Analysis of Structure–Activity Relationships of Food-Derived DPP IV-Inhibitory Di- and Tripeptides Using Interpretable Descriptors

Abstract

This study aimed to analyze the structural requirements for di- and tripeptides exhibiting a DPP IV-inhibitory effect. The sequences of 46 di- and 33 tripeptides, including their bioactivity (IC₅₀; μM), were implemented from the BIOPEP-UWM database, whereas modeling was performed using SCIGRESS Explorer: Version FJ 3.5.1 software. Models included 336 (dipeptide dataset) and 184 descriptors (tripeptide dataset). The values of the determination coefficient (R²) defining model reliability were 0.782 and 0.829 for di- and tripeptides, respectively. Based on the implemented descriptors, it was concluded that increased numbers of nitrogen atoms, as well as the methyl groups, are required for dipeptides to enhance the DPP IV-inhibitory effect. This was indicated by the presence of amino acids with an aliphatic side chain (e.g., Leu, Val, Ile) and an aromatic ring (Trp). In the case of tripeptides, a correlation was found between their molecular weight (MW) and studied bioactivity. A tripeptide with a molecular weight of up to 500 Da was found suitable for the sequence to act as the DPP IV inhibitor. Although there is still a gap in explaining the relations between the structural nature and the DPP IV-inhibitory activity of peptides, and certain issues related to this topic still remain unknown, the results are in line with those reported by other authors. Additionally, the suitability of the SCIGRESS tool in the QSAR analysis of peptides derived from foods can be confirmed. Interpretable descriptors enabled the achievement of more unequivocal results concerning the main structural factors affecting the DPP IV inhibition of di- and tripeptides.

1. Introduction

It has recently been highlighted that food, as well as its main function, i.e., providing essential nutrients that regulate bodily functions, is also expected to prevent, or even cure, diseases. Its health potential results from the presence of bioactive compounds like, e.g., vitamins, phytochemicals, and peptides [1]. The latter ones have attracted the attention of scientists as non-toxic and multi-active molecules [2]. They were proven to act as, i.a., glucose level reductants, blood hypotensors, cholesterol level reductants, anti-oxidants [3,4], anti-inflammatory, and anti-obesity agents [5].

Reduction of the blood glucose level is related to, i.a., the action of peptides that inhibit dipeptidyl peptidase IV (DPP IV). DPP IV itself (EC 3.4.14.5) is an enzyme involved in the degradation of GLP 1(glucagon-like peptide 1) and GIP (glucose-dependent insulinotropic peptide). These two molecules are released into the gastrointestinal tract and are responsible for glucose homeostasis. The action of DPP IV briefly relies on the cleavage of N-terminal dipeptides from GLP 1 and GIP to produce inactive incretins; hence, it triggers the loss of insulinotropic activity of GLP 1 and GIP. Thus, peptides that inhibit DPP IV action display the anti-diabetic function [6,7].

DPP IV inhibitors were identified in, e.g., milk and dairy products, rice bran, salmon, tuna, and azuki beans [8]. The hydrolysis of proteins followed by the determination of biological activity of a hydrolysate and the identification and/or synthesis of peptides is the classical procedure of studying bioactive peptides derived from foods. Another research strategy includes an in silico analysis of biopeptides. It involves the application of, i.a., bioinformatic and cheminformatic databases as sources of information about peptide sequences, tools for the molecular docking of peptides to predict the possible mechanism of their action, and the prediction of peptide bioactivity based on its structural nature [9]. The latter is called a QSAR, i.e., the quantitative structure–activity relationship, and, so far, it has been applied in the study of peptides showing a one [9,10,11] and dual biological effect [12,13].

