Dissecting yield associations through genotypic and phenotypic correlation and path analysis in upland rice

Introduction

In rice, grain yield is assembled through the joint expression of phenology, plant architecture, tillering, panicle productivity and grain attributes. Simple correlation is useful for identifying traits that vary with yield, but a large correlation can arise from indirect effects through a third character. Path coefficient analysis addresses this limitation by partitioning an observed correlation into direct and indirect components, an approach introduced by Wright (1921) [4] and adapted for crop selection by Dewey and Lu (1959) [1] . Previous rice studies have shown that the most useful selection traits can differ among populations and environments. In upland rice, grains or spikelets per panicle, tillering and panicle traits have often been associated with yield (Seyoum et al., 2012) [3] , while recent work has emphasized the value of combining correlation, path coefficients and genetic parameters before defining selection criteria (Faysal et al., 2022) [2] . The present paper focuses exclusively on character association and path analysis in the 40-genotype upland rice population evaluated at Parbhani. India. Forty upland rice genotypes from CRRUS, Hazaribagh, and ICAR-IRRI, Hyderabad, were grown in a randomized block design with three replications. Each entry was planted in two rows in a 4.5 m × 1.20 m plot at 30 cm × 5 cm spacing. Five competitive plants per genotype per replication were used to derive plot-level means. Genotypic and phenotypic covariances were used to calculate correlation coefficients. Significance was assessed in the thesis using a t-test. Path coefficients were obtained by solving the simultaneous relationship rᵢy = pᵢy + Σrᵢⱼpⱼy, where rᵢy is the correlation between trait i and yield, pᵢy is the direct effect of trait i, and rᵢⱼpⱼy represents the indirect contribution through trait j. The present article reports the complete correlation and direct-effect values supplied in the thesis summary. Complete numerical tables were available for 11 traits; filled and unfilled spikelets described in the methods chapter were not reconstructed because their and path coefficients were not supplied.

Materials and methods

Results and discussion

Correlations with grain yield

Genotypic correlations with grain yield were generally stronger than phenotypic correlations, especially for tiller number, panicle length and grain breadth. This pattern suggests that environmental variation attenuated the expression of several underlying genetic relationships. Tiller number had the strongest positive genotypic association with yield (r = 0.956), followed by total spikelets panicle-1 (0.846) and grain breadth (0.670). At the phenotypic level, total spikelets panicle-1 had the strongest positive association (0.818), followed by grain breadth (0.641) and tiller number (0.664). Days to 50% flowering had a small but significant negative association with grain yield at both levels (-0.203 genotypic; -0.202 phenotypic), suggesting that earlier flowering may be advantageous in this moisture-limited production context. Days to maturity, plant height, test weight and length/breadth ratio were not meaningfully associated with yield. However, correlation should not be interpreted as causation: the high inter-trait association between tillers and total spikelets panicle-1 (r = 0.963 at the genotypic level), and between tillers and panicle length (0.796), indicates substantial overlap among candidate selection traits.

Direct and indirect effects revealed by path analysis

Total spikelets panicle-1 had the largest positive direct effect on grain yield at both the genotypic (0.997) and phenotypic (0.584) levels. Grain breadth was the second most consistent direct contributor, with coefficients of 0.826 and 0.721. These direct effects agree with the positive correlations and support their use as primary selection criteria in this population. Panicle length also had a positive direct effect, although the phenotypic coefficient was small (0.025), indicating that its favourable association may operate mainly through panicle productivity.

The contrast between correlation and direct effect was particularly important for tiller number and grain length. Tiller number showed a very strong positive genotypic correlation with yield but a negative genotypic direct effect (-0.580); the thesis summary attributes its positive overall relationship chiefly to indirect effects through total spikelets panicle-1 (0.960), grain breadth (0.511) and panicle length (0.197). Grain length had positive correlations with yield but negative direct effects at both levels (-0.294 genotypic; -0.359 phenotypic), indicating that its apparent yield benefit was mediated through other grain and panicle traits.

Therefore, selection for tiller number or grain length in isolation could be misleading.

Test weight showed negligible correlations with yield and negative direct effects at both levels. Length/breadth ratio had positive direct effects but weak negative total correlations, demonstrating how opposing indirect effects can reverse the direction of a simple association. These findings are biologically and statistically plausible for a population in which grain number, grain size and panicle structure compete for assimilates. They also echo the broader rice literature, where the most reliable yield criteria vary with genetic background and test environment (Seyoum et al., 2012; Faysal et al., 2022) [2,3] .

Selection model for upland rice yield improvement

The combined evidence supports a hierarchical selection model. Total spikelets panicle-1 and grain breadth are the most dependable primary criteria because their correlations and direct effects are positive at both genetic and phenotypic levels. Panicle length is a complementary structural trait, while productive tillering should be retained only when it results in productive panicles and higher spikelet-bearing capacity. Earlier flowering may be useful for drought escape, but it must be combined with adequate grain filling and acceptable grain quality. Test weight should not be prioritized as a yield criterion in this population unless quality objectives independently justify it.

Conclusions

Correlation and path analysis revealed that grain yield in the evaluated upland rice population was most directly associated with total spikelets panicle-1 and grain breadth. Tiller number was strongly correlated with yield, but its negative genotypic direct effect showed that the association was primarily indirect through panicle productivity and grain breadth. Grain length showed the same general pattern of positive correlation but negative direct contribution. The breeding implication is to select for a coordinated phenotype rather than for a single correlated trait: high spikelet-bearing capacity, favourable grain breadth, adequate panicle length, and sufficient productive tillering under a suitable flowering duration.

The results are population-and environment-specific. The trial was conducted in one season and one location, and the source summary did not provide residual effects, standard errors of path coefficients or multicollinearity diagnostics. These should be reported in the final journal analysis, ideally using the original plot-level data and multi-

References

1. Dewey DR, Lu KH. A correlation and path-coefficient analysis of components of crested wheatgrass seed production. Agronomy Journal. 1959;51:515-518. DOI:10.2134/agronj1959.00021962005100090002x.
2. Faysal ASM, Ali L, Azam MG, Sarker U, Ercisli S, Golokhvast KS, Marc RA. Genetic variability, character association, and path coefficient analysis in transplant Aman rice genotypes. Plants. 2022;11(21):2952. DOI:10.3390/plants11212952.
3. Seyoum M, Alamerew S, Bantte K. Genetic variability, heritability, correlation coefficient and path analysis for yield and yield related traits in upland rice (Oryza sativa L.). Journal of Plant Sciences. 2012;7:13-22. DOI:10.3923/jps.2012.13.22.
4. Wright S. Correlation and causation. Journal of Agricultural Research. 1921;20:557-585.