Infrared spectroscopy for the quality assessment of Habanero chilli: a proof-of-concept study

Habanero chillies (Capsicum chinense cv Habanero) are a popular species of hot chilli in Australia, with production steadily increasing. However, there is limited research on this crop due to its relatively low levels of production at present. Rapid methods of assessing fruit quality could be greatly beneficial both for quality assurance purposes and for use in breeding programs or experimental growing trials. This work investigated the potential of using infrared spectroscopy for the quality assessment of 20 Australian Habanero chilli samples, including the prediction of dry matter content, total phenolic content, ferric reducing antioxidant potential and capsaicin/dihydrocapsaicin content. Near-infrared spectra (908–1676 nm) taken from the fresh fruit showed strong potential for the estimation of dry matter content, with a root mean standard error of cross-validation (RMSECV) of 0.50% FW. A moving-window partial least squares regression model was applied to optimise the spectral window used for dry matter content prediction, with the best-performing window being between 1224 and 1422 nm. However, the near-infrared spectra of fresh fruit could not be used to estimate the total phenolic content or capsaicin/dihydrocapsaicin content of the samples. Mid-infrared spectra (4000 − 400 cm− 1) collected from the dried, powdered material showed slightly more promise for the prediction of total phenolics and the ratio of capsaicin:dihydrocapsaicin, with a R2cv of 0.45 and RMSECV of 0.32 for the latter parameter. Although further refinement is required, this method may be able to detect samples with high/low contents of total phenolics or for the estimation of the capsaicin:dihydrocapsaicin ratio.


Introduction
Habanero chillies (Capsicum chinense cv Habanero) are some of the hottest commonly consumed chillies in Australia. In addition to having a Scoville Heat rating of up to 200,000 Scoville Heat Units (SHU), they have a sweet and fruity flavour (AustChilli 2019) and contain exceptionally high levels of vitamin C and other phytonutrients such as phenolic and antioxidant compounds (Olatunji and Afolayan 2018;Hamed et al. 2019;Palma-Orozco et al. 2021). Compounds found in the chilli fruit, including phenolics, Joel B. Johnson joel.johnson@cqumail.com 1 capsaicinoids present in chilli are capsaicin and dihydrocapsaicin (Aza-González et al. 2011); however, a number of other capsaicinoids may be present in minor amounts (Kaiser et al. 2017). Both capsaicin and dihydrocapsaicin comprise a vanillyl group that is connected by an amide and ketone group to an extended alkyl chain ending in an isobutyl group (Fig. 1). Capsaicin possesses an alkene bond adjacent to the isobutyl group, while the alkyl chain of dihydrocapsaicin is saturated. The general structure is quite reminiscent and analogous of other natural pungent compounds, such as gingerols from ginger (Johnson et al. 2020c). Capsaicinoids are synthesised through a combination of the phenylpropanoid and branched-chain fatty acid pathways (Mazourek et al. 2009;Kaiser et al. 2017).
In Australia, production of chilli mainly occurs in Queensland (Orchard Tech 2021). Although currently only worth around $21 million p.a. (2013-14 figures), the value of the Australian chilli industry is rapidly expanding (AusVeg 2016). There is an ongoing interest in developing new Habanero chilli varieties with higher capsaicin contents, as these form a niche high-value market sector. Canto-Flick et al. (2008) reported a wide range of variability in the pungency of 18 germplasm accessions of Mexican Habanero pepper, with capsaicinoid contents of the whole fruit ranging from 9.7 to 59.5 mg/g (145,950 − 892,719 SHU). This highlights the spectrum of genetic potential existing in the Habanero crop for breeding more pungent chilli cultivars. However, environmental conditions also have a strong influence on capsaicinoid content (Naves et al. 2019), underscoring the importance of understanding the impact of growing conditions on pungency. For example, drought stress increases capsaicin and dihydrocapsaicin contents (Kopta et al. 2020), particularly in cultivars with initially low or medium pungencies (Phimchan et al. 2012). Similarly, capsaicinoid content is affected by the growing temperature (González-Zamora et al. 2013) and light intensity and wavelength (Gangadhar et al. 2012;Nagy et al. 2017).
Capsaicinoid contents are generally measured by highperformance liquid chromatography (HPLC), which provides a high level of specificity and accuracy (Davis et al. 2007). However, this technique is time-consuming and expensive, which means that it may not be suitable for routine assessment of large number of samples. Hence there is recent interest in using rapid analytical techniques such as infrared spectroscopy for the quality assurance/analysis of chilli (Sánchez et al. 2019).
Near-infrared spectroscopy has previously been used for estimating the capsaicinoid content in chilli powder (Bae et al. 1998;San Park et al. 2008;Lim et al. 2015;Bonifazi et al.2019) and in anti-rheumatical plasters (Czarnik-Matusewicz and Korniewicz 1998). Additionally, NIRS has been used for the measurement of total phenolic content, total carotenoid content and vitamin C content (Toledo-Martín et al. 2016), again in dried chilli powder. Similarly, Domínguez-Martínez et al. (2014) used mid-infrared spectroscopy to quantify the capsaicin, ascorbic acid, total phenol content, and antioxidant activity in dried serrano chilli powder.
However, there is more limited work using NIRS on fresh chilli fruit. A number of studies applying NIRS to the fresh crop have focused on the prediction of dry matter content and soluble solid content (Toledo-Martín et al. 2016;Sánchez et al. 2019). Nevertheless, some studies have also measured the vitamin C, total chlorophyll and/or carotenoid contents in the intact fruit (Penchaiya et al. 2009;Ignat et al. 2012;Ignat et al. 2014;Kasampalis et al. 2021).
Typical RMSECVs found in these studies for the prediction of dry matter content in the intact fruit range from 0.4 to 0.7%. For example, Sánchez et al. (2019) recently used a handheld MicroPhazir instrument (spectral range 1600-2400 nm) to predict dry matter content in bell peppers, reporting an R 2 cv of 0.62, RMSECV of 0.66%, and crossvalidation ratio performance deviation (RPD cv ) of 1.64. In addition, they were able to quantify soluble solid content in the fresh pepper with a similar level of accuracy.
However, there does not appear to be any work investigating the use of infrared spectroscopy for the determination of capsaicinoids in fresh chilli fruit, or for the quality determination of Habanero chilli samples. Consequently, the aim of this study was to conduct a small-scale, proof-of-concept investigation into the potential application of infrared spectroscopy for the quality analysis of Habanero chilli, and to compare the relative performance of near-infrared (NIR) and mid-infrared (NIR) spectroscopy for this purpose.