A literature search using the Scopus [14] platform (accession date: October 2023) revealed dozens of records (i.e., 51, including reviews and original papers as well as conference abstracts) dedicated to QSAR of DPP IV inhibitors using “QSAR and dipeptidyl peptidase 4 and inhibitors” as a query. Further analysis of these records showed that none of them concerned peptides with such an activity. The articles mostly concerned the QSAR analysis of non-peptide DPP IV inhibitors like, e.g., 4-fluoropyrrolidine-2-carbonitrile and octahydrocyclopenta[b]pyrrole-2-carbonitrile [15], tetralin-sulfonamide derivatives [16], and polyphenols [17]. This literature search via Scopus has shown that there is paucity of literature data related to the QSAR of DPP IV peptide inhibitors, especially those derived from food proteins. The query “QSAR and dipeptidyl peptidase-IV and inhibition or inhibitors and peptides” revealed 15 records (as of October 2023), but barely 2 of them were related to the food-originating DPP IV inhibitors. They included an original article concerning the identification of DPP IV inhibitors from camel milk hydrolysates [18] and a review of the efficiency of the QSAR of food-derived bioactive peptides [19]. One of the latest articles concerned the impact of certain structural features on the DPP IV-inhibitory activity of dipeptides. However, the models were based on several attributes describing protein structures (long-chain molecules) which are not applicable to dipeptides (short-chain compounds) [20]. Other articles were related to DPP IV inhibition described by means of programs useful for reporting the DPP IV inhibition of compounds like, e.g., iDPPIV-SCM and StackDPPIV [21,22]. Hence, even though there is ample literature data on the potential of bioactive peptides, including DPP IV inhibitors, there is a still a need to revise and/or systemize knowledge about DPP IV inhibitors. This knowledge gap has stemmed from: (a) the not-fully elucidated mechanism of DPP IV inhibition induced by peptides [6]; (b) the not-fully explained structure–activity relationship of these peptides [6]; and c) an increase in the number of new biopeptides (including DPP IV inhibitors) deposited in databases [9]. Therefore, taking into account the latter of the above-mentioned issues in particular, this study aimed to predict the DPP IV-inhibitory activity of di- and tripeptides based on their structural nature using interpretable descriptors.

2. Materials and Methods

2.1. Sequences of DPP IV Inhibitors

The sequences of di- and tripeptides with DPP IV-inhibitory activity were acquired from the BIOPEP-UWM database of protein and bioactive peptide sequences [23], available at: https://biochemia.uwm.edu.pl/en/biopep-uwm-2/ (accessed on 12 June 2023). The complete list of DPP IV inhibitors, including their bioactivity measure expressed as the concentration of peptides corresponding to their half-maximal activity (IC₅₀; μM), is shown in Results and Discussion (see the appropriate Tables). To summarize, 46 dipeptides and 33 tripeptides were used for QSAR modeling. Data on peptides available in the BIOPEP-UWM database were acquired in June 2023.

2.2. Analysis of Structure–Activity Relationship of DPP IV Inhibitors

The procedure of QSAR modeling of di- and tripeptide DPP IV inhibitors using BIOPEP-UWM database (see Section 2.1) and SCIGRESS Explorer: Version FJ 3.5.1 is shown in Figure 1.

Briefly, QSAR models calculated using SCIGRESS software were based on MLR. All the structures of di- (46 sequences) and tripeptides (33 sequences) were drawn using the above-mentioned software by means of the following tabs available in the applied program: Action → Protein → Sequence. They were protonated to obtain the zwitterion by exemplifying the occurrence of peptides in a solution at a neutral pH. The geometries of peptides were optimized using the SCIGRESS option Experiment → Run by selecting “chemical sample” and the procedure called “MO-G PM 6 in water”. QSAR analyses were initially computed using a method called “Complete Topological QSAR” (see

Table 1. Statistical parameters of QSAR models obtained for di- and tripeptides with DPP IV-inhibitory activity (understood as logIC₅₀).

Parameter	DPP IV Inhibitors
	Dipeptides	Tripeptides
R2	0.782	0.829
Degrees-of-freedom adjusted R2	0.755	0.798
CVR2	0.737	0.710
Average error	0.255	0.209
RMSEC	0.343	0.296
RMSECV	0.377	0.386
F-statistics	28.773	26.228
p-value	0.0000	0.0000
α	0.05	0.05

R²—determination coefficient; CVR²—correlation coefficient of leave-one-out cross-validation; RMSEC—root mean square error of calibration; RMSECV—root mean square error of cross-validation.

