Strains, media, and culture conditions

The strain A. pullulans CCTCC M2012223 was isolated by our laboratory and can be obtained from the China Center for Type Culture Collection (Wuhan, China). This strain was maintained on potato dextrose agar slants. The seed culture medium contained 60-g/L glucose, 2-g/L NH₄NO₃, 0.1-g/L KH₂PO₄, 0.1-g/L MgSO₄, 0.1-g/L ZnSO₄, 0.5-g/L KCl, and 20-g/L CaCO₃. The seed culture was grown in a 500-mL shake flask containing 50-mL liquid medium and incubated at 25 °C on a rotary shaker (180 rpm) for 2 days. The fermentation medium contained 90-g/L sugar (glucose, sucrose, xylose, arabinose, or fructose), 2-g/L NH₄NO₃, 0.1-g/L KH₂PO₄, 0.1-g/L MgSO₄, 0.1-g/L ZnSO₄, 0.5-g/L KCl, and 20-g/L CaCO₃. Escherichia coli DH5α cells were employed for routine DNA manipulations. E. coli was grown in LB medium (5 g/L yeast extract, 10 g/L tryptone, 10 g/L NaCl, pH 7.0) at 37 °C, and kanamycin or ampicillin (50 mg/L) was added when required. Agrobacterium tumefaciens AGL1 was grown on YEB medium (10-g/L yeast extract, 5-g/L tryptone, 5-g/L sucrose, 0.5-g/L MgSO₄·7H₂O, pH 7.0) at 28 °C, and kanamycin or carbenicillin (50 mg/L) was added when required.

Sampling, quenching, and extraction of intracellular metabolites

The cells of strain A. pullulans CCTCC M2012223 were collected after culturing for 48 and 72 h with glucose- and sucrose-based fermentation, respectively, and intracellular metabolism was quenched by immediately adding five volumes of prechilled 60% (v/v) methanol. After quenching at − 40 °C, cells were pelleted in a centrifuge (4000×g, 4 °C, 10 min). The pellets were washed with phosphate-buffered saline (pH 7.4), frozen in liquid nitrogen, and stored at − 80 °C. Next, 60 mg per sample with six biological replicates was ball-milled into a fine powder under frozen conditions with 1 mL of extraction mix consisting of deionized water (50%), methanol (50%), and the internal standards nonadecanoic acid (0.2 mg/mL) and isotope alanine (10 mM), and then lyophilized. The samples were derivatized by vortexing with 60 μL of 15-mg/mL methoxypyridine hydrochloride for 30 s and then incubating for 2 h at 37 °C. The samples were then trimethylsilylated by adding 60-μL bis(trimethylsilyl) trifluoroacetamide (BSTFA) with 1% trimethylchlorosilane and incubating for 90 min at 37 °C. Subsequently, an extract containing all intracellular metabolites was collected via centrifugation at 12,000×g for 10 min at 4 °C [22].

Metabolite detection with gas chromatography–mass spectrometry (GC–MS)

GC–MS analysis was performed to detect the metabolites in the samples. Briefly, 1 μL sample was injected into an Agilent 7890A/5975C GC–MS system (Agilent Technologies, CA, USA) with a 20:1 split injection ratio. The system was equipped with an HP-5MS capillary column (5% phenyl methyl silox: 30 m × 250 μm i.d., 0.25 μm; Agilent J&W Scientific, Folsom, CA, USA). The inlet temperature was maintained at 280 °C, the interface temperature was 150 °C, and the ion source temperature was 250 °C. The GC temperature program was as follows: hold at 70 °C for 2 min, ramp at 10 °C/min to 300 °C, and a final hold at 300 °C for 5 min. Helium was used as carrier gas at a constant flow rate of 1 mL/min, with a total run time of 30 min. MS conditions were as follows: electron ionization (EI); full-scan mode (m/z 35–780); electron energy, 70 eV [23]. The chromatographic differences between sample groups were obvious, and the retention time was reproducible and stable, demonstrating the reliability of metabolomics analysis.

