A preliminary hydroponic screening experiment was conducted in 2022 to evaluate nutrient-use efficiency of 35 sweetpotato genotypes and to identify candidates for subsequent studies. The plant materials consisted of 35 sweetpotato germplasm resources (Supplementary Table S1), which exhibited divergent morphological and agronomic traits, provided by the Key Laboratory of Potato Biology and Genetic Breeding of Chongqing Municipality. Uniform vine cuttings (30 cm in length) were established in a hydroponic system. Four nutrient treatments were established based on prior studies (Deng et al., 2020; Li et al., 2014a; Sun et al., 2021) with slight modifications: (1) Control (CK) with sufficient supply of N, P, and K; (2) Low N (LN, 0 mmol L^–1 NO₃^-); (3) Low P (LP, 0 mmol L^–1 PO₄^3-); (4) Low K (LK, 0 mmol L^–1 K⁺). The composition of the full-strength nutrient solution for CK was detailed in Supplementary Table S2. The LN, LP, and LK treatments followed a randomized complete block design with three replications, each containing two plants per genotype. During the establishment phase, seedlings were pre-acclimatized in distilled water for 10 days to facilitate root initiation and the development of three fully expanded leaves newly formed during the acclimation period. Nutrient stress treatments were subsequently initiated through daily replenishment of the respective solutions each morning (8:00-9:00) by adding solution to maintain the initial 8 cm depth, which were replaced every three days with fresh solution of the same composition to maintain nutrient stability. Each hydroponic container had a volume of 2 L, and the 8 cm depth corresponded to approximately 1.6 L of solution per container. Growth chambers were maintained at 25 ± 1 °C, 70 ± 5% relative humidity, and a 12-hour light/dark cycle with a photosynthetic photon flux density (PPFD) of 250 μmol m^-2 s^-1 provided by white LED growth lights. After 28 days of treatment, photosynthetic parameters and agronomic characteristics were measured.

Based on the comprehensive evaluation value (Y) derived from the hydroponic screening, we selected eight genotypes that represent a broad spectrum of efficiency, ranging from high to low, for field validation in 2023. This selection was made to investigate genotype-specific responses to varying fertilizer levels. The basic soil nutrient properties at a depth of 0–20 cm are presented in Supplementary Table S3. A randomized complete block design with three replications was utilized. The six fertilizer treatments included: Low Nitrogen (LN), High Nitrogen (HN), Low Phosphorus (LP), High Phosphorus (HP), Low Potassium (LK), and High Potassium (HK), respectively. This design resulted in a total of 144 experimental plots. Each plot measured 1.8 m², with a plant spacing of 22.5 cm (equivalent to 2,964 plants per 667 m²). Vine cuttings were planted on June 2, 2023. Fertilizers were applied via the hole-application method 10 days after planting, using urea (46% N), calcium superphosphate (12% P₂O₅), and potassium sulfate (52% K₂O). The specific application rates for each treatment are provided in Supplementary Table S4. The “high” level (HN, HP, and HK) treatments were set at or slightly above the locally recommended optimum fertilizer rates for sweetpotato production, thereby representing a condition of non-limiting nutrient supply. In contrast, the “low” level (LN, LP, and LK) treatments were designed to induce moderate nutrient stress by applying zero additional N, P, or K fertilizer, respectively, beyond the basal soil background.

In the hydroponic screening trial, plant growth traits were measured for each replication. Following established protocols (Wera, 2014; Yildirim et al., 2011), we measured the following traits: main stem length, adventitious root length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, and shoot dry matter increase. Each measurement was replicated three times and the mean value was used for subsequent analysis. In the field trials, all sweetpotato plants were harvested at the appropriate harvest stage. The fresh weight of storage roots was recorded for each plot to determine yield. Five plants were randomly selected from each plot to assess shoot and the fresh weight of storage root. The shoot and storage root samples were cleaned, sliced, and heated at 105 °C for 30 minutes to deactivate enzymes, followed by drying at 80 °C until a constant weight was achieved. The storage root dry weight (g), shoot dry weight (g) and whole plant dry weight (g) were documented as described by Huang et al. (2020).

