The potential to breed for genetic legacy effects in sustainable farming systems

16 Pulse crops provide protein-rich food, critical disease breaks in cereal rotations, and contribute 17 to soil fertility through symbiotic nitrogen fixation . However, crop improvement programs 18 typically only focus on individual crop performance rather than system -wide benefits. We 19 propose a novel approach: breeding pulses and cereals to optimize rotational effects and 20 enhance farming system sustainability in addition to crop-specific traits . In this proof -of-21 concept study, we evaluated how genetically diverse mungbeans influence subsequent wheat 22 performance. A mini-core panel of 309 mungbean accessions was grown at a nitrogen depleted 23 site, followed by a single wheat cultivar sown in the same plot locations. Remarkably, wheat 24 yield ranged from 2.51 to 3.49 t ha-1 and displayed moderate broad-sense heritability (H2: 0.42) 25 as a trait of the proceeding mungbean genotype. This variation demonstrates untapped genetic 26 potential for breeding mungbean varieties that enhance subsequent wheat performance. 27 Integrated analyses of mungbean -wheat traits, soil properties, and microbiome composition 28 revealed mechanisms underlying these rotational relationships. Haplotype mapping identified 29 genomic regions in mungbean associated with wheat yield and quality. To explore how legacy 30 traits could be leveraged in breeding programs, we conducted forward genetic simulations 31 using empirically derived haplotype effects to compare strategies targeting mungbean yield, 32 wheat yield, or both simultaneously. T hese simulations reveal both challenges and 33 opportunities in breeding crops collectively for rotational benefits , offering a pathway toward 34 more sustainable and profitable farming systems with reduced input requirements. 35

36 Modern crop breeding programs typically aim to optimize a single species, focusing on traits 37 that enhance yield and quality, irrespective of the preceding or subsequent crops in rotation 38 (Bennett et al., 2012). This approach achieved remarkable success during the Green 39 Revolution, transforming wheat and rice production through high -yielding cultivars that fed 40 billions globally, though often requiring increased fertilizer and irrigation inputs (Evenson and 41 Gollin, 2003). Importantly, breeding remains effective even in low -input systems, as 42 demonstrated by continued genetic gains across diverse production environments (V oss-Fels et 43 al., 2019). However, the global population will approach 10 billion by 2050 and agricultural 44 systems have become increasingly reliant on external inputs (nitrogen fertilizer use has 45 increased over 40% globally since 1990; Li et al., 2025), and is expected to increase further to 46 feed the growing population despite the growing pressure to reduce them due to environmental 47 and health concerns. Therefore this calls for innovative breeding strategies to accelerate genetic 48 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 3 gain and improve sustainability (Hickey et al., 2019). The single-crop breeding paradigm may 49 be missing critical genetic opportunities when the farming system is holistically considered. 50 Emerging evidence suggests that genetic variation within crop species creates measurable 51 effects on subsequent crops in rotation (Rose et al., 2010; Ellouze et al., 2013), yet the genetic 52 basis of these "legacy effects" remains largely unexplored. Legacy effects occur when the 53 genetic characteristics of one crop influence the performance of subsequent crops through 54 persistent changes to soil conditions that carry over between growing seasons (Somenahally et 55 al., 2018; Lilley and Kirkegaard, 2016). While the benefits of crop rotations at the cross-species 56 level are well established , for example legumes fixing nitrogen for cereals (e.g. Zhao et al., 57 2022), crop rotations to break disease cycles (e.g. Kirkegaard et al., 1997 ), and diverse root 58 systems improving soil structure (e.g. Yu et al., 2025), the genetic variation within species that 59 could enhance these rotational benefits through targeted breeding remains unknown and is 60 untapped in current breeding programs . Lilley and Kirkegaard (2016) highlight this concept, 61 demonstrating that wheat cultivars with deeper, more extensive root systems extracted 62 additional subsoil water but left soils drier at subsequent crop sowing, reducing predicted yield 63 benefits when legacy effects were considered. Thus, selection for deeper root architecture 64 within wheat breeding germplasm could lead to higher yielding wheat cultivars, but they may 65

in unintended negative consequences for subsequent crops. Integrating legumes into crop 66 rotations enriches soil nitrogen through symbiotic nitrogen fixation, benefiting subsequent 67 crops with yield increases often exceeding 10% and with substantial reductions in fertilizer 68 requirements (Cernay et al., 2018; Zhao et al., 2022) . However, the extent to which genetic 69 variation within legume crop species could amplify these benefits remains unexplored. We 70 hypothesize that the genetic architecture of crops directly modulates the phenotypic expression 71 and yield of subsequent crops, creating heritable legacy effects that can be leveraged through 72 breeding to optimize integrated farming system productivity (Fig 1). To test this, we conducted 73 a proof-of-concept study using a wheat-based rotation system, and mungbean (Vigna radiata) 74 as a preceding legume crop. Mungbean contributes to the system by fixing nitrogen and 75 providing a disease break thereby enhancing the productivity and health of the subsequent 76 wheat crop (Ilyas et al., 2018; Foyer, Nguyen & Lam, 2019 ). We evaluated how ~300 77 genetically diverse mungbean accessions influenced performance of a subsequent wheat crop 78 rotation, combined with haplotype mapping to identify genomic regions influencing both direct 79 mungbean traits and legacy effects on wheat productivity and quality . Through genetic 80 simulations using real haplotype effects, we further evaluated the future potential of breeding 81 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 4 programs that systematically consider legacy traits alongside traditional performance metrics, 82 demonstrating that such integrated approaches are not only feasible but could improve genetic 83 gain and productivity for both crops through enhancing resource use efficiency for sustainable 84 agriculture. 85 86 Fig 1. Conceptual framework for integrating legacy traits into farming systems breeding. 87 Schematic overview comparing traditional breeding approaches with a farming systems 88 breeding strategy that could consider legacy trait selection. The traditional breeding program 89 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 5 model focuses on optimising crop performance in isolation, whereas the farming systems 90 breeding approach could also consider potential genetic effects of one crop on the next, 91 termed legacy traits, within rotational systems. In the farming systems approach, Crop 1 92 undergoes evaluation for both direct performance traits and legacy effects on Crop 2 (Stage 93 1a), followed by assessment of Crop 2's legacy trait expression (Stage 1b), with this dual 94 evaluation continuing through subsequent breeding stages. The lower panel illustrates the 95 potential flow-on impact of considering legacy traits in breeding pipelines, which could 96 enable the selection of genotypes that enhance overall system productivity through improved 97 water-use efficiency, nutrient-use efficiency, and beneficial soil properties, including 98 microbes. This framework represents a potential paradigm shift toward multi-crop, multi-99 season breeding that could support sustainable agricultural intensification while reducing 100 input requirements. 101