), taking into account the function called the Enhanced Replacement Method (ERM) (denoted as the ERM model) [24] implemented in the SCIGRESS. A parameter called Descriptor Count was set to 5 in order to obtain an acceptable ratio of the number of descriptors to the number of experimental points. The initial models were refined using the “Custom QSAR” method in order to manually remove descriptors which are difficult to interpret mechanistically while retaining satisfying statistical performance parameters. Finally, the QSAR model created for dipeptides was built up from 46 sequences and 336 potentially usable descriptors, whereas 33 sequences and 184 descriptors were used for tripeptides modeling. The list of descriptors, as well as the complete reports concerning the models, may be found in the Supplementary Materials. The IC₅₀ values (µM), defining the DDP-IV-inhibitory activity of peptides, were converted to log₁₀. The software applied in our study automatically adjusts the measure of bioactivity to produce optimized results.

3. Results and Discussion

The structure–analysis relationship approach relies on the construction of mathematical models that enable computing and finding (if possible) the correlation between the structure of a compound and its bioactivity [6]. In the case of peptides, the structure is understood as their amino acid sequences, whereas bioactivity refers to the measure of a molecule activity [25]. Usually, it is the concentration of a compound that corresponds to its-half maximal effect (like, e.g., IC₅₀). Moreover, such a measure is usually transformed into logarithms [26].

The calculations were made using SCIGRESS software, as the QSAR model development in SCIGRESS is GUI-driven (i.e., graphical user interface), that is, by a user-interactive process with the possibility of manual removal of hard-to-mechanistically-interpret descriptors. This is an advantage of this software over other tools, like, e.g., RDKit/Scikit-learn Python scripts [27,28], where the feature (descriptor) selection process is mostly automated and based on statistical parameters, with no simple way of taking descriptor mechanistic interpretability.

Analyses concerned 46 di- and 33 tripeptide DPP IV inhibitors. According to Tropsha [29], the number of samples (i.e., cases, objects) in the datasets should not be too small or too big. The low number of samples results in false-positive or false-negative results, which may lead to uncertain conclusions in biological and medical sciences [30]. The upper limit of the samples is often defined by the computer and the method by which the QSAR model was constructed [29]. The analysis using SCIGRESS was based on multilinear regression (MLR), i.e., a traditional approach in QSAR [31]. According to Cargnelutti Filho et al. [32], in MLR, the sample size depends on effect size, whereas the number of independent variables and recommendations for the number of chemicals are associated with the different criteria adopted by each researcher. Based on information from the theoretical model selection, Jenkins and Quintana-Asencio [30] tried to recommend the minimum N (understood as the number of samples) to correctly match a model to a data shape in regression models. In the words by Jenkins and Quintana-Asencio [30], “statistical limbo may be better avoided and reproducibility may be improved if research based on regressions and meta-regressions uses N ≥ 25” [30]. Apart from the number of molecules (samples), the number of descriptors (variables) in a model is important. To conclude, the number of peptides taken for QSAR analysis in the present study is in line with the above statement. However, due to a relatively small dataset, there was no possibility of performing a typical train/test dataset split. Instead, leave-one-out validation was used to measure the performance of the models (see discussion below).

Table 1 shows the statistical results of analyses obtained for the di- and tripeptide DPP IV-inhibitory models. For both models, similar statistics were obtained in terms of the R² values, which were R² = 0.782 and R² = 0.829 for di- and tripeptides, respectively. R² is a statistical parameter that shows which percent of endogenous variable is explained by exogenous variable(s) [33]. Endogenous and exogenous variables refer to the bioactivity of peptides and descriptors expressing their properties, respectively [14]. On the other hand, the parameter called adjusted R² (see Table 1) includes degrees of freedom, and its value is usually smaller than that of R² [16] (0.755 and 0.798, respectively (see Table 1)).

