2.4.2. PLSR

The quantitative PLSR method was used for developing predictive training models of ◦Brix and sucrose contents by combining features from multiple linear regression and PCA. Cross-validation (leave-one-out) was used for internal validation of the training model. All syrup samples (*n* = 37) were randomly separated into calibration (~80% of the total samples) and external validation (~20% of the total samples) sets to evaluate the robustness of the trained models. Triplications of the same sample were used either in the training set or in the external validation set. The performance of the PLSR model was assessed with a correlation coeffect of cross-validation (Rcal) and predictions (Rval), standard error of cross-validation (SECV) and predictions (SEP), outlier diagnostics, leverage, and residual analysis [25]. Samples with high residuals and leverage were re-analyzed and excluded from the model if needed.

#### **3. Results and Discussion**

#### *3.1. Characterization of Maple Syrup Samples*

Reference analysis results for total soluble solids (◦Brix), sugar (sucrose, fructose, and glucose), and total phenolics for all samples, including traditional maple syrup, bourbon barrel (BBL)-aged maple syrup, and table syrups (corn, cane, and mixture—consisted of cane, maple, and agave syrups) are summarized in Table 1.


**Table 1.** Reference analysis results of total soluble solids, sugar (sucrose, fructose, and glucose), and total phenolics in traditional, bourbon barrel (BBL)-aged maple syrup and commercial table syrups.

a Total phenolics, expressed as micrograms of gallic acid equivalent (GAE) per 1 mL of distilled water. Three unusual maple syrups were excluded from this analysis due to containing of interferences. b *p* value, based on one-way ANOVA test; there were significant differences in total phenolics between three types of products (*p* < 0.05). Based on post hoc LSD, all samples were significantly different, except for BBL and dark maple syrup (*p* = 0.69). c Table syrups were excluded from total phenolic analysis.

Traditional maple syrups and BBL maple syrups showed similar total soluble solids ( ◦Brix) contents (Table 1). The ◦Brix values of maple syrups (65.4–68.7◦ with an average of 66.6 ± 0.7◦) were within the range reported by Stuckel and Low (62.2–74.0◦ with an average of 67.0 ± 1.6◦) [4] and Perkins (66–68◦) [26]. Sucrose, fructose, and glucose contents of traditional and BBL maple syrups are summarized in Table 1. We found no significant difference (*p* = 0.98, *p* > 0.05) in the sugar content between traditional and BBL maple syrups. Most sucrose contents of maple syrups agree with the reported literature (51.7–75.6% with an average of 68.0 ± 4.0%) [4], while four labeled as traditional maple syrup samples were far below the range (22.0, 23.5, 36.1, and 50.6%). These same four samples have much higher fructose (9.5–17.1%) and glucose (9.7–17.1%) contents than the literature reports (fructose 0.3 ± 0.5%, and glucose 0.4 ± 1.1%) [4]. In Morselli's study, fructose content in maple syrup was undetectable, and glucose content ranged from 0–7.3% [27]. The glucose to fructose ratio of 3 of the suspect samples was ~1:1, while the other had a 1:1.6 ratio. Invert sugar in maple syrups can be produced from sucrose hydrolysis during thermal processing or microbial contamination of the sap [26]. However, the abnormally high invert sugar contents and low sucrose contents in the samples indicate the potential adulteration of maple syrup with inexpensive table syrups. We evaluated commercial table syrup blends that showed similar levels of fructose (13.3 ± 0.8%), glucose (12.3 ± 1.9%), and sucrose 21.7 ± 7.8%) content to the suspect maple syrups.

