3) was due to transcriptional disruption reported in the human L1 (9). The molecular masses of ORF2p of S1-3666HA (Fig. ​(Fig.2C,2C, lane 1) and S1-4304HA (Fig. ​(Fig.2C,2C, lane 2) were close to the estimated sizes, 27 kDa and 56 kDa, respectively, assuming that the first AUG in ORF2 was used for the initiation of translation. If ORF1-ORF2 fusion proteins were expressed, the estimated molecular masses were 133 kDa for S1-3666HA (Fig. ​(Fig.2C,2C, lane 1) and 162 kDa for S1-4304HA (Fig. ​(Fig.2C,2C, lane 2). However, we could not find any bands above 100 kDa (Fig. ​(Fig.2C,2C, lanes 1 and 2). We detected ORF1-ORF2 fusion proteins at the expected size (133 kDa) in S1-F3666HA (Fig. ​(Fig.2C,2C, lane 4). If ORF1-ORF2 fusion proteins were translated and processed in S1-3666HA and the band in lane 1 represented proteins after proteolysis, the ORF1-ORF2 fusion proteins in S1-F3666HA would also be processed (Fig. ​(Fig.2C,2C, lanes 1 and 4). Therefore, ribosomal frameshifting and proteolysis are unlikely mechanisms for the production of ORF2p.

The first AUG in ORF2 is used as a start codon.

To investigate whether the first AUG in ORF2 is the actual start codon of ORF2, we introduced a substitution of the first AUG in ORF2 to ACG (Fig. 3, b) or AGC (Fig. 3, c) in S1-3666HA (Fig. ​(Fig.2A,2A, ​,1).1). ACG has a weak activity for translation initiation in eukaryotic cells (30), and thus the order of translation initiation potency is AUG > ACG > AGC. When ACG is used for translation initiation, ACG is recognized by tRNA bound to methionine, as with AUG (30). The ORF1p was detected as an expected 106-kDa band in all three constructs (Fig. ​(Fig.4A,4A, a to c). The substitution from AUG to ACG reduced, but did not completely diminish, the ORF2 translation, whereas the substitution from AUG to AGC appeared to eliminate it (Fig. ​(Fig.4B,4B, a to c). To exclude the possibility of unexpected mutations in plasmid construction and recombinant AcNPV generation, we sequenced the mutant AcNPV genome. There were no unexpected mutations in ORF2 (data not shown). The reduction of ORF2 translation in mutants indicated that the first AUG in ORF2 is used as a start codon. The translation efficiency of ORF2 accorded with the potency of translation initiation of each codon. Judging from the ORF2p/ORF1p ratio quantified by densitometry, the translation efficiency of ORF2 in mutant b was approximately 4% of the wild type (a) (the results are summarized in Fig. ​Fig.3).3). Unexpectedly, in mutant b, ORF2p was translated from the mutated start codon ACG (3018), not from the second in-frame AUG (3096) codon 78 nt downstream from the native start codon (Fig. 3, b). If the translation started at the second AUG codon, the expected size of ORF2p would be 24 kDa, 3 kDa smaller. There was no band below 27 kDa (Fig. ​(Fig.4B,4B, b).

The overlapping stop-start codon is essential for ORF2 translation.

The next question is whether the overlapping stop-start codon is important for ORF2 translation. We made three other mutants. In mutant d, the substitution from UAA at position 3016 to 3018 to CAA formed another overlapping stop-start codon, AUGA (AUG plus UGA) at 3018 to 3021 (Fig. ​(Fig.3,3, d). The same overlapping stop-start codon, AUGA, is observed in some non-LTR retrotransposons (Table ​(Table2).2). In mutant e, the second stop codon, UGA at 3019 to 3021, was changed to UGG in addition to the first stop codon mutation, and the start codon of ORF2 was not mutated (Fig. ​(Fig.3,3, e). In this mutant, the stop codon of ORF1 is positioned at 3085 to 3087, 69 nt downstream from the native stop codon and 67 nt downstream from the native start codon of ORF2. If the 69-nt-downstream stop codon in mutant e is used, ORF1p is expected to be 108 kDa, which is 2.64 kDa larger than ORF1p in constructs a to d. In mutant f, two consecutive stop codons and the start codon were all destroyed (Fig. ​(Fig.3,3, f). In this mutant, the stop codon of ORF1 is also 69 nt downstream from the native stop codon, as in mutant e. The molecular weight of ORF1p in f is also expected to be 108 kDa. As the native start codon is mutated, AUG at 3096 to 3099 becomes the first AUG codon in ORF2 in mutant f, 8 nt downstream from the stop codon of ORF1.

