Investigating the diverse potential of a multi-purpose legume, Lablab purpureus (L.) Sweet, for smallholder production in East Africa

Climate change is posing severe challenges in Africa, where resilient crops are urgently needed to withstand drought periods and unreliable rainfall. Multi-purpose legume species, such as lablab (Lablab purpureus (L.) Sweet), have been under-utilized yet have the potential to overcome climate challenges. While lablab is native to Africa, there are few characterized varieties and it is under-utilized by smallholder farmers due to a lack of information and access to varieties. Knowledge is especially lacking on the performance of this crop by genotype, management, and environment. We conducted a two-year field study at two sites to evaluate 29 lablab cultivars under sole and maize intercrop management, with 14 cultivars selected for in-depth study. Cultivars were evaluated on vegetative biomass and grain yield production, with N fixation assessed for one site year. Biomass and grain production differed across environments and cultivars, with only biomass affected by intercropping. Average grain yield was substantially reduced to only 37 kg ha-1 in environments with maximum temperatures greater than 33°C, but biomass production yielded comparable amounts across high temperatures and in dry (<500 mm rainfall) environments. Tradeoffs were found between biomass and grain yield across high yielding cultivars, with the top three grain accessions averaging 612 kg ha-1 of grain and 1.97 Mg ha-1 biomass whereas the top three biomass accessions produced 327 kg ha-1 grain and 2.52 Mg ha-1 biomass across all environments. In a comparison of production and N fixation measurements, cultivars were identified which may have high performance in both. Suitability of lablab for grain and biomass production were visualized across Tanzania in a map comparing max temperature thresholds for grain and biomass against average regional livestock populations. This provides a way forward for identifying potential areas for lablab cultivation as a novel means to enhance fodder and pulse production with smallholder farmers.

Yes -all data are fully available without restriction continued to be promoted to smallholder farmers, at the expense of legume production. 41 Challenges to increasing legume production, such as limited availability of legume seed, 42 pest problems, lack of markets, and low yields, have held farmers back from adopting more 43 legume intensive systems, and must be addressed for sustainable intensification to occur. 44 Challenges to legume production have been exacerbated by the limited nature of 45 legume research and minimization of the multi-purpose nature of legumes. Legume studies 46 often prioritize either the grain or forage potential of the study crops, with less focus on the 47 tradeoffs or interactions of these traits [4,5]. Studies on farmer objectives in growing 48 legumes however confirm that farmers have multiple production objectives in growing 49 legumes and consider other benefits besides just grain yield in choosing to cultivate 50 legumes, such as improving soil fertility [2,6,7]. A legume's ability to improve soil fertility 51 through nitrogen (N) fixation depends on many different factors, including genetics, 52 management and the environment [3]. N contribution by a legume therefore must be 53 considered across these factors to fully understand the legume's effect within a system. The 54 diverse objectives of farmers must be considered in order for appropriate legumes to be 55 identified that will meet farmer production needs and improve farming systems. As such, 56 studies should evaluate legume potential from multiple angles, such as grain, forage, and N 57 fixation potential, across diverse cultivars and environments for a more robust assessment 58 of these crops. 59 Typical legume studies, such as those for common bean and cowpea, focus on sole 60 cropping and singular productivity measurements in assessing crop potential and potential 61 across cultivars [8,9,10,11]. There is therefore a lack of quantifying multiple production 62 traits and understanding tradeoffs of these traits within cultivars and in systems that 63 resemble local farmer context, such as in an intercrop with maize. Many cultivar studies 64 have instead focused on finding a few top grain producing types that fit across 65 environments [12]. However, identifying appropriate legume cultivars that fit within 66 different farming systems requires testing diverse cultivars for multiple production 67 qualities and testing their performance under different environment and management 68

