NMR and Alanine Scan Studies of Glucose-dependent Insulinotropic Polypeptide in Water*

Glucose-dependent insulinotropic polypeptide (GIP) is an incretin hormone that stimulates the secretion of insulin after ingestion of food. GIP also promotes the synthesis of fatty acids in adipose tissue. Therefore, it is not surprising that numerous literature reports have shown that GIP is linked to diabetes and obesity-related diseases. In this study, we present the solution structure of GIP in water determined by NMR spectroscopy. The calculated structure is characterized by the presence of an α-helical motif between residues Ser11 and Gln29. The helical conformation of GIP is further supported by CD spectroscopic studies. Six GIP-(1–42)Ala1–7 analogues were synthesized by replacing individual N-terminal residues with alanine. Alanine scan studies of these N-terminal residues showed that the GIP-(1–42)Ala6 was the only analogue to show insulin-secreting activity similar to that of the native GIP. However, when compared with glucose, its insulinotropic ability was reduced. For the first time, these NMR and modeling results contribute to the understanding of the structural requirements for the biological activity of GIP.

Glucose-dependent insulinotropic polypeptide (GIP) is an incretin hormone that stimulates the secretion of insulin after ingestion of food. GIP also promotes the synthesis of fatty acids in adipose tissue. Therefore, it is not surprising that numerous literature reports have shown that GIP is linked to diabetes and obesity-related diseases. In this study, we present the solution structure of GIP in water determined by NMR spectroscopy. The calculated structure is characterized by the presence of an ␣-helical motif between residues Ser 11 and Gln 29 . The helical conformation of GIP is further supported by CD spectroscopic studies. Six GIP-(1-42)Ala 1-7 analogues were synthesized by replacing individual N-terminal residues with alanine. Alanine scan studies of these N-terminal residues showed that the GIP-(1-42)Ala 6 was the only analogue to show insulin-secreting activity similar to that of the native GIP. However, when compared with glucose, its insulinotropic ability was reduced. For the first time, these NMR and modeling results contribute to the understanding of the structural requirements for the biological activity of GIP. GIP 3 (YAEGTFISDY SIAMDKIHQQ DFVNWLLAQK GKKND-WKHNI TQ) is an incretin hormone synthesized and secreted in the gut (1). In addition to its prominent glucose-lowering action, GIP promotes pro-insulin gene expression, proliferation of ␤-cells, and ␤-cell survival (2,3). GIP also plays a role in fat metabolism stimulating the synthesis of fatty acids (4). Furthermore, a direct relationship between GIP suppression and obesity has also been found (5,6). However, GIP is rapidly degraded by the enzyme dipeptidyl peptidase IV immediately after secretion (7)(8)(9). The resulting fragment, GIP- , has been shown to be a GIP antagonist in vivo (10). These results have encouraged the development of GIP analogues (11,12) and dipeptidyl peptidase IV inhibitors (13,14) as potential new targets for the treatment of type 2 diabetes and obesity-related disorders. We are interested in assessing the importance of the structure of GIP and its analogues in determining their biological activity. In this study, we present the three-dimensional solution structure of native GIP in water, as determined by NMR spec-troscopy and molecular modeling. The structure of GIP consists of an ␣-helical motif between residues Ser 11 and Gln 29 , unlike the other major incretin hormone, GLP-1, which mainly adopts a random coil conformation, under physiological conditions (15). Previous solution structural studies carried out in our research laboratory showed that the GIP-(1-30)amide also adopts a random coil conformation when dissolved in water, whereas in a TFE-d 3 /water-mixed solvent, the peptide showed a well defined full-length ␣-helix (16). The results presented here could help in the design of new drugs for diabetes and obesityrelated disorders.
CD Spectroscopy-CD experiments were carried out at 20°C on a JASCO J-810 spectropolarimeter, with wavelengths in the ranges of 180 -250 and 250 -340 nm, for the far-and near-UV, respectively. For studies in the far-UV, GIP samples were dissolved in acetate buffer (20 mM, pH 4.0, 5.0), phosphate buffer (20 mM, pH 6.0, 7.0, 8.0), or in water (unbuffered, pH 6.8, uncorrected) for a final peptide concentration of 30 M. For the near-UV, GIP samples were dissolved in phosphate buffer (20 mM, pH 7.0) or in water (unbuffered, pH 3.0, uncorrected) for a final peptide concentration of 0.95 mM. The peptide concentration was determined using the UV absorbance at 280 nm and an extinction coefficient of 13,980 M Ϫ1 cm Ϫ1 , as calculated with ProtParam (21). Experiments in the far-UV were performed using a response of 2 s and a scanning speed of 50 nm/min, by accumulation of 15 scans. In the near-UV, a response of 0.5 s, a scanning speed of 20 nm/min, and an accumulation of 50 scans were used. In all cases, a bandwidth of 1 nm and a data pitch of 0.2 nm were used. All samples were analyzed in a 0.1-cm cell. All the spectra were baseline-corrected with their corresponding buffer solution or water. The calculation of the helical content was carried out with the program K2D (22) using the DICHROWEB web interface (23)(24)(25).