The ORF1p bands in d to f could be detected (Fig. ​(Fig.4A,4A, a to f). The strong ORF2p band was observed in d, but no bands were detected in e and f (Fig. ​(Fig.4B,4B, d to f), even though the start codon of ORF2 was intact in e (Fig. ​(Fig.3,3, e). The molecular weight of ORF2p in d is equal to that in a, which supports that ORF2p in d is also translated from the first AUG codon at 3018 in ORF2. Judging from the ORF2p/ORF1p ratio quantified by densitometry, the translation efficiency of ORF2 in d was approximately 13% of wild type. This result indicates that another stop-start codon, AUGA, was used in d in place of UAAUG.

Greater distance between the stop codon and the start codon diminishes ORF2 translation.

The stop codon of ORF1 and the start codon of ORF2 are at their closest when they form overlapping stop-start codons. We considered that ORF2 was not translated in mutant e because the stop codon of ORF1 was 67 nt from the start codon of ORF2 (Fig. ​(Fig.3,3, e). Therefore, we further analyzed the effect of the distance between the stop codon of ORF1 and the start codon of ORF2. In mutant g, the stop codon of ORF1 is positioned at 2803 to 2805, 215 nt upstream from the start codon of ORF2, and there are no differences around the start codon of ORF2 (Fig. ​(Fig.3,3, g). In mutant h, the first AUG in ORF2 was mutated, and a new in-frame AUG was created at 2967 to 2969, 49 nt upstream from the stop codon of ORF1 (Fig. ​(Fig.3,3, h). The sequence near the stop codon of ORF1 in h is the same as that in c.

Although the ORF1p bands were observed in mutants g and h at the expected sizes of 98 kDa and 106 kDa, respectively (Fig. ​(Fig.4A,4A, g and h), we could not detect ORF2p (Fig. ​(Fig.4B,4B, g and h). Because the sequence around the first AUG in ORF2 of g is identical to the wild type (Fig. ​(Fig.3,3, g), the premature translation termination of ORF1 considerably reduced ORF2 translation. In addition, the artificial upstream AUG in h could not substitute for the native start codon.

Overlapping stop-start codon is necessary but not sufficient for ORF2 translation.

Next, we examined whether the overlapping stop-start codon is sufficient for ORF2 translation, introducing an artificial stop-start codon, UAAUG, 51 nt upstream from the native UAAUG. In both mutant i and mutant j, UAA at 2965 to 2967 is the stop codon of ORF1 and AUG at 2967 to 2969 is the first AUG in ORF2 (Fig. ​(Fig.3,3, i and j). There are no mutations near the native start codon at 3018 to 3020 in i. However, in mutant i, UAA at 3016 to 3018 is not the stop codon of ORF1, and AUG at 3018 to 3020, the start codon of ORF2 in wild-type SART1, is not the first AUG codon in ORF2. In mutant j, the native start codon, AUG at 3018 to 3020, is changed to AGC, which is the same mutation as in c and h. If artificial UAAUG were sufficient for the translation of ORF2, the 29-kDa ORF2p would be expressed.

We detected the slightly smaller ORF1p bands in both mutants (Fig. ​(Fig.4A,4A, i and j), but we could not detect clear ORF2p bands in either construct (Fig. ​(Fig.4B,4B, i and j). Thus, the overlapping stop-start codon is essential but is not enough for the translation of ORF2. These results showed that an artificial overlapping stop-start codon cannot replace the native overlapping stop-start codon. There is another factor necessary for the translation of ORF2.

Downstream RNA secondary structures also affect ORF2 translation.

