Food Chemistry 274 (2019) 137–145

Contents lists available at ScienceDirect

Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem

Volatile and sensory characterization of roast coffees – Effects of cherry T

maturity
⁎
Sebastián Velásqueza, Nestor Peñaa, Juan Carlos Bohórqueza, Nelson Gutierrezb, Gavin L. Sacksc,
a
Department of Electrical and Electronics Engineering, Universidad de los Andes, Bogota, Colombia
b
Department of Agriculture Engineering, Universidad Surcolombiana, Neiva Huila, Colombia
c
Department of Food Science, Cornell University, Ithaca, NY, USA

A R T I C LE I N FO A B S T R A C T

Keywords: Immature coffee cherries produce roast coffees with lower hedonic scores than those produced from mature
Coffee cherries, but variation in volatile and sensory characteristics over a range of maturities is not well studied. In this
Fruit maturity work, cherries from two coffee cultivars (Caturra, Catimor) were sorted into seven maturity stages from fully
Volatile profiling immature (Stage 1, green) to fully overripe (Stage 7, purple). Volatile profiles of Stage 1 roast coffee had lower
Cupping
concentrations of carbohydrate degradation products and higher concentrations of N-heterocycles and phenols.
Sensory characterization
PCA
Differences in volatiles among Stage 2 (partially immature, yellow-green) and subsequent stages were insig-
nificant (p > 0.05) or else minor. Principle component analysis of the volatile data set also distinguished Stage
1 from other stages. Similarly, a trained cupping panel reported significantly lower sensory scores for Stage 1 as
compared to Stages 2–7, but few differences among Stages 2–7. Thus, partially mature and overripe cherries may
be appropriate for specialty coffee.

1. Introduction of specialty coffee have been identified, including harvesting coffee

cherries at an appropriate maturity level, monitoring and controlling
The international market for coffee (Coffea spp.) has undergone coffee fermentation conditions, and ensuring appropriately low
major transformations in the last decade due to rapidly growing de- moisture content of parchment coffee during drying (Arcila, Farfán,
mand for ‘specialty coffees’ (Castle, 2016). The definition of specialty Moreno, Salazar, & Hincapié, 2007). Of these factors, ensuring uniform
coffee is subject to debate, but typically refers to coffees that possess and appropriate cherry maturity is arguably the most challenging for
desirable flavor qualities specific to a particular geographic region small producers to achieve, for reasons explained below.
(Fold & Pritchard, 2011). Specialty coffee designations are generally Cherry maturity may be described in terms of “days post-bloom”
given by certifying organizations, which employ formalized ‘cupping’ (Marín, Montoya, & Oliveros, 2004), but is more commonly and easily
protocols for preparation and hedonic evaluation of the coffee. For classified by the color of the cherries (Arcila et al., 2007). Other ma-
example, the cupping protocol of the Specialty Coffee Association – the turity metrics, such as cherry firmness, titratable acidity, sugar content,
major specialty coffee trade organization – utilizes a 100-point scale for and density are well correlated with color (Arcila et al., 2007) and are
total quality, and a coffee must achieve an cumulative score of 80 or typically not as easy to assess in the field. Immature coffee cherries are
better to be considered specialty coffee (Specialty Coffee Association of green, and ripe cherries may be red, orange, pink or yellow depending
America, 2015). on to the variety (Flament & Bessière-Thomas, 2002). The deleterious
The majority of specialty coffee is produced through the so-called effects of immature green coffee cherries on eventual coffee quality can
“wet process”, in which the cherry skin and pulp are removed me- be striking – one report showed that the presence of only 2.5% green
chanically, the pulped coffee is fermented and washed to remove the cherries resulted in rejection of almost 30% of cups due to sensory
mucilage, and the seeds dried to produce parchment coffee (Poltronieri defects, purportedly because green cherries are harder to pulp and more
& Rossi, 2016). In many countries, including Colombia, these initial prone to spoilage issues (Puerta-Quintero, 2000). Post-harvest sorting
steps typically occur at the coffee plantation or at nearby processing to remove green cherries can limit these defects, but results in lost yield
cooperatives. Several quality control points for growers and processors (Marín et al., 2004). A separate study demonstrated that total quality

⁎
Corresponding author.
E-mail addresses: [email protected] (S. Velásquez), [email protected] (N. Peña), [email protected] (J.C. Bohórquez),
[email protected] (N. Gutierrez), [email protected] (G.L. Sacks).

https://doi.org/10.1016/j.foodchem.2018.08.127
Received 31 May 2018; Received in revised form 25 August 2018; Accepted 28 August 2018
Available online 29 August 2018
0308-8146/ © 2018 Elsevier Ltd. All rights reserved.
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