Figure 1 shows the representative HPLC chromatograms of traditional maple syrup, suspicious maple syrup (which has high invert sugar and low sucrose content), BBL maple

syrup and table syrup (specifically corn syrup). The sugar profiles of both traditional and BBL maple syrups obtained by HPLC showed sucrose as the dominant sugar, while the suspicious maple syrup had noticeably high fructose and glucose contents as well as a detectable but low maltose content. In the literature, it has been stated that authentic maple syrup should not have any detectable maltose content [1]. Furthermore, it has been reported that syrup sweeteners, including molasses, high fructose corn syrup, and honey, have wide maltose composition variability, from 3.0–14.4% [28].

**Figure 1.** Representative HPLC-RID chromatograms of sugar profiles for traditional maple syrup, BBL maple syrup, suspicious maple syrup and corn syrup.

The total phenolic contents of traditional maple syrups and BBL maple syrups are summarized in Table 1. In previous reports, total phenolic contents in maple syrups ranged from 200–900 μg/mL, which agreed with our findings [29,30]. Since the FC method is based on the reagent's chemical reduction, the most problematic assay interference could be the presence of reducing sugars and samples with high protein levels [19]. In traditional maple syrups (except for suspicious ones) and BBL maple syrups, reducing sugars were undetectable. In addition, according to the literature, protein contents in maple syrups are in low concentration (~0–50 ppm) [1]. Therefore, the FC method can be considered a suitable method for analyzing total phenolics in maple syrups. The suspicious maple syrups were excluded from this analysis due to a high-level of reducing sugar (glucose and fructose) content.

The total phenolic content of traditional maple syrups correlated with their color grade and can be separated into golden and amber and dark groups. The dark traditional and BBL maple syrups had significantly higher total phenolic content than the golden and amber maple syrups (*p* < 0.001) according to one-way ANOVA and post hoc LSD tests. Dark maple saps are collected in the later production season when the temperature is

warmer (usually at warm springs) and sucrose is converted to invert sugar due to higher microbial activity [1]. Higher invert sugar contents in maple saps result in a stronger Maillard reaction during sap evaporation, giving darker color and stronger flavor in the final maple syrup products. In addition, a higher cultivation temperature of plants and higher activity of beneficial microbe/pathogen/insect feeding increase the total phenolic compounds, which also explain higher phenolics in dark maple syrups [31]. The higher phenolic content of BBL maple syrups could be associated with the aging process in the barrels, resulting in volatile and non-volatile phenolic compounds from the oak wood being absorbed [6] and contributing to their richer, more complex smell and flavor than traditional pure maple syrups.

The unique volatile profile of all maple syrup samples was characterized by GC-MS analysis, and two out of thirteen BBL-aged maple syrups were flagged as having a different volatile profile than the other BBL-aged maple syrup samples. These two samples did not have a different sugar and total phenolic profile than the other BBL samples. As shown in Table A1, a total of 18 volatile compounds were tentatively identified using the NIST 14.L database through a Mass Spectral Library search and were shown to be present in either maple or liquor products. The representative chromatograms for traditional maple syrup, BBL and suspicious BBL maple syrup are shown in Figure 2. There are several noticeable peak differences between BBL and traditional maple syrups. All authentic BBL maple syrups have one unique peak that other traditional and suspicious BBL (*n* = 2) maple syrups do not have, which corresponds to 1,1-diethoxy-2-methylpropane. Previous studies found 1,1-diethoxy-2-methylpropane in aged bourbon whiskey [32]. Therefore, authentic BBL maple syrups could absorb this volatile compound from the bourbon residue in barrels during the long aging process. In addition, one of the suspicious BBL had a similar volatile pattern as traditional samples in that they all had significantly lower contents in ethanol, oxalic acid, isoamyl alcohol, furfural and phenylethyl alcohol than authentic BBL samples, indicating that this suspicious BBL sample might have a minimum or no aging process, or the aging barrel does not contain any bourbon residuals [33]. While the other suspicious BBL sample had a similar volatile pattern as traditional samples, except for contents of isoamyl alcohol and oxalic acid, which were even higher than authentic BBLs, indicating that instead of aging, it might be added with bourbon flavor.