Photosynthetic parameters were measured between 9:00 and 11:00 A.M. on the day prior to sampling. On each of three representative plants per plot, the youngest fully expanded leaf was selected for the measurement of the Pn, stomatal conductance (Gs), intercellular carbon dioxide (CO₂) concentration (Ci), and transpiration rate (Tr), using a LI-6400 portable photosynthesis system (LI-COR Biosciences, Lincoln, NE, USA) following standard procedures (Wellburn, 1994).

The concentrations of N, P, and K in plant samples were determined after digestion with H₂SO₄-H₂O₂. Total N in the samples was determined using the sulfuric acid-hydrogen peroxide digestion method in conjunction with an automatic Kjeldahl apparatus (Chen et al., 2016; Du et al., 2024). Total P was quantified by the vanadium-molybdenum yellow colorimetric method using a spectrophotometer (UV-1500, Macylab, Shanghai, China) (Lu, 1999). Total K was analyzed using the sulfuric acid-hydrogen peroxide digestion method with a flame photometer (AA-7000, SHIMADZU, Kyoto, Japan) (Deng et al., 2021). The soil alkaline-N content was determined by the alkali diffusion method described by Chen et al. (2016). Available P and available K were measured using a Top nutrient analyzer (TPY-16A, Zhejiang Top Cloud-Agri Technology Co., Ltd., China) following the standard methodology.

Nutrient accumulation and utilization parameters were calculated following established methods from previous studies. The formulas employed are outlined below:

N accumulation: N concentration × Dry matter weight (Deng et al., 2021).

N physiological use efficiency (%): Dry matter increase/N accumulation value (Liu et al., 2023).

Root-shoot ratio: Root dry weight/Shoot dry weight (Liu et al., 2023; Tang et al., 2014).

N harvest index: N accumulation in storage root/N accumulation in all organs (Deng et al., 2021).

N uptake efficiency (kg·kg^-¹): N accumulation in plants/N applied (Deng et al., 2021).

Storage root N utilization efficiency (kg·kg^-¹): storage root yield (fresh weight)/total plant N accumulation (Hitz et al., 2017).

Plant N utilization efficiency (kg·kg^-¹): whole-plant yield (fresh weight)/total plant N accumulation (Hitz et al., 2017).

Relative yield: storage root yield of LN/storage root yield of HN (Deng et al., 2021).

N sensitivity index: storage root yield of N treatment/storage root yield of no N treatment (Tang et al., 2014).

Yield reduction rate (%): (storage root yield of HN – storage root yield of LN)/storage root yield of HN × 100%.

N agronomic use efficiency (kg·kg^-¹): (Yield in N applied area – Yield in non-N applied area)/N applied (Du et al., 2024; Lemma et al., 2023).

N partial factor productivity (kg·kg^-¹): Yield in N applied area/N application rate (Du et al., 2024).

N fertilizer contribution rate (%): (Yield of N fertilizer application area – Yield of no N fertilizer application area)/Yield of N fertilizer application area × 100% (Wu et al., 2016).

N utilization index and Low-N tolerance index was calculated according to the Group Standard (T/GDSMM0009-2021) as below:

N utilization index: (storage root yield of no N treatment/Average storage root yield of no N treatment) × (storage root yield of N treatment/Average storage root yield of N treatment).

Low-N tolerance index: (Storage root yield of no N treatment/Average storage root yield of no N treatment) × (Storage root yield of no N treatment/Average storage root yield of N treatment).

The calculations of P and K follow the same procedures.