scores for coffees from different plantations were correlated with cherry approaches.
maturity (based on sugar content) (Silva, de Queiroz, Pinto, & Santos,
2014). Several other studies have evaluated differences in physical 2. Materials and methods
properties or non-volatile constituents between roast mature and im-
mature beans (Cossignani, Montesano, Simonetti, & Blasi, 2016; 2.1. Coffee samples – raw materials, processing, and roasting
Defernez et al., 2017; Franca & Oliveira, 2008; Monakhova et al.,
2015). Coffee cherries of the Catimor (CA) and Caturra (CU) varieties were
Studies on the effects of environmental or processing decisions on sourced from an independent smallholding farm in Pitalito, Colombia in
volatiles in saffron, tea, wine, and other foodstuffs are common in the the department of Huila. This farm is at 1350 amsl. The mean tem-
recent literature, with many reports relying on headspace solid-phase perature is 20.7 °C, with yearly precipitation of 1520 mm. Coffee
microextraction (HS-SPME) followed by gas chromatography mass cherries were sampled at seven different maturity stages, based on
spectrometry (GC–MS) (Braga et al., 2018; Martínez-Gil, del Alamo- appearance, with three biological replicates per stage (Fig. 1). Each
Sanza, Gutiérrez-Gamboa, Moreno-Simunovic, & Nevares, 2018; Mu biological replicate consisted of 10 kg of fresh coffee cherries. Stage 1
et al., 2018; Urbani, Blasi, Chiesi, Maurizi, & Cossignani, 2015). Al- represents immature green cherries, which appear 196 days after
though the volatile composition of premium coffee has been described flowering (Marín et al., 2004). Stage 2 corresponds to green-yellow
in recent reports (De Melo-Pereira et al., 2019; Inocencio-Monteiro cherries (i.e., 203 days), Stages 3 and 4 to almost ripe or “pintón”
et al., 2018), there is relatively little information regarding which cherries (i.e., 208–215 days. Stages 5 and 6 represent ripe cherries (i.e.,
odorants in roast coffees change with cherry maturity. One study 217–224 days) while Stage 7 (i.e., after 224 days) represents overripe
compared GC–MS volatile profiles among roast coffees from four types cherries.
of beans (black, sour, immature and defect-free), and reported higher From each 10 kg sample, 500 g were set aside for colorimetric
concentrations of certain pyrroles and pyrazines in coffees produced analyses, and the remainder underwent a wet processing approach si-
from immature vs. defect-free beans (Agresti, Franca, Oliveira, & milar to standard regional practices (Arcila et al., 2007). Each coffee
Augusti, 2008). Interestingly, GC-olfactometry/MS and reconstitution sample was pulped with a Gaviota 300 pulping machine (Bogotá, Co-
studies have shown that lower quality C. robusta coffees also have lombia), and 5 kg of pulped beans were fermented in plastic containers
higher concentrations of ‘earthy’ smelling pyrazines than C. arabica for 18–20 h. Beans were washed of mucilage, and sun-dried until a
coffees, along with higher concentrations of volatile phenols and lower moisture content of 9–10% was achieved. Moisture content was mea-
concentrations of “roast, sweet” smelling furans (Grosch, 1998). Non- sured with a Gehaka Agri Moisture Tester G600 (Sao Paulo, Brazil).
targeted volatile analyses on coffees of different origins have reported Finally, the beans were hulled in an Ingesec laboratory hulling machine
similar inverse correlations between certain N-heterocycles (including (Bogotá, Colombia).
pyrazines) and total quality (Ribeiro, Augusto, Salva, & Ferreira, 2012). For sensory analysis, samples (150 g) were lightly roasted in a
Other odorants are also known to affect cup quality, e.g. the appearance Quantik TC-150A R/G laboratory roaster (Armenia, Colombia).
of unwanted “stale” aromas in coffee is credited to oxidative losses of 2- Roasting conditions were set according to the Specialty Coffee
furfurylthiol and other thiols (Dulsat-Serra, Quintanilla-Casas, & Vichi, Association of America (SCAA) cupping protocol (Specialty Coffee
2016), but it is unclear if this phenomenon is related to differences in Association of America, 2015), and the color was used to verify the
quality among roasts of different bean maturities. roasting level required by the protocol (i.e., the L luminosity colori-
Although there is strong sensory evidence (and, to a lesser extent, metric axis). The heating curve was as follows: preheating at 200 °C for
chemical evidence) in support of avoiding immature coffee cherries 90 s, followed by equilibration and crepitation at 170 °C for 300 s, fol-
during specialty coffee production, this is often difficult to achieve in lowed by cooling starting at 192 °C down to room temperature over a
practice due to economic constraints. In larger processing operations, period of 90 s. The roast beans were ground, at medium particle setting,
sorting of cherries by maturity can be done (semi-)mechanically using in a Bunn G3 HD grinder (Springfield, IL) to achieve 70–75 pass in a 20-
flotation devices (density sorters) (Oliveros-Tascón, Sanz-Uribe, mesh sieve. For preparation of samples for volatile analyses, conditions
Montoya-Restrepo, & Moreno-Cárdenas, 2009). However, as mentioned were slightly altered to minimize volatile loss of the samples. Samples
earlier, this results in loss of yield. Additionally, mechanical sorting is were roasted in a FreshRoast SR500 30 g roaster (Bradford, CT) at
too expensive for regions with predominantly small-holding farmers, 195 °C for 360 s and subsequently cooled to room temperature over a
e.g. in Colombia, where 95% of growers manage plots smaller than 2 ha 90 s period. Samples were then sealed inside a beaker during 24 h, then
(Federación Nacional de Cafeteros, 2010). In smaller operations, ma- ground with a variable particle size grinder used for espresso pre-
turity sorting is typically performed manually in the field based on color paration prior to GC–MS analysis.
(Federación Nacional de Cafeteros, 2010). This practice is labor-in-
tensive, often requiring 4 to 7 separate harvests due to non-uniform 2.2. Colorimetric analysis of samples
fruit development (Arcila et al., 2007), and can dramatically reduce the
profitability of specialty coffee. Thus, there is a strong incentive to Colorimetric analysis was performed on harvested samples to con-
determine if the number of harvest dates can be decreased without firm the accuracy of the manual sorting protocol. Randomly selected
compromising cup quality, e.g., by using coffee cherries at stages other cherries (n = 3) from each cultivar × maturity stage combination were
than mature (Stage 5 or 6 in Fig. 1). Currently, growers are cautioned to analyzed with a Konica Minolta CR-410 Chroma Meter colorimeter
avoid cherries with even the slightest presence of green of the surface (New Jersey, USA) to obtain CIE L*a*b* parameters, in which L* in-
(Flament & Bessière-Thomas, 2002), but to our knowledge there is little dicates the luminosity of the sample, a* represents the red to green
chemical or sensory work to establish that partially-ripe (Stages 2–4) or ratio, and b* represents the blue to yellow ratio.
overripe (Stage 7) cherries present the same danger to coffee quality as
fully immature cherries (Stage 1). 2.3. Volatile quantitation by HS-SPME-GC–MS
We hypothesized that significant differences in volatile profiles and
cupping quality scores existed between Stages 5–6 (mature) roast cof- Roast, ground coffee samples (500 mg) were diluted in 4 mL satu-
fees and immature (1–4) and overripe (7) maturity stages. In this work rated NaCl solution in brown 20 mL solid-phase microextraction vials.
we report the first study of sensory assessments and volatile profiles for Vials were spiked with a set of four deuterated internal standards
coffees produced from cherries harvested over a range of maturities (hexanal-d12, naphthalene-d8, isobutyric acid-d7, and IBMP-d3).
(Fig. 1). Two different varieties from two sites in Colombia were used in Headspace solid-phase microextraction (HS-SPME) followed by gas
the study, and results were evaluated using multivariate statistical chromatography mass spectrometry (GC–MS) was performed on a

138
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

Cherry Pulped Green Roasted

1 2 3 4 5 6 7

Maturity
stages
Color GC/MS
1 2 3 4 5 6 7 • SPME
• Manual
BR1 BR2 BR1 BR2 BR1 BR2
• Targeted
BR1 BR2 BR1 BR2 BR1 BR2 BR1 BR2 classifica on
approach
Replicates • 10kg of
BR3 BR3 BR3 BR3 BR3 BR3 BR3
(per stage) coffee SCAA cupping
berries • 10 individual
• Wet a ributes
Caturra (CU) Catimor (CA) processing and Total
• Range: 6-10
Origin: Huila, • Konica
• Seven
Colombia Minolta CR a ributes for
Origin 410 modeling

Fig. 1. (Left) Depiction of sampling stages. Samples were collected from Stage 1 (immature, fully green) to Stage 7 (purple, overripe) for two cultivars, Catimor (CA)
and Caturra (CU). BR = biological replicate. (Right) Overview of data collection. Following harvest, sub-samples were characterized by colorimetry. The remainder
were processed and roast prior to volatile profiling by SPME-GC–MS or sensory analysis by the SCAA cupping protocol.