The validity of the model was calculated by default using the leave-one-out (LOO) cross-validation procedure. The LOO procedure is the one of the simplest statistical procedures for model validation and relies on excluding each sample once and constructing the model without this sample to predict the value of a dependent variable (bioactivity) [34]. Moreover, the LOO procedure enables nearly unbiased estimation of the true generalization performance [35]. The statistical parameters of LOO cross-validation are represented by CVR² (correlation coefficient of leave-one-out cross-validation, also called Q₂), RMSEC (root mean square error of calibration), and RMSECV (root mean square error of cross-validation). For di- and tripeptides, the CVR² values were 0.817 and 0.722, respectively (see Table 1), whereas the respective RMSECV values were 0.377 and 0.387. Usually, CVR² > 0.5 is found sufficient to prove model accuracy [36]. Kiralj and Fereira [34] reported that relations of the above-mentioned statistics when using the LOO procedure should be as follows: R² > Q² and RMSEC < RMSECV. The statistics present in Table 1 meet these requirements. Moreover, according to Kiralj and Fereira [34], the sufficient requirements for regression models include the following conditions: CVR² > 0.5 and R² > 0.6. The difference between R² and CVR² exceeding 0.2–0.3 indicates the model overfitting. Based on these findings, the models presented in this study could be considered statistically acceptable (see Table 1). The statistical significance of the estimated QSAR models was analyzed based on F-statistics (α = 0.05), the values of which reached 28.773 and 26.228 for di- and tripeptides, respectively. The p-value was lower than α. To summarize, the results of computations made for CVR², F-statistics, and p-values indicate some significant relationships between descriptors and bioactivity (logIC₅₀) of both DPP IV-inhibitory QSAR models and enable developing equations (see Figure 2a,b) for predicting the relationship between the structure and DPP IV-inhibitory activity of dipeptides/tripeptides. The names of the descriptors found in these equations are provided in the exact words as are given in the program used.

The impact of individual descriptors on the DPP IV activity of di- and tripeptides is shown in

Table 2. The relative weight of each descriptor contributing to the structure–activity prediction of DPP IV-inhibitory di- and tripeptides (the statistical parameters affecting the improvement of bioactivity are in bold).

DPP IV inhibitors	Dipeptides	Descriptor	Normalized Coefficient	Partial-F
		hydrogen donor partial surface area/total accessible surface area/MW	0.7559	51.3355
		guanidine count/MW	0.5086	14.5636
		nitrogen count2	−1.0000	53.3368
		methyl count2	−0.3414	9.8481
		ring count 5 member2	0.6070	28.3351
	Tripeptides	Csp3 bonded to 3 C	0.0053	3.4655
		methyl count2	−0.0091	11.0810
		ln(molecular weight)	1.0000	20.7152
		1.0/molecular weight	0.3657	24.0262
		sqrt(molecular weight)	−0.6413	19.3041

. The respective plots referring to the correlations between the observed and the predicted activity of di- and tripeptide DPP IV inhibitors are shown in Figure 3 and Figure 4, respectively. The predicted and observed values of logIC₅₀ for di- and tripeptides (tripeptides) are provided in

Table 3. Observed and predicted logIC₅₀ values obtained for dipeptide DPP IV inhibitors.