Under various nutrient treatments, 35 sweetpotato genotypes exhibited differential responses in seedling phenotypic and physiological indices (Table 1). Specific nutrient deficiencies triggered distinct adaptive responses. Compared to the control (CK), the LN treatment significantly increased root length by 15.2%, and enhanced root fresh weight by 9.8%. However, it also resulted in a notable reduction in stomatal conductance by 45.94%. Shoot N physiological use efficiency increased by 22.6% under LN stress. LP conditions enhanced root length by 18.3% but caused a sharp reduction of 69.77% in shoot P accumulation. The LK treatment specifically improved shoot K physiological use efficiency by 19.1%, with a concomitant decrease of 67.4% in Tr.

The coefficients of variation (CVs) for all traits, calculated to assess their discriminative power, ranged from 5.36% to 80.79% for CK, 12.50% to 94.99% for LN, 12.55% to 96.25% for LP, and 18.19% to 99.17% for LK. Under all stress conditions, the shoot N, P, and K contents, along with the Ci, exhibited low variability (less than 30%), indicating minimal impacts from nutrient stress. In contrast, traits such as leaf number per plant, root morphology (including root length and fresh weight), shoot biomass (both fresh and dry weight, as well as dry matter increase), photosynthetic parameters (Pn, Gs, and Tr), and nutrient use efficiency indices (N, P, and K accumulation values, physiological use efficiency) showed significant variation, making them effective indicators for assessing nutrient use efficiency in sweetpotato germplasm.

Given the complexity of nutrient stress responses, PCA was employed to integrate 14 phenotypic and physiological indices. PCA identified four principal components (PCs) with eigenvalues >1 for each nutrient stress (Table 2), which collectively explained 83.22%, 82.79%, and 76.19% of the total variance under N, P, and K deficiency, respectively, indicating effective dimensionality reduction.

In the analysis of N responses, critical positive loadings were identified for shoot N physiological use efficiency (0.371 in PC1), Gs (0.405 in PC2), and root fresh weight (0.513 in PC4). Conversely, the shoot N accumulation value displayed a negative loading (-0.419 in PC3). The components accounted for variance contributions of 33.00% in PC1, 29.08% in PC2, 13.53% in PC3, and 7.62% in PC4, respectively, collectively explaining 83.22% of the variance. This confirms the effectiveness of dimensionality reduction in characterizing N responses.

In the P analyses, significant positive loadings were observed for shoot fresh weight (0.368 in PC1), shoot P physiological use efficiency (0.473 in PC2), and shoot phosphorus content (0.442 in PC3), while a negative loading was noted for root length (-0.460 in PC4). The variance contributions were ranked at 43.77% for PC1, 17.40% for PC2, 12.68% for PC3, and 8.96% for PC4, achieving a cumulative variance of 82.79% and validating the model’s capacity to distill the complexity of P response.

Regarding K assessments, key positive loadings were observed for shoot dry weight (0.398 in PC1), Gs (0.505 in PC2), and shoot K physiological use efficiency (0.447 in PC4), whereas Ci negatively influenced PC3 (-0.409). The component contributions were 31.32% (PC1), 21.84% (PC2), 12.52% (PC3), and 10.51% (PC4), collectively explaining 76.19% of variance and demonstrating a robust extraction of K response patterns.

To establish a simplified model for predicting the comprehensive evaluation value (Y), stepwise regression was performed between Y and the stress tolerance indices of the 14 agronomic traits (Table 3). In the N tolerance response, we found that the tolerance indices for eight traits (including main stem length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, shoot dry matter increase, Pn, and shoot N physiological use efficiency) exhibited highly significant correlations with Y. After addressing multicollinearity by eliminating the shoot dry matter increase, a stepwise regression model was established with the remaining seven traits. The resulting equation Z = 4.121 + 0.541X₁ + 0.677X₃ + 0.592X₄ + 0.366X₅ + 1.122X₆ + 1.188X₈ + 0.164X₁₄, showed strong predictive capability with R² = 0.985, where the regression coefficients indicated the contribution weights of the variables. Through comprehensive variation, principal component, and regression analyses, seven key evaluation criteria were identified for N deficiency tolerance at the seedling stage: main stem length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, net photosynthetic rate, and shoot N use efficiency.