conditions. 69
Overall there is a need for better understanding of environment and management 70 parameters of legumes, especially those with multipurpose traits. Previous legume studies 71 have been too empirical and fail to look at legume management as a system within which 72 growing parameters may be established [13]. This is especially true for understudied 73 multipurpose legumes, and there are few systematic studies that present ways of 74 introducing novel crops. Lablab (Lablab purpureus (L.) Sweet) is one such understudied 75 legume with limited study of its diverse genetic collection and evaluation of its 76 multipurpose qualities [14]. Our study takes a multi-dimensional approach to assessing 77 lablab amongst different genetic sources, environments, and management, across which 78 these effects are not well understood for lablab. Our overall objective was to identify 79 promising lablab accessions and suitable growing conditions to inform lablab integration 80 into smallholder farming systems. Specifically, we aimed to identify lablab accessions that 81 are high yielding and stable across environments as well as those that perform best in 82 terms of grain yield and biomass in specific environments and sole cropped or 83 intercropped with maize. We further wanted to assess accession performance across 84 vegetative biomass, N fixation, and grain yield to determine whether some accessions have 85 high multipurpose potential or if accessions are more likely to perform well in one trait 86 over another. 87 with preliminary performance assessment. The 14-accession subset was chosen based on 109 those that had shown promise from early observations of the full set of accessions, with the 110 goal of selecting cultivars with a range of growth types ( Table 1). Four of these accessions 111

Study Sites
were subsequently chosen for further study through on-farm trials with the purpose of 112 selecting a final set of accessions for registration. The cropping system factor consisted of 113 each lablab accession sole-cropped or intercropped with maize (Pannar 15). To simplify 114 field operations, cropping system was randomly arranged within blocks in strips of 115 consecutive intercropped or sole-cropped plots. Each strip had either 8 plots (SARI) or 7 116 (TPRI). One sole maize plot was also included in each block. Individual plots were 4.5 by 117 5.4 m with 1.5 m unplanted borders between plots within strips. Lablab spacing was 0.9 m 118 between rows and 0.5 m within rows with five rows per plot and two seeds planted per 119 station (4.4 seeds m -2 ). Cowpea spacing was half that of lablab, with 0.45 m between rows 120 and 0.5 m within rows. Planting was done in an additive design, where lablab and cowpea 121 spacing was the same intercropped with maize as it was sole cropped. Maize was planted 122 between rows with lablab or cowpea, with six rows per plot at 0.9 m between rows and 0.5 123 m within rows and two seeds planted per station for a seeding rate of 4.4 seeds m -2 . One 124 maize row was planted at the borders of all sole cropped plots to ensure uniform shading 125 regardless of whether sole-cropped plots were adjacent to intercropped plots.   fertilizer (18-46-0) was applied to maize at planting with a rate of 77 kg ha -1 . Urea (46-0-0) 139 was side dressed at 110 kg ha -1 . Fields were tilled with a disc plow in 2016, but planted 140 without tillage in 2017. Weed control was achieved using a pre-plant glyphosate 141 application (2.5 L ha -1 ) at planting, and by hand-hoe throughout the growing season as 142 needed. Insecticide was applied as needed at both sites as significant insect pest damage 143 was observed. 144 Tukey-Kramer test used to identify accession mean differences (alpha=0.05). 218

Plant and Soil Measurements
Analysis of multivariate stability statistics was done with the accession main effect 219 plus accession by environment interaction for grain yield and biomass using the 220 GGEBiplotGUI package with RStudio in R statistical software. Two biplot views, "mean vs. 221 stability" and "which-won-where" were used to visually assess accession performance 222 across environments for grain yield and biomass as well as to determine tradeoffs among 223 high performing accessions for both traits. These biplots have been identified as best 224 capturing genotype by environment effects for multi-environment variety trials [20]. 225 Accession measurements were averaged across management practices for each 226 environment to obtain mean performance for each trait which was subjected to the GGE 227 biplot analysis. The data in "Mean vs. Stability" view were not scaled (Scaling = 0), 228 environment-centered (Centering = 2) and based on genotype-focused singular value 229 partitioning (SVP = 1). The "which-won-where" model parameters were also set on un-230 scaled data (Scaling=0), environment-centered (Centering=2) and environment focused 231 singular value partitioning (SVP=2) [21]. 232

