The experiment was conducted in a greenhouse of the Research Centre for Vegetable and Ornamental Crops, Council of Agricultural Research and Economics, in Pescia, Italy, during the spring season 2024. The substrate was a mixture of 60/40 v/v dark peat/blond peat (TecnoGrowth^® Professional, Tercomposti S.p.A., Calvisano, BS, Italy) with an average N concentration of 12.4 g kg^-1, determined by a FlashSmart™ NC Soil elemental analyzer (Thermo Fisher Scientific, Waltham, MA, USA). Broccoli (Brassica oleracea L. var. italica cv. Naxos F1) seedlings were planted in seed trays the 10^th of April 2024 and then grown in a greenhouse under different N fertilization treatments at 0, 4.5, 9.0, 15.0 and 22.5 mM N (N0, N1, N2, N3 -standard dosage according to local nursery grower practices, N4 respectively) maintaining a constant N-NO₃/N-NH₄ ratio of 2.75 and using a standard nutritive solution for leafy vegetables fertigation regarding to phosphorous, potassium and other micro and meso-elements ( Figure 1 ). No other treatment was provided to the analyzed plants. Each treatment consisted of 4 replicates for a total of 84 seedlings per treatment. The measurement session occurred 35 d after the sowing date (DAS). Leaf collection occurred within the 10:00 – 11:00 am time interval. Water was supplied through fertigation as a function of microclimate conditions (roughly at least once a day), maintaining constant substrate moisture. Microclimate conditions were monitored every 5 minutes by a portable weather station (Zentra ZL6 datalogger, Meter Group., Pullman, WA, USA) equipped with an air temperature (T) and relative humidity sensors (RH) (VP-3; Decagon Em50; Decagon Devices Inc., Pullman, WA, USA) and a pyranometer for solar radiation measurement (Pyr sensor, Meter Group., Pullman, WA, USA). Temperature was in the range of 5.8 – 41.6°C (medium T in the period, 22°C), medium RH amounting to 60% and light intensity amounting as medium value to 246 W m^-2 (maximum value recorded 1,007.4 W m^-2 close to the end date of the experiment).

Each leaf was analyzed with the Dualex Scientific+ (Force-A, Orsay, France) and the Multiplex (Force-A, Orsay, France) optical sensors. The leaves were measured under ambient lighting at 25°C and within 15 min of sampling. Dx chlorophyll content (DxChl) is determined by the ratio between the transmittance at 850 nm (reference signal that is not absorbed by chlorophyll) and at 710 nm. Epidermal flavonoids (Flav) are calculated by the logarithm of the ratio between chlorophyll fluorescence excited at 650 nm and 375 nm (which is attenuated by the epidermal compounds). The ratio of DxChl over DxFlav represents the nitrogen balance index (NBI), which has been shown to account for N changes in the leaf. In our experiment, two Dx readings per leaf from the upper leaf blade and lateral to the midrib of both adaxial and abaxial sides of the leaf were carried out and averaged, and the relevant DxChl, DxFlav, and DxNBI values were used for further consideration. Multiplex chlorophyll content (MxChl) is calculated from the SFR_G index accounting from the ratio FRF_G/RF_G (between far-red and red fluorescence signals under a green excitation), while the flavonoid content (MxFlav) is calculated as the logarithm of the FRF_R/FRF_UV ratio (between far-red fluorescent signals obtained under a red and a UV excitations). The MxNBI is achieved by the FRF_UV/RF_G ratio (between far-red fluorescence signal under a UV excitation and the red fluorescence signal under a green excitation). A single reading from the adaxial side of a whole leaf was collected and used for further consideration. A total of 10 leaves per replica were analyzed.

Raman spectra within the 400 ÷ 2500 cm^-1 range were collected with a portable spectrometer (miniRaman Dual, LightNovo, Denmark) using a 785 nm laser source coupled with a laptop for parameter setting and data visualization under the proprietary software. The spectrometer was equipped with a f=30 mm lens (NA=0.05) producing a 50 mm wide spot and to which it was screwed a custom-made aluminum perforated spacer to allow collecting the maximum signal at the leaf plane surface and maintain a 90° geometry between the irradiation direction and the leaf surface. An acquisition time of 500 ms with 10 accumulations and 97.32 mW of output power was employed. The leaves, previously measured by Dx and Mx, were analyzed by Raman spectroscopy under the same conditions. Similar to the Dx measurements, a single spectrum was acquired per leaf from the upper leaf blade and lateral to the midrib of the adaxial side. A total of 10 measurements (one per leaf) per replica for a total of 40 measurements per treatment, were collected, averaged, and used for further evaluation. This approach helps to mitigate spatial variability and improves the representativeness of the spectral data. Raman spectra were preprocessed before analysis according to Ref (Matteini et al., 2025). Raman spectra of kaempferol and quercetin and their glucoside derivatives (Extrasynthese, Genay, France) were collected under a LabRAM HR Evolution spectrometer (Horiba, Lille, France) working in back-scattering geometry and equipped with an excitation laser source at 785 nm (de Angelis et al., 2025), focused through a ×10 objective from 0.1 M stock solutions in dimethyl sulphoxide.