Shimadzu TQ-8040 GC–MS with AOC-5000 autosampler (Columbia, 2.5. Statistical analysis
MD). SPME vials were incubated for 20 min at 60 °C prior to extraction,
agitating at 500 rpm. Extraction was performed for 15 min with a 2 cm All statistical analyses were performed using R v3.4.4 with the
50/30 µm DVB/CAR/PDMS fiber (Supelco, Bellefonte, PA). The SPME mdatools package (Sergey, 2017). The correlation of maturity stage and
fiber was then desorbed, splitless, for 2 min at 240 °C onto a VF-WAXms CIE L*a*b* data was evaluated by principal component analysis (PCA) of
column (30 m x 0.32 mm x 0.50 µm; Agilent Technologies, Santa Clara, the CIE L*a*b* data, using mean centered values of all variables for each
CA). The column temperature program was as follows: initial tem- maturity × variety combination.
perature of 40 °C, no hold; increased to 240 °C at a rate of 5 °C/min, and For the volatile data (n = 103 volatiles), a two-way ANOVA (fac-
held for 10.0 min for a total run time of 50.0 min. The carrier gas was tors = variety, maturity stage) was performed on normalized, log2
helium, set at a constant linear velocity of 44.0 cm/s, pressure of transformed peak areas for each volatile. When significant effects of
11.3 kPa, total flow of 34.2 mL/min, and column flow of 1.49 mL/min. maturity stage were noted for a given volatile, pairwise comparisons
The ion source and interface temperature were both 230 °C, with data among maturity stages were performed by Tukey test. The normalized,
collected from 2.1 to 50.0 min. Q1 was operated in transmission mode, transformed mean values for each volatile were also evaluated by PCA.
and Q3 was operated in scan mode (m/z = 29–350), with a scan every For the sensory data, a multi-level model was used for each sensory
0.3 s. attribute, with maturity and variety as fixed effects and judges as a
Linear retention indices were calculated based on an alkane stan- random effect. As with volatile analyses, pairwise comparisons by
dard. When possible, compound identification was performed by Tukey test were performed for attributes which were significantly de-
comparison of mass spectra and retention indices against authentic pendent on maturity stage. Then, the mean-centered sensory values of
standards. The compounds for which there were no available standards the seven mentioned attributes and stages 2–7 were evaluated by PCA.
were tentatively identified based on a similarity search of a NIST mass Finally, the volatile measurements were used to create least-square
spectral library (2014 version) and comparison to literature retention models of sensory scores. For modeling sensory from the volatile data
indices (National Institute of Standards and Technology, 2018). Com- set, Partial Least Squares (PLS) was used (Wehrens, 2011). The X data
pounds with odor activity values (OAV) correspond to the ones reported matrix contained the relative concentration of the 103 volatile com-
in other references (Grosch, 1998; Piccino, Boulanger, Descroix, Shum, pounds detected. This matrix was mean-centered but not normalized,
& Sing, 2014; Steen, Waehrens, Petersen, Münchow, & Bredie, 2017). and not otherwise pre-processed. The Y matrix contained the results for
Quantification was performed by integrating the peak area of the main each of the sensory attributes results and was mean centered but not
ion, and normalization was performed using the average peak area of normalized. Individual PLS1 models were used. The variables (i.e.,
the following internal standards: hexanal-d12, naphthalene-d8, iso- compounds) were selected according to VIP (Variable Importance
butyric acid-d7, and IBMP-d3. Projection). Each variable that presented values lower than 1 was
iteratively extracted if the root mean square error (RMSE) was reduced.
As three samples for 14 different categories (cultivar × maturity stage
2.4. Sensory evaluation
combinations) were available, the sample set was divided into 28
samples for training and 14 for testing. The sample division was ran-
The sensory analysis consisted of a four judge panel trained on the
domized and tested 15 times, and the number of latent variables for PLS
SCAA cupping protocol (Specialty Coffee Association of America,
was selected by cross-validation to avoid overfitting. From this, the
2015). This protocol involves evaluation of ten coffee qualities, each
cross-validated RMSE, R2 , variable loadings for the first two compo-
graded on a scale from 6 to 10 at 0.25 point intervals. The total score is
nents of PLS, and estimated coefficients for PLS were obtained.
calculated as the sum of the ten attributes, with a final mark which
results from the deductions for possible defects. A total score of at least
80 is required for a specialty coffee certification. All ten attributes were 3. Results and discussion
evaluated during the cupping sessions, and seven of these attributes
(aroma, flavor, aftertaste, acidity, body, overall, and total) were used in 3.1. Colorimetric analysis
further statistical analyses, as described in Section 2.5. The other three
attributes (clean cup, uniformity, sweetness) were not further con- Coffee cherries were initially harvested and manually sorted from
sidered because their values showed little variation across samples. The Stage 1 (green immature) to Stage 7 (purple overripe) based on visual
tests were performed on two different days, with three cupping sessions appearance. Visual assessment of coffee cherry maturity is standard
per day, and coffee samples prepared freshly for each evaluation. The practice in the Colombian industry. To confirm that sorting was exe-
judges were not informed that samples varied in maturity, nor that cuted accurately, a colorimeter was used to evaluate for color para-
biological replicates were included. meters (L*, a*, b*) replicate cherry samples. Raw CIE L*a*b* data is