Similar analytical procedures for P deficiency tolerance (Table 3) revealed highly significant correlations between yield (Y) and twelve indices, including main stem length, root length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, shoot dry matter increase, Pn, Gs, Ci, shoot P accumulation, and shoot P physiological use efficiency. Subsequent regression modeling, after addressing multicollinearity effects produced the equation Z = 4.284 + 0.690X₁ + 0.443X₂ + 0.735X₃ + 0.408X₄ + 0.267X₅ + 0.286X₆ + 0.139X₈ + 0.113X₉ + 0.642X₁₀ + 2.419X₁₃ + 0.031X₁₄ (R² = 0.999, P < 0.001). This analysis ultimately identified ten critical indices for evaluating phosphorus tolerance: main stem length, root length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, Pn, Gs, shoot P accumulation value, and shoot P physiological use efficiency through multidimensional analysis.

In the assessment of K deficiency (Table 3), ten parameters were identified as exhibiting significant importance: root length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, shoot dry matter increase, Pn, Gs, Tr, and shoot K accumulation. The optimized regression mode, represented by the equation Z = - 3.590 + 0.482X₂ + 0.191X₃ + 0.519X₄ + 0.303X₅ + 0.318X₆ + 0.263X₇ + 0.610X₈ + 0.780X₉ + 0.810X ₁₁ + 1.433X₁₃ demonstrated an exceptional fit (R² = 0.999, P < 0.001). This analysis enabled the selection of these ten metrics for evaluating K tolerance through integrated analytical approaches.

Cluster analysis utilizing Euclidean distance was conducted on the evaluation indices for low N, P, and K tolerance, with the results illustrated in Figure 1. In the LN assessment (Figure 1A), the 35 germplasms were categorized into two clusters. Cluster I, comprising 20 germplasms, characterized by superior root fresh weight and shoot N physiological use efficiency, was identified as low-N tolerant (mean Y = 0.33). Conversely, Cluster II, consisting of 15 germplasms, despite higher shoot dry weight and Pn, was less efficient and classified as low-N sensitive (mean Y = -0.44).

In the LP analysis (Figure 1B), three clusters were identified. The majority of germplasms (Clusters I and II, encompassing 29 germplasms), exhibited greater root length and enhanced shoot P use efficiency and were classified as low-P tolerant. In contrast, Cluster III, comprising 6 germplasms, was sensitive to P deficiency. Regarding the LK evaluation (Figure 1C), two clusters were formed. Cluster I, consisting of 27 germplasms, demonstrated larger root length, greater shoot dry weight, increased leaf number per plant, higher root fresh weight, and improved N use efficiency, was determined to be the low K-tolerant type. Conversely, Cluster II, comprising 8 germplasms, exhibited a higher Pn and Gs but lower stress tolerance, was sensitive to K deficiency.

Germplasms were further classified into four nutrient use efficiency types based on their performance under CK and stress conditions (Figure 2). Comprehensive N efficiency values were calculated under both LN and CK conditions. Scatter plot analysis (Figure 2A) categorized the germplasms into four groups: inefficient-efficient (IE type, nutrient-efficient in high nutrient treatment, 11.43%, 4 germplasms), efficient-efficient (EE type, nutrient-efficient in both high and low nutrient treatments, 31.43%, 11 germplasms), inefficient-inefficient (II type, nutrient-efficient in normal or high nutrient treatments, 48.26%, 15 germplasms) and efficient-inefficient (EI type, nutrient-efficient in low nutrient treatment, 14.29%, 5 germplasms).

For P efficiency, indices such as main stem length, root length, leaf number per plant, shoot fresh weight, root fresh weight, shoot dry weight, Pn, Gs, shoot P accumulation value, and shoot P physiological use efficiency were assessed. Under LP and CK treatments with K supplementation, scatter plot analysis (Figure 2B) classified the germplasms into EE, (25.71%, 9 germplasms), EI (25.71%, 9 germplasms), II (28.57%, 10 germplasms), and IE (20%, 7 germplasms).