No.1	Sequence	ID 2	IC50	Log IC50	Predicted log IC50 3
1	AA	8637	9400.00	3.97	3.77
2	AL	8559	882.13	2.95	2.72
3	AP	3177	7950.00	3.90	3.71
4	EK	8558	3216.75	3.51	3.00
5	FL	8555	399.58	2.60	2.33
6	FP	8506	363.00	2.56	2.99
7	GL	8561	2615.03	3.42	3.37
8	GP	3169	9690.00	3.99	4.12
9	HL	8557	143.19	2.16	1.71
10	HP	8520	2820.00	3.45	3.14
11	IP	8501	410.00	2.61	2.96
12	KA	3174	6270.00	3.80	3.53
13	KP	8519	2540.00	3.41	3.54
14	LP	3180	2370.00	3.38	2.98
15	LW	8688	993.40	3.00	2.13
16	ML	8832	91.00	1.96	2.11
17	MM	8833	93.00	1.97	2.33
18	MP	3171	870.00	2.94	3.02
19	MW	8690	1691.40	3.23	2.25
20	NH	8844	69.00	1.84	1.85
21	PP	3170	5860.00	3.77	4.01
22	RP	8518	2240.00	3.35	3.44
23	SL	8560	2517.08	3.40	3.16
24	SP	8505	5980.00	3.78	3.97
25	TH	8902	49.00	1.69	2.21
26	TP	8503	2370.00	3.38	3.52
27	TW	8913	84.00	1.92	2.55
28	VA	3172	168.24	2.23	2.72
29	VL	8922	74.00	1.87	1.87
30	VP	3181	880.00	2.94	3.07
31	VR	8594	826.10	2.92	2.83
32	WC	8684	420.00	2.62	2.71
33	WI	8679	138.70	2.14	2.16
34	WK	8676	40.60	1.61	2.29
35	WL	8677	43.60	1.64	2.13
36	WM	8682	243.10	2.39	2.30
37	WN	8680	148.50	2.17	2.25
38	WP	8504	4530.00	3.66	3.49
39	WQ	8678	120.30	2.08	2.15
40	WR	8675	37.80	1.58	1.57
41	WS	8687	643.50	2.81	2.76
42	WT	8685	482.10	2.68	2.60
43	WV	8556	65.69	1.82	2.15
44	WW	8686	554.80	2.74	2.69
45	WY	8683	281.00	2.45	2.40
46	YP	8521	3170.00	3.50	3.16

¹ Numbers used in Figure 3. ² ID in the BIOPEP-UWM database of bioactive peptides (see Section 2). ³ Denoted as Log IC₅₀ (ERM).

and

Table 4. Observed and predicted logIC₅₀ values obtained for tripeptide DPP IV inhibitors.

No 1	Sequence	ID 2	IC50	Log IC50	Predicted log IC50 3
1	APG	8500	40,000	4.60	4.50
2	GPA	8522	40,000	4.60	4.50
3	GPM	9117	417.90	2.62	2.43
4	GPV	9116	794.80	2.90	3.26
5	IPA	8304	49.00	1.69	2.03
6	IPI	3167	7.40	0.87	1.45
7	IPM	9233	69.50	1.84	1.81
8	LPL	8616	241.40	2.38	1.45
9	LPQ	9339	82.00	1.91	2.41
10	LQP	8689	1181.10	3.07	2.41
11	PPG	8653	2252.68	3.35	3.32
12	PPL	8652	390.14	2.59	2.42
13	VPL	8347	15.80	1.20	1.47
14	WRA	8608	690	2.84	2.58
15	WRD	8596	376	2.58	2.79
16	WRE	8595	350	2.54	2.79
17	WRF	8599	413	2.62	2.76
18	WRG	8600	473	2.68	2.64
19	WRH	8606	670	2.83	2.78
20	WRI	8610	730	2.86	2.82
21	WRK	8598	406	2.61	2.79
22	WRL	8612	903	2.96	2.82
23	WRM	8607	673	2.83	2.67
24	WRN	8597	403	2.61	2.79
25	WRP	8611	780	2.89	2.77
26	WRQ	8609	720	2.86	2.79
27	WRR	8604	570	2.76	2.72
28	WRS	8601	483	2.68	2.75
29	WRT	8603	526	2.72	2.66
30	WRW	8602	487	2.69	2.57
31	WRY	8605	640	2.81	2.70
32	WWW	8681	216	2.33	2.32
33	YPY	8617	243.70	2.39	2.73

¹ Numbers used in Figure 4 and Figure 6. ² ID in the BIOPEP-UWM database of bioactive peptides (see Section 2). ³ Denoted as Log IC₅₀ (ERM).

, respectively.