102 High and heritable legacy trait variation across mungbean genotypes 103 In the mungbean mini-core panel, substantial phenotypic variation with moderate to high broad 104 sense heritability was observed across all measured mungbean traits (Fig. 2a; Supplementary 105 Table 1 and 2). Mungbean yield varied 11-fold from 0.21 to 2.19 t ha⁻¹, with the check cultivar 106 Crystal producing 1.31 t ha⁻¹. Mungbean aboveground biomass at mid canopy development 107 showed 2.6 -fold variation ( 151.07 to 393 .31 g m⁻²), with Crystal reaching 280 .73 g m⁻². 108 Vegetation index values derived from UA V-based imaging (mungbean NDRE) at 90% black 109 pod exhibited 2.5-fold variation (0.18 to 0.45; Crystal: 0.23), indicating considerable genotypic 110 diversity in canopy vigour and greenness. All mungbean traits exhibited high heritability (H² = 111 0.63–0.84). 112 To quantify legacy effects after harvest of the mungbean genotypes, we measured wheat yield 113 for the single cultivar LongReach Raider. Wheat yield varied substantially by preceding 114 mungbean genotype (2.52 to 3.49 t ha⁻¹), with wheat grown after mungbean cultivar Crystal 115 yielding 3.06 t ha⁻¹. Wheat protein content varied from 7.36% to 8.48% (wheat crop grown on 116 preceding Crystal crop : 8.14%), while wheat NDRE ranged from 0.38 to 0.43 ( wheat crop 117 grown on preceding Crystal crop: 0.41). Critically, wheat legacy traits showed moderate to high 118 heritability (H² = 0.43 –0.65), indicating that a substantial portion of variation in wheat 119 performance was attributable to mungbean genotype. Several genotypes enhanced subsequent 120 wheat yield by up to 45% relative to the commercial mungbean cultivar Crystal, while others 121 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 6 reduced yield by as much as 50%, underscoring the substantial impact of mungbean genetic 122

and growth on rotational crop productivity (Fig. 2b). 123 124 Fig 2. Genotypic variation and heritability of mungbean traits and their legacy effects on 125 subsequent wheat yield. a) Distribution of phenotypic values for direct mungbean 126 physiological traits (yield, biomass, NDRE; purple) and wheat legacy traits (yield, grain 127 protein, NDRE; orange) measured across diverse mungbean genotypes. Broad-sense 128 heritability (H²) estimates for each trait are indicated in the top-right corner of each panel. b) 129 Yield performance of wheat plots, ordered by increasing yield (t ha-1), reflecting legacy effects 130 from preceding mungbean genotypes. Each bar represents the yield best linear unbiased 131 estimates and standard error of wheat planted on individual mungbean genotype. Red bars 132 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 7 denote wheat plots that have a significantly different legacy-induced yield response relative to 133 wheat grown on the check mungbean cultivar ‘Crystal’ (blue). 134 Interactions between direct and legacy traits reveal trade-offs 135 Relationships between mungbean and wheat legacy traits varied in directionality and strength 136 when evaluated across phenotypic, genetic, and haplotype levels (Fig. 3). At the phenotypic 137 level (Fig 3a), mungbean yield showed a moderate negative correlation with wheat yield ( r = 138 −0.39, p = <0.05), suggesting potential trade -offs between mun gbean productivity and 139 subsequent crop performance. In contrast, wheat NDRE measured at flowering and wheat yield 140 were positively correlated ( r = 0.75 , p = <0.05), indicating consistency among wheat 141 physiological performance metrics. Wheat NDRE at flowering also showed a significant 142 negative correlation with mungbean yield (r = -0.48, p = <0.05). 143 Genetic correlations revealed stronger associations, highlighting underlying genetic 144 relationships not evident at the phenotypic level (Fig 3b). Mungbean NDRE at 90% black pod 145 exhibited strong negative genetic correlations with all wheat traits (wheat yield r = −0.90; 146 wheat protein r = −0.95; wheat NDRE r = −0.92), suggesting that genetic architecture 147 underlying mungbean chlorophyll concentration and photosynthetic capacity captured by 148 NDRE late in the season may be antagonistic to wheat legacy performance. Conversely, 149 mungbean yield showed moderately low negative genetic correlation with mungbean biomass 150 (r = -0.32) and wheat yield ( r = -0.29), but a strong positive genetic correlation with wheat 151 protein ( r = 0.73) . Wheat protein displayed positive genetic relationships with both wheat 152 NDRE at flowering (r = 0.68), wheat biomass (r = 0.88) and wheat yield (r = 0.67), reinforcing 153 coherence among wheat performance traits, as well as a strong positive genetic correlation with 154 mungbean biomass (r = 0.88). 155 Haplotype-level variance correlations were uniformly positive across all mungbean and wheat 156 traits (r = 0.28–0.78, p = <0.05), indicating that shared genomic regions contribute to trait 157 variation across the mungbean–wheat system (Fig 3c). However, haplotype effect correlations 158 revealed more nuanced patterns, including directional shifts between traits (Fig 3d). Mungbean 159 yield showed strong negative haplotype effect corre lations with wheat yield ( r = −0.61, p = 160 <0.05) and wheat NDRE ( r = −0.77, p = <0.05), but positive associations with mungbean 161 biomass ( r = 0.60 , p = <0.05 ) and wheat protein ( r = 0.70 , p = <0.05). These contrasting 162 directions suggest trade -offs at the genome level and highlight key haplotypes that 163 differentially influence mungbean performance and legacy benefits to wheat. 164 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 8 165 166 Fig 3. Correlations among mungbean and wheat legacy traits at phenotypic, genetic, and 167 haplotype levels. A) Phenotypic correlations (r) among mungbean and wheat legacy traits, 168 based on BLUEs across all genotypes. B) Genetic correlations estimated from mixed-model 169 variance components, revealing shared genetic architecture between traits. C) Correlation of 170 haplotype-level variance across traits, reflecting the shared genomic regions contributing to 171 trait variation. D) Correlation of haplotype effects, quantifying the concordance in direction 172 and magnitude of haplotype influences across traits. The divergent colour key ranges from 1 173 (pink - positive correlation) through -1 (yellow - negative correlation). Non-significant 174 correlations are indicated in red text. 175 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 9 Following haplotype mapping, the top 12 haploblocks showing highest variance were selected 176 and reported as most important for each trait. A total of 33 unique haploblocks were identified 177 across all traits (Supplementary Table 3). Seventeen haploblocks were unique to single traits, 178 with eight showing high variance for wheat legacy effects but not for any measured mungbean 179 traits (Fig 4a and Supplementary Fig 1). Of the other 16 haploblocks all were found to have 180 high variance for at least two traits across both wheat and mungbean. Notably, one haploblock 181 (b000055, chr 1) was identified as high variance for all six traits across both species. Three 182 additional haploblocks, b000160 (chr 2), b000221 (chr 3), and b000509 (chr 5), showed high 183 variance for five of the six traits, suggesting that these genomic regions harbor loci with 184 pleiotropic effects within mungbean, and genetically mediated legacy effects influencing wheat 185 traits in rotation systems. 186 Among the top haploblocks, five were shared across mungbean yield and wheat yield (Fig. 4a), 187 including b000022 (chr 1), b000055 (chr 1), b000221 (chr 2), b000253 (chr 2), and b000509 188 (chr 5), while seven high -variance haploblocks did not overlap between crops. This pattern 189 indicates both shared and independent genetic control of direct versus legacy traits. 190 Functional genes associated with nitrogen fixation in legumes were positioned within five top 191 haploblocks identified in this study (Supplementary Table 4). Interestingly, co-located genes 192 were associated with broader root development processes, including gibberellin signa lling 193 (MtDELLA2), root meristem maintenance ( MtPLT1–4), and auxin transport ( MtPIN2–4), 194 suggesting that root architecture may be a key mechanism underlying legacy effects. 