Data processing and statistical analysis

XCMS software was used to identify and quantify mass spectral peaks. Data preprocessing, including raw data filtering, peak detection, alignment, normalization, and identification, was performed automatically. Metabolites were identified by comparison of mass spectra with the National Institution of Standards and Technology (NIST), Wiley metabolome, Golm metabolome (GM), and in-house databases (Additional file 1: Table S1, red color) [24]. Six independent experiments were performed for each conclusion. Pattern recognition methods based on principal component analysis (PCA) and orthogonal-partial least squares-discriminant analysis (OPLS-DA) were then carried out using SIMCA-P V13.0 (Umetrics AB, Umea, Sweden). All variables were unit variance scaled before PCA and OPLS-DA. Hierarchical clustering analysis was carried out using R software (version 3.1.3) to visualize and group metabolite profiles. A list of variance metabolites that contributed mostly to the model was generated from the P values obtained using Student’s t tests (P < 0.05) and variable importance in the projection values (> 1). In addition, the false discovery rate (FDR) significance criterion was performed with an FDR limit of 0.05 to avoid false-positive results in analyzing differential metabolite correlations. Metabolic pathways were constructed according to pathway analysis of the Metaboanalyst and KEGG metabolic databases [25, 26].

In silico analysis of a genome-scale metabolic model

The reconstructed genome-scale metabolic network (iZX637) of A. pullulans, containing 1347 reactions and 1133 metabolites, was used for constraint-based flux analyses [27] to determine the most effective route for maximizing PMA production and minimizing the influence of biomass accumulation. The synthesis of PMA is closely related to MA synthesis in A. pullulans. Therefore, three routes of MA synthesis, the oxidative branches of the TCA cycle, the reductive branches of the TCA cycle, and the glyoxylate shunt, were examined. In silico, constrained reactions of the oxidative branches of the TCA cycle contained oxoglutarate dehydrogenase, aconitate hydratase, fumarase, succinate dehydrogenase, and succinate-CoA ligase. Constrained reactions of the reductive branches of the TCA cycle contained malate dehydrogenase, and constrained reactions of the glyoxylate shunt contained fumarase, isocitrate lyase, and glyoxylate lyase. For in silico simulation, the reaction fluxes corresponding to one route of MA synthesis (the oxidative branches of the TCA cycle, the reductive branches of the TCA cycle, or the glyoxylate shunt) were constrained from 0 to 140 mmol/gDW/h (the reaction fluxes of other two routes of MA synthesis were set to 0 gDW/h), and the cell growth rate was constrained from 0 to 1 gDW/h. The specific PMA synthesis rate was then maximized as an objective function and the substrate uptake rates for glucose in the simulation were constrained by 10 mmol/gDW/h. The specific PMA synthesis rate at different flux values of the cell growth rate and route of MA synthesis were plotted on the Z-axis of the flux profile graph. In addition, because the reductive branches of the TCA cycle were assumed to be located only in the cytoplasm in silico, the flux of malate dehydrogenase located in the mitochondria was set to 0 mmol/gDW/h in simulation of three routes. The basic tools used for the model analysis were flux balance analysis (FBA) [28]. GLPK (freeware) (http://www.gnu.org/software/glpk/) was used as a linear programming solver. The COBRA Toolbox 2.0 [29] was used for calculation and analysis in the MATLAB environment.

Gene cloning and expression in the host strain

The DNA of A. pullulans was extracted using a TIANamp Yeast DNA Kit (TIANGEN, China) and was used as a template to amplify the pyruvate carboxylase (pyc, EC: 6.4.1.1) gene. First, one 3.748-kb pyc fragment was amplified using SmaI-pyc-S and MunI-pyc-A, as listed in Additional file 1: Table S1. Second, the pyc fragment obtained was double digested with SmaI and MunI and then ligated into plasmid pBARGPE1, which was also treated with SmaI and EcoRI (MunI isocaudarner) to generate the plasmid pBARGPE1-pyc, which contained a gpdA promoter (PgpdA) and a trpC terminator (TtrpC) flanking the pyc gene. The PgpdA-pyc-TtrpC cassette was amplified by primers PgpdA.pyc.TtrpC-S and PgpdA.pyc.TtrpC-A. The PgpdA-pyc-TtrpC cassette was cloned into the EcoRI site of the pK2-PtrpC-hyg-TtrpC plasmid using a One Step Cloning Kit (Vazyme, USA), forming the plasmid pK2-hyg-pyc for A. tumefaciens-mediated transformation (ATMT), as described in our previous study [30]. T-DNA insertion sites were analyzed using a Genome Walking Kit (TaKaRa, Japan) in ATMT-derived clones. Polymerase chain reaction (PCR) primers were used to amplify the T-DNA flanking sequence (Additional file 1: Table S1).