Statistical parameters present in Table 2 can be explained as the weights showing the contribution of each descriptor as a factor influencing the activity of DPP IV inhibitors. Partial-F indicates the importance of a descriptor in the SAR prediction. The higher its value is, the higher the significance of the descriptor. The relative weights of the normalized coefficients are the measures of the impact of the specific property on the peptide structure. The negative values (bold in Table 2) of normalized coefficients suggest that the property improves the bioactivity of peptides, whereas the positive values mean that the property should rather be minimized (and/or eliminated) from the structure of a molecule in order to “improve” its biological function [37]. To recapitulate, the advantageous impact on dipeptide DPP IV-inhibitory activity was correlated with the descriptors defined as “nitrogen count²” and “methyl count²” (having been squared). Disadvantageous impact was correlated with the “ring count 5” (square value). In the case of tripeptides, the descriptors improving their bioactivity were: “methyl count²” (having been squared) and “sqrt(molecular weight)”. Based on the normalized coefficient value, the importance of the first descriptor can be ignored as similar to that of the descriptor called “Csp³ bonded to 3C” (the higher its value, the lesser its impact). Some descriptors summarized in Table 2 are counts of simple structure features, possess only integer values, and may thus be understood as molecular quantum numbers according to a definition proposed by Nguyen et al. [38]. Descriptors proposed here include, i.a., nitrogen count, methyl count, ring count 5 member, and Csp³ bonded to 3 C. Among them, the ring count 5 member, i.e., the number of rings containing five atoms, is one of the molecular quantum numbers proposed by Nguyen and coworkers [38]. The above authors have divided nitrogen atoms into two classes: acyclic and cyclic nitrogens. In this work, it was enough to include the total count of nitrogen atoms in a molecule. The set of descriptors proposed by Nguyen et al. [38] is designed to enable the classification of all classes of low-molecular compounds, whereas the set analyzed in this study is adapted to work with a limited number of short peptides, with limited variability of backbones and a limited number of possible side chains. Peptide fragments taken into account while constructing a set of descriptors, summarized in Table 2, are presented in Figure 5 using four peptides as examples.

Focusing on the descriptors enhancing the DPP IV-inhibitory activity of dipeptides, it can be concluded that peptides rich in methyl groups as well as nitrogen might be greater enhancers of DPP IV inhibition. Data presented in Table 3 show 23 dipeptides out of the 46 in total whose predicted (ERM variant (see Section 2)) and observed values of logIC₅₀ were comparable. These peptides were as follows: GL, GP, KP, ML, MP, NH, RP, SL, TP, VL, VP, VR, WC, WI, WM, WN, WP, WQ, WR, WS, WT, WW, and WY. When looking at their sequences, some of them contained an amino acid with a non-polar side chain (e.g., G, V, L, I), which can be represented as the presence of methyl groups (the higher their number, the better their activity) in a dipeptide chain. Several dipeptides were composed of tryptophan (W; 11 peptides out of 23), asparagine (N), and/or glutamine (Q). Compared to other amino acids, these three residues contain “additional” nitrogen. Thus, the presence of one of the above-mentioned amino acids promoting the DPP IV-inhibitory activity can be explained by the descriptor defined as “nitrogen count2” (the higher the number, the more favorable the impact on dipeptide bioactivity). Three dipeptides (RP, VR, WR) were composed of arginine (R). Although the presence of guanidine (R) increases the number of nitrogens in a dipeptide chain, the statistical parameters (see Table 3), as well as the equation (Figure 2a), indicated that the number of guanidine groups should be reduced (or eliminated) to improve the DPP IV-inhibitory effect of dipeptide. Moreover, RP and WR peptides—numbers 22 and 40 in the plot (see Figure 3)—were located out of the confidence interval line, whereas VR (number 31) was found in the confidence interval line (dashed line in Figure 3). Some of the dipeptides were built up of proline (P), a residue containing a five-atom aliphatic ring. According to the statistics (see Table 3), the lesser the number of such rings in the peptide chain, the stronger the impact on the DPP IV-inhibitory effect.