In retroviruses, the frameshift signal is composed of a slippery sequence and a downstream stem-loop RNA secondary structure (16). Although there are no stable stem-loop structures in the ORF1-ORF2 overlapping region (2952 to 3015) of SART1, the RNA secondary structure prediction program at GeneBee-NET server (6) predicts a very stable stem-loop structure at position 3055 to 3111, downstream of the overlapping stop-start codon UAAUG (Fig. ​(Fig.5A,5A, left). We predicted another possibility for a pseudoknot structure manually (Fig. ​(Fig.5A,5A, right). Sequences 3055 to 3079 and 3088 to 3111 are predicted to form a stem, and this stem formation is necessary for both the stem-loop and the pseudoknot. If CUGCUG at 3081 to 3086 in the loop annealed to CAGCAG at 3023 to 3028, the stem-loop structure would change into the pseudoknot structure.

Open in a separate window

FIG. 5.

ORF2 translation assay for the downstream RNA secondary structure. (A) Predicted secondary structure downstream from the overlapping stop-start codon of SART1. Mutagenized sequences are shown below. (B) Conserved stem-loop structures of three related retrotransposons. Two bold lines in the SART1 structure indicate sequences assumed to be annealed in the pseudoknot model. (C) AcNPV-expressed ORF1p visualized by Western blotting with anti-His antibody. Letters of the lanes indicate constructs shown in panel A and Fig. ​Fig.3.3. (D) AcNPV-expressed ORF2p visualized by Western blotting with anti-HA antibody.

We also predicted RNA stem-loop structures just downstream of UAAUG in HOPEBm1 and HOPEBm2 (Fig. ​(Fig.5B),5B), which are close relatives of SART1 (17) and have the SART1-type overlapping stop-start codons (Table ​(Table22 and Fig. ​Fig.5B).5B). The spacing between UAAUG and the stem-loops in HOPEBm1 and HOPEBm2 is 34 nt, equal to SART1. The nucleotides forming stems are highly conserved among these three retrotransposons. However, HOPEBm1 and HOPEBm2 cannot form the pseudoknot structures that could be predicted in SART1 (Fig. ​(Fig.5B5B).

We made four constructs to investigate whether stem-loop or pseudoknot structures are important for ORF2 translation. In mutant k, the downstream stem-loop structure (3055 to 3111) was entirely deleted from S1-3666HA (Fig. ​(Fig.5A,5A, k). In this construct, the downstream region from UAAUG can form neither the stem-loop nor the pseudoknot. The expected size of ORF2p in k was 24 kDa. In mutants l and m, the presumed pseudoknot would be weakened but the stem-loop structure would not (Fig. ​(Fig.5A,5A, l and m). The presumed pseudoknot and the stem-loop structure would be as stable in mutant n as in the wild type because two sets of mutated nucleotides can form base pairs (Fig. ​(Fig.5A,5A, n).

We observed ORF1p bands of expected sizes in all constructs (Fig. ​(Fig.5C).5C). The ORF2p band appeared to be completely eliminated in k and reduced in l, m, and n (Fig. ​(Fig.5D).5D). Judging from the ORF2p/ORF1p ratio quantified by densitometry, the translation efficiency of ORF2 was low in l and m (approximately 6 to 10% of the wild type) but was recovered in n (approximately 20%), even though the efficiency was still lower than in the wild type (Fig. 5C and D). These results suggested that the stem formation is essential for the translation of ORF2 and that annealing between CUGCUG at 3081 to 3086 in the loop and CAGCAG at 3023 to 3028 is involved in the translation efficiency. It is also possible that the primary sequence is important for the ORF2 translation instead of the pseudoknot formation.

Translational mechanism of SART1 ORF2: translational coupling.

Several eukaryotic mechanisms for internal translation initiation have been reported (20). The scanning model predicts that the ribosome enters at the 5′ terminus of mRNA and migrates on mRNA from 5′ to 3′. One mechanism for internal translation initiation is leaky scanning, in which the ribosome bypasses AUG codons in a weak context for translation initiation (20). As there are 24 AUG codons before the ORF2 translation initiation site, leaky scanning can be excluded for the translational mechanism of SART1 ORF2. Another possible mechanism is nonlinear movement of the ribosome from the 5′ end of RNA to the first AUG in ORF2 according to ribosome shunting (8). If SART1 used ribosome shunting, the shunting ribosome would translate ORF2 and the remaining ribosome would translate ORF1, and the ORF2 translation would not be affected by the position of ORF1 translation termination. Thus, the above result that the premature translation termination of ORF1 diminished ORF2p translation (Fig. ​(Fig.33 and ​and4,4, g and i) excludes the possibility of ribosome shunting.