139
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

1 2 CHO degrada on Other/Unknown Phenol deriva ves

3 4 5 6 7 Strecker S-compounds
N-heterocycles

2,3-Pentanedione
5-Methylfurfural

2,3-Butanedione
Phenylacetaldehyde 1-Ethyl-3-methylbenzene
2-Ethyl-3,5-dimethylpyrazine
3-Ethyl-2,5-dimethylpyrazine
2-furfurylthiol ß-Pinene
PC2 :11.5% explained

2,3-Dimethyl-5-ethylpyrazine
2-Ethyl-5-methylpyrazine
(E,E)-2,4-Nonadienal
2,6-Diethylpyrazine
2-Methyl-5-propylpyrazine
3-methylbutanal
2-Acetyl-3-methylpyrazine
3,4-Dimethylcyclopentenolone
4-vinyl-guaiacol
(E)-ß-damascenone
Cyclotene
Homofuraneol 4-Ethylguaiacol
3-Mercapto- 3-methylbutylformate Vanillin
2-Furanmethanol o-Guaiacol
Furaneol

γ-Butyrolactone

1H-Pyrrole-2-carboxaldehyde
CU CA

PC1: 56.2% explained

Fig. 2. PCA biplot of volatile data set (n = 103 volatiles). Each point in the scores plot (left) represents a biological replicate from one of the maturity × variety
combinations. The loadings plot (right) shows only the 32 volatiles reported in the literature to have concentrations 10-fold or greater than their odor thresholds are
included (i.e. OAV > 10; see Supplementary Table S1).

reported in Fig. S3 of the supplementary data, and the PCA results are that a total of 103 volatiles were eventually identified and quantitated.
shown in Fig. S4 of the supplementary data. The first principal com- When possible, identification was based on comparison of chromato-
ponent (PC1) explained the majority of the variance (82.3%) and graphic and mass spectral properties with authentic standards, with the
showed good segregation for Stage 2 (more positive PC1) through Stage remaining volatiles identified based on comparison to literature values
7 (more negative PC1). As expected, the more positive PC1 values (less and library matches. A list of volatiles and their observed retention
mature fruit, e.g. Stages 2–4)) were associated with more positive L* indices and normalized peak areas are reported in Table S1, along with
values (lighter colored cherries), more positive b* values (higher literature retention indices and (where available) literature odor ac-
yellow-to-blue ratio), and more negative a* values (lower red-to-green tivity values (Grosch, 1998; Piccino et al., 2014; Steen et al., 2017).
ratio). PC2 was most strongly associated with the a* value, and the most A representative HS-SPME-GC–MS chromatogram of a coffee
green and immature fruit, Stage 1, was separated from other stages by sample, including the four non-native deuterated internal standards, is
having a more negative value for PC2. Good discrimination was ob- shown in Supplementary Fig. S1. One concern associated with HS-SPME
served among most stages (Fig. 2), although small overlaps in con- volatile profiling studies is that SPME extraction efficiency can show
fidence intervals could be observed between Stage 3 and 4 cherries, and strong compound × volatile interactions, such that the use of surrogate
between Stage 5 and 6 cherries. In summary, the CIE L*a*b* data in- standard leads to large quantitative errors (Gómez-Cortés, Brenna, &
dicated that visual sorting gave comparable results to what would be Sacks, 2012). However, in our current study, we observed little varia-
expected from sorting based on colorimetry. Further statistical analysis tion in the average signal for each of the four deuterated standards
of the CIE L*a*b* data as has been described elsewhere, e.g. for defect among the different samples – standard deviations ranged from 0.25 to
recognition in green and roast coffees by PCA analysis (Craig, Franca, & 2.9% for hexanal-d12, 0.1–6.1% for isobutyric acid-d7, 0–2.1% for
Oliveira, 2012) was not explored, since our primary interest was in IBMP-d3, and 0.01–0.6% for naphthalene-d8 (Supplementary Fig. S2).
evaluating volatile and sensory differences within the visual sorting This result suggested that matrix effects were consistent across samples,
categories. and that semi-quantitative comparison of normalized peak areas was
appropriate. Based on this, we chose to use the average peak area of the
four standards within each run as the internal standard peak area for
3.2. Volatile analyses
normalizing analyte signal areas.
Most of the significant differences observed among stages were
A target list of volatiles was initially developed based on about 60
between Stage 1 (immature green) and later maturity stages – of the
compounds previously reported in volatile analyses of coffee
103 volatiles characterized, 90 differed significantly between Stage 1
(Bressanello et al., 2017; Grosch, 1998). HS-SPME-GC–MS analyses of
and the other stages (p < 0.05, ANOVA; Table S1). By comparison,
freshly ground coffee samples were then performed, and additional
differences in volatiles for the other six stages were limited – for
target compounds added based on manual inspection of the data, such