NMR Spectroscopy-NMR experiments were performed on Bruker DRX 500, 800, and 900 NMR spectrometers equipped with 5-mm inverse probe heads, operating at 298 K and at proton resonance frequencies of 500.13, 800.13, and 900.27 MHz, respectively. An ϳ2 mM peptide sample was prepared using a 9:1 mixture of H 2 O:D 2 O (600 l, pH 3.3, uncorrected). Two-dimensional phase-sensitive DQF-COSY (26), TOCSY (27), and NOESY (28) spectra were acquired with a relaxation delay of 1.4 s, an acquisition time of 0.40 s, and a spectral width of 10 kHz. TOCSY spectra were acquired with 30-and 80-ms mixing times. NOESY spectra were acquired with mixing times of 200 ms. All DQF-COSY, TOCSY, and NOESY experiments were performed with 16, 8, and 16 transients for each of 2,048 t 1 increments into 8,192 complex data points, respectively. All spectra were zero-filled to 4,096 data points in F 1 and apodized using a shifted squared sinebell window function in both dimensions. The signal of 3-(trimethylsilyl) propionic acid was used as an internal chemical shift reference at 0.0 ppm. All data were acquired and processed using Bruker XWINNMR (version 3.5).
Structure Calculations-Two-dimensional NMR spectra were analyzed with SPARKY (version 3.110) (29). The integrals of the NOEs were converted into distance constraints with CALIBA (30) according to three proton classes, namely backbone, flexible side chain, and methyl protons. Pseudoatoms were also introduced for protons that could not be stereospecifically assigned (31). Structure calculations were carried out using CYANA (version 1.0.6) (30). 200 random structures were generated, and their energy was minimized using the anneal protocol in CYANA, including 10,000 steps of simulated annealing, as well as 5,000 steps of conjugate-gradient minimization. During the CYANA calculations, distance constraints were weighted using the force constant k NOE ϭ 1 kJ mol Ϫ1 Å Ϫ2 . The 20 structures with lowest target function values were subjected to 5,000 steps of unconstrained Powell minimization with SYBYL (version 6.8.1) using the Tripos SYBYL Force Field (32). MOLMOL (33) and PROCHECK-NMR (34) were used for further structure analysis.
Back Calculation-One of the low energy GIP models from the final 20 conformations and the individual peak positions and their assignments from the NOESY experiment with a mixing time of 200 ms were used to simulate the fingerprint region of the NOESY spectrum using the CORMA module (35) within SYBYL. The calculation was performed using a model-free approach, overall correlation times of 1-10 ns, and a Gaussian line shape. A fixed normalization method was also used along with the three-site methyl jump model to calculate methyl distances. Further CORMA simulations were carried out for the five lowest energy conformations. Fig. 1 shows the effects of GIP and GIP peptides on insulin secretion, from clonal pancreatic BRIN BD11 cells. From this figure, it is clear that native GIP dose-dependently (10 Ϫ12 -10 Ϫ6 M) stimulated insulin secretion when compared with control incubations (5.6 mM glucose alone) (p Ͻ 0.001). Similarly, GIP-(1-42)Ala 6 also stimulated insulin secretion when compared with control (p Ͻ 0.01 to p Ͻ 0.001), however not to the same extent as native GIP (p Ͻ 0.05 to p Ͻ 0.01). All other GIP peptides displayed decreased insulinotropic activity (1.2-1.9-fold decrease; p Ͻ 0.05 to p Ͻ 0.001) when compared with both glucose control and GIP.

RESULTS
The influence of pH on the secondary structure of GIP was studied by means of CD spectroscopy. 30 M GIP samples in 20 mM acetate buffer (pH 4.0, 5.0) and phosphate buffer (pH 6.0, 7.0, 8.0) were analyzed in the far-UV ( Fig. 2A), as well as an unbuffered 30 M sample of GIP in water (pH 6.8, uncorrected) (Fig. 2B). The helical content of GIP gradually increased from 10 to 26%, from pH 8.0 to 4.0, respectively. The helical content of the buffered sample at pH 7.0 was 20%, whereas that of the unbuffered sample was 7%.