Internal ribosome entry represents another way to initiate translation far downstream of the 5′ end of RNA (32). Internal ribosome entry needs a specific RNA secondary structure, named an internal ribosome entry site (IRES). Insect picorna-like viruses have IRESs in their intercistronic region to translate their bicistronic RNA (38), although most viral and cellular IRESs are positioned in the 5′ untranslated regions (20, 32). The independence of the expression of a downstream ORF from translation of a preceding ORF represents a key feature of internal ribosome entry. Therefore, the disappearance of ORF2p by premature translation termination of ORF1 (Fig. ​(Fig.33 and ​and4,4, g and i) rejects the possibility of internal ribosome entry for the mechanism of SART1 ORF2 translation.

Another possibility of internal translation initiation is translation reinitiation after the translation of an upstream ORF. Several examples of translation reinitiation in eukaryotes have been reported (18, 23), but the mechanisms of reinitiation seem distinct from SART1 internal translation initiation by two points. First, reinitiation is efficient when the upstream ORF is less than 30 codons long (18). In the well-known reinitiation system for the yeast GCN4, each of four upstream ORFs consists of only three or four codons (10). In contrast, SART1 ORF2p is translated after the translation of ORF1p, which is 712 amino acids in length. Second, reinitiation is most efficient when the upstream ORF terminates some distance before the start codon of the downstream ORF (18, 23). Reinitiation of the yeast GCN4 is inefficient when the stop codon of the upstream ORF is 56 nt upstream from AUG of the downstream ORF but efficient when the stop codon is 176 nt or 201 nt upstream from AUG in a natural (nonstarvation) condition (10). The distance between the upstream ORF and downstream ORF is necessary for ribosomes to bind the eIF2-GTP-Met-tRNA_i^Met ternary complex (10). In the translation of SART1 ORF2, the first AUG of ORF2 at 3018 to 3020 served as a site for translation initiation only when it was 2 nt downstream (Fig. ​(Fig.33 and ​and4,4, a) or 1 nt upstream (d) from the stop codon of ORF1 but did not do so when it was 67 nt upstream (e), 215 nt downstream (g), or 53 nt downstream (i). The change of translation efficiency following the change of distance between two ORFs in SART1 ORF2 translation is opposite to that in reinitiation.

The overlapping stop-start codon UAAUG plays a key role for the ORF2 translation of SART1. Such an overlapping stop-start codon is also observed in bacterial genes. The UGAUG stop-start codon is observed at the cistronic junction between trpB and trpA in Escherichia coli (3), and UAAUG is situated between the coat gene and the lysis gene in GA RNA phage (13). Expression of these genes is translationally coupled through their overlapping stop-start codons. AUGA also has a capacity for effective translational coupling (33). In the case of translational coupling, the following ORF is translated exclusively by the ribosome that translates the preceding ORF. In addition, increasing the distance between the stop codon of the preceding ORF and the start codon of the following ORF decreases the translation of the following ORF (33). The effect of the distance between two ORFs of SART1 is similar to that in translational coupling.