195 To dissect these relationships at higher resolution, we examined two haploblocks in detail: 196 b000055 (chr 1), the only haploblock with high variance for all six traits, and b000942 (chr 197 10), a haploblock with high variance for wheat yield but not ranking among the top 12 for 198 mungbean yield. At haploblock b000055, haplotypes with the strongest positive effects for 199 mungbean yield were associated with greater negative effects on wheat yield, consistent with 200 the overall negative genetic correlation (Fig. 4b). This pattern illustrates the genomic basis of 201 trade-offs between direct crop performance and legacy benefits. In contrast, block b000942 202 was not among the highest -variance regions for mungbean yield but nonetheless exhibited 203 haplotype variance comparable to that observed for wheat, where it ranked among the top 204 variance haploblocks. Importantly, this haploblock harbored haplotypes with positive effects 205 for both mungbean and wheat yield, as well as variants with neutral effects, indicating 206 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 10 opportunities to select haplotypes beneficial for system -wide productivity without 207 compromising individual crop performance. 208 209 Figure 4. Haploblocks associated with mungbean yield and legacy effects on wheat yield. 210 a) Circos plot displaying haploblock effect variance for yield in mungbean (inner ring) and 211 wheat (outer ring) following mungbean cultivation. Haploblocks are displayed on the 11 212 chromosomes of mungbean. The top 1% of high-variance haploblocks associated with yield 213 are highlighted in purple (mungbean) and orange (wheat). Red lines connect shared top 214 haploblocks with high haplotype effect variance for yield in both crops, indicating potential 215 loci underpinning legacy effects. b) Haplotype effects for mungbean and wheat yield at two 216 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 11 representative haploblocks: b000055 (chr 1) and b000942 (chr 10). Contrasting or aligned 217 effect directions across species highlight key genomic regions contributing to trade-offs or 218 co-benefits in system-level productivity. 219 Simulated selection strategies reveal challenges and opportunities for system-level yield 220 optimization 221 To evaluate the potential for simultaneous genetic improvement of mungbean and its legacy 222 effects on wheat, we simulated haplotype -based selection strategies using variable weighting 223 schemes for the two crops over 25 breeding cycles, using the real haplotype effects as a starting 224 point. Simulated progeny were generated by sampling chromosomes with recombination from 225 selected parents (e.g. Villiers et al, 2022). When haplotypes were selected solely to maximize 226 mungbean yield (100% mungbean weighting), predicted genomic estimated breeding values 227 for mungbean increased by 36.86% over the 25 cycles. However, this improvement came at a 228 cost to the subsequent crop, with legacy wheat yield showing a 7.71% reduction, highlighting 229 the presence of antagonistic haplotypes that benefit mungbean but negatively impact wheat 230 performance within this environment. 231 Adjusting the selection index to include moderate wheat emphasis (70:30 mungbean:wheat 232 weighting) maintained high genetic gain in mungbean (36.32%) while moderating the negative 233 legacy impact on wheat (reduced to 4.59% loss), demonstrating a more balanced trade -off 234 between direct and legacy benefits. Conversely, placing greater emphasis on wheat legacy 235 performance (30:70 weighting) yielded strong improvements in wheat legacy yield (13.44% 236 increase) with a modest decline in mungbean yield (22.56% reduction), indicating that 237 prioritizing legacy traits can substantially benefit the subsequent crop , albeit at a substantial 238 cost to the legume crop. 239 Notably, selection based on equal weighting (50:50) resulted in simultaneous yield gains for 240 both mungbean and wheat (19.45% and 7.55%, respectively), though the overall magnitude of 241 improvement was lower than in crop-specific optimization scenarios. This balanced approach 242 highlights the potential of system -wide genetic improvement through integrated breeding 243 strategies that consider both direct performance and legacy effects. The simulation results 244 reveal that while trade -offs exist between optimizing individual crop performance versus 245 rotational benefits, strategic selection can achieve meaningful gains across the entire cropping 246 system, potentially providing a pathway for sustainable intensification through breeding. 247 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 12 248 Fig 5. Simulated selection strategies reveal trade-offs and optimistion potential for 249 breeding mungbean with system-level legacy benefits. Simulations of genomic selection 250 under five haplotype stacking strategies, illustrating their effects on genomic estimated 251 breeding values (GEBVs) for mungbean yield (purple) and legacy wheat yield (orange). 252 From left to right: 100% selection for mungbean yield, 70% mungbean + 30% wheat 253 weighting, equal weighting (50:50), 30% mungbean + 70% wheat weighting, and 100% 254 selection for wheat legacy yield. 255 Analysis of soil properties reveals genetic variation for biological drivers of legacy effects 256 To explore potential traits contributing to legacy effects on subsequent crops, we sampled and 257 analysed soil from a subset of 10 genetically diverse mungbean genotypes after mungbean 258 harvest and prior to wheat sowing. Among the soil nutrient parameters assessed, nitrate (NO₃⁻) 259 levels differed significantly between genotypes ( p = 0.01), indicating genotype -dependent 260 variation in soil nitrogen dynamics. In contrast, no significant genotypic differences were 261 detected for other soil nutrient components, including ammonium, phosphorus, or 262 exchangeable cations (Suppleme ntary Table 5). To investgate possible differences in soil 263 microbiome due to mungbean genotype, soil volatile components were estimated for the same 264 subset (Supplementary Table 6). Out of a total of 143 volatile organic compounds (VOCs) 265 detected in the soil samples (data not shown), 25 VOCs past the filter criteria of which eleven 266 showed a significant genotypic effect (Supplementary Fig 2). Dehydrocamphor and 2 -267 methylisoborneol showed the largest significant difference among the el even VOCs with 268 Crystal 8A displaying a significant increase in both VOCs’ abundance compared to the bare 269 soil control (Supplementary Fig 3 and Supplementary Table 7) . Further, Crystal 8A was 270 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 13 associated with a significantly greater accumulation of dehydrocamphor compared to 271 AusTRC324159 and CPI62672 (Supplementary Fig 3 and Supplementary Table 7). A similar 272 trend follows for 2-pentylfuran, -fenchyl-methylester, and (5Z)-1-octan-5,7-dienol. However, 273 regarding -fenchyl-methylester M10452 (Onyx) also displays a significantly higher 274 accumulation compared to bare soil, while (5Z)-1-octa-5,7-dienol was detected at significantly 275 higher levels in AusTRC324159, Celera II AU, Moong, and Crystal 8A compared to bare soil 276 (Supplementary Fig 3 and Supplementary Table 7). Berken displays a distinct soil VOC profile 277 with significantly higher levels of 1,4 -cis-dimethylcyclooctane, 4,5-dimethylnonane, and 3,7-278 dimethyloctan-1-ol compared to multiple Mungbean genotypes and the bare soil control 279 (Supplementary Fig 3 and Supplementary Table 7). Interestingly, the potential for Mungbean 280 cultivars to regulate/control the soil microbiome diversity may be best seen in 3 -methoxy-281 3methylbutanol and styrene, where multiple mungbean cultivars are associated with 282 signifcantly reduced accumulation of those VOCs compared to bare soil (Supplementary Fig 3 283 and Supplementary Table 7). 284 285