The analysis of tripeptide DPP IV inhibitors showed that the following 16 tripeptides out of the 33 sequences analyzed had comparable values of the observed and predicted logIC₅₀ (ERM variant): APG, GPA, PPG, PPL, WRF, WRG, WRH, WRI, WRL, WRP, WRQ, WRR, WRS, WRT, WRY, and WWW (see Table 4 and Figure 4). As was mentioned above, due to the minor impact of such descriptors as the number of -CH₃ groups squared (“methyl count²”), as well as the number of sp³ carbon atoms bound to the other three carbon atoms (“Csp³ bonded to 3 C”), they were not further considered as factors affecting the DPP IV-inhibitory activity of tripeptides. Finally, the bioactivity of tripeptides was found to be related to their molecular weight (MW). Moreover, based on the normalized coefficients (see Table 3), it can be concluded that the higher molecular weight of a tripeptide is more beneficial to its DPP IV-inhibitory effect. When looking first at the sequences of tripeptides with comparable values of the predicted and observed logIC₅₀, the majority of them contained tryptophans (W), prolines (P), and/or tyrosines (Y), which made them heavier than the other tripeptide sequences. Nevertheless, the next step of the research was to create the non-linear function of three descriptors related to molecular weight (see Figure 6a), with the plot (see Figure 6b) showing the relationship between the predicted logIC₅₀ and the molecular weight of a tripeptide. This function was applied for the first time to analyze the impact of specific descriptors of DPP IV inhibition by peptides. The plot enables the distinguishing of the minimum and the maximum of the function. The minima and maxima of the functions have a long history in terms of their practical applications in life to solve the problems of the so-called “optimizations” [39]. According to the plot, the minimum of the function was reached by peptides IPI and LPL (number 6 and 8, respectively). It can be concluded that the minimal structural requirements for a tripeptide to act as a DPP IV inhibitor is for it to be composed of hydrophobic amino acids (Leu, Ile) and the ones with rings (Pro). According to Guasch et al. [40], the presence of hydrophobic residues may enhance the interactions between the ligand (peptide) and the enzyme (DPP IV). Moreover, the S1 subsite of DPP IV was composed of hydrophobic pockets, which are important in enzyme inhibition by peptides [40]. Liu et al. [6] reported that DPP IV possessed two pockets, namely, S1, consisting of Tyr631, Val656, Trp659, Tyr662, Tyr666, and Val711; and S2, which was charged due to the presence of Arg125, Glu205, Glu206, Phe357, Ser209, and Arg358. According to other scientific data, three pockets can be distinguished in DPP IV: S1 (Tyr547, Ser630, Tyr631, Val656, Trp659, Tyr662, Asn710, Val711, His740), S2 (Glu205, Glu206, and Tyr662), and S3 (Ser 209, Arg358, Phe357) [6]. The tripeptides from the studied dataset that reached the maximum on the plot were WRD, WRI, WRL, and WRN (numbers 15, 20, 22, and 24 in the plot, respectively). The impact of Trp and/or aliphatic amino acids, like, e.g., Ala or Ile, has already been discussed (see above). In the case of hydrophilic residues, like Arg (R) and/or Asn (N), their role in DPP IV inhibition remains unknown [21]; but, as it could be seen above, Arg is a residue forming the charged pocket of DPP IV [6].

The molecular weight of the tripeptides that reached the minimum of the function (see Figure 6b) was ca. 345 Da, whereas the MW of the tripeptides reaching the maximum of the function ranged from ca. 473 to 476 Da. According to Liu et al. [6], the molecular weight of the peptides is one of the factors affecting, e.g., the bioavailability of peptides, mode of action, and/or bioactivity. Liu et al. [6] analyzed the molecular weight of 222 peptides with different chain lengths for their IC₅₀ values (μM). It was found that peptide DPP IV inhibitors with MW lesser than 500 Da had IC₅₀ lower than 1000 μM (logIC₅₀ = 3). These findings are in line with the results obtained in the present study, indicating that predicted logarithmic values of IC₅₀ of tripeptides located in the minimum and maximum of the plot (Figure 6b) ranged from 1.45 (IPI; minimum of the function) to 2.82 (WRI; maximum of the function).