Chilli samples
Twenty samples of Habanero chilli were sourced from Austchilli (Bundaberg, Queensland) in January 2020. Samples were chosen from different within-field locations of a commercial chilli crop to incorporate a wide range of environmental variability. After being shipped to the laboratory overnight on ice, NIR spectra were collected from the fresh chillies. Following this, the chillies were dried at 50 °C in a forced air oven (Sunbeam Food Lab Dehydrator) until reaching a constant mass. The dry matter content of the samples was calculated from the loss in mass and expressed on a % fresh weight basis. Using a Retsch ZM1000 centrifugal grinding mill (Sydney, Australia), the dried samples were ground to < 1 mm particle size.

Collection of NIR spectra
Near-infrared (NIR) spectra between 908 and 1676 nm were collected from the fresh, intact chillies using a MicroNIR OnSite handheld spectrometer (Viavi, Santa Rosa, CA, USA), calibrated with black and white references every 10 samples. Duplicate spectra were collected from opposite sides of each chilli, providing four spectra per sample (n = 80 spectra in total). All spectra were collected approximately 3 cm down from the chilli stem for consistency.

Collection of MIR spectra
The mid-infrared (MIR) spectra were collected from the dried, ground chilli powder using a Bruker Alpha Fourier transform infrared (FTIR) spectrophotometer (Ettlingen, Germany) fitted with an attenuated total reflectance (ATR) module (Johnson et al. 2020d). Spectra were collected in triplicate for each sample (n = 60 spectra in total), at wavenumbers between 4000 and 400 cm − 1 (24 scans at 4 cm − 1 resolution).

Total phenolic content and ferric reducing antioxidant potential
Polar compounds were extracted from the dried, powdered samples in duplicate using 90% methanol, following previously described protocols (Johnson et al. 2020b). The total phenolic (TP) content of the extracts was measured as previously reported for other matrices (Johnson et al. 2020a), with results expressed as gallic acid equivalents (GAE) per 100 g (dry weight basis). The ferric reducing antioxidant potential (FRAP) was also measured on the extracts as previously described (Johnson et al. 2020a), with results expressed as Trolox equivalents (TE) per 100 g.