286 While agricultural research has long demonstrated the value of crop rotations, this study reveals 287 for the first time the potential for rotational benefits to be enhanced through targeted plant 288 breeding by selecting genotypes that shape the environment in ways that influence the 289 performance of subsequent crops. To date, research has focused on species-level benefits (e.g., 290 Li et al., 2019; Lago-Olveira et al., 2023; Salvagiotti et al., 2024), such as nitrogen fixation or 291 disease suppression, rather than legacy variation between genotypes within the initial crop. Our 292

show that crop legacy effects, defined as the influence of one crop’s genotype on the 293 productivity of the next, are under genetic control. In our mungbean –wheat rotation system, 294 substantial phenotypic variation in wheat NDRE, yield and protein content was observed as a 295 function of the preceding mungbean genotype. Importantly, these legacy traits exhibited 296 moderate to high heritability (H² = 0.43 –0.65), demonstrating that they are amenable to 297 selection. Remarkably, mungbean genotypes enhanced subsequent wheat yield by up to 45% 298 or reduced it by 50% relative to the commercial check variety , revealing substantial genetic 299 variation in legacy effects beyond environmental variation alone. These results highlight a real 300

in traditional breeding that focuses on maximising the performance of a single crop, 301 making an implicit assumption that any effect on subsequent crops is negligible. In contrast, 302 our results demonstrate that legacy effects vary across genotypes and can have a substantial 303 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 14 effect on following crops. This reveals an opportunity for a fundamental shift in breeding 304 philosophy, from exploiting fixed species -level rotational benefits to harnessing genetic 305 variation between species for enhanced system-wide productivity. 306 Another novel concept demonstrated in this study is that legacy effects can be genetically 307 mapped to the genome of the prior rotation crop. The legacy traits investigated in this case 308 study exhibited complex genetic architecture, where many agronomic traits could have shared 309 genetic control. We identified both shared and unique haploblocks influencing direct mungbean 310 traits and legacy effects on wheat, with 1 6 high-variance haploblocks showing overlapping 311 effects across both crops. Notably, the magnitude of haplotype effects was consistently greater 312 for mungbean traits than for legacy impacts on wheat, reflecting the range of other factors 313 contributing to wheat performance beyond legacy effects, including the underlying genetics of 314 the wheat cultivar and seasonal growing conditions. Several top haploblocks contained 315 functional genes associated with root development processes, including gibberellin signaling 316 (MtDELLA2), root meristem maintenance (MtPLT1–4), and auxin transport (MtPIN2–4), rather 317 than nodulation-specific pathways. This suggests that some of the legacy traits identified in this 318 study, may be mediated through root growth and architecture, which influence soil nutrient and 319 water status, and rhizosphere interactions, providing a potential mechanistic link between 320 mungbean genotype and subsequent wheat performance. 321 A trade-off was observed between mungbean yield and its legacy effects on wheat performance, 322 indicating that traits conferring high productivity in one crop may compromise the next crop 323 in rotation. Specifically, high -yielding mungbean genotypes were associated with reduced 324 wheat NDRE at flowering, which in turn was linked to lower wheat yield, consistent with a 325 source–sink limitation driven by environmental modification. This suggests that mungbean 326 genotypes may alter the soil environment through mechanisms such as differential nutrient 327 extraction, altered water availability, or rhizosphere -mediated shifts in the soil microbiome. 328 These changes may reduce resource availability for the subsequent crop, thereby impacting 329 biomass accumulation and final yield. Genomic analysis revealed negative correlations 330 between mungbean haplotype effects on mungbean and wheat traits, with particularly strong 331 trade-offs for yield. For instance, haploblock b000055 showed that haplotypes associated with 332 the highest mungbean yield also had the most detrimental effect on legacy wheat yield, whereas 333 b000942 included haplotypes with neutral or even positive effects on both crops. Importantly, 334 these ‘correlation breakers’ occurred mostly in the lower effect size range, suggesting that 335 system-level improvement may be best achieved through genome-wide strategies like weighted 336 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 15 genomic selection or haplotype stacking, rather than marker-assisted selection for large-effect 337 loci only. Although the precise biological drivers of these legacy effects remain unclear, our 338 targeted soil analyses indicate that there is measurable variation in both nutrient availability 339 and volatile organic compound profiles among a small set of diverse mungbean genotypes 340 (Supplementary Table 5 and 7). It is important to note th at while the concept of microbes 341 utilizing VOCs for interkingdom communication is fairly youn g (i.e. emerging in the last 15 342 years) soil volatilomic analysis hold s enormous potential as a cost-effective means to assess 343 soil health and microbial biodiversity (Weisskopf , et al., 2021). Therefore, while Mungbean 344 genotype-associated manipulations of the soil VOC profile were exposed, the ability to directly 345 link these VOCs to a specific microbe or plant requires additional work outside of the scope of 346 this study. That said, there is promise in the VOCs identified herein. For instance, 2-pentylfuran 347 is known to be emitted from various crops’ roots and acts as an attractant for certain insect pests 348 (la Forgia, et al., 2023); 3 -methoxy-3-methylbutanol is likely a product of a ligation reaction 349 with the common microbial primary metabolism VOC 3 -methylbutanol (Weisskopf et al., 350 2021); and 2-methylisoborneol and related terpenes (i.e. dehydrocamphor and -fenchyl-351 methylester) may indicate the presence of a specific type of bacteria (e.g. actinobacteria, 352 myxobacteria, and cyanobacteria) (Weisskopf et al., 2021; Martin-Sanchez, et al., 2019), This 353 highlights the potential for genotypic influence on the soil environment, potentially via root 354 architecture, nutrient and water acquisition, or microbial community structuring. In particular, 355 legacy effects mediated through persistent changes in the rhizosphere microbiome are a 356 promising avenue, as legume genotypes are known to establish distinct microbial communities 357 that can enhance nutrient cycling and disease suppression for subsequent crops (Yu et al., 2025; 358 Kirkegaard et al., 1997). Together, these findings underscore the need for further biological 359 characterisation of legacy mechanisms and demonstrate the feasibility of incorporating legacy 360 traits into breeding strategies to optimize productivity across crop rotations. 361 To evaluate the practical implications of these trade-offs, simulated haplotype-based selection 362 strategies revealed clear consequences when breeding solely for single -crop performance 363 within a rotation system. When selection was fully weighted toward mungbean yield, 364 substantial genetic gain in mungbean (36.86%) was observed but consistent reductions in wheat 365 legacy yield (7.71%) were noted , reflecting negative genetic correlations between traits 366 expressed in the two crops. Conversely, selection indices designed to balance genetic gain in 367 both crops moderated these trade -offs when using a 70:30 weighting or even eliminated any 368 tradeoffs in the case of a 50:50 weighting scheme. While absolute genetic gain in each crop 369 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 16 was reduced, overall system performance improved by minimizing negative legacy effects. 