140
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

example, over 50% of volatiles showed no significant variation among lower in roast C. robusta coffees. In summary, lower quality beans are
Stages 2–7 for each cultivar, and only 11 volatiles were classified into characterized by higher concentrations of phenols, higher concentra-
more than three groups. These results agree with previous analytical tions of N-heterocycles, and lower concentrations of CHO degradation
studies in the cherry maturity which focused on extremes of cherry products, whether those beans are produced from immature fruit (as in
maturity – total or partially green cherries as compared to almost ripe, the current study) or from C. robusta.
ripe, or overripe cherries (i.e., stage 1–2 vs. stage 3–7) (Smrke, The major N-heterocycles (pyrazines) reportedly have less desirable
Kroslakova, Gloess, & Yeretzian, 2015). For most volatiles that differed (“earthy”) aroma characteristics as compared to key CHO degradation
significantly, the difference between the minimum and maximum products (“sweet, caramel”) (Grosch, 1998), and their relative con-
concentrations were typically between 2- and 4-fold centrations could directly affect the sensory qualities of immature vs.
(1 < [log2(Max) − log2(Min) < 2; Table S1). The maximum range mature beans. Alternatively, differences in N-heterocycle (or phenol) to
observed for a compound was for pyrrole, which was 5.2-fold higher in CHO degradation product ratios may be correlated with other factors
Stage 1 Catimor than in Stage 3 Catimor (Table S1). responsible for differences in sensory characteristics. Interestingly, we
Because of the large number of volatiles, and because many of these saw no significant differences in concentrations of S-compounds be-
compounds were expected to arise from related pathways, we chose to tween Stage 1 coffee vs. the remaining stages (Fig. 3, p > 0.05). No-
evaluate the data by PCA (Fig. 2). Similar approaches have been used tably, we observed no significant effect of maturity stage on 2-furfur-
for evaluation of volatile profile data sets from other foodstuffs or fra- ylthiol (FFT) formation (Table S1). Furfurylthiol is an important
grance e.g. classifying citrus oil fractions or products from thermal contributor to the “roast” aroma of coffee (Grosch, 1998; Ribeiro et al.,
oxidation of edible oils (Cossignani, Giua, Simonetti, & Blasi, 2014; 2012), and oxidative losses of FFT during storage of roast beans or
Zhang et al., 2017). To simplify the loadings plot, only the 32 volatiles following brewing is correlated with a loss of coffee quality (Dulsat-
previously reported to have OAV > 10 are depicted (Supplementary Serra et al., 2016). However, based on our current work, FFT formation
Table S1). PC1 discriminated Stage 1 from all other maturity stages does not appear to be related to coffee maturity stage. We also observed
(Stage 2–7). The loadings plot indicates that the higher PC1 scores were no clear correlation between maturity stage and Strecker aldehydes (2-
associated with higher concentrations of N-heterocycles, particularly methylbutanal, 3-methylbutanal, phenylacetaldehyde; Fig. 3).
pyrazines (e.g. trimethylpyrazine and 2-methyl-5-propylpyrazine) and PC2 of the PCA explained only a small amount of variance (11%;
phenols (e.g. 4-ethylguaiacol, o-guaiacol) Conversely, compounds as- Fig. 2). PC2 appeared to discriminate the varieties, with most Catimor
sociated with CHO degradation (Belitz & Grosch, 2013) (e.g. cyclotene, samples having positive PC2 scores and most Caturra samples having
furaneol, 5-methylfurfural) have negative loadings on PC1. negative PC2 scores (Fig. 2). However, compounds with high PC2
To evaluate if certain categories of volatiles were significantly de- loadings (e.g. 1H-pyrrole-2-carboxaldehyde, γ-butyrolactone) did not
pendent on either maturity or cultivar, we classified volatiles into one differ significantly across varieties (p > 0.05, Table S1). A few com-
of six categories: Phenol derivatives, N-heterocycles, S-compounds, pounds, such as 2,3-dimethyl-5-ethylpyrazine, did differ significantly,
CHO degradation products, Strecker aldehydes, and Other/Unknown. but differences between varieties for this and other volatiles were small
Peak areas for each compound were then normalized to the maximum (|log2(Caturra Max) − log2(Catimor Max)| < 0.8 for all volatiles).
value for that compound across all stages and cultivars (max Larger differences in roast coffee volatiles among varieties have been
value = 100%). Average normalized concentrations for each compound reported (Toledo, Pezza, Pezza, & Toci, 2016), although the results are
classes (excluding Other/Unknown) are shown as a function of maturity often confounded because the varieties were grown in different regions,
stage and cultivar in Fig. 3. as opposed to the same location as in this study.
For both the Caturra and Catimor varieties, we observed that N-
heterocycles in Stages 2–7 were significantly lower than their corre- 3.3. Sensory analysis
sponding Stage 1 values; in the case of Caturra, N-heterocycle con-
centrations averaged only 60–75% of the Stage 1 maximum. Similarly, An unexpected outcome of the volatile analysis study was that the
volatile phenols in Stages 2–7 were 65–85% of their Stage 1 maxima. volatile profiles of all stages excluding the immature Stage 1 samples
Higher concentrations of N-heterocycles (pyrazines, pyrroles) have (Stages 2–7; from partially ripe to overripe) showed minimal differ-
been previously reported in coffees produced from immature vs. mature ences – in particular, we had expected to be able to distinguish Stage 2
beans (Agresti et al., 2008). Similarly, higher pyrazines and volatile from other stages, as previous work reported that yellow-green cherries
phenols, along with lower furans, are reported in lower quality C. ro- produced coffee with diminished sensory quality (Marín et al., 2004).
busta coffees as compared to C. arabica (Grosch, 1998). Pyrazines and To evaluate if the poor discrimination in volatile characteristics ex-
phenols have also been reported to be elevated in lower quality cups of tended to coffee sensory characteristics, coffee samples were evaluated
the cultivar “Bourbon Pointu” (Piccino et al., 2014). Pyrazines are by certified panelists using the widely-employed SCAA cupping pro-
formed from amino acids via Maillard reaction pathways, and the tocol. This protocol evaluates ten attributes on a numerical scale ran-
higher pyrazine concentration formed by roasting immature beans may ging from 6 to 10 for each attribute, with a 0.25 point step size and
be related to their higher free amino acid content (Marín et al., 2004). lower scores possible for faulted coffees (Puerta-Quintero, 2000;
The explanation for higher concentrations of phenol derivatives in Specialty Coffee Association of America, 2015). The “Total” score is
immature beans is not clear. Volatile phenols are presumably formed by calculated as the sum of the ten attributes, and a Total score of 80 or
degradation of the major phenolic species in green coffee beans, the above is required to receive a “specialty coffee” designation.
chlorogenic acids; however, chlorogenic acid content changes negli- Mean and standard deviations for all quality attributes across stages
gibly in beans during maturation (Smrke et al., 2015). and cultivars are listed in Supplementary Table S2. Values for four re-
In contrast, CHO degradation products at Stage 1 were approxi- presentative attributes (Aroma, Flavor, Acidity, and Total) are plotted
mately 2-fold lower than the maximum concentrations observed at later as a function of maturity stage are shown in Fig. 4. A PCA of sensory
stages (Fig. 3). Among Stages 2–7, average concentrations of CHO de- data is presented in Supplementary Fig. S5. Most sensory attributes are
gradation products did not vary significantly. The higher concentra- highly covariant, and the first component represents almost 70% of the
tions of CHO degradation products may arise from the higher con- variance (see Fig. S5). For all attributes, mean values were significantly
centration of free sugars in mature coffee beans (Ribeiro et al., 2012). lower in Stage 1 (fully green) than all other stages for both the Caturra
Interestingly, lower concentrations of CHO degradation products have and Catimor cultivars (Tukey test, p < 0.05). Stage 1 coffees had
been reported in lower quality C. robusta coffees as compared to C. identical mean and standard deviations for each attribute, reported in
arabica (Semmelroch, Laskawy, Blank, & Grosch, 1995), e.g. furaneol Table S2 (5.42 ± 0.51). This statistically unlikely result was a con-
(4-hydroxy-2,5-dimethyl-3(2H)-furanone) was approximately 2-fold sequence of each panelist assigning either a 5 or 6 to every attribute for