The evaluation of K efficiency evaluation incorporated root length, Tr, and shoot K accumulation value. Scatter plot analysis (Figure 2C) under LK and CK treatments revealed EE (25.71%, 9 germplasms), EI (8.57%, 3 germplasms), II (48.57%, 17 germplasms), and IE (14.29%, 5 germplasms). The II category predominated across all nutrients, representing 48.26% for N, 28.57% for P, and 48.57% for K while EE genotypes constituted 25.71 to 31.43% of the population.

The 35 sweetpotato germplasms were systematically classified into two categories: nutrient stress resilience and nutrient use efficiency types, as detailed in Supplementary Table S5. A comprehensive cluster analysis revealed that 13 germplasms exhibited combined tolerance to low N, P, and K stress, while one germplasm demonstrated complete intolerance to low N, P, and K stress, designated as the low N-, P-, and K-intolerant type. Nutrient use efficiency evaluation categorized 4 germplasms as N-, P-, and K-EE types and 5 as N-, P-, and K-II types.

An integrated analysis of low N tolerance indices and N efficiency values identified certain germplasms as low N-tolerant and N-efficient, in contrast to 5 germplasms classified as low N-intolerant and N-inefficient types. A parallel assessment for P utilization distinguished 3 germplasms as low P-tolerant and P-efficient, compared to 5 germplasms identified as low P-intolerant and P-inefficient types. Similarly, K analysis revealed 4 germplasms as low K-tolerant and K-efficient, in comparison to 5 germplasms classified as low K-intolerant and K-inefficient types.

Critical findings identified germplasm XN2153–5 as a low N-, P-, and K-tolerant and highly efficient type, which maintained robust performance across all three nutrient limitations. Conversely, XN2153–1 was characterized as a low N-, P-, and K-intolerant and low-efficient type, demonstrating compromised viability under nutrient stress conditions.

Field trials validated and extended the seedling-stage findings. At the harvest stage, the responses of various indices among different sweetpotato germplasm resources under two N levels exhibited discrepancies (Supplementary Table S6). The analysis of agronomic traits and nutrient use efficiency-related indices of sweetpotato germplasm resources at the harvest stage revealed that, compared to the HN treatment, the LN treatment resulted in significant declines in most indices, particularly in the storage root N accumulation, storage root N content, storage root fresh weight, N harvest index, shoot dry weight, whole plant dry weight, storage root dry weight, total N accumulation value, total N content, and shoot N accumulation value. This indicated that these traits are highly sensitive to N supply. The root-shoot ratio remained stable across N treatments, demonstrating its insensitivity to N levels. Both storage root N utilization efficiency and plant N utilization efficiency showed increased values under LN conditions, indicating that N limitation triggers adaptive physiological responses that enhance N acquisition and utilization efficiency in sweetpotato. Under LN conditions, the CVs significantly increased for storage root N utilization efficiency, plant N utilization efficiency, root-shoot ratio, shoot N content, storage root N content, total N content, shoot N accumulation value, storage root N accumulation value, total N accumulation value, and N harvest index, indicating that these traits exhibit greater variation among germplasm resources under low N conditions. This reflects enhanced expressivity of genetic variation under N stress and provides an effective screening opportunity for identifying N-efficient genotypes.

The response of various indices to two P levels varied among different sweetpotato germplasms (Supplementary Table S7). Compared to the HP treatment, the LP treatment resulted in declines in most indices, particularly the root-shoot ratio, storage root fresh weight, storage root dry weight, storage root P accumulation, total P accumulation, whole plant dry weight, and storage root P use efficiency. This indicated that these biomass-related traits are highly sensitive to P availability. Conversely, shoot P content, total P content, shoot P accumulation, storage root P accumulation, and P harvest index remained relatively stable across P treatments, demonstrating their insensitivity to P levels. Notably, storage root P content exhibited increased values under LP conditions. Under LP conditions, the CVs highly increased for storage root dry weight, P harvest index, storage root P content, shoot P content, total P content, and storage root P utilization efficiency, compared to HP. This suggests that these traits exhibit greater variation among germplasm resources under low P conditions and could serve as critical indices for selecting P-efficient germplasm resources.