After optical measurements, each leaf was weighed to determine the fresh weight (FW) and dry weight (DW), by oven drying until reaching constant weight at 70°C, and then milled for subsequent destructive analyses. Before drying, the same leaves were used for the measurement of leaf area (LA) by a leaf area meter (WinDIAS Image Analysis System, Delta-T Devices, Cambridge, UK) and for the determination of the Leaf Mass per Area (LMA), namely the DW on LA ratio. Chlorophyll a (Chl a) and b (Chl b) and total phenols concentrations were determined by spectrophotometer analysis (Evolution™ 300 UV–Vis Spectrophotometer, Thermo Fisher Scientific Inc., Waltham, MA, USA) after fresh leaf samples extraction in methanol (99%), keeping samples in the dark at -20°C for 48 h, renewing the solution after 24 h, and reading the absorbance at 665.2, 652.4, 470.0 and 320.0 nm, respectively. Chl a, Chl b, and total chlorophyll (Chl_a+b) were then calculated as described by (Lichtenthaler and Buschmann, 2001), while total phenols (Total_Phenols) were calculated according to (Maggini et al., 2018). Total Kjeldahl Nitrogen (TKN) concentration was measured by Kjeldahl method after phospho-sulphuric acid digestion (Massa et al., 2016). Nitrate nitrogen (N-NO₃) leaf concentration was determined by the spectrophotometric assay using the salicylic-sulfuric acid method described by (Cataldo et al., 1975) after 1 h water extraction of dry leaf tissue.

Statistical analysis was carried out with OriginPro 2024 version 10.1.0.178 software (OriginLab Corporation, Northampton, MA). Mean data values were analyzed using ANOVA and compared with the all-pairwise multiple comparison Tukey’s test. P values of <0.05 were considered statistically significant.

The Raman spectra of broccoli seedling leaves exhibit a distinctive signature characteristic of green leafy vegetables when excited under a 785 nm light source ( Figure 2 ). This signature is primarily defined by the vibrational modes of carotenoids, chlorophyll, and polyphenols. Carotenoids contribute bands at 1003, 1156, 1188, 1218 and 1526 cm^-1 (Bhosale et al., 2004; Baranski et al., 2005; Ibarrondo et al., 2016; Park et al., 2023). Chlorophyll (a and b) is represented by major bands at 747, 917, 992, 1188, 1220, and 1328 cm^-1, with additional shoulders at 1141 and 1555 cm^-1 (Kagan and Mccreery, 1995; Ishihara and Takahashi, 2023). Polyphenols display bands mainly in the ranges of 550 – 650, 1245 – 1321, 1370 – 1420, 1586–1670 cm^-1 (Jurasekova et al., 2006, Jurasekova et al., 2014; Gamsjaeger et al., 2011; Pompeu et al., 2018; Bock et al., 2021).

The heatmap in Figure 3 illustrates the relationships between Raman spectral frequencies and various parameters, including biometric metrics (FW, LA, and LMA), concentration of plant compounds (Chl_a+b, Total_Phenols, TKN, and N-NO₃) obtained through destructive analyses, and indices derived from Dx and Mx measurements (DxChl, DxChl/LMA, DxFlav, DxNBI, MxSFR_G, MxFlav, MxNBI_G) on the same samples. A notable observation is the close correlation pattern between FW and LA on one side, and the concentrations of extracted chlorophyll (Chl_a+b), TKN, and N-NO₃ on the other. Higher N levels are often positively correlated with increased total chlorophyll content, as N is a fundamental component of chlorophyll molecules and essential for photosynthesis (Wen et al., 2019). This increase in chlorophyll typically enhances photosynthetic efficiency, leading to greater leaf expansion (larger LA) and higher biomass accumulation (FW). Optimal N levels generally promote balanced growth, while insufficient or excessive nitrogen can negatively impact chlorophyll synthesis, leaf area, and fresh weight (Boussadia et al., 2010). Accordingly, we observe an increase in LA and FW with higher fertilization levels up to N2, followed by a slight decrease due to excessive N supply ( Figure 4 ). While N-NO₃ levels continue to rise from N0 to N4, TKN reaches a saturation stage starting at N2 ( Figure 5 ), negatively affecting biomass performance ( Figure 4 ), and chlorophyll content (Chl_a+b plateaus for N > 2, Figure 6 ). LMA clearly distinguished N0 from the other doses ( Supplementary Figure S1 ), showing a significantly higher value at N0 and stabilizing at a lower, statistically similar level for N1–N4. This trend aligns with the observed increases in LA and FW up to N2, as nitrogen availability promotes leaf expansion and biomass accumulation, leading to thinner leaves and a lower LMA. Beyond N2, excessive N negatively impacts biomass production without further affecting LMA.