141
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

100%
80%
60% CA CU
40%
Phenol derivatives (n=9)
20%
0

100% 100%
80% 80%
60% 60%
40% 40%
N-heterocycles (n=37) 20%
CHO degradation (n=36)
20%
0 0

100% 100%
80% 80%
60% 60%
40% 40%
S-compounds (n=7) Strecker (n=3)
20% 20%
0 0
1 2 3 4 5 6 7 1 2 3 4 5 6 7

Fig. 3. Average normalized concentrations (calculated as a percentage of maximum concentration for a given compound) for five compound classes: phenol deri-
vatives, CHO degradation, N-heterocycles, S-compounds, Strecker aldehydes. Each bar represents a compound class × maturity × variety combination, and error
bars represent standard deviations.

a given cup, as an indication that they thought the coffee was un- among attributes were observed among Stages 2–7 – for example,
acceptable and not worthy of nuanced judging. Flavor was rated as significantly lower in Caturra, Stage 3
Differences among other quality attributes from Stage 2 (yellow- (overall = 6.85) than in either Stage 6 or 7 (overall = 7.56 or 7.52
green, underripe) to Stage 7 (purple, overripe) for each cultivar were respectively). The Total quality score among Stages 2–7 ranged from 76
mostly non-significant (Fig. 4). Some small significant differences to 85. These differences were not significant and were also not

CA CU

8
Aroma

7
6
5
8
7
Flavor

6
5
8
7
Acidity

6
5

80
70
Total

60
50
1 2 3 4 5 6 7

Fig. 4. Average scores (n = 4 judges) for Aroma, Flavor, Acidity, and Total attributes. Error bars represent standard deviations.

142
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

CHO degrada on Other/Unknown Phenol deriva ves

1 2 3 4 5 6 7 N-heterocycles Strecker S-compounds

CU CA 2,3-Diethyl-5-methylpyrazine
Aroma
2,6-Diethylpyrazine 4-Ethylguaiacol

Trimethylpyrazine 2,3-Pentanedione
LV2 :9% explained (X) / 15.9 % (Y)

2,3-Dimethyl-5-ethylpyrazine

3,4-Dimethylcyclopentenolone
Linalool 2,3-Butanedione
Cyclotene

2-Acetyl-3-methylpyrazine
5-Methylfurfural
2,4,5-Trimethylthiazole

o-Guaiacol

LV1: 46.9% explained (X) / 74.2% (Y)

Fig. 5. First two latent variables from PLS regression based on volatiles (predictors) and Aroma (response) for each maturity × cultivar × field replicate combi-
nation. Coffee scores are shown at left, and volatile loadings for compounds with high VIP coefficients are shown at right.

significantly below the Total score (80) required for the SCAA specialty combined volatile and sensory data set. PLSR was chosen instead of
coffee designation. linear discriminant analysis (LDA) because our primary interest was in
Our observation that fully green coffee (Stage 1) received low modeling variation in sensory attributes among samples rather than
cupping scores was expected, as immature green beans are widely cited discriminating between samples. The RMSE and cross validated R2 were
as a contributor to low cup quality (Marín et al., 2004). However, our obtained for each sensory attribute are reported in Table S5. The esti-
observation that partially mature yellow-green fruit (Stage 2) presented mated coefficients for each compound, and VIP for all cupping attri-
cups of similar quality to red-ripe mature (Stage 5) was not expected butes are included in Table S4. An alternative multivariate modeling
based on previous work (Marín et al., 2004), although these results approach (LASSO) was also investigated (data not shown), but results
agree our observation that volatiles in Stages 2–7 showed only small were not readily interpretable, likely because of the high level of cov-
differences (Section 3.2). Our observation that coffees produced from ariance within the data sets.
cherries at Stage 3 (yellow-orange, i.e., pintón maduro) and Stage 7 Because volatile compounds are responsible for aroma, we were
(overripe, purple) agree with a previous report (Marín et al., 2004). particularly interested in the identifying volatiles which could predict
Finally, we evaluated if there was any evidence of greater inter- the Aroma attribute. A plot of the PLS regression based on volatiles
judge variance for sensory qualities among maturity stages (Table S3). (predictors) and Aroma (response) for each maturity × cultivar × field
Significantly higher inter-judge variance was observed for Total scores replicate combination is shown in Fig. 5. Similar to results from the PCA
for Stage 7 as compared to Stages 2–6 (Tukey test, p < 0.05), as well as of volatiles (Fig. 3), the first latent variable (LV1) separated Stage 1
for some sub-attributes, e.g. Balance. This indicates that there was less from Stages 2–7. The volatiles with high loadings in LV1 were also si-
agreement among judges regarding over-ripe cherries. Other than Stage milar to those observed for PC1 in the PCA plot; N-heterocycles (par-
1 and Stage 7, only a few other significant differences in variance for ticularly pyrazines) and phenols had negative LV1 coefficients and were
quality attributes among stages were observed, and no differences for associated with Stage 1, and CHO degradation products had positive
Overall and Total, although the strength of this observation was limited LV1 coefficients and were associated with Stages 2–7. The second latent
by the small number of judges. As an additional caveat, the SCAA variable (LV2) appeared to be associated with differences between the
cupping protocol is largely a hedonic evaluation, and thus our results two cultivars, with higher scores for Catimor, and lower scores for
indicate that there were minimal significant differences among Stages Caturra. LV2 was characterized by a positive contribution from several
2–7 in the judges’ liking of different attributes. These coffees could still pyrazines (2,3-diethyl-5-methylpyrazine) and 4-ethylguaiacol, and a
have differed in sensory descriptions, although this was not assessed. negative contribution from o-guaiacol. As with PCA, Stages 2–7 were
The use of descriptive analysis for coffee assessment has been described not well differentiated by the first two latent variables from PLS re-
(Chambers et al., 2016) and may be better suited to distinguishing gression.
among coffees from Stages 2–7.
4. Conclusion
3.4. Modeling of sensory qualities from volatile data
We have rejected out hypothesis that significant differences in vo-
To determine if variation in sensory qualities – particularly Aroma – latile profiles and cupping quality scores exist between Stages 5–6
could be correlated with specific volatiles, PLSR was performed on the (mature) roast coffees as compared to immature (1–4) and overripe (7)