**Figure 2.** Representative GC-MS chromatograms of volatile compound profiles for BBL maple syrup, suspicious BBL maple syrup and traditional maple syrup.

#### *3.2. Spectral Information of Maple Syrup Samples*

The characteristic FT-IR absorption spectra of traditional maple syrup, BBL maple syrup, and corn syrup (as an example of table syrups) and their corresponding band assignments for specific functional groups are shown in Figure 3a. Key absorbance signals included the band at 2929 cm<sup>−</sup><sup>1</sup> associated with C-H stretching of the CH2 group in carbohydrates [8]. The band at 1637 cm<sup>−</sup><sup>1</sup> may be mainly related to O-H bonding in water, with minor contributions to C-O stretching in saccharides [34]. The band at 1415 cm<sup>−</sup><sup>1</sup> related to C-H bending [35] and the band at 1327 cm<sup>−</sup><sup>1</sup> related to O-H bending of the C-OH

group might attribute to organic acids. The band at 1110 cm<sup>−</sup><sup>1</sup> was associated with C-O stretching of C-O-C linkage, which could be the glycosidic linkage in sucrose. The bands at 1042 and 990 cm<sup>−</sup><sup>1</sup> were associated with C-O stretching in the C-OH group and C-C stretching in carbohydrates, and the band at 927 cm<sup>−</sup><sup>1</sup> was related to C-H stretching [8]. The broadband located around 3600–3000 cm<sup>−</sup><sup>1</sup> was mainly related to O-H bonds stretching in water, which has been reported previously as the major infrared bands of water located at 3490 and 3280 cm<sup>−</sup><sup>1</sup> for O–H stretching [36,37]. The range from 1200 to 800 cm<sup>−</sup><sup>1</sup> could be assigned to the carbohydrates absorption region, mainly related to sucrose, fructose, and glucose absorption bands [8,36].

**Figure 3.** (**a**)FT-IR spectrum band positions and corresponding wavenumbers of traditional maple syrup, BBL maple syrup, and corn syrup at a frequency of 4000–700 cm<sup>−</sup><sup>1</sup> collected using a portable five-reflections ZnSe crystal ATR system. (**b**) Raman spectrum, band positions and corresponding wavenumbers of traditional maple syrup, BBL maple syrup and corn syrup at a frequency of 350–1500 cm<sup>−</sup><sup>1</sup> collected using benchtop Raman with 1064 nm excitation laser.

The characteristic Raman signal of traditional maple syrup, BBL maple syrup, and corn syrup (as an example of table syrups) and their corresponding band assignments for specific functional groups are shown in Figure 3b. The major bands in the Raman spectra were centered in the range of 500–1500 cm<sup>−</sup>1. One major band at 522 cm<sup>−</sup><sup>1</sup> was associated with the deformation of C-C-O and C-C-C [38], while another major band at 542 cm<sup>−</sup><sup>1</sup> is related to an unassigned vibration [8]. The band at 590 cm<sup>−</sup><sup>1</sup> is associated with skeletal vibration [38], and the band at 629 cm<sup>−</sup><sup>1</sup> corresponded with sugar ring deformation [28]. The minor band at 740 cm<sup>−</sup><sup>1</sup> could be due to C-C, C-O stretching in the carbohydrate molecules [13]. The dominant peak at 835 cm<sup>−</sup><sup>1</sup> is responsible for C-C stretching, which is an intense band found in sucrose [39]. The high Raman signal at 835 cm<sup>−</sup><sup>1</sup> band is associated with the high sucrose content (~68%) in maple syrup [4]. Both peaks at 923 and 1067 cm<sup>−</sup><sup>1</sup> are responsible for the combination of vibration C-H bending, especially the C-H bond at C1 position and COH bending [8]. The peak at 1127 cm<sup>−</sup><sup>1</sup> could be due to the deformation of C-O-H, as well as the vibration of C-N, which is found in protein or amino acid [28,38]. The band at 1265 cm<sup>−</sup><sup>1</sup> is associated with the deformation of C-C-H, O-C-H, C-O-H, and the vibration of Amide III, which is a peptide bond, and the band at 1460 cm<sup>−</sup><sup>1</sup> is related to the symmetric deformation in the plane of CH2 [38].

