Carbon isotope discrimination (Δ13C) in grain is
a potentially useful trait in breeding programs that aim to increase the yield
of wheat and other cereals. Near infrared reflectance spectroscopy (NIRS) is
used in routine assays to determine grain and flour quality. This study
assesses the ability of NIRS to predict Δ13C in
mature kernels of durum wheat. Plants were grown in north-west Syria as this
location provided 3 distinct Mediterranean trials that covered a wide range
for Δ13C values in grains (from about
12.9‰ to 17.6‰). We measured the spectral reflectance signature
between 1100 and 2500 nm in samples from the same flour used in the
conventional (i.e. mass spectrometry) determinations of
Δ13C. By using principal components regression
and partial least squares regression (PLSR), a model of the association
between conventional laboratory analysis and these spectra was produced.
Global regressions, which included samples from all 3 trials, and local
models, which used samples from only one trial, were built and then validated
with sample sets not included in calibration procedures. In global models,
strong significant correlations (P < 0.001) were
found between NIRS-predicted Δ13C and measured
Δ13C values. PLSR gave r
2 values of 0.86 and 0.82 for calibration and validation
sets, respectively. Although less strongly correlated, all local models
selected for a subset of samples with significantly higher
Δ13C values. Local models also performed well
when selecting samples from the other 2 trials. The advantages and possible
limitations of NIRS are further discussed.