Three recently published rapid methods, namely the NIRS-litter-bag technique13, the raw foliar pH14,15,16 and foliar NIR-Tomoscopy17, have been used together, in a holistic model, for a symbiotic corn yield production4. The aim of this study has been to investigate the quality of symbiotic corn. First, the spectral NIR signature of the single kernels was obtained and elaborated in order to assess whether this method could fingerprint the source of the corn (Control vs. Symbiotic), as induced by a biofertilizer in different corn cultivars. Moreover, some commercial qualities of the integer grain were determined, by means of a bench NIRS instrument, and elaborated as univariate. Finally, by chaining the results of this second part to the previous results, a correlation has emerged which has highlighted the relationships among the phenotype variables that affect the quantity of the corn as well as their consequences on the variation in the quality traits.

Twenty-six pairwise comparisons, namely Symbiotic inoculated (S) vs. Control non-inoculated (C) were obtained from a total of 44 plots in 4 centers, with four cultivars and six AM types, over a period of 2 years.

In 2018, three centers collaborated in the set up and the realization of the calibration experiments (Table 1). The inoculation was performed using a Micosat F ® bio-fertilizer, as coating (1 kg ha^-1) or granulate (10 kg^-1). In 2019 validation experiments were performed using a Micosat F ® bio-fertilizer and four other AFM types as coatings (1 kg ha^-1) at a University Center in soil where cation exchange capacity appears to be average, while the phosphorus supply assimilable is low, as is that of exchangeable potassium (Table 2).

1Biota composition: finely ground cultivated Sorghum sudanensis roots, containing spores and ifae of Funneliformiscoronatus GO01 and GU53, F. caledonium GM24, F. intraradices GB67 and GG32, F.mosseaeGP11 and GC11, F. viscosum GC41; saprotrophic fungi: Streptomyces spp. ST60, Streptomyces spp. SB14, Streptomyces spp. SA51, Beauveria spp. BB48, Trichoderma viride, Trichoderma harzianum TH01, Trichoderma atroviride TA28, Trichoderma spp.; rhizosphere bacteria: Bacillus subtilis BA41, Pseudomonas fluorescens PN53, Pseudomonas spp. PT65 and Pochoniachlamidosporia, in a relative percentage of 40% crude inoculum and 21.6% bacteria and saprotrophic fungi; 2Rhizophagus intraradicesCA502;3Gigaspora rosea NY328A;4Sclerocystis sinuosa MD126; 5Claroideoglomus claroideum ON393.

An NIR-SCiO mod. 2 (Consumer Physics inc, Herzliya, Tel Aviv, Israel) (Figure 1), was used to scan about twenty-five kernels selected from corn samples from each plot. Each grain was put in the center of a reverberant pill sample holder, with the embryo facing downward (Figure 2), and then scanned from 740 to 1070 nm (331 points at a 2 nm interval).

Chemometric elaborations were carried out, by means of “The SCiO-Lab” software, which operates by means of AKA (As Known As) matrices and provides a percentage recognition of the matrix cells. The method used for the calibrative classification was the Random Forest algorithm. The percentages of fingerprinting for the Kernel Control (K_CC) and the Kernel Symbiotic (K_SS) classes were analysed according to the free MedCalc online software. The interaction Cultivar* AM type was tested, by means of a Friedman test for paired comparisons (StatBox V6.5, Grimmersoft, Paris). The predictivity of the models was established, by means of a leave-one-out validation procedure, within the four cultivars and considering the different AM types.

A wide range of Yield responses to the Symbiotic treatments was observed during the experiments developed for the present paper. Is it possible that the NIR spectra of the kernels produced in plants treated with a BF can contain some information on the degree of the Yield result? For this purpose, the PLS procedure of the SCiO-Lab software was applied to the collection, for over 1338 spectra.

Duplicate spectra of the whole grain plots were obtained using a DA1650 NIRS-FOSS^TM instrument, provided with a calibration model to predict five components on a DM basis: Ash, Protein, Starch, Fiber, Fat and the NIRS undetermined Residual.

Kernel quality data from the Control and paired Symbiotic plots were analysed using the Friedman test for paired comparisons.