143
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

maturity stages. Immature green cherries (Stage 1) produce roast cof- Craig, A. P., Franca, A. S., & Oliveira, L. S. (2012). Evaluation of the potential of FTIR and
fees with lower concentrations of carbohydrate degradation products chemometrics for separation between defective and non-defective coffees. Food
Chemistry, 132(3), 1368–1374. https://doi.org/10.1016/j.foodchem.2011.11.121.
and higher concentrations of N-heterocycles and phenols than other De Melo-Pereira, G. V., De Carvalho-Neto, D. P., Magalhães-Júnior, A. I., Vásquez, Z. S.,
maturity stages, as well as significantly lower sensory quality scores. Medeiros, A. B. P., Vandenberghe, L. P. S., & Soccol, C. R. (2019). Exploring the
However, cherries harvested at later stages, from yellow-green to impacts of postharvest processing on the aroma formation of coffee beans – A review.
Food Chemistry, 272, 441–452. https://doi.org/10.1016/j.foodchem.2018.08.061.
overripe purple (Stage 2 to Stage 7), produce coffees that are not readily Defernez, M., Wren, E., Watson, A. D., Gunning, Y., Colquhoun, I. J., Le Gall, G., ...
distinguished by volatile profiles or sensory evaluations, either through Kemsley, E. K. (2017). Low-field 1 H NMR spectroscopy for distinguishing between
comparison of individual metrics or by PCA. This work suggests that arabica and robusta ground roast coffees. Food Chemistry, 216, 106–113. https://doi.
org/10.1016/j.foodchem.2016.08.028.
only fully immature green cherries produce roast coffee with distinctive Dulsat-Serra, N., Quintanilla-Casas, B., & Vichi, S. (2016). Volatile thiols in coffee: A
volatile profiles and inferior sensory attributes for specialty coffee review on their formation, degradation, assessment and influence on coffee sensory
production. Potentially, specialty coffee growers could reduce their quality. Food Research International, 89(1), 982–988. https://doi.org/10.1016/j.
foodres.2016.02.008.
number of selective harvests (and thus production costs) while still
Federación Nacional de Cafeteros. (2010). Sustainability that matters 1927-2010. FNC –
achieving specialty coffee designation. However, before changing ex- Management reports. Retrieved from https://www.federaciondecafeteros.org/static/
isting practices, additional validation with more sites and varieties files/informe_sostenibilidad_eng.pdf.
would be prudent, as our study considered only two C. arabica cultivars, Flament, I., & Bessière-Thomas, Y. C. N.-M. L. T. F. (2002). Coffee flavor chemistry.
Chichester; New York: Wiley.
and future work will be necessary to confirm that our conclusions can Fold, N., & Pritchard, B. (2011). Cross-continental agro-food chains: Structures, actors and
be generalized to all coffee cultivars. Furthermore, the current study dynamics in the global food system. Routledge.
focused differences in sensory qualities using the SCAA cupping pro- Franca, A. S., & Oliveira, L. S. (2008). Chemistry of defective coffee beans. Food Chemistry
Research Developments (pp. 105–138). (1st Edition). Belo Horizonte: Nova Science.
tocol, which is primarily a hedonic evaluation. Future studies using Gómez-Cortés, P., Brenna, J. T., & Sacks, G. L. (2012). Production of isotopically labeled
descriptive analyses to characterize sensory differences among later standards from a uniformly labeled precursor for quantitative volatile metabolomic
maturity stages (and the potential for product differentiation) are of studies. Analytical Chemistry, 84(12), 5400–5406. https://doi.org/10.1021/
ac300933d.
interest. Grosch, W. (1998). Flavour of coffee. A review. Nahrung-Food, 42(6), 344–350. https://
doi.org/10.1002/(sici)1521-3803(199812)42:06<344::aid-food344>3.0.co;2-v.
Acknowledgments Inocencio-Monteiro, P., Sousa-Santos, J., Alvarenga-Brizola, V. R., Pasini-Deolindo, C. T.,
Koot, A., Boerrigter-Eenling, R., ... Granato, D. (2018). Comparison between proton
transfer reaction mass spectrometry and near infrared spectroscopy for the authen-
Physical analysis of coffee and sample collection was performed by tication of Brazilian coffee: A preliminary chemometric study. Food Control, 91,
the members of Cesurcafé group, at Universidad Surcolombiana. This 276–283. https://doi.org/10.1016/j.foodcont.2018.04.009.
Marín, Arcila, J., Montoya, E. C. C., & Oliveros, C. E. E. (2004). Relación entre el estado de
project was funded by the Colombian Science Council, Colciencias
madurez del fruto del café y las características de beneficio rendimiento y calidad de
[grant number 617-20] (The science, technology, and development la bebida. Cenicafé, 54(4), 297–315.
management department). Martínez-Gil, A. M., del Alamo-Sanza, M., Gutiérrez-Gamboa, G., Moreno-Simunovic, Y.,
& Nevares, I. (2018). Volatile composition and sensory characteristics of Carménère
wines macerating with Colombian (Quercus humboldtii) oak chips compared to
Conflict of interest wines macerated with American (Q. alba) and European (Q. petraea) oak chips. Food
Chemistry, 266, 90–100. https://doi.org/10.1016/J.FOODCHEM.2018.05.123.
The authors have no conflict of interest to declare. Monakhova, Y. B., Ruge, W., Kuballa, T., Ilse, M., Winkelmann, O., Diehl, B., ...
Lachenmeier, D. W. (2015). Rapid approach to identify the presence of Arabica and
Robusta species in coffee using 1 H NMR spectroscopy. Food Chemistry, 182, 178–187.
Appendix A. Supplementary data https://doi.org/10.1016/j.foodchem.2015.02.132.
Mu, B., Zhu, Y., Lv, H.-P., Yan, H., Peng, Q.-H., & Lin, Z. (2018). The enantiomeric dis-
tributions of volatile constituents in different tea cultivars. Food Chemistry, 265,
Supplementary data associated with this article can be found, in the 329–336. https://doi.org/10.1016/J.FOODCHEM.2018.05.094.