Among the results presented in Figure 6, four peptides with the lowest molecular weight (MW) had the following predicted log IC₅₀ values: 1.45 (IPI and LPL), 1.47 (VPL), and 2.41 (LQP). There are no significant differences in predicted log IC₅₀ (MW) between peptides with a molecular weight lower or higher than 500 Da. The molecular weight of 500 Da is the upper limit for compounds recommended as orally delivered drugs according to the Rule of Five [41]. This rule may also be applied to the bioactive components of food. Recommendation takes into account the fact that molecules with too high a molecular weight and containing too many groups able to form hydrogen bonds have lower bioavailability than the smaller ones. On the other hand, peptides too long to fulfill the Rule of Five are considered to be compounds with relatively high bioavailability [42,43]. Peptides with molecular weight exceeding 500 Da may thus be considered objects of interest as bioactive compounds together with the smaller ones.

Comparison of Figure 4 and Figure 7 shows that the prediction of log IC₅₀ based on a wider variety of descriptors provides better correlation between predicted and observed values than the calculation taking into account predictors associated with molecular weight only. Log IC₅₀ (ERM) reveals a better correlation with experimental results than log IC₅₀ (MW).

When discussing the results, it is worth mentioning that the SCIGRESS program was applied for the first time to study the SAR of DPP IV-inhibitory peptides. Its advantage is the possibility of refining the models by removing the descriptors which are difficult to interpret in order to produce reliable statistics (see Section 2). Nevertheless, the outputs obtained using SCIGRESS and the implemented descriptors are comparable with the results of Nongonierma et al. [44], who confirmed that some characteristic residues occurred in different locations of peptide DPP IV inhibitors. Dipeptide DPP IV inhibitors were often rich in Trp, which was also confirmed in this study. Additionally, amino acids—like Ala, Leu, and/or His—were typical for dipeptides capable of DPP IV inhibition [44]. This regularity was also found in the present study, in which the presence of the above-mentioned amino acids in a dipeptide sequence was defined by descriptors related to the number of methyl groups and the number of nitrogens (the higher this number is, the better the impact on bioactivity).

The analyses of tripeptide SAR were focused on the impact of their molecular weight on DPP IV inhibition. Nongonierma et al. [44] demonstrated more the diversified composition of tripeptides acting as DPP IV inhibitors, but some specific residues were indicated as well, including, Gly, Ala, Leu, Ile, Gln, Met, Phe, and Pro, which were found in different locations of the sequence. The results achieved using SCIGRESS software showed a more specific focus on the presence of Trp in the tripeptide chain to enhance the DPP IV-inhibiting effect.

To summarize, QSAR studies of compounds have been known since 1962 and are still being developed with increasingly sophisticated methods of computation to become a standard approach in drug discovery due to their efficiency in prediction and low cost of analyses [45]. In the area of the QSAR of peptides, particularly those of food origin, studies were undertaken on the sequences showing ACE-inhibitory [46], bitter-tasting [47], and anti-oxidative [11] effects. To date, there has been little research on QSAR of peptide DPP IV inhibitors. This may be due to the fact that although new peptides with such an activity are being identified, it is still not enough to merit creating a dataset for QSAR analysis. The results obtained using SCIGRESS bring more insight into the structural nature of peptides acting as DPP IV inhibitors.

4. Conclusions

The applied descriptors proved useful in revealing some structural requirements for food-derived di- and tripeptides to act as DPP IV inhibitors. Compared to the findings reported by other authors, this study results indicate more unequivocal factors affecting this particular bioactivity of peptides. For example, an increasing number of nitrogen atoms, as well as methyl groups, enhanced the DPP IV inhibition of dipeptides, whereas molecular weight turned out to be crucial in the case of tripeptides. Based on the latter, the requirement for tripeptide DPP IV inhibitor was a molecular weight up to 500 Da.