In both FT-IR and Raman spectra, corn syrup was easily differentiated from traditional maple syrup and BBL maple syrup using only visual assessment due to maple syrups' unique patterns. However, between traditional maple syrup and BBL maple syrup, the spectral differences were not noticeable via visual evaluation due to their similarity. Therefore, a supervised classification method (SIMCA) was used to analyze the spectral data and to determine the class belongings, including traditional maple syrups, BBL maple syrups, and suspicious samples.

#### *3.3. Multivariate Data Analysis*

#### 3.3.1. SIMCA Classification Model of GC-MS

The GC-MS data of volatile compounds in traditional and BBL-aged maple syrup samples were analyzed and grouped using the Soft Independence Modeling of Class Analogy (SIMCA), and the class projection plot is shown in Figure A1. All the authentic BBL maple syrups were successfully discriminated from the traditional maple syrups based on their volatile composition, having an interclass distance (ICD) of 4.1. Furthermore, authentic BBL samples were also successfully differentiated from the suspicious BBL maple syrups (ICD = 2.2), and the classification pattern agreed with the GC-MS data that one of the suspicious BBL grouped with traditional samples, while the other one did not fall into either traditional or authentic sample group. Overall, the five most critical volatile compounds that have the highest impact on SIMCA model discrimination are the order of ethanol, isoamyl alcohol, isobutanol, oxalic acid, and acetoin, which are found to exist in bourbon whiskey or maple sap [33,40]. Therefore, these compounds are significant in authenticating qualified BBL maple syrups from suspicious BBL and traditional maple syrups.

#### 3.3.2. SIMCA Classification Models of FT-IR and Raman Spectroscopy

Collected FT-IR and Raman spectra were analyzed using SIMCA classification analysis to discriminate traditional and BBL maple syrups from suspicious maple syrups. The multiple-class approach was applied for both FT-IR and Raman spectral data by having two well-established classes existing (BBL and traditional maple syrups) in the training model. The projection plots of training sets are shown in Figure 4a,c. The training sets were developed using 11 BBL maple syrups (two suspicious BBL samples were excluded) and 15 traditional maple syrups (four suspicious traditional samples were excluded). All the BBL maple syrups were assigned to class number 1, and traditional maple syrups were assigned to a different class (#2). Suspicious maple syrups that were found according to the HPLC and GC-MS analysis were assigned as non-target samples and were not represented by the classes. For the FT-IR model, five factors were employed and explained 99.8% of the variances. In the Raman model, six factors were used and explained 98.1% of the variances. In this approach, the training models have ICDs of 4.8 and 2.5, classifying BBL maple syrups into traditional maple syrups based on the FT-IR and Raman methods, respectively.

The SIMCA discriminating power plot interprets variables that have a predominant effect on the sample classification [41]. The fingerprint region of 800–1200 and 800–1000 cm<sup>−</sup><sup>1</sup> was used to discriminate BBL and traditional maple syrups using FT-IR and Raman spectrometers, respectively. For the FT-IR system, most of the model variance was explained by intensity differences of bands located at 878 cm<sup>−</sup>1, which is closely related to the symmetric stretching of the primary alcohol group, and 1034 cm<sup>−</sup>1, which is related to the C-O bond stretching [42,43]. For the Raman system, most of the model variance was explained by the band at 879 cm<sup>−</sup>1, which was also related to the alcohol group's concentration [44]. Therefore, both FT-IR and Raman methods indicated that differences in compounds with alcoholic groups could explain the variance between BBL and traditional maple syrups. This finding agrees with our GC-MS results since ethanol, isoamyl alcohol, and isobutanol are the top three compounds, assisting BBL maple syrups' differentiation from traditional maple syrups.