Chlorophyll variations across fertilization levels are well-tracked by DxChl/LMA as well as by the 1328 cm^-¹ Raman chlorophyll band (I₁₃₂₈, Figure 6 ). Notably, both DxChl/LMA and I₁₃₂₈ show improved differentiation among N levels, surpassing even the spectrophotometric analysis. We note that the correlation between Raman and DxChl significantly improves when DxChl is normalized by LMA (R² = 0.31 for DxChl vs. I₁₃₂₈; R² = 0.84 for DxChl/LMA vs. I₁₃₂₈, Supplementary Figure S2 ). A similar improvement of DxChl index is observed in the correlation with total chlorophyll concentration (R² = 0.10 for DxChl vs. Chl_a+b; R² = 0.60 for DxChl/LMA vs. Chl_a+b), highlighting a dependence on leaf morphology (e.g., thickness or area) with DxChl/LMA offering a clearer representation of chlorophyll concentration per unit leaf mass, as previously pointed out (Bracke et al., 2019a; Tuccio et al., 2022). It is noteworthy that other Raman bands associated with chlorophyll exhibit similar predictive behavior, albeit with lower R² values (R² < 0.80, Supplementary Figure S2 ). Further taking into consideration medium-high covariance between I₁₃₂₈ and extracted Chl_a+b (R² = 0.64) and the almost absence of other competing bands in that Raman region, the 1328 cm^-¹ frequency, which is ubiquitarian in the Raman spectrum of chlorophyll (Koyama et al., 1986; Sato et al., 1995), is here suggested as a valuable indicator for predicting chlorophyll concentration in green leaves using Raman spectroscopy, while also offering a non-destructive alternative to LMA-based assessments.

High positive correlations ( Figure 3 ) are observed between several Raman spectral regions and the reference Dx and Mx indices for flavonoids (DxFlav, MxFlav). Specifically, these regions are: 838–848 cm^-¹, 1010–1030 cm^-¹, 1092–1096 cm^-¹, 1244–1248 cm^-¹, 1272–1276 cm^-¹, 1364–1422 cm^-¹, 1474–1502 cm^-¹, 1580–1612 cm^-¹, and 1645–1674 cm^-¹. A similar correlation pattern, albeit weaker, is observed with the total phenols concentration obtained through leaf destructive analysis (Total_Phenols, Figure 3 ). Dx and Mx indices are limited to detecting epidermal flavonoids, particularly those attenuating chlorophyll fluorescence due to their optical absorption in the UV-A region (specifically at 375 nm, as implemented in Dx and Mx optical sensors) (Diago et al., 2016). Broccoli are rich in flavonols such as quercetin and kaempferol and their 3O-glycoside derivatives (Vallejo et al., 2004; Cartea et al., 2011). The Raman spectrum of quercetin, kaempferol, and their 3O-glucosilates are shown in Figure 7 , along with evidenced bands involved in the correlation with Dx and Mx. Interestingly, some main bands, such as those at ~1310 cm^-¹, lack a positive correlation with Dx and Mx indices ( Figure 3 ). This can be explained by their proximity with strong chlorophyll and carotenoid bands ( Figure 2 ), which diminish or reverse their correlation with DxFlav and MxFlav in Figure 3 . Conversely, the spectrophotometric measurement (i.e., Total_Phenols) represents a broad estimate of the overall phenolic content. This explicates its additional positive covariance with certain Raman frequencies (i.e., at 950 and 1684 cm^-1 associated with phenolic compounds beyond epidermal flavonoids, such as epicatechin, gallic acid, and p-coumaric acid, which were also found in mature broccoli (Rahman et al., 2022)). These compounds typically exhibit an optical absorption below 370 nm (Csepregi et al., 2013), making them undetectable by Dx and Mx sensors. A decrease in phenolic content (Total_Phenols) is observed with increasing N fertilization dosages ( Figure 8 ), which also reflects the DxFlav and MxFlav trend as well as that of selected Raman bands, such as the 1020, 1246, and 1584 cm^-¹. Notably, DxFlav, MxFlav, and the intensities of these Raman bands align closely in differentiating among N doses, whereas variations in the 1606 cm^-¹ intensity more closely mirror those of total phenols. This distinction arises because the first three frequencies correspond to Raman bands of quercetin and kaempferol and their glucoside derivatives ( Figure 7 ), explaining their predictive value for the flavonoid content (mainly flavonols) with a strong correlation with DxFlav (R² 0.65 ⨫ 0.80), and moderately high correlation with MxFlav (R² 0.53 ⨫ 0.65). In contrast, the 1606 cm^-¹ band (R² = 0.52 vs. DxFlav and MxFlav) lies in a spectral region less specific to flavonoids but includes Raman features of the main part of polyphenols (Pompeu et al., 2018; Espina et al., 2022; Krysa et al., 2022). Consequently, this band (ascribed to the aromatic C=C stretching) is more ubiquitous across various polyphenol species and, therefore, appears well-suited for deriving information about the total phenolic content of the sample, as suggested elsewhere (Pompeu et al., 2018).