Batch and fed-batch fermentation in a 5-L stirred-tank fermentor

Batch fermentation kinetics were studied in a 5-L stirred-tank fermentor (Shanghai Baoxing Co., Ltd., China) containing 3.5 L of fermentation media (with glucose as carbon resource) or only diluted sugarcane molasses. The fermentation medium was inoculated with 350 mL of the seed culture grown in a shake flask for 48 h and operated at 25 °C with agitation and aeration at 400–700 rpm and 0.8 vvm, respectively. During fermentation, the agitation was controlled to maintain the dissolved oxygen level at greater than 20%. For sugarcane molasses fermentation, the crude sugarcane molasses (containing ~ 420 g/kg of total sugar and 14.29 ± 0.27 g/kg of total nitrogen) were diluted by water to ~ 90 g/L total sugar. For fed-batch fermentation, when the total sugar concentration was lower than 30 g/L, sugarcane molasses (approximately 270-g/L reducing sugars) was continuously added to maintain the total sugar concentration at approximately 10–20 g/L. Broth samples were collected periodically for the analysis of residual sugar, biomass, and PMA titers. All trials were performed in triplicate.

Quantitative reverse transcription (RT)-PCR

Total RNA was extracted using a Fungal RNA Kit (Omega, USA) and reverse transcribed into cDNA using reverse transcriptase (TIANGEN). The quantitative PCR assay was performed using the SYBR Green method (FastQuant RT Super Mix; TIANGEN). Primers for the pyc gene are shown in Additional file 1: Table S1. The gene encoding β-actin was used as the reference gene.

Analytical methods

Biomass was determined by the dry cell weight method. Prior to measurement, excessive CaCO₃ was eliminated from the broth by adding 1-M HCl. The cell suspension was centrifuged at 4000×g and then dried overnight at 105 °C [31]. PMA was analyzed by centrifuging the fermentation broth and the resulting supernatant was mixed with an equal volume of 2-M H₂SO₄ in an incubator at 85 °C for 8 h. The hydrolyzed PMA sample was analyzed by high-performance liquid chromatography (Agilent 1260, USA) using a Spursil C18-EP organic acid column at 40 °C and eluted with 5-mM H₂SO₄ at a rate of 0.6 mL/min to determine its malic acid content [32]. The glucose concentration was measured using a fully automatic residual sugar analyzer (Biology Institute of Shandong Academy of Sciences, China).

Evaluation of carbon sources in shake-flask culture

Sugarcane molasses is the main by-product of sugar production, and mainly comprises sucrose and some glucose and fructose. In this study, fermentation with different carbon sources in shake-flask culture was investigated (Fig. 1). Of the carbon sources used, sucrose and fructose showed superior PMA production of 28.30 ± 2.58 and 22.58 ± 1.08 g/L, respectively, which was higher than that (20.38 ± 2.32 g/L) from glucose as the carbon source. Moreover, five-carbon sugar (arabinose and xylose) fermentation was also superior to that with glucose for PMA production. In our previous study, a novel PMA-producing strain, A. pullulans YJ6-11, was also found to utilize xylose as a superior carbon source compared with glucose [33]. The fermentation kinetics of sucrose and glucose in shake-flask culture are shown in Fig. 2. A total of 33.91 ± 3.70 g/L PMA (~ 38.98 g/L MA after hydrolysis) were produced from ~ 90 g/L sucrose in 96 h via batch fermentation; this was increased by 40.7% compared with that of glucose (24.10 ± 3.30 g/L). These results showed that sucrose present in sugarcane molasses was the most suitable sugar for PMA biosynthesis. However, the underlying factors responsible for the differences between sucrose and glucose were still unclear.