**Figure 4.** Soft independent modeling of class analogy (SIMCA) projection plots of classification of traditional and BBL maple syrups with (**a**) FT-IR and (**c**) Raman; prediction of external validation sets, including authentic traditional and BBL samples and suspicious samples by (**b**) FT-IR and (**d**) Raman.

The performances of the supervised multiple-class FT-IR and Raman models were evaluated through an independent external validation set, which comprised four traditional and four BBL maple syrups, four suspicious traditional maple syrups, and two suspicious BBL maple syrups. All four traditional and four BBL maple syrups in the external validation set were tested with all reference analyses, and no abnormal pattern was found. The projection plots of validation sets are shown in Figure 4b, d and displayed well-separated clusters in both methods.

Both FT-IR and Raman models accurately predict all traditional and BBL maple syrups in the correct class (*n* true positive = 8, *n* false negative = 0, sensitivity = 100%), except for two traditional samples with one replication predicted as No Match in the Raman model. In addition, all suspicious traditional maple syrups were predicted as non-pure, and all suspicious BBL samples were predicted as traditional maple syrup, which was consistent with our expectations (*n* false positive = 0, *n* true negative = 6, specificity = 100%). Therefore, all traditional and BBL maple syrups were successfully authenticated by FT-IR and Raman with the multiple-class approach based on their unique chemical composition, and our results agreed with the reference analysis. Our FT-IR and Raman systems displayed a better performance than previous studies of detecting cheap sweetener adulteration in maple syrups, which had 88–100% correctness of discrimination with FT-IR and 98% correctness of discrimination with Raman [8]. Since there is no previous peer-reviewed study investigating BBL maple syrups' characterization and no formal regulation about the quality control of BBL maple syrups, a larger sample size of BBL maple syrup samples is needed for generating a more comprehensive and representative prediction model in the future.

#### 3.3.3. Regression Models

It is important to monitor the ◦Brix and sucrose contents in maple syrup to ensure product quality and stability [26]. Partial least square regression (PLSR) prediction models were developed with FT-IR and Raman spectra and reference values of ◦Brix and sucrose contents (Figure A2). Performance statistics of the PLSR models developed using training (*n* = 26) and external validation (*n* = 11) data sets are listed in Table 2. The number of samples and the range in training models are not all the same due to outlier exclusion. Four and five factors were selected to generate FT-IR and Raman training models, respectively, according to the standard error of cross-validation (SECV) (leave-out-out) result from carrying out the best quality of the models as well as to avoid possible overfitting.

**Table 2.** Statistics of partial least square regression (PLSR) models developed using a training (*n* = 30) and an external validation (*n* = 7) data set based on FT-IR and Raman spectral data for estimating Brix and sucrose contents in traditional maple syrups, BBL maple syrups, and table syrup samples.


a Sample number in the training models. b Standard error of cross validation. c Sample number in the external validation models. d Standard error of prediction.

Our PLSR models showed strong correlations (Rcal > 0.98 and Rval > 0.95) in predicting ◦Brix and sucrose contents in traditional maple syrups, BBL maple syrups, and table syrup samples. The standard error of prediction (SEP) values were 0.88% and 1.66% for FT-IR validation models for ◦Brix and sucrose, respectively, and were 1.23% and 1.67% for Raman validation models for ◦Brix and sucrose, respectively. Similar SECV and SEP were obtained, indicating the robustness of the models. Standard errors of laboratory (SEL) for reference methods of ◦Brix and sucrose were 0.21% and 0.62%, respectively. The SEL values were compared with the prediction performances of the models (SEP), and we found that the SEP values (Table 2) were always higher than those of SEL because the SEP includes not only the sampling and analysis errors but also the spectroscopy and model errors [45]. The SEP obtained for the FT-IR and Raman models were 2.7 times those of the SEL for sucrose, representing good precision of the models [46]. Conversely, the models predicting ◦Brix had a SEP/SEL ratio of 4.2 (FT-IR) and 5.9 (Raman), which were higher than the SEP/SEL threshold of 2 [46] for acceptable precision compared to the referenced method. However, our models show superior performance compared to reported ◦Brix predictions for honey using FTIR (R2val = 0.86, SEP = 1.84%) and Raman (R2val = 0.87, SEP = 1.32%) [47,48]. Nickless et al. quantified the total sugar contents in Manuka nectar using FT-IR, reporting Rval = 0.95 and SEP = 1.17% values [15].