One-and two-dimensional NMR data were obtained using 500-, 800-, and 900-MHz spectrometers. The DQF-COSY and TOCSY spectra provided a basis for the identification of individual residue spin systems. Except for the N-terminal residue Tyr 1 , all other resonances were clearly observed in the TOCSY spectrum. The chemical shifts of methyl resonances were used to identify the alanine, isoleucine, and leucine residues, as well as a unique valine residue. The 2 serine residues, Ser 8 and Ser 11 , were identified by their distinctive ␤H proton resonances. Other spin systems could not be distinguished due to the similarity of their AMX spin pattern. Discrimination of non-unique residues and identification of all other unidentified spin systems was achieved by direct comparison of the TOCSY and NOESY spectra. The 2 glycine residues, Gly 4 and Gly 31 , showed strong, prominent cross-peaks in the TOCSY and NOESY spectra. Alanine, glycine, isoleucine, serine, threonine, and valine residues were used as starting points for the identification of sequence-specific resonance assignments. The connectivities of Ala 2 to Asp 9 were clearly identified using the ␣ i H/N iϩ1 H and ␤ i H/ N iϩ1 H cross-peaks in the fingerprint region of the NOESY spectrum (Fig. 3). Distinctive ␤H/NH intra-residue and methyl connectivities were used to identify the connectivities of Ser 11 /Ile 12 , Ile 12 /Ala 13 , Met 14 / Asp 15 , and Lys 16 /Ile 17 . The 2 glutamine residues, Gln 19 and Gln 20 , were identified using the backbone and ␤H/NH connectivities of Asp 21 . The were used to identify Gln 29 and Lys 32 , respectively. Identification of the Asn 34 , Asp 35 , and Trp 36 spin systems was made via the backbone and ␤ i H/NH iϩ1 resonance connectivities of their neighboring residues. Aromatic and side-chain protons to inter-residue ␣H and ␤H cross-peaks were observed for Trp 36 and His 38 . The prominent methyl resonances of Ile 40 and Thr 41 were used to identify the C-terminal end of the polypeptide chain. Sequence-specific resonance assignments were completed by identification of backbone, ␣H, ␤H, aromatic, and side-chain resonance connectivities between immediately neighboring residues.
Other parts of the NOESY spectrum were then searched for secondary structure correlations. Strong ␣ i H/␤ iϩ3 H medium range connectivities were observed suggesting the presence of an ␣-helix for residues Ser 11 -Gln 29 , except those between Ala 13 /Lys 16 , Asp 15 /His 18 , His 18 / Asp 21 , Asp 21 /Asn 24 , due to chemical shift degeneracy or signal overlap. Clear medium range ␣ i H/N iϩ3 H connectivities for residues between Ser 11 and Lys 30 were also identified, except between Ala 13 /Lys 16 and Gln 20 /Val 23 , due to overlapping. Distance constraints derived from the NOESY data were used for structure calculations (Fig. 4). 200 random structures were generated using the CYANA protocol, and the 20 conformations that best fit the experimental constraints were subjected to additional unconstrained energy minimization. The main structural feature of the peptide is a right-handed ␣-helix between residues Ser 11 and Gln 29 , as determined by MOLMOL and PROCHECK-NMR. Best fit superposition of the backbone atoms between residues Ser 11 and Gln 29 of the 20 calculated structures is presented in Fig. 5.
Simulated spectra were obtained for the five conformations with the lowest energy from the final set of 20 energy-minimized conformations, with overall correlation time values of 1-10 ns. Only the fingerprint region of the NOESY spectrum was simulated and compared with the experimental spectrum. The spectrum simulated with a correlation time of 2 ns showed the highest level of agreement with the experimental results (Fig. 6).