Capsaicin and dihydrocapsaicin content
The HPLC method for the analysis of capsaicin and dihydrocapsaicin was adapted from Waite and Aubin (2008). The 90% methanol extracts were syringe filtered (Livingstone 0.45 μm PTFE) prior to direct HPLC analysis. The capsaicinoids were separated using a reversed phase C 18 column (Agilent Eclipse XDB-C18; 150 × 4.6 mm; 5 μm pore size) on an Agilent 1100 HPLC system (Waldbronn, Germany). The injection volume was 25 µL, with the mobile phase comprised of water (A) and methanol (B) at a flow rate of 1 mL/min. The gradient began at 40% B, increasing to reach 85% B at 8 min, then 99% B at 10 min. It was held at 99% B for a further 3 min, before returning to 40% B by 15 min. The total run time was 16 min, with a post-run equilibration time of 2 min between samples. The detection wavelength was set at 280 nm.

Data processing and analysis
The NIR spectra were exported in ASCII (*.csv) format and subsequently imported into the Unscrambler X software (version 10; Camo Analytics; Oslo, Norway) for chemometric analysis. The MIR spectra were exported in Opus format and read directly into the Unscrambler X software.
Partial least squares regression (PLS-R) was performed in the Unscrambler X, using leave-one-out (LOO) crossvalidation. Various spectral pre-processing treatments were trialled, including Standard Normal Variate (SNV) normalisation and first and second derivative treatments using differing numbers of smoothing points for the Savitzky-Golay algorithm.
Moving window PLS-R was conducted using a custom script in R Studio running R 4.0.2 (R Core Team 2020). The spectrolab, prospectr and plsr packages were used for spectral importation, pre-processing and PLS-R analysis, respectively.

HPLC method
The HPLC method showed good linearity for the standards of both capsaicin and dihydrocapsaicin (R 2 > 0.998), with a limit of detection (LOD) and limit of quantification (LOQ) of approximately 0.35 and 1.17 mg L − 1 , respectively ( Table 1). The reproducibility of the method, measured by triplicate injections of 165 mg L − 1 capsaicin standard, was much lower than most of the Habanero samples reported by Canto-Flick et al. (2008).

NIR spectra
The NIR spectra of the chilli samples are shown in Fig. 3. Following Standard Normal Variate (SNV) treatment, two outlier spectra were removed. The major peaks were centred at approximately 1447, 1193 and 976 nm, corresponding to the OH first overtone, CH second overtone stretch, and OH second overtone, respectively (Workman and Weyer 2007;Lapcharoensuk et al. 2020).
Only the prediction of dry matter gave acceptable model statistics when using the NIR spectra, with a R 2 cv of 0.65 and RMSECV of 0.50% FW (Table 3). Although the model statistics could be further improved (Fig. 4), it could find use for the rapid, in-field estimation of dry matter content. As dry matter content has been found to increase significantly during the ripening process (Kasampalis et al. 2021), this could potentially be used as a fruit maturity index to guide harvest decisions.
As all the chilli samples included in this study were mature and hence had a relatively similar range of dry matter contents (Table 2), using a wider range of sample maturities with a greater range of dry matter content would undoubtedly improve the statistical accuracy of the model (i.e. R 2 , RPD). Nevertheless, the performance found here was better than that reported by Sánchez et al. (2019), who 0.2% relative standard deviation (RSD). The inter-day precision of the same standard, measured across two days, was < 0.2% RSD.
A typical chromatogram obtained for one of the Habanero chilli extracts is shown in Fig. 2.

Descriptive statistics
Descriptive statistics relating to the parameters measured in this study are provided in Table 2. The capsaicin content ranged from 1474 to 3916 mg/kg (DW), while the dihydrocapsaicin content ranged from 638 to 1757 mg/kg. The most pungent sample found was found to have 88,480 SHU, The results of the moving window analysis are shown in Fig. 6. The optimum model was found between wavelengths of 1224-1422 nm, with a R 2 val of 0.67, RMSECV of 0.46% and RPD of 2.20. This showed a moderate improvement over the full-wavelength model (Table 3).