233
Weather 234 Rainfall across both study years was below average and unevenly distributed across 235 months for both sites. In 2016 the SARI site had 315 mm rainfall between January and 236 September, with the majority of rainfall occurring between January and April (Fig 1A). For 237 the same time period the TPRI site had 252 mm of rainfall with the majority of rainfall 238 occurring in April after the field trials were planted (Fig 1B). including lablab grain yield, biomass, and maize yield (Fig 2, S2 Table). All lablab accessions 252 produced low to nil grain yield at the TPRI site across both years, with averages of 31 kg ha -253 1 in 2016 and 42 kg ha -1 in 2017 (Fig 2c,d) Table). Biomass production differed across environments, accession, and by management 270 (Fig 2, S2 Table). In contrast to grain yield, biomass produced at TPRI was comparable to 271 that produced at SARI. Sole cropped lablab generally produced greater biomass than  Maize yield was also considered in assessing the productivity of intercropped lablab. 292 Maize yield was not affected by accessions and only differed across environments (S2. 293 Table). Maize yield was highest at SARI in 2016, with 3.0 Mg ha -1 and lowest at TPRI in 15 2016 with 1.1 Mg ha -1 . Maize yields in 2017 were within this range, with 2.3 Mg ha -1 at TPRI 295 and 1.6 Mg ha -1 at SARI. 296 Intercrop systems were overall more productive than sole cropped plots as shown 297 by LER values greater than 1.7 across environments for both lablab grain LER and lablab 298 biomass LER (S3 Table). An LER greater than 1 is indicative of an intercrop advantage over 299 sole cropped production of the crops. Accessions of lablab performed in a highly similar 300 manner, with no differences detected between accessions in terms of LER for either grain 301 or biomass. 302