DISCUSSION
GIP-(1-42) is a known stimulator of insulin secretion in the presence of glucose (36). In this study, seven synthetic peptides with a singleresidue replacement with an alanine in each of the first seven positions were synthesized to evaluate specific residues that play a role in insulin secretion. In comparison with GIP-(1-42), GIP-(1-42)Ala 6 was the only analogue to show insulin-secreting activity similar to that of the native peptide (p Ͻ 0.001) when compared with glucose control; however, its insulinotropic ability was noticeably reduced. When compared against glucose control, the remaining alanine scan peptides (GIP-(1-42)Ala 1 , GIP-(1-42)Ala 3 , GIP-(1-42)Ala 4 , GIP-(1-42)Ala 5 , and GIP-(1-42)Ala 7 ) showed reduced or no insulinotropic activity, leading us to believe that the amino acid residues replaced are in fact vital to the insulin-releasing activity of GIP and that the N terminus is of great importance in glucose-dependent insulin release. Interestingly, GIP-(1-42)Ala 3 and GIP-(1-42)Ala 4 showed such a reduced level of insulinotropic activity that these may be considered for antagonistic studies in the future, especially since it is well known that GIP-(3-42) is a GIPreceptor antagonist (10). With such an inactivation of insulin release when glutamic acid in position 3 is replaced, it may suggest further that position 3 is of great importance in the insulinotropic potency of GIP.
A sample of native GIP was also subjected to basic solubility studies. The peptide was solubilized in Tris buffer (10 mM Tris, 100 mM NaCl, 5 mM MgCl 2 , and 1 mM dithiothreitol, pH 7.5) and analyzed by spectrophotometry combined with light scattering. The results showed a single UV peak at 280 nm, revealing that GIP is monomeric under these conditions.  As can be seen from Fig. 2A, the CD spectra obtained in the far-UV at pH values 4.0 -8.0 are characteristic of samples with some degree of helical secondary structure, whereas the spectrum acquired for the unbuffered sample (Fig. 2B) suggests an irregular secondary structure (37). Analysis of the spectra with the program K2D reveals an increase in helical content with decreasing pH. We have also found that GIP is 20% helical in 20 mM phosphate buffer, pH 7.0, increasing up to 26% when in 20 mM acetate buffer, pH 4.0. These results are in contrast with studies of Manhart et al. (38), in which they reported a helical content of 11% for GIP in 20 mM phosphate buffer, pH 7. Studies in the near-UV (results not shown), carried out for 0.95 mM GIP samples in phosphate buffer (20 mM, pH 7.0) and unbuffered (pH 3.0, uncorrected), show spectra similar to those found for folded proteins (37). The results could be interpreted as self-association, the adoption of a regular tertiary structure, or a combination of both. However, as our spectrophotometric and light-scattering experiments show that GIP is monomeric, then the CD results should reflect the adoption of a regular tertiary structure by GIP. The near-UV CD results could be valuable in future studies as a template of the tertiary structure of native GIP (39).  The NMR data obtained with the 800-and 900-MHz spectrometers were used for resonance identification and structure calculation. Twodimensional DQF-COSY and TOCSY spectra were used for spin system identification. TOCSY and NOESY data were used for full and unambiguous assignment of individual and sequence-specific resonances. Some side-chain and most of the aromatic protons could not be assigned due to signal overlap. Identification of ␤H/NH, N i H/N iϩ1 H cross-peaks, and aromatic proton connectivities to neighboring residues supported the reliability of the sequence-specific resonance assignments. The large number of well spread NOEs observed suggested that GIP is in a folded state. The presence of ␣ i H/N iϩ3 H NOE connectivities in the fingerprint region of the NOESY spectrum and of stronger ␣ i H/ ␤ iϩ3 H NOEs in the range of Tyr 10 -Gln 29 revealed the occurrence of elements of ␣-helical secondary structure.
The helical character of GIP in water is further supported by the chemical shift index (40), which suggested the presence of two helical segments between residues Ser 8 -Lys 16 and Gln 19 -Lys 30 . It was noted that the chemical shifts of all the isoleucine residues did not follow the predicted chemical shift index pattern. However, the absence of medium range ␣ i H/N iϩ3 H and ␣ i H/␤ iϩ3 H connectivities in the NOESY spectrum in the range Ser 8 -Tyr 10 , and their presence between residues Lys 16 and Gln 19 , leads us to suggest a continuous ␣-helical segment between residues 11 and 29. In addition, all of the calculated structures showed a continuous helical segment in this region of the peptide chain. Analysis of the calculated structures revealed that the GIP in water adopts an ␣-helical motif between residues Ser 11 and Gln 29 for most of the calculated conformations. These results were confirmed by the secondary structure predicted within the MOLMOL package. Further analysis of the structures with PROCHECK-NMR showed that more than 98% of the residues lie within allowed regions of the Ramachandran plot (Table 1).