370 Most remarkably, equal weighting (50:50) achieved simultaneous yield gains for both 371 mungbean and wheat (19.45% and 7.55%, respectively), demonstrating that direct and legacy 372 traits under negative correlation can be successfully managed through weighted selection 373 indices. In practice, optimal weighting schemes could consider economic factors including 374 relative commodity prices, input costs such as nitrogen fertilizer, and farm-specific production 375 costs, as the economic value of legacy effects may vary substantially depending on market 376 conditions and production systems (Melton et al., 1979) . These simulations provide a 377 framework for integrating legacy traits into breeding pipelines, offering insights into how 378 genomic prediction can be optimized to support multi-season, multi-crop production. 379 The implications of breeding for legacy traits extend far beyond the immediate case study, 380 presenting a promising pathway to enhance agricultural sustainability by reducing input 381 dependency and improving system efficiency. By identifying legume genotypes that promote 382 beneficial downstream effects on cereals, breeders can select varieties that contribute residual 383 nitrogen, improve soil structure, modify soil microbiome, and maintain moisture availability, 384 thereby reducing resource inputs in subsequent crops. This approach can also extend beyond 385 sequential rotations to intercropping systems where genetic variation in root architecture, 386 nitrogen fixation efficiency, and rhizosphere interactions could enhance complementarity and 387 resource-use efficiency. Taking these benefits into consideration, the availability of modern 388 genomic approaches now enable predictive breeding decisions that can harness these potential 389 legacy effects for the first time (Hayes et al., 202 4), and the ability to map these effects to 390 defined genomic regions provides a foundation for incorporating them into genomic prediction 391 pipelines. However, implementing this approach at scale requires a paradigm shift that aligns 392 breeding priorities across multiple crops or within large breeding programs that manage 393 multiple crop portfolios simultaneously, as current breeding programs and funding models are 394 structured around single -crop optimization with economic incentives favouring direct trait 395 improvements over legacy effects. Transitioning to legacy trait breeding would require new 396 collaborative frameworks and revised intellectual property models that recognize contributions 397 to farming systems productivity, alongside logistically complex multi -location, multi-season 398 trials to validate genetic effect stability and build industry confidence. Legacy trait expression 399 is likely to exhibit significant genotype -by-environment interactions, particularly when soil 400 properties vary across target production environments, necessitating comprehensive testing to 401 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 17 establish breeding value stability. However, this investment in comprehensive testing will be 402 essential to build industry confidence and drive adoption of system-level breeding approaches. 403 As with any trait, studying and improving legacy effects requires sufficient genetic diversity. 404 Elite breeding materials may lack the variation necessary to detect and exploit these effects 405 compared to the mungbean diversity panel used in this research, necessitating the introgression 406 of required genetic variation into breeding populations. This case study also was conducted at 407 a single location with mungbean as the initial crop under nitrogen-deficient, rainfed conditions, 408 which does not fully capture the complexity of legacy dynamics in diverse commercial farming 409 systems. Nonetheless, yield from our mungbean mini-core experiments was similar to that of 410 previous research experiments under different management conditions and years 411 (Supplementary Fig 4). Furthermore, testing legacy effects across multiple wheat genetic 412 backgrounds would reveal whether these effects are consistent across cereal genotypes or if 413 specific legume-cereal genetic combinations maximize rotational benefits. Therefore, future 414 studies involving multi -location trials, diverse cropping systems, and multiple species with 415 appropriate genetic diversity are critical to explore the potential of this approach. Overall, the 416 ability to map legacy effects to defined genomic regions in the initial crop provides a 417 foundation for incorporating these traits into genomic prediction pipelines, enabling more 418 efficient breeding for sustainable farming system outcomes. This work represents a significant 419 advance toward breeding programs that aim to optimize cropping systems rather than a single 420 crop, offering a pathway for improved agricultural sustainability that harnesses genetic 421 diversity to reduce external inputs while maintaining productivity. 422

423 Field preparation and site characteristics 424 Prior to sowing the mungbean panel, soil nitrogen was depleted by sequentially sowing grain 425 sorghum and oat crop s to maturity before harvest (Supplementary Fig 5). This depletion 426 addressed the characteristically high nitrogen levels at research farms and created nitrogen -427 limiting conditions that would maximize expression of genotypic differences in nitrogen 428 fixation capacity and enhance legacy effects on the subsequent wheat crop. 429 Plant material and experimental design 430 A diverse mungbean panel of 309 genotypes was sown on 22nd January 2024 at the Department 431 of Primary Industries research station in Gatton, Australia (27.55 °S; 152.33 °E) 432 (Supplementary Fig 6). This panel consisted of the mungbean mini-core collection, which is a 433 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 18 subset of diverse mungbean germplasm collected from the World Vegetable Center gene bank 434 representing 19 countries across Asia, Australia, Africa and the Americas (Schafleitner et al., 435 2015). An additional 25 diverse genotypes and one black gram cultivar were included 436 (Supplementary Table 8). The trial used an unbalanced replicated design with the mini -core 437 collection replicated twice and the additional genotypes replicated four times. The experiment 438 was designed using a model-based row-column design incorporating genetic relatedness using 439 the R package 'od' (Cullis et al., 2020). 440 Mungbean genotypes were sown in 2-row 1.52 m × 4 m plots at a density of 25 plants m⁻². 441 Mungbean seeds were inoculated with Group I inoculum (EasyRhiz ™, New Edge Microbials 442 Pty. Ltd) at sowing to promote nitrogen fixation through rhizobia establishment in root nodules. 443 Starter fertilizer was applied as a basal application of 25 kg ha⁻¹. When the check cultivar 444 Crystal reached 90% black pod stage, the mungbean trial was desiccated with glyphosate and 445 machine harvested to determine grain yield. The remaining mungbean shoot and stubble was 446 left in the paddock and no tillage occurred prior to planting the wheat trial. 447 A single wheat cultivar (LongReach Raider) was sown in the same paddock 23 days after 448 harvesting the mungbean trial (23rd May 2024). The wheat was sown in f ive-row 1.52 m × 4 449 m plots and a seeding rate of 120 plants m⁻², where plot positions overlapped exactly with the 450 previous mungbean plots. Traits measured on the wheat plots were therefore mapped to the 451 corresponding mungbean plot information. Upon maturity, wheat was machine harvested for 452 grain yield determination. 