Accession performance, stability, and environmental niches 303
Lablab accession performance across environments was ranked for grain yield and 304 biomass production through the "Mean vs Stability" view of the GGE biplot (Fig 3). This 305 view is based on mean performance and stability across environments within a mega-306 environment. The single arrowed axis is the average-environment coordination (AEC) 307 abscissa and represents the average environment against which the accession 308 performances are ranked. The arrow indicates the direction of higher mean performance 309 and thus shows the rank of each accession. Stability of each lablab accession is represented 310 by its location along the AEC ordinate (axis perpendicular to AEC abscissa), with the most 311 stable accessions located on the AEC abscissa. The GGE biplots explained 98% of genotypic 312 and genotype by environment variation across locations for grain yield performance and 313 79% of variation for biomass production (Fig 3). Accessions with above average grain yield 314 in order of magnitude are DL1002 (#4), Karamoja Red (#17), Q 6880B (#22), ILRI 14437 315 (#14), CIAT 22759 (#1) and SARI Rongai (#26). Of these, DL1002 had the highest grain 316 yield but was the most unstable as its rank was inconsistent across environments. Q 6880B 317 was the most stable of the accessions that had above average grain yield (Fig 3A). 318 Accession performance in relation to biomass production shows DL1001 (#3), Rongai 319 (#23), and ILRI 6930 (#16) as having above average biomass yields, with Rongai also being 320 the most stable (Fig 3B). In general, those accessions with above average grain yields were 321 among the lowest in biomass production. No accessions had both above average grain yield 322 and biomass. Similarly, no accession had low stability in both grain yield and biomass. 323 The which-won-where view of the GGE biplot identifies the accessions which 329 performed best in different environments as measured by grain yield and biomass (Fig 4). also best adapted to this environment for grain (Fig 4a). Karamoja Red (#17) was the best 340 performer in SARI 2016 with Q 6880B (#22) also well adapted to this environment. TPRI 341 2016 and TPRI 2017 had low grain yields overall and were not clearly identified in a sector, 342 suggesting that these environments are not well suited to lablab grain production. The 343 remaining accessions did not clearly align to a test environment, which indicates that the 344 environments in this study did not necessarily provide ideal conditions for grain 345 production of these accessions. 346 In the which-won-where view for biomass, SARI 2016 and TPRI 2016 were 352 identified as having similar accession performance, and thus were similar environments for 353 biomass production (Fig 4b). Within these two environments DL1001 (#3) was the top 354 biomass producer, with Rongai (#23) and ILRI 6930 (#16) performing well in these 355 environments as well. TPRI 2017 was identified as a unique environment for biomass 356 production within which Karamoja Red (#17) did best. Echo Cream (#6), PI 195851 (#21), 357 Highworth (#8), DL1002 (#4) also did well in this environment. SARI 2017 did not align to 358 a sector, suggesting it was not a suitable environment for maximizing biomass production. 359 The remaining accessions that fell in different sectors without a clear environment signal 360 are consistent with study environments as being not well suited to high biomass 361 production for these accessions. nodule weight assumed to be associated with increased N fixation and δ 15 N values closer to 371 zero associated with higher N fixation given that δ 15 N signature of atmospheric N2 is 372 defined as zero [3,17]. 373 The correlation matrix showed that grain yield was positively correlated with 374 biomass (r=0.387; p<0.05) and %N negatively correlated with grain (r=-0.518; p<0.001) 375 and biomass (r=-0.647; p<0.001) (S4 . Table). The variables were grouped into two 376 components with eigenvalues greater than 1 and which explained 52.3% of the total 377 variability among the variables ( Table 2). The first component accounted for 33.6% of the 378 variability and represented plant production as it was dominated by large loadings by grain 379 yield, biomass, plant population and negatively with %N. The second component accounted 380 for 18.7% of the variability and was associated with the nitrogen fixation variables nodule 381 weight and δ 15 N (negatively correlated) and soil nitrate (Table 2). 382 Biplots of the first two components with the variable loadings shows the 385 distribution of accessions and block across productivity/growth (PC1) and N fixation (PC2) 386 (Fig 5). Multivariate accession and block effects were found for PC1 but not PC2 (S5 . Table). 387 ILRI 14437 (#14) was found to have the highest productivity whereas SARI Nyeupe (#25) 388 had the lowest. This suggests some accessions may be able to maintain high growth (yield, 389 biomass) and N fixation, but for others N fixation may come at a cost to low growth. 390 [23] tested lablab over a moisture gradient and found lablab grain yields as high as 1271 kg 407 ha -1 with 190 mm of rainfall. Our results are generally consistent with these studies as 408 lablab yield ranged widely and was often higher than 500 kg ha -1 under low precipitation 409 (<500 mm), suggesting that drought-tolerance is a common trait in lablab. 410 We further found evidence for an interaction of temperature and precipitation in 411 lablab grain production. The TPRI site in our study had between 252-315 mm of rainfall 412 but was not suitable for grain production. Given that a similar amount of rainfall was seen 413 at SARI in 2016 but with yields upwards of 1000 kg ha -1 , hot temperatures at the TPRI site 414 seems to be the limiting factor for grain production. The TPRI site had both higher 415 minimum and maximum temperatures than SARI, with maximum temperatures averaging 416 lablab found that flowering was delayed at temperatures higher than 28°C. Our results 420 support this finding and suggest that if grain yield is a priority, environments with 421 maximum temperatures >33°C may not be suitable for lablab cultivation. 422 A third environment effect, intercrop versus sole crop management, was found to 423 not affect grain yield (Fig 2, S2 Table). Previous descriptions of lablab suggest average 424 grain yields around 1500 kg ha -1 when sole cropped, but only 450 kg ha -1 when 425 intercropped [25]. Interestingly, in the highest grain yield environment, SARI 2017, 426 average intercropped grain yields were higher (980 kg ha -1 ) than these previous reports. In  Lablab biomass production by environment 443 Biomass production by rainfall gradient demonstrated unclear trends. While many 444 studies show biomass production increases with increased water availability, Sennhenn et 445 al. [23] found that the increase in lablab biomass with increased water amounts was 446 gradual in an irrigation gradient. In our study, the highest rainfall environment, SARI 2017 447 with 463 mm rainfall, had the lowest biomass production overall (1.1 Mg ha -1 ). In contrast, 448 SARI 2016 (315 mm) had the highest biomass overall (3.1 Mg ha -1 ) but TPRI 2017 (311 449 mm), biomass was substantially lower (2.3 Mg ha -1 ). Disease prevalence amongst legumes 450 is also well known, and increased moisture may increase the severity of disease, thus 451 negatively affecting biomass production [28]. 452 Biomass production was less affected by high temperatures than grain in our study.