To assess the quality of the calculated structures, the fingerprint region of the NOESY data was simulated and compared with the experimentally obtained spectrum. Simulated spectra showed an optimum level of agreement when a correlation time of 2 ns was used (Fig. 6). The fingerprint region of the simulated spectrum showed few additional cross-peaks with varying intensities. These peaks are mainly ␣ i H/N iϩ4 H connectivities that arise from protons of residues that are part of the helical segment, such as ␣ 17 H/N 21 H, and ␣ 20 H/N 24 H. Also, several ␣ i H/ N iϩ3 H medium range connectivities from residues within the helical  region appeared stronger in the simulation, whereas ␣ i H/N iϩ2 H crosspeaks were weaker. This could be due to the fact that CORMA calculates NOE intensities by assuming a rigid backbone structure. Therefore, the simulated spectrum showed weak connectivities, such as ␣ i H/ N iϩ4 H, that are difficult to observe or can be seen at a lower threshold in a standard NOESY spectrum for smaller peptides. Cross-peaks with varying intensities were also observed for some side-chain resonances in the simulated spectrum. This is due to the motion of the side chain protons, which would normally be expected to be averaged in the experimental data. Since the fit between the experimentally and the theoretically calculated spectra is good, it may be concluded that the low energy structure used for the simulation satisfies the experimentally derived distance constraints; hence, the NOESY peak assignments and distance constraint calibration were also accurately carried out. These results confirmed that the calculated structure shows a true picture of the peptide in water. The root mean square deviation value of 0.67 Ϯ 0.26 Å obtained after superposition of the backbone atoms between residues Ser 11 and Gln 29 could suggest that the helical segment of the peptide chain is somewhat flexible under these conditions. This may be due in part to the high mobility of both the N-terminal and C-terminal ends at both sides of the helix. This was supported by the lack of strong NOEs observed beyond the N-and C-terminal ends of the helical region. However, NOEs observed at the C-terminal end of the helix are marginally stronger than those of the N-terminal end. In general, short peptides adopt mainly random coil conformations when dissolved in pure water. In contrast, we observed 10 NOEs per residue, which is much higher than the average of seven NOEs generally observed for peptides of similar size.
Ongoing studies in our laboratory showed that the biologically active GIP fragment, GIP-(1-30)amide (16) and the parent GIP 4 adopted a tighter and better defined helical conformation when TFE was introduced in the sample. This leads us to suggest that the lack of helixinducing and -stabilizing solvents, such as TFE, trifluoroacetic acid, and methanol, results in the solution structure of GIP in water being somewhat less ordered. As a result, the lower number of short and medium range NOE connectivities observed in the NOESY spectrum, in comparison with the GIP-(1-30)amide and GIP in TFE, contributes to the lower precision of the resulting structures and slightly higher average energies (Table 1).
It seems obvious to attempt a direct comparison of the solution structure of GIP with that of the other major incretin hormone GLP-1. However, we believe that, in this case, the length of the peptide chain plays a crucial role toward the adoption by these peptides of regular secondary structural features in a purely aqueous solvent. On the basis of NMR studies, the 29-amino-acid residue GLP-1 has been shown to adopt a random coil conformation in water (15), as well as the 30-residue GIP fragment, GIP-(1-30)amide (16). It was also noted that the formation of the secondary structure was observed to start at the C-terminal end to then spread toward to the N terminus by the addition of helix-favoring solvents. On the other hand, exendin-4, a 39-residue agonist of GLP-1 that shows a 50% sequence homology to GIP, adopts a helical conformation in aqueous medium (41). Therefore, we believe that the residues at the C-terminal end (Gly 31 -Gln 42 ) of GIP appear to provide significant stability to its structure in neat aqueous solvent. However, the C-terminal segment does not seem to be important for insulinotropic activity since GIP-(1-30)amide shows a similar insulin-releasing activity to that of the parent GIP (16). On the other hand, the N-terminal end, which appears to be important for the activity of the peptide, remains disordered under these conditions.
The proposed NMR structures are in good agreement with the results of our CD spectroscopic studies. In view of the CD results, we may conclude that the secondary structure of GIP depends strongly on the buffering capacity of the solvent. The helical content of GIP increases with decreasing pH. Since the sample studied by NMR was unbuffered and its uncorrected pH was 3.0, we can also conclude that, at higher concentrations, GIP acts as its own buffer, resulting in a more helical conformation, as shown in our NMR studies.
In conclusion, we have shown that GIP adopts an ␣-helical conformation between residues Ser 11 and Gln 29 when dissolved in pure water. All 20 structures agree with the experimentally derived NMR constraints as indicated by the low number of NOE violations. These results may help to gain further insight into the biologically relevant receptorbound structure of GIP and so facilitate the design of new analogues and peptidic drugs to treat type 2 diabetes and obesity.