453 Both trials were conducted under rainfed conditions with appropriate weed, insect and disease 454 control implemented as required. Daily meteorological data (minimum and maximum 455 temperature, rainfall) were obtained from an onsite weather station (Supplementary Fig 7). 456 Data Collection 457 A diverse range of agronomic and physiological traits, as well as soil properties, were captured 458 prior and during both mungbean and wheat trials on either the entire trial or on a subset of 459 diverse genotypes (Supplementary Fig 8). 460 Soil water and nutrient analyses 461 Prior to the commencement of the mungbean field trial, soil samples were taken from five soil 462 cores across the paddock which were split into two depths (0-20 and 20-40cm) (Supplementary 463 Table 9). These samples were subset into two samples, the first was weighed and dried in 105°C 464 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 19 oven to be weighed again to calculate gravotometic water content (%). A sample of the dried 465 soil was also used for analysis of extractable phosphorus (Colwell P), ammonium nitrogen 466 (NH₄⁺), and nitrate nitrogen (NO₃⁻) using standard protocols described by Rayment and Lyons 467 (2011). All analyses were performed by an accredited external laboratory using a SEAL AQ400 468 discrete analyser for colorimetric detection. All final concentrations were converted to mg/kg 469 dry soil for reporting. 470 Colwell P was extracted using a 0.5 M sodium bicarbonate solution at a 1:50 soil -to-solution 471 ratio. Samples were shaken for 16 hours, centrifuged, and filt ered. Phosphorus concentrations 472 in the extracts were determined colorimetrically using the ammonium molybdate–ascorbic acid 473 method, with potassium antimonyl tartrate added to control the reaction rate. Concentrations 474 measured in mg/L were converted to mg/kg based on extraction volume and soil mass. 475 Ammonium nitrogen (NH₄⁺) was extracted using a 2 M potassium chloride (KCl) solution at a 476 1:10 soil-to-solution ratio and mixed for 1 hour. The resulting extracts were centrifuged and 477 filtered before analysis. Ammonium was quantified colorimetrically throu gh its reaction with 478 sodium salicylate and nitroprusside in a mildly alkaline buffer, forming a coloured complex in 479 the presence of chlorine and detected at 650 nm. 480 Nitrate nitrogen (NO₃⁻) was extracted following the same procedure as ammonium using 2 M 481 KCl. Nitrate was reduced to nitrite via a cadmium reduction column, and the resulting nitrite 482 was measured colorimetrically using a reaction with sulfanilamide and NED, with absorbance 483 read at 540 nm. 484 Additional soil nutrient analyses were undertaken prior to harvesting the mungbean trial, from 485 two replicates of 10 diverse mungbean genotypes (Supplementary Fig 9; AGG324187 (2B), 486 AusTRC 324159, Berken, Celera II-AU, CHIH-CO, CPI62672, Crystal 8A, M10403, M10452 487 and Moong). In addition to P, NH₄⁺ and NH3- being captured, exchangeable bases (Ca, K, Mg, 488 Na) were also measured by using a 1 M ammonium chloride solution at a 1:10 soil-to-solution 489 ratio, shaken for 1 hour. The extracts were centrifuged, filtered, and analysed using a Thermo 490 iCAP PRO XP ICP -OES instrument. Concentrations (mg/L) were converted to mg/kg using 491 extraction volume and soil mass. These values were further converted to milliequivalents 492 (meq/100g) by dividing the mg/kg concentration by the atomic weight of each ion, and then 493 adjusting for charge: divided by 10 for monovalent ions (Na⁺, K⁺) and by 5 for divalent ions 494 (Ca²⁺, Mg²⁺). Cation exchange capacity (CEC) was estimated from the sum of these 495 exchangeable cations. 496 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 20 Soil microbiome analyses 497 Using a subset of soil samples collected from the ten genotypes described above, soil volatile 498 analyses was undertaken. 499 HS-SPME-GC/MS analysis 500 The optimized HS-SPME-GC/MS methodology in Rivers et al. (2019) was modified as follows 501 for the volatilomic analysis of soil samples with the Orbitrap Exploris GC 240 Mass 502 Spectrometer fitted with a TriPlus RSH Multipurpose (i.e. Liquid and Headspace Injection, 503 SPME Arrow, and ITEX -DHS) Extended Rail (Thermo Fisher Scientific, Massachusetts, 504 USA). Splitless mode on the was employed for application of the VOCs via a 1.0mm straight 505 no-wool liner to a TRACE 1610 series Gas Chromatograph (Thermo Fisher Scientific). Soil 506 samples, QC samples, laboratory blanks, and standards were block randomised throughout 507 each batch. Each sample was place on 2ml of 20% NaCl solution and incubated at 80oC for 40 508 min with continuous agitation at 500 rpm for enhance VOC volatilization and homogenization. 509 VOC extraction (adsorption) was carried out by exposing a broad specificity SPME Arrow 510 fibre (1.10mm DVB/C-WR/PDMS 110m x 20mm, Thermo Fisher Scientific) to soil sample 511 head space for 40 min with continuous agitation at 1000 rpm. Extracted VOCs underwent 512 desorption from the SPME Arrow on to the GC column (TraceGOLD TG-5SilMS GC Column, 513 Thermo Fisher Scientific) via the inlet at 250oC for 10 min. Between samples the SPME Arrow 514 fibre undergoes a pre- and post-conditioning at 250oC for an additional 10 min to prevent cross 515 sample contamination. Chromatographic separation of the soil VOCs was achieved employing 516 ultra-pure Helium as a carrier gas and via a 38.5 min oven temperature program: 40 oC for 2 517 mins, increasing to 150oC at a ramp rate of 5oC /min and held for 2 min, and then increased to 518 320oC at 15oC/min and held for 1 min. Lastly, the Orbitrap Exploris 240 MS was operated in 519 full MS scan acquisition using electron ionisation at 70eV from m/z 40-500 with a lag time of 520 5min, and a Orbitrap resolution of 60,000. Temperatures were 320oC for the Auxillary transfer 521 line and 250oC for the ion source. 522 Volatile Data Analysis 523 Thermo Compound Discoverer (ver. 3.3.2.31) was used for deconvolution and data analysis. 524 The NIST/EPA/NIH Mass spectral libraries (ver. 2023) were used for mass spectral matching 525 and peak annotations. Using an n-alkanes (C7 -C40) standard mix Kovats non -isothermal 526 retention indices (RIs) were calculated and compared to reference semistandard non-polar RIs 527 from the NIST and Pubchem databases when available. Peak intensity values were collected 528 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 21 and corrected based on the internal standard nitrobenzene, background peak detection in blank 529 samples, and sample weight. Subsequent statistical analyses were carried out using 530 MetaboAnalyst 6.0 (www.metaboanalyst.ca). 531 Total aboveground biomass 532 In the mungbean trial, biomass samples were harvested from 1-4 replications of a subset of 47 533 diverse lines. Sampling occurred when the commercial check cultivar Crystal reached 534 flowering (50% of the plot had open flower), which represented mid canopy development . 535 Samples were taken from 0.5 x 1.52 m of the two rows per plot . For the wheat trial, biomass 536 cuts were taken from the same plots as the preceeding mungbean trial at heading. Samples were 537 harvested from 0.5 x 0.75 m of the two middle rows per plot and dried at 65°C for 5 days 538 before weighing for total biomass. 