The responses of various indices among different sweetpotato germplasms to two K levels also exhibited significant variation (Supplementary Table S8). Compared to the HK treatment, the LK treatment resulted in declines in most indices, such as storage root K accumulation, storage root dry weight, storage root fresh weight, root-shoot ratio, storage root K use efficiency, K harvest index, whole plant dry weight, shoot K content, storage root K content, total K accumulation, and plant K utilization efficiency. This indicated that these traits are highly sensitive to K supply. Interestingly, shoot dry weight and shoot K accumulation increased under LK conditions, demonstrating the plant’s adaptive physiological strategies through regulating resource allocation and enhancing nutrient utilization efficiency. Under LK conditions, CVs among germplasms increased for most of the tested indices, with a dramatic increase in root-shoot ratio, K harvest index, total K accumulation, shoot K accumulation, whole plant dry weight, shoot K content, storage root K content, and storage root K utilization efficiency. This suggests that these traits exhibit greater variation among germplasm resources under low K conditions, providing critical targets for selecting K-efficient genotypes. Furthermore, the tested agronomic traits and nutrient use efficiency-related indices varied distinctly under LN, LP, and LK stress, indicating that sweetpotato exhibited different adaptive physiological strategies under different nutrient stress.

Correlation analysis revealed conserved relationships among traits across nutrients (Figure 3). Under N treatments (Figure 3A), biomass traits exhibited a strong association with N accumulation. Both storage root fresh weight and dry weight demonstrated highly significant positive correlations (P < 0.01) with storage root N accumulation. Similarly, shoot dry weight and whole plant dry weight revealed strong positive correlations with shoot N accumulation, total N accumulation, and plant N uptake efficiency. A notable finding was the significant negative correlation (P < 0.01) between plant N utilization efficiency and both total N content and shoot N content. Furthermore, storage root N accumulation was positively correlated with both plant N uptake efficiency and plant N utilization efficiency.

Under P treatments (Figure 3B), a similar pattern emerged where biomass and accumulation were closely linked. The storage root fresh weight exhibited a positive correlation with storage root P accumulation. Additionally, the shoot dry weight and the whole plant dry weight demonstrated significant positive correlations with the shoot P accumulation as well as the total P accumulation. Root-shoot ratio demonstrated significant positive correlations with P harvest index, storage root P utilization efficiency and plant P utilization efficiency. Moreover, total P content showed a positive correlation with shoot P accumulation and total P accumulation, while exhibiting a highly significant negative correlation with the P harvest index, storage root P utilization efficiency and plant P utilization efficiency. Furthermore, shoot P accumulation and total P accumulation revealed highly significant negative associations with P harvest index, storage root P use efficiency and plant P use efficiency. Storage root P use efficiency and plant P use efficiency exhibited highly significant positive correlations with P harvest value.

In the context of K conditions (Figure 3C), storage root fresh weight, storage root dry weight maintained highly significant positive correlations with storage root K accumulation and total K accumulation. Root-shoot ratio demonstrated highly positive correlations with K harvest index, storage root K utilization efficiency and plant K utilization efficiency, while showing negative correlations with shoot K accumulation value. Shoot K content and total K content demonstrated significant negative correlations with plant K utilization efficiency. Shoot K accumulation value and total K accumulation value exhibited significant negative correlations with K harvest index, storage root K utilization efficiency and plant K utilization efficiency. Additionally, K harvest index demonstrated highly significant positive correlations with storage root K utilization efficiency and plant K utilization efficiency.