online version, at https://doi.org/10.1016/j.foodchem.2018.08.127. National Institute of Standards and Technology. (2018). NIST Chemistry WebBook, SRD
69. Retrieved March 8, 2018, from http://webbook.nist.gov/chemistry/#Search.
Oliveros-Tascón, C. E., Sanz-Uribe, J. R. R., Montoya-Restrepo, E. C., & Moreno-Cárdenas,
References
E. L. (2009). Dispositivo hidráulico de bajo impacto ambiental para limpieza y
clasificación del café en cereza. Cenicafé, 60(3), 229–238.
Agresti, P. D. C. M., Franca, A. S., Oliveira, L. S., & Augusti, R. (2008). Analytical, Piccino, S., Boulanger, R., Descroix, F., Shum, A., & Sing, C. (2014). Aromatic composi-
Nutritional and Clinical Methods Discrimination between defective and non-defective tion and potent odorants of the “specialty coffee” brew “Bourbon Pointu” correlated
Brazilian coffee beans by their volatile profile. Food Chemistry, 106(1), 787–796. to its three trade classifications. Food Research International, 61, 264–271. https://doi.
https://doi.org/10.1016/j.foodchem.2007.06.019. org/10.1016/j.foodres.2013.07.034.
Arcila, J., Farfán, F., Moreno, A., Salazar, L., & Hincapié, E. (2007). Sistemas de producción Poltronieri, P., & Rossi, F. (2016). Challenges in specialty coffee processing and quality
de café en Colombia. FNC-Cenicafé (1st Edition). Chinchina; Caldas: CENICAFE. assurance. Challenges, 7(2), 19. https://doi.org/10.3390/challe7020019.
Belitz, H. D., & Grosch, W. (2013). Food chemistry (2nd ed.). Springer. Puerta-Quintero, G. I. (2000). Influencia de los granos de café cosechados verdes en la
Braga, S. C. G. N., Oliveira, L. F., Hashimoto, J. C., Gama, M. R., Efraim, P., Poppi, R. J., & calidad física y organoléptica de la bebida. Cenicafé, 51(2), 136–150.
Augusto, F. (2018). Study of volatile profile in cocoa nibs, cocoa liquor and chocolate Ribeiro, J. S., Augusto, F., Salva, T. J. G., & Ferreira, M. M. C. (2012). Prediction models
on production process using GC × GC-QMS. Microchemical Journal, 141, 353–361. for Arabica coffee beverage quality based on aroma analyses and chemometrics.
https://doi.org/10.1016/J.MICROC.2018.05.042. Talanta, 101, 253–260. https://doi.org/10.1016/j.talanta.2012.09.022.
Bressanello, D., Liberto, E., Cordero, C., Rubiolo, P., Pellegrino, G., Ruosi, M. R., & Bicchi, Semmelroch, P., Laskawy, G., Blank, I., & Grosch, W. (1995). Determination of potent
C. (2017). Coffee aroma: Chemometric comparison of the chemical information odourants in roasted coffee by stable isotope dilution assays. Flavour and Fragrance
provided by three different samplings combined with GC–MS to describe the sensory Journal, 10(1), 1–7. https://doi.org/10.1002/ffj.2730100102.
properties in cup. Food Chemistry, 214, 218–226. https://doi.org/10.1016/j. Sergey, K. (2017). mdatools: Multivariate Data Analysis for Chemometrics. Retrieved
foodchem.2016.07.088. from https://cran.r-project.org/web/packages/mdatools/mdatools.pdf.
Castle, T. J. (2016). Colombia’s arc over the specialty coffee landscape (& Vice Versa) | Silva, S. D., de Queiroz, D. M., Pinto, F. D. C., & Santos, N. T. (2014). Coffee quality and its
The Specialty Coffee Chronicle. Retrieved October 14, 2016, from http://www.scaa. relationship with Brix degree and colorimetric information of coffee cherries.
org/chronicle/2016/04/27/colombias-arc-over-the-specialty-coffee-landscape-vice- Precision Agriculture, 15(5), 543–554. https://doi.org/10.1007/s11119-014-9352-y.
versa/. Smrke, S., Kroslakova, I., Gloess, A. N., & Yeretzian, C. (2015). Differentiation of degrees
Chambers, E., Sanchez, K., Phan, U. X. T., Miller, R., Civille, G. V., & Di Donfrancesco, B. of ripeness of Catuai and Tipica green coffee by chromatographical and statistical
(2016). Development of a “living” lexicon for descriptive sensory analysis of brewed techniques. Food Chemistry, 174, 637–642. https://doi.org/https://doi.org/10.1016/
coffee. Journal of Sensory Studies, 31(6), 465–480. https://doi.org/10.1111/joss. j.foodchem.2014.11.060.
12237. Specialty Coffee Association of America. (2015). Cupping standards. Retrieved January 1,
Cossignani, L., Giua, L., Simonetti, M. S., & Blasi, F. (2014). Volatile compounds as in- 2015, from http://www.scaa.org/?page=resources&d=cupping-standards.
dicators of conjugated and unconjugated linoleic acid thermal oxidation. European Steen, I., Waehrens, S. S., Petersen, M. A., Münchow, M., & Bredie, W. L. P. (2017).
Journal of Lipid Science and Technology, 116(4), 407–412. https://doi.org/10.1002/ Influence of serving temperature on flavour perception and release of Bourbon
ejlt.201300205. Caturra coffee. Food Chemistry, 219, 61–68. https://doi.org/10.1016/j.foodchem.
Cossignani, L., Montesano, D., Simonetti, M. S., & Blasi, F. (2016). Authentication of 2016.09.113.
Coffea arabica according to Triacylglycerol Stereospecific Composition. Journal of Toledo, P., Pezza, L., Pezza, H. R., & Toci, A. T. (2016). Relationship between the different
Analytical Methods in Chemistry, 2016, 1–7. https://doi.org/10.1155/2016/7482620. aspects related to coffee quality and their volatile compounds. Comprehensive Reviews

144
S. Velásquez et al. Food Chemistry 274 (2019) 137–145

in Food Science and Food Safety, 15(4), 705–719. https://doi.org/10.1111/1541-4337. Wehrens, R. (2011). Chemometrics with R: Multivariate data analysis in the natural sciences
12205. and life sciences. Springer.
Urbani, E., Blasi, F., Chiesi, C., Maurizi, A., & Cossignani, L. (2015). Characterization of Zhang, H., Xie, Y., Liu, C., Chen, S., Hu, S., Xie, Z., ... Xu, J. (2017). Comprehensive
volatile fraction of saffron from Central Italy (Cascia, Umbria). International Journal of comparative analysis of volatile compounds in citrus fruits of different species. Food
Food Properties, 18(10), 2223–2230. https://doi.org/10.1080/10942912.2014. Chemistry, 230, 2316–2326. https://doi.org/10.1016/j.foodchem.2017.03.040.
968787.

145