MIR spectra
The MIR spectra of the chilli samples are shown in Fig. 7.
The spectra were pre-processed using a Savitzky-Golay algorithm (2nd order polynomial; 11 pts) prior to subsequent processing (SNV, 1st and 2nd derivative). The preprocessing treatments trialled included: SNV smoothing, 1d11, 1d21, 2d11, 2d1, SNV + 1d11 and SNV + 2d11. As shown in Table 4, the model performance for all parameters aside from dry matter was slightly higher when using the MIR spectra compared to the NIR spectra; nevertheless, model performance remained quite poor. While the MIR spectra were collected from the dried chilli samples and hence dry matter cannot be predicted from the moisture content of these samples, it may be correlated with other matrix constituents such as cellulosic carbohydrates that can be detected using MIRS, explaining the moderate correlation observed among the MIR prediction models. The best performing MIR model was for the ratio of capsaicin:dihydrocapsaicin, with a R 2 val of 0.45, RMSECV of 0.32 and RPD of 1.34 (Fig. 8). Key wavelengths loading onto this PLS-R model included 1641 cm − 1 , potentially resulting from the C = C stretch of the alkene bond in dihydrocapsaicin, and 1064 cm − 1 , which may result from C = C bending and/or the C-O stretch of the alcohol group. Domínguez-Martínez et al. (2014) reported successfully using MIR spectroscopy for the estimation of capsaicin content in serrano chilli (C. annuum var. serrano). These used a handheld MicroPhazir instrument (spectral range 1600-2400 nm) to predict dry matter content in bell peppers and found an R 2 cv of 0.62, bias-corrected RMSECV of 0.66% and RPD cv of 1.64. Examination of the loadings plots revealed the key predictive wavelengths were 1391 and 1137 nm, corresponding to the second and third overtones of the OH bond (Fig. 5).
However, PLS-R prediction of other parameters showed poor R 2 val values and high RMSECV values, indicating that the NIR spectral range used was not able to detect the functional groups responsible for these compound classes (i.e., total phenolics, capsaicinoids). This contrasted with previous work which found that NIRS may show promise for the estimation of capsaicinoid content (San Park et al. 2008;Lim et al. 2015), likely due to the presence of water in the fresh samples overwhelming the spectral signals from these more minor compounds. Other factors which could vary between instruments include the much narrower band of NIR wavelengths used by the handheld NIR instrument in this study (~ 900-1700 nm), excluding much of the informative regions in the mid to far NIR portions of the spectrum, as well as the optical geometry and signal-to-noise ratio.
One of the challenges in creating robust non-invasive prediction models is the selection of infrared wavelengths that provide the most useful spectral information, while excluding wavelengths that do not contain any information relating to the parameter of interest. Hence a moving spectral window procedure was implemented to determine the optimum range of wavelengths to use for the prediction of dry matter content, following the method of Anderson et al. (2020). The procedure used the SNV + 1d5 pre-processed spectra, with a step increment of 6 nm throughout the NIR spectrum.

Conclusion
The proof-of-concept study presented here sought to investigate the potential of NIR and MIR spectroscopy towards the rapid quality assessment of Habanero chillies, including the prediction of dry matter content, total phenolic content, ferric reducing antioxidant potential and capsaicin/dihydrocapsaicin content. Spectra collected using a handheld NIR instrument showed strong potential for the estimation authors used 25 samples with capsaicin contents ranging from 4.4 to 15.3 mg/g (DW), which are much higher concentrations compared to those found in the Habanero chilli samples used in this study (0.7-1.5 mg/g DW). Hence the poorer results from this preliminary study may be due to the lower analyte concentrations and relatively small sample size. Further work would be required to determine if infrared spectroscopy can be used to detect and quantify the lower levels of capsaicinoids present in Habanero chilli.   . 7 The MIR spectra of the Habanero chilli samples between 4000 − 400 cm − 1 of DM content, but not for TP, FRAP or capsaicinoid content. Similarly, MIR spectroscopy did not perform well for the estimation of capsaicinoid content, although it showed slightly more promise for the estimation of TP and FRAP content, and the ratio of capsaicin:dihydrocapsaicin.