To enhance the robustness of the field validation, eight selected germplasms were incorporated from the initial screen. Field screening revealed an average yield reduction of 19.74% under LN (Table 4). Based on the yields of germplasms under HN, LN and their average value, the germplasms could be classified into four categories: EE type, including XN1985–7 and XN1878-7, which yield exceeded the average yield under both HN and LN; EI type, including XN2154–1 and XN2155-1, which yield exceeded the average yield under LN, but lower than the average yield under HN; II type, including XN17104–132 and XN2153-1, which yield was lower than the average both under HN and LN; IE type, including XN2153–5 and XN2141-3, which yield was higher than the average under HN but lower than the average under LN. Notably, XN1878–7 exhibited a superior N sensitivity index, yield increase rate, and agronomic use efficiency, thereby confirming its classification as low N-tolerant and high N- efficient type. Conversely, XN2153–1 was identified as low N-sensitive and inefficient germplasm.

Field trials (Table 5) indicated an average yield decline of 16.98% under LP conditions. Likewise, the germplasms could be categorized as EE (XN2153-5, XN2141-3), EI (XN2153-1, XN2155-1), II (XN17104-132, XN2154-1), and IE (XN1878-7, XN1985-7) types. Notably, XN1985–7 exhibited the highest values across all P efficiency indices, including relative value, P sensitivity index, yield increase rate, P agronomic use efficiency and P contribution rate. Integrated analysis of both seedling-stage and field experiments identified two germplasms, XN2153–5 and XN2141-3, which demonstrated superior low P tolerance and high P efficiency, rendering them ideal germplasm resources for investigating P utilization mechanisms in sweetpotato.

K treatments (Table 6) indicated an average of 8.97% reduction in yield under LK conditions. The germplasms also could be categorized into four groups: EE (XN2153-5, XN17104-132), EI (XN1985-7), II (XN1878-7, XN2154-1, XN2155-1, XN2153-1), and IE (XN2141-3). XN17104–132 demonstrated the highest values across K efficiency indices, including K agronomic use efficiency, K partial factor productivity, K utilization index, and low K tolerance index. An integrated analysis of hydroponic and field experiments identified XN2153–5 as a low K-tolerant and K-efficient germplasm, while XN2153–1 was characterized as a low K sensitive and K-inefficient genotype. These contrasting phenotypes provide ideal materials for deciphering K utilization mechanisms in sweetpotato.

Comprehensive analysis of the two phases (hydroponic screening and field trial) results revealed significant differentiation in nutrient utilization characteristics among the tested germplasm resources (Supplementary Table S9). XN2153–5 showed tolerance to low N, P, and K conditions at the seedling stage, and its good yield performance under LN, LP and LK also confirmed its low-N/P/K tolerance. XN1985–7 was identified as N-efficient type during two phases and could be identified as low N- tolerant and N-efficient genotype. XN2141–3 showed N/P/K-tolerance at seedling stage, and P-efficiency under the two phases, suggesting it’s a low P-tolerant and P-efficient genotype. XN17104–132 showed N/P-inefficiency but exhibited low K-tolerance and efficiency, and showed the highest K agronomic use efficiency, partial factor productivity, utilization index, and low-K tolerance index, and could be identified as a K-efficient genotype.

XN2153–1 was identified as a low N-, P-, and K-intolerant and II genotype during the seedling stage, and field experiments further confirmed its low N/P/K utilization index and low-N/P/K tolerance index. Therefore, XN2153–1 could be identified as a low N-, P-, and K-intolerant and -inefficient germplasm. In contrast, XN2153–5 exhibited outstanding comprehensive traits. This germplasm was consistently classified as a low N-, P-, and K-tolerant and EE type at the seedling stage. Particularly under LP conditions, its yield reached 1240 kg·ha^-1, significantly exceeding that of other germplasm resources. Although its yield was relatively low under LN, it showed superior N agronomic use efficiency, partial factor productivity and utilization index, indicating robust N utilization efficiency. Thus, XN2153–5 could be identified as a low N-, P-, and K-tolerant and nutrient-efficient genotype.