The regression vector plots, shown in Figure A3, help to identify the functional groups whose variance is the highest for correlating between reference values and spectral data. The key FT-IR region for the ◦Brix and sucrose predictions was in the 1750–700 cm<sup>−</sup><sup>1</sup> range, with distinguished bands centered at 1635 (OH bending vibration characteristic of absorbed water) and the 1125 to 900 cm<sup>−</sup><sup>1</sup> related to C-O stretching and ring vibrational modes of sugars [8,34]. The regression vector plots for Raman data indicated that the bands at 835, 990, 1100 cm<sup>−</sup><sup>1</sup> explained most of the variance for the Brix model, and the bands at 424, 600, and 890 cm<sup>−</sup><sup>1</sup> explained for the Sucrose model. The scattering bands in the vicinity of 424 and 600 cm<sup>−</sup><sup>1</sup> are associated with the deformation of C-C-O and C-C-C [38]. The bands near 990 and 1100 cm<sup>−</sup><sup>1</sup> are related to the deformation modes of saccharides functional groups [28,38].

## **4. Conclusions**

In summary, FT-IR and Raman techniques fingerprinted maple syrup products based on their unique chemical composition, allowing for BBL and traditional maple syrup authentication. Both FT-IR and Raman systems combined with SIMCA provided nondestructive, fast, and accurate determination of quality traits in BBL and traditional maple

syrups and detected potential maple syrup adulterants. Our results showed that 15% of commercial maple syrup (traditional and/or BBL) samples that were tested and labeled as "pure" exhibited unusual sugar and/or volatile profiles, and both FT-IR and Raman equipment discriminated these suspicious samples from the pure ones. Furthermore, both FT-IR and Raman, combined with PLSR, showed good predictions for all samples' total ◦Brix and sucrose contents.

**Author Contributions:** K.Z.: methodology, validation, formal analysis, investigation, writing— original draft, visualization; D.P.A.: methodology, validation, writing—review and editing, supervision, project administration; L.E.R.-S.: conceptualization, validation, resources, writing—review and editing, supervision. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research did not receive any specific gran<sup>t</sup> from funding agencies in the public, commercial, or not-for-profit sectors.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data used to support the findings of this study can be made available by the corresponding author upon request.

**Acknowledgments:** We would like to thank Bissell Maple Farm, Jefferson, Ohio, for kindly providing maple syrup and BBL samples for our research.

**Conflicts of Interest:** The authors declare no conflict of interest.


**Appendix A**

**Figure A1.** Soft independent modeling of class analogy (SIMCA) projection plots of classification of authentic BBL samples from suspicious BBL and traditional maple syrup samples by GC-MS.

**Figure A2.** PLSR calibration and validation plots for Brix (**<sup>a</sup>**,**b**), and sucrose (**<sup>c</sup>**,**d**) in traditional maple syrups, BBL maple syrups, and table syrup samples utilizing 4500 FT-IR and Raman data, respectively. Black circles represent calibration set samples; gray circles represent external validation set samples.

**Figure A3.** PLSR regression vectors for Brix (**<sup>a</sup>**,**b**) and sucrose (**<sup>c</sup>**,**d**), utilizing 4500 FT−IR and Raman data, respectively.