In order to explore the relationships between the quality traits and the quantity variations in yield, the regressions of the main variables, namely the Kernel spectral fingerprint (K_CC, K_SS and their average K_ ), were computed on the results of measurements previously obtained from the phenotype of the plants. The independent variables were expressed in terms of a plot effect-size, computed as the Ln of the response ratio (S/C), where the mean of the inoculated treatment (S) was divided by the mean of the non-inoculated control (C), namely d_Y = Ln(S/C). This mode of expression is arithmetically equivalent to calculating the relative prevalence of S over C (d_Y= S/C -1). Only the selected co-variables that were significant at a Pearson test were included in the regression study of the main variables.

The average reflectance spectra of the Symbiotic kernels were close to the Control ones (Figure 3). The brilliant values obtained in the reverberant chamber should be noted, especially compared to the average NIR spectra of the leaves.

The spectral fingerprinting of the Kernel collection was able to perfectly discriminate the four cultivars (Table 3).

The fingerprinting of the Control and Symbiotic types produced the same value of 73% for the whole collection of kernel spectra (Table 4 and Table 2), but only when Micosat F was used. The fingerprint values otherwise increased for the S (88%), but decreased to 55% for the C, when the other five types of AM were included in the Symbiotic treatment: these two values are biased by the different numbers in the two classes. As far as the cultivars are concerned, higher values characterized DK4316, which was only tested in one field, with respect to P1547, which was tested in 16 plots. When different AMs were combined with two different cultivars, the results showed variable fingerprint responses to the five types (Figure 4 and Figure 5), and an interaction cultivar * AM appeared for Friedman’s test (P 0.05, Figure 4 and Figure 5 combined).

In general, Kernel symbiotic models built for one cultivar cannot be extrapolated to a different cultivar. In fact, although the calibration models that excluded each cultivar were apparently satisfactory (Table 5, left), the leave-one-out validation (Table 5, right) showed high inaccuracies, especially in the K_CC values, which were systematically underestimated in the DK4316 and P1547 cultivars, while the K_SS values over-performed in the previous cultivars, but were inefficient (three cases) or underestimated (one case).

Only the fat content was slightly increased in the Symbiotic kernel (P 0.05; Table 6). On average, the starch was increased (P 0.12) together with the fiber, while the protein, ash and undetermined residuals were decreased. The within constituent correlations of the C and S values were all significant, except for the fiber and the residual undetermined by NIRS. The two spectral fingerprints of the kernels were highly correlated (r 0.83).

The between-trait printout (Table 5) showed significant correlation coefficients between the three main kernel variables and fourteen co-variables, out of a total of sixty, obtained from different sources in the growth phase 4.

The induced respiration (SIR) from the soil source variables was negatively correlated with the Control kernel fingerprint level (r -0.46) and with the yield results.

As far as the plant sources are concerned, all the foliar pH records (S, C, S/C) were negatively correlated with the NIRS C and S fingerprints of the kernels, clearly showing that a higher (protonic) energy charge in the leaves promoted the kernel diversification.

Among the kernel components, the fat and the starch contents were positively related to the high NIRS fingerprinting and characterization, while the protein and the fiber levels reduced the spectral originality of the C and the S kernels.

In short, the quantitative results, in terms of yield from the BF management were significantly and positively correlated with a higher diversification of the kernel, thereby permitting a higher fingerprinting in the Control groups (r 0.69) as well as in the Symbiotic one (r 0.55), where a higher productive level was available for the plants (Table 7).

A few relationships can be considered to highlight how the NIRS kernel fingerprint – the average of the two K_CC and K_SS values - depended on the main plant variables. The main regression was the one on Yield dressing (Figure 6).

By removing the two outliers and reversing the variables, a plausible parabolic model was obtained to estimate the yield size-effect from the NIRS fingerprint of the Control group, as shown in Figure 7 (R² 0.70).

As shown in Figure 8, despite the presence of two outliers, the variation in foliar pH, due to the increase in acidity of the leaves of the plants treated with BF, was responsible for a progressive increase in yield as well as for a higher spectral level of characterization.

Moreover, the foliar protein may be considered a sign of AM symbiosis or parasitism. A close parabolic relationship (Figure 9) linked the increase in foliar protein - from a BF treatment - to the NIRS fingerprint (R² 0.43) rather than to the Yield performances (R² 0.29), and a sparse response dispersed the points, with two presumed outliers in the negative outcome zone.

The calibration of the symbiotic response in yield was successful (Figure 10). The R² 0.56 did not represent the more instructive relationships that emerged from a quadrant analysis (Table 8).

In fact, a cut-off around the zero-crossing point showed that 90% of the kernels could be correctly classified, correlating the productive outcomes from the field (Table 8).