The inverse relationship between flavonoid content and N availability in plants has been highlighted by several studies. Flavonoid content increases under N (nitrate and ammonium) deficiency because the plant reallocates resources from protein synthesis and growth to secondary metabolite production, enhancing defense mechanisms against environmental stress (Prinsi et al., 2020; Ma et al., 2023). This relationship has been explored not only to monitor plant responses to varying fertilization levels (Tremblay et al., 2007, Tremblay et al., 2010; Agati et al., 2016), as also evidenced in this study ( Figure 8 ), but also as a potential tool to quantify N content (Cartelat et al., 2005; Bracke et al., 2019b; Tuccio et al., 2022). The correlation coefficients between DxFlav and MxFlav vs. N content (N-NO₃ and TKN) in broccoli as listed in Table 1 reveal a moderate potential (R² ~0.50), which is here surpassed using the DxChl index or the NBI values, reaching excellent performances in the case of N-NO₃ (R² = 0.81 and 0.71 for DxChl and DxNBI, respectively). The chlorophyll content is often used as an indicator of available nitrogen levels (Padilla et al., 2018) due to the almost linear relationship between these parameters ( Figure 3 ), as previously mentioned. Another chance is to take advantage of NBI reflecting the ratio between chlorophyll and flavonoids, which can sometimes improve N estimates, especially under conditions of high nitrogen levels (Shi et al., 2024).

Several Raman frequencies exhibit a correlation of R² >0.65 with N-NO₃ and/or TKN (see Figure 9 , Table 1 ). Specifically, bands at 820, 1065 and 1336 cm^-1 effectively correlate with both N-NO₃ and TKN. These bands are linked to cellulose and structural cell wall components, representing the C-C and C-O stretching, the C–O–C stretching, and the C-O stretching or C-O-H bending modes of polysaccharides, respectively (Chylińska et al., 2014; Szymańska-Chargot et al., 2016; Bock et al., 2021; Payne and Kurouski, 2021). Overall, their association with N content underscores the dependence of these species on N availability, which regulates plant growth and cell wall biosynthesis (Ogden et al., 2018). The TKN demonstrates a strong correlation with bands at 692, 1004, 1224 cm^-1 (R² >0.65) and a moderate correlation (R² = 0.55-0.60) with bands at 746, 916, 1160, 1530 cm^-1 of chlorophyll and carotenoids, as well with the more generic 1446 cm^-1 band (CH₂ bending). This can be explained by ammonium’s direct role in metabolic pathways that enhance and regulate photosynthetic pigment synthesis and the production of CH-rich structural and storage compounds in photosystems (Chenard et al., 2005; Wen et al., 2019) (in fact TKN is limiting for chlorophyll, see Figure 5 and Figure 6 ). Conversely, the 938 and 1268 cm^-1 bands correlate preferentially with N-NO₃ and may tentatively reflect C-H bending modes in aliphatic chains of lipids and fatty acids, or C-O stretching vibrations in phenolic compounds (Pompeu et al., 2018; Saletnik et al., 2022). The accumulation of these compounds can align with nitrate availability, which regulates lipid biosynthesis and phenolic metabolism (Cartea et al., 2011; Pereira et al., 2024).

Overall, the above considerations highlight the potential of Raman bands as effective tools for probing available nitrogen in broccoli, with performance comparable to or exceeding that of portable meters ( Table 1 ). Moreover, we have evidence that a strategic selection of band intensities could provide specific insights into N-NO₃ or TKN levels, offering valuable guidance for the optimized management of fertilizer doses. For example, the 1446 cm^-1 band can represent an excellent reference to follow the variation in TKN, while the 1268 cm^-1 band is willing to resemble variations in N-NO₃ with enough accuracy, as emerges comparing Figure 10 with Figure 5 .