539 Vegeatation indices 540 Vegetation indices were extracted from a multispectral sensor mounted on an unmanned aerial 541 vehicle platform flown across the mungbean trial when check cultivar Crystal was at the 542 flowering and 90% black pod growth stages. The UA V imagery approach for mungbean is 543 described in detail in Van Haeften et al (2025). Briefly, a MicaSense Altum ™ sensor 544 (MicaSense Inc., Seattle, Washington, USA) was flown onboard a Matrice 300 RTK (DJI, 545 USA) UA V platform. Appropriate flight conditions and calibrations were maintained for each 546 flight to ensure high -quality image capture. Images were stitched to generate georeferenced 547 orthomosaics and digital surface models using Pix4D software. Plot boundaries were defined 548 in ArcMap for the mungbean trial and applied to the wheat trial to ensure corresponding wheat 549 plots aligned with previous mungbean genotype plots. Normalized difference red edge index 550 (NDRE), height, thermal and canopy coverage were extracted using orthomosaics, digital 551 surface models , and shape files in a Python environment -based program developed at The 552 University of Queensland (Das et al., 2022). These traits were used to estimate aboveground 553 biomass in mungbean during mid-canopy development (determined when Crystal had reached 554 50% flowering), as described in the ‘mungbean biomass prediction models’ section below. In 555 addition, raw NDRE values were extracted for analysis at 90% black pod stage in mungbean 556 and at flowering in the subsequent wheat crop. 557 Grain quality characteristics 558 Wheat grain protein content was measured using near-infrared spectroscopy with an Inframatic 559 9500 (IM9500) Grain Analyser (Perten Instruments). The instrument warmed up for a 560 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 22 minimum of 30 minutes and self -calibrated using the internal reference standard per 561 manufacturer recommendations. Approximately 200 g sub-samples were loaded into the 562 IM9500 sample hopper for analysis. The IM9500 conducted three sub -sample readings and 563 averaged spectra before predicting constituent values. Each sample was analysed in triplicate; 564 between replicates, the hopper was emptied and refilled with fresh sample portions. Protein 565 content was reported on a 12% moisture basis. 566 Statistical Analyses 567 Mungbean biomass prediction models 568 Mungbean biomass predictions used the UA V flight corresponding to mid-canopy development 569 which was determined by flowering time for check cultivar Crystal and ground-based biomass 570 measurements collected on the same day. This stage represented mid-canopy development, and 571 a prediction model previously developed for mungbean at this stage (Van Haeften et al., 2025) 572 was applied. This stepwise regression model incorporated height, thermal, NDRE and 573 coverage. Model validation used an 80/20 training/testing split. 574 Single-site spatial analyses 575 Single site spatial analysis accounted for spatial variation across the site for all traits. Both 576 fixed and random row and column terms were tested for model improvement and retained as 577 needed. Wald statistics determined significance of fixed effects, whilst AIC, BIC and Log -578 likelihood assessed model fit changes due to added random components. Best Linear Unbiased 579 Estimators (BLUEs) and Best Linear Unbiased Predictors (BLUPs) were calculated using 580 equation 1: 581 𝑦 = 𝑋𝑏 + 𝑍𝑢 + 𝑒 (1) 582 where b is a vector of fixed effects, Z is a vector of random effects, X and Z are the associated 583 design matrices and e is the residual. 584 Generalised broad sense heritability for each trait was calculated per Cullis et al (2006) using 585 equation 2: 586 𝐻𝑔 2 = 1 − 𝐴𝑡𝑡 (2𝛾𝑣) (2) 587 where 𝐴𝑡𝑡 is the average prediction error variance and 𝛾𝑣 is the genetic variance estimated by 588 the model. 589 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 23 Correlations 590 Phenotypic correlations between traits were calculated using Pe arson’s correlation coefficient 591 based on BLUEs for each genotype. Statistical significance was assessed using two -tailed p-592 values, with non-significant correlations (p > 0.05) flagged in visualisations. The same 593 methodology calculated correlations between haplotype variances and haplotype effects. 594 Genetic correlations were calculated for each trait pair using mixed linear model s with 595 unstructured variance structures applied to trait-to-trait interaction terms as per equation 3: 596 [𝑦1 𝑦2 ] = [𝑋1 0 0 𝑋2 ] [𝛽1 𝛽2 ] + [𝑍1 0 0 𝑍2 ] [𝑢1 𝑢2 ] + [𝜀1 𝜀2 ] (3) 597 where y represents phenotype vectors for traits 1 and 2 respectively, X and Z are design 598 matrices for the fixed and random components of t he model, 𝛽 represents coefficient of fixed 599 effects, 𝑢 represents the random genetic effects , and 𝜀 is the residual. Covariance between 600 genetic effects is represented as 𝑉𝑎𝑟(𝑢) = 𝐺 ⊗ 𝐾, where K is the genomic relationship matrix 601 (GRM) and G is the genetic covariance matrix between traits. 602 Genomic analyses 603 DArTseq markers were curated based on a minor allele frequency greater then 0.05, 604 heterozygosity of less then 0.2 , and <20% missing calls , resulting in 5 ,516 high quality 605 markers. Linkage disequilibrium was explored to identify suitable threshold s for haploblock 606 definition. The R package ‘Selection Tools’ defined haploblocks with markers in LD >0.3 and 607 a marker tolerance of 4. Marker tolerance represents the maximum number of markers that can 608 be skipped during haploblock creation, chosen based on the potential for misalignment. This 609 approach defined 1,126 haploblocks with an average of 4.9 markers per block. 610 Marker effects were calculated using ridge-regression BLUP to determine simultaneous effects 611 of all SNPs per equation 4. 612 𝑦 = 𝑊𝐺𝑢 + 𝜀 (4) 613 where u is a vector of marker effects, G is the genotype matrix, W is a design matrix relating 614 genotypes to observations (y) , and 𝜀 is the error. Relevant SNP effects were summed to 615 calculate haplotype effects at each haploblock. Effect variance was calculated, with the top 1% 616 of haploblocks by variance (rounded up to 12) selected for each trait and considered to be a 617 related block for that trait. 618 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 24 Identifying and co-locating mungbean othologues for functional legume genes 619 Minimap2 (Li, 2018) was used to extract regions corresponding to identified haploblocks in a 620 Crystal reference genome enhanced with additional PacBio HiFi long -read sequencing data 621 (Mens & Udvardi, unpublished). Mungbean orthologues of known nodulation and nitrogen 622 fixation genes were identified using reciprocal tBLASTn searches (E -value threshold ≤ 1e-5) 623 against a compiled list of over 200 characterised genes in model legume systems (Roy et al., 624 2022). Orthologues were retained if overlapping with atleast one of the five haploblocks. 625 Weighted selection strategies simulations 626 Three mungbean breeding programs targeting genetic improvement were simulated: 1) 627 mungbean yield alone, 2) legacy wheat yield alone, and 3) simultaneous improvement of both. 628 For simultaneous improvement, selection used trait indices calculated as weighted sums of 629 breeding values with weights of 0.7:0.3, 0.5:0.5, and 0.3:0.7 for mungbean:wheat yield, 630 respectively. Simulations used SNP effects estimated from genomic prediction models to 631 calculate genomic estimated breeding values (GEBVs). Each program ran 25 cycles, selecting 632 30 parents per cycle based on program-specific criteria. Selected parents were randomly mated 633 to create 50 crosses, with 100 progenies per cross, generating 5000 selection candidates for the 634 next cycle. Genetic improvement was assessed by population mean GEBV changes per cycle. 635 Each program was simulated five times, reporting average values and 95% confidence intervals 636 of population mean GEBVs using the genomicSimulation R package (Villiers et al., 2022). 637 Comparison of genotypic differences for soil characteristics 638 To evaluate whether soil characteristics differed significantly among the subset of 10 diverse 639 mungbean genotypes, a one -way analysis of variance (ANOV A) was performed for each 640 measured soil variable. Genotype was treated as a fixed effect, and significance was assessed 641 to determine whether plant genetic variation contributed to differences in soil nutrient 642 concentrations and moisture content. 