501
Results observed for individual accessions indicated high plasticity, with accession 502 performance varying by environment. This is consistent with many test environments 503 being desirable to identify ideal environments for dual-purpose legumes. Generally in 504 cultivar assessment the presence of genotype by environment interactions necessitates 505 multiple test environments to identify suitable varieties for various production areas [36]. 506 In assessing test environments for common bean across Africa using a GGE Biplot analysis, 507 Kang et al. [8] identified redundant test environments for bean cultivars with implications 508 for regional breeding centers. While an excessive use of test environments may be possible 509 with a heavily studied crop such as common bean, dual-purpose legumes with a diverse 510 genetic background such as lablab may well require many test environments [37]. This is 511 supported by the GGE biplots of lablab accessions included in this study (Fig 4), where nine 512 accessions for grain and six for biomass did not clearly align with the test environments, 513 suggesting that further environments are needed to identify suitable growing niches for 514 grain and biomass production, in addition to areas that are suitable for both. 515 An initial step towards identifying suitable environmental niches for lablab by 516 mapping maximum temperature thresholds across Tanzania shows that the niche for high 517 lablab biomass performance is substantially larger than it is for grain yield (Fig 6). 518 Furthermore, these areas have substantial overlap with high livestock production areas. 519 Future lablab performance studies in Tanzania should focus on these areas of overlap 520 between high livestock and hot environments to expand the test environments used for 521 lablab and thus gain additional insight on lablab accessions' environmental parameters. 522 Optimal grain areas are also suitable for biomass, but biomass areas are unsuitable for 527 grain production. Livestock populations calculated as mean pixel TLU per region. 528 temperatures, which could explain how Q 6880B was one of the few accessions to produce 539 grain at the hot TPRI site and suggests that this accession may be best suited for promotion 540 as a heat-tolerant grain variety. In our study, only three of the lablab accessions (DL1001, 541 Rongai, ILRI 6930) had above average biomass yields (Fig 3B), and these accessions were 542 best suited to both SARI 2016 and TPRI 2016 environments. The Rongai accession is a 543 common lablab variety used for forage production, and previous studies have also noted 544 it's high forage potential [31,37]. Interestingly, another lablab accession often promoted for 545 forage, Highworth, was not found to be a top biomass producer in our study environments 546 with high biomass production in hot environments may be desirable for farmers in these 560 locations, especially given widespread livestock husbandry. Farmers require expanded 561 crop options, and a wide range of lablab accessions could help address these needs [14]. 562

Lablab performance tradeoffs 563
The SARI 2017 sub-study provided the first systematic assessment in lablab that we 564 know of to quantify variability in accession N fixation traits, biomass, grain yield and soil N 565 status. In this environment, tradeoffs were modest between biomass and grain, with some 566 accessions identified that had high productivity and similar N fixation as those with low 567 productivity. In the only other study of lablab genetic variation in N fixation, Ewansiha et al. 568 [33], note that late maturing lablab varieties generally were associated with copious 569 growth and large amounts of accumulated biologically fixed N, yet generally had low 570 nodulation compared to earlier accessions. Our study found a similar trend in nodulation as 571 those with higher nodule weight and low d15N (suggesting higher N-fixation) were early-572 mid maturing accessions. However, previous studies that estimated N fixation rates in 573 lablab report percentages from 35 -89% [32,43,44], suggesting that total N amounts in 574 biomass might not imply greater amounts of N2, as Ewansiha et al. indicate especially if N 575 fixation rate differences are due to accession type. Further study is needed to assess lablab 576 N fixation potential across accession types and to understand the relationship between 577 maturity type and N-fixation [33], with clear implications for sustainability of multi-578 purpose legumes in smallholder farming systems. 579 Lablab potential in smallholder farming systems 580 Lablab accessions provide unique options that address the multiple needs of 581 farmers who are managing complex cropping systems, with clear potential to expand dual 582 use legume production in hot environments across Tanzania (Fig 6). This study provides a 583 methodology for identifying lablab accessions suitable to current farming systems, with the 584 potential to improve overall sustainability. Accessions were identified as high performers 585 in terms of grain or biomass, with particularly strong forage biomass performers identified 586 for hot, dry environments, which could be introduced to support sustainable intensified 587 livestock production in Tanzania [45]. At the same time, high environmental plasticity was 588 observed for dual use strong performers, consistent with the need for broader 589 environmental testing of accessions for dual use. Further, the study provided evidence that 590 incorporating lablab into maize cropping systems as an intercrop would allow farmers to 591 achieve sufficient grain and forage yield without having to commit land solely to lablab. The 592 maize-lablab system was also suitable for accession screening, providing consistent results 593 to sole lablab under hot dry environments. A recommendation coming out of our study is 594 that lablab production and accession evaluations be conducted using intercrop rather than 595 sole conditions, as this is most applicable to small-scale farming systems. 596

Conclusion 597
While common bean is the most widely grown grain legume in Tanzania, it's 598 production area is limited by temperature and precipitation, thus limiting current legume 599 production [46,47]. Expanding the temperature and rainfall range in which legumes are 600 produced would therefore increase Tanzania's legume production area. Lablab accessions 601 in our study produced substantial amounts of grain and biomass in hot, dry environments 602 that were 6°C above common bean's 25°C max temperature threshold [47]. In a review of 603