Although translational coupling is observed mainly in prokaryotes, the existence of eukaryotic translational coupling has been proved by some experiments and the observation of translational mechanisms of several RNA viruses (2, 11, 28, 31). Peabody and Berg showed the existence of translational coupling machinery in mammalian cells using artificial bicistronic RNA (31). Influenza B virus RNA segment 7 contains a UAAUG overlapping stop-start codon, which plays a central role in the translation of the downstream ORF (11). Termination-dependent translation reinitiation with short intercistronic sequences was reported for two viruses, human respiratory syncytial virus (HRSV) (2) and rabbit hemorrhagic disease virus (RHDV) (28). Although these viruses do not contain overlapping stop-start codons, the translational mechanism of downstream ORF could be similar to that of SART1 and influenza virus because some of their related viruses have overlapping stop-start codons. Feline calicivirus and Manchester virus, which are related to RHDV, and pneumonia virus of mice, which is related to HRSV, contain AUGA overlapping stop-start codons (2, 28). Norwalk virus has UAAUG (28). In RHDV, the substitution from AUG to ACG or AUC resulted in about one-fourth of the wild-type expression level of the downstream ORF; in contrast, the substitution to UGU reduced it to nearly zero (28). In our experiments, the substitution from AUG to ACG reduced the SART1 ORF2 translation, whereas the substitution from AUG to AGC appeared to eliminate it (Fig. ​(Fig.4B,4B, a to c). The effects of substitutions at AUG are similar between SART1 and RHDV.

Associated with the mechanism of translational coupling, the downstream RNA secondary structure reinforces ORF2 translation in SART1. The most likely role of the RNA secondary structure of SART1 is blocking the movement of ribosomes to allow sufficient time for recognizing a start codon. The spacing between the start codon and the stem-loop is 34 nt and between the start codon and the presumed pseudoknot is 2 nt (Fig. ​(Fig.5A).5A). It was reported that a downstream RNA secondary structure facilitates translation initiation most efficiently when it is separated from the AUG codon by 14 nt (19). This indicates that the downstream RNA secondary structure of SART1 could not directly lead ribosomes to recognize the ORF2 start codon. Adhin and van Duin proposed that the loosened ribosome may slip forward or backward to locate a reinitiation site in prokaryotes (1). This ribosomal scanning-like movement has a range of action of more than 40 nt. Kozak reported that the backward movement of eukaryotic ribosomes after the translation of short ORFs is limited (21), but it was the case of translation termination reinitiation, not the case of translational coupling. There remains the possibility that the downstream RNA secondary structure of SART1 obstructs the ribosome forward movement and leads the backward movement in order to recognize the ORF2 start codon.

Do other non-LTR retrotransposons use translational coupling?

HOPEBm1 and HOPEBm2 are the close relatives of SART1 and have overlapping stop-start codons (UAAUG) and downstream RNA secondary structures (Table ​(Table22 and Fig. ​Fig.5B).5B). Translational coupling is likely to have been acquired in the common ancestor of SART1, HOPEBm1, and HOPEBm2. Some non-LTR retrotransposons that are phylogenetically distant from SART1 also have overlapping stop-start codons (Table ​(Table2).2). These retrotransposons could also use translational coupling for their ORF2 translation. However, not all non-LTR retrotransposons have overlapping stop-start codons. For example, SARTPx1, another close relative of SART1, has ORF2 overlapping ORF1 in the −1 frame but does not have an overlapping stop-start codon (17). Some non-LTR retrotransposons, like human L1 or I factor of Drosophila, have two separate ORFs. In these elements, internal translation initiation was indicated because there was no detection of ORF1-ORF2 fusion proteins (5, 12, 27). Many non-LTR retrotransposons do not have overlapping stop-start codons, but most of them encode methionines at the beginning of endonuclease domains (data not shown). We consider that the ORF2 protein of many non-LTR retrotransposons could be translated from the first AUG in ORF2. However, we could not exclude the possibility of ribosomal frameshifting for the ORF2 translation of some non-LTR retrotransposons. TRAS1 and TRAS3, which belong to a family of the silkworm telomeric repeat-specific retrotransposons other than SART1, have ORF2 overlapping ORF1 in the +1 frame, and their first AUG in ORF2 is positioned inside the conserved endonuclease domains (22). We further characterized the ORF1-ORF2 junction sequences of other TRAS family retrotransposons, TRAS4, TRAS5, and TRAS6. However, no TRAS family retrotransposons have methionines at the beginning of endonuclease domains (unpublished data). The TRAS family is likely to apply ribosomal frameshifting for their ORF2 translation, although further analysis is necessary to ascertain this possibility.

This work was supported by grants from the Ministry of Education, Science and Culture of Japan and by a Grant-in-Aid from the Research for the Future Program of Japan and Research Fellowships for Young Scientists of the Japan Society for the Promotion of Science (JSPS).