643 644 645 646 647 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 25 Acknowledgments 648 We thank the World Vegetable Center (WorldVeg), particularly Roland Schafleitner and Ram 649 Nair, for providing the genomic data for the mungbean mini core collection that formed the 650 foundation of this study. We are grateful to Queensland Department of Primary Industries for 651 Gatton field trial support, and provision of the Hermitage 2018 yield datasets. We particularly 652 acknowledge the technical staff at the Gatton Research Facility for their assistance with field 653 operations, and trial management. We thank Long Reach Plant Breeders for generat ing the 654 wheat grain protein data. 655 Data availability 656 All primary data supporting the findings of this study are available via The University of 657 Queensland eSpace repository at https://doi.org/10.48610/cee8b66. 658 Funding 659 This work was supported by The University of Queensland start-up funds awarded to MS. 660 Mungbean research to MS, LH , MU at The University of Queensland was supported by the 661 Australian Centre for International Agricultural Research International Mungbean 662 Improvement Network Phase 2 (CIM2014/079) and Grains Research & Development 663 Corporation Genetic Initiative to Transform Symbiotic Nitrogen Fixation in Australian Pulse 664 Crops (UOQ2403 -012RTX) .LH was supported through an ARC Future Fellowship 665 (FT220100350). ED is supported by the Queensland Government’s Industry Research 666 Fellowships program (AQIRF096 -2023RD6). SVH, ME were recipients Grains Research & 667 Development Corporation Research Scholarship ( UOQ2101-003RSX; UOQ 2410-011RSX). 668 SVH, S MB were recipients of an Australian Government Research Training Program 669 Scholarship. SMB was supported as a PhD student by the Deutsche Forschungsgemeinschaft 670 (DFG) and University of Queensland International Research Training Group 2843 – 671 Accelerating Crop Genetic Gain. ME was supported as a PhD student of the Australian 672 Research Council Training Centre in Predictive Breeding for Agricultural Futures 673 (IC230100016). 674 Authors contributions 675 Shanice Van Haeften: Writing – Original Draft Preparation , Investigation, Data Curation , 676 Formal analyses, Visualisation. Stephanie Brunner : Writing – Original Draft Preparation , 677 Data Curation , Formal analyses, Visualisation . Eric Dingalsan: Methodology, Writing – 678 Review & Editing. Edward Fabreag: Formal analyses, Writing – Review & Editing. Joseph 679 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 26 Eyre: Methodology, Investigation, Writing – Review & Editing. Celine Mens: Investigation. 680 Ben J. Hayes : Supervision, Writing – Review & Editing . Michael Udvardi : Supervision, 681 Writing – Review & Editing . Samir Alahmad: Investigation. Mitchell Eglinton: 682 Investigation, Writing – Review & Editing. Ryan McQuinn: Investigation, Formal analyses, 683 Writing – Review & Editing. Merril Ryan: Resources, Writing – Review & Editing. Sarah 684 van der Meer: Investigation. Millicent R. Smith: Conceptualisation, Funding Acquisition, 685 Project Administration, Writing – Review & Editing . Lee T. Hickey: Conceptualisation , 686 Project Administration Writing – Review & Editing. 687 Competing interests 688 The authors declare no competing interests. 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 27

707 Bennett, A. J., Bending, G. D., Chandler, D., Hilton, S. & Mills, P. Meeting the demand for 708 crop production: the challenge of yield decline in crops grown in short rotations. Biol. Rev. 87, 709 52–71 (2012). 710 Cernay, C., Makowski, D. & Pelzer, E. Preceding cultivation of grain legumes increases cereal 711 yields under low nitrogen input conditions. Environ. Chem. Lett. 16, 631–636 (2018). 712 Ellouze, W. et al. Chickpea genotypes shape the soil microbiome and affect the establishment 713 of the subsequent durum wheat crop in the semiarid North American Great Plains. Soil Biol. 714 Biochem. 63, 129–141 (2013). 715 Evenson, R. E. & Gollin, D. Assessing the impact of the Green Revolution, 1960 to 2000. 716 Science 300, 758–762 (2003). 717 Foyer, C. H., Nguyen, H. & Lam, H. M. Legumes —the art and science of environmentally 718 sustainable agriculture. Plant Cell Environ. 42, 1–5 (2019). 719 Hayes, B. J. et al. Potential approaches to create ultimate genotypes in crops and livestock. Nat. 720 Genet. 56, 2310–2317 (2024). 721 Hickey, L. T. et al. Breeding crops to feed 10 billion. Nat. Biotechnol. 37, 744–754 (2019). 722 Ilyas, N. et al. Contribution of nitrogen fixed by mung bean to the following wheat crop. 723 Commun. Soil Sci. Plant Anal. 49, 148–158 (2018). 724 Kirkegaard, J. A., Hocking, P. J., Angus, J. F., Howe, G. N. & Gardner, P. A. Comparison of 725 canola, Indian mustard and Linola in two contrasting environments. II. Break-crop and nitrogen 726 effects on subsequent wheat crops. Field Crops Res. 52, 179–191 (1997). 727 La Forgia, D., Martin, C., Turlings, T. C. & Verheggen, F. 2 -Pentylfuran: an aggregation 728 attractant for wireworms. Arthropod-Plant Interact. 17, 465–472 (2023). 729 Lago-Olveira, S., Rebolledo -Leiva, R., Garofalo, P., Moreira, M. T. & González -García, S. 730 Environmental and economic benefits of wheat and chickpea crop rotation in the 731 Mediterranean region of Apulia (Italy). Sci. Total Environ. 896, 165124 (2023). 732 Li, J. et al. Diversifying crop rotation improves system robustness. Agron. Sustain. Dev. 39, 38 733 (2019). 734 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 28 Li, L., Awada, T., Shi, Y ., Jin, V . L. & Kaiser, M. Global greenhouse gas emissions from 735 agriculture: pathways to sustainable reductions. Glob. Change Biol. 31, e70015 (2025). 736 Lilley, J. M. & Kirkegaard, J. A. Farming system context drives the value of deep wheat roots 737 in semi-arid environments. J. Exp. Bot. 67, 3665–3681 (2016). 738 Martín-Sánchez, L. et al. Phylogenomic analyses and distribution of terpene synthases among 739 Streptomyces. Beilstein J. Org. Chem. 15, 1181–1193 (2019). 740 Melton, B. E., Heady, E. O. & Willham, R. L. Estimation of economic values for selection 741 indices. Anim. Sci. 28, 279–286 (1979). 742 Rivers, J. Y ., Truong, T. T., Pogson, B. J. & McQuinn, R. P. V olatile apocarotenoid discovery 743 and quantification in Arabidopsis thaliana: optimized sensitive analysis via HS-SPME-GC/MS. 744 Metabolomics 15, 79 (2019). 745 Rose, T. J., Damon, P. & Rengel, Z. Phosphorus-efficient faba bean (Vicia faba L.) genotypes 746 enhance subsequent wheat crop growth in an acid and an alkaline soil. Crop Pasture Sci. 61, 747 1009–1016 (2010). 748 Salvagiotti, F., Biassoni, M. M., Magnano, L. & Bacigaluppo, S. Enhanced seed yield of full -749 season soybean when rotated with cereals and cover crops as compared to monoculture in a 750 long-term experiment. Eur. J. Agron. 161, 127382 (2024). 751 Somenahally, A. et al. Microbial communities in soil profile are more responsive to legacy 752 effects of wheat-cover crop rotations than tillage systems. Soil Biol. Biochem. 123, 126 –135 753 (2018). 754 Villiers, K., Dinglasan, E., Hayes, B. J. & V oss-Fels, K. P. genomicSimulation: fast R functions 755 for stochastic simulation of breeding programs. G3 12, jkac216 (2022). 756 V oss-Fels, K. P., Cooper, M. & Hayes, B. J. Accelerating crop genetic gains with genomic 757 selection. Theor. Appl. Genet. 132, 669–686 (2019). 758 Weisskopf, L., Schulz, S. & Garbeva, P. Microbial volatile organic compounds in intra -759 kingdom and inter-kingdom interactions. Nat. Rev. Microbiol. 19, 391–404 (2021). 760 Yu, S., Wang, T., Wang, L., Yao, S. & Zhang, B. Preceding crop rotation systems shape the 761 selection process of wheat root-associated bacterial communities. J. Integr. Agric. 24, 739–753 762 (2025). 763 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint 29 Zhao, J. et al. Global systematic review with meta-analysis reveals yield advantage of legume-764 based rotations and its drivers. Nat. Commun. 13, 4926 (2022). 765 .CC-BY-NC-ND 4.0 International licenseperpetuity. It is made available under a preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in The copyright holder for thisthis version posted November 5, 2025. ; https://doi.org/10.1101/2025.11.03.686422doi: bioRxiv preprint