Analysis of key metabolites and pathways associated with PMA biosynthesis

To explain the differences in metabolites of A. pullulans, the metabolic states of sucrose- and glucose-based fermentation were compared by metabolomics analysis. As shown in Additional file 2: Table S2, in total, 81 intracellular metabolites were identified and divided into eight categories. These groups mainly included organic acids, amino acids, and sugar metabolism that contained 21, 15, and 14 metabolites, respectively. Among these metabolites, compared with culture at 48 h, 13 and 3 differential metabolites were increased at 72 h in glucose and sucrose, respectively. 27 and 30 differentially abundant metabolites from glucose versus sucrose after culture for 48 and 72 h, respectively, were validated. Of these metabolites, both at 48 and 72 h, 16 in sucrose fermentation, including fructose, ribose, glycolic acid, oleic acid, and heptanoic acid, were higher than that of glucose (Additional files 3 and 4: Figure S1 and Table S3). These results indicated that the differences of metabolites were varied in the different sugar fermentation. Moreover, compared with glucose fermentation at 48 h, 13 differential metabolites were increased at 72 h. Notably, at 72 h, citric acid was increased by 1.84-fold, whereas other intermediate metabolites associated with the TCA cycle (such as pyruvic acid, succinic acid, and fumaric acid) did not differ significantly at 48 h (Additional file 4: Table S3). However, at 72 h, MA was decreased by 1.04-fold compared with that at 48 h, which was negative correlated with citric acid (Fig. 3). This may induce a decrease in PMA biosynthesis, because accumulation of citric acid resulted in carbon flux that was not efficiently channeled to the objective product. Pyruvate is a key node in central carbon metabolism and plays a major role in pathways related to organic acid synthesis [34]. The carbon flux can be channeled into different products through pyruvate, including succinate [35], lactate [36], and malate [37]. The reductive TCA cycle has been identified as the most efficient pathway for many products, such as succinate [38] and malate [37]. Thus, the accumulation of citric acid also indicated that metabolic flux could not effectively direct pyruvate to PMA synthesis or the high-efficiency PMA synthesis route in native metabolic regulation. Therefore, the enhanced carboxylation of pyruvate to oxaloacetate, catalyzed by pyruvate carboxylase, may be an important target for improving PMA synthesis.

In addition, pyruvate is also a major component in pathways related to amino acid metabolism. Amino acids are crucial to microorganism metabolism and have roles in synthesis of proteins (e.g., leucine, threonine, and phenylalanine), regulation of gene expression (e.g., arginine and leucine), osmoregulation (e.g., citrulline and proline), control of enzyme activity (e.g., alanine and leucine), transamination (e.g., aspartate, alanine, and glutamate), and ammonia detoxification (e.g., citrulline, arginine, and glutamate) [39]. As shown in Fig. 3, amino acids present at higher levels, particularly leucine and proline, could be directly or indirectly transformed from pyruvate and α-oxoglutarate [40]. Most amino acids were observed at higher levels in sucrose than in glucose because of the abundant precursors in the glycolytic pathway and TCA cycle. The redox balance is maintained by modifying the distribution of the metabolic flux of pyruvate [41].

In addition, when sucrose was used as the carbon source, relative concentration of trehalose was significantly decreased 2.37- and 2.70-fold at 48 and 72 h, respectively, compared with that when glucose was used as the carbon source (Additional file 4: Table S3). Accumulation of trehalose in yeasts has been shown to be an important mechanism mediating tolerance against adverse stress conditions, and trehalose functions as a reserve carbohydrate that can be synthesized when the exogenous energy exceeds the cellular needs for growth and biosynthesis [42, 43]. These results suggested that glucose as a rapidly metabolizable carbon source can induce a regulatory response (e.g., osmotic stress) to restore excess exogenous energy, which would cause some of the carbon source to be used for other processes, not PMA synthesis, thereby decreasing PMA titers.

In silico analysis of a genome-scale metabolic model

PMA is synthesized from MA, which is an intermediate in the TCA cycle in aerobic metabolism. Genome-scale flux sensitivity analyses were performed to determine the most effective route for maximizing PMA synthesis. Three metabolic routes involved in MA synthesis, including oxidative branches of the TCA cycle, reductive branches of the TCA cycle, and glyoxylate shunt, were simulated using the genome-scale metabolic model (iZX637). As shown in Fig. 4, the optimal PMA synthesis rate, using glucose or sucrose as the carbon source, reached 15 mmol/gDW/h (indicated by an arrow), increasing the reductive TCA malate synthesis rate to 15 mmol/gDW/h and the biomass growth rate to 0 mmol/gDW/h. When the carbon flux turned to oxidative branches of the TCA cycle and the glyoxylate shunt, the optimal PMA synthesis rates reached only as high as 12.89 mmol/gDW/h (sucrose as a carbon source: 12.58 mmol/gDW/h) and 14.67 mmol/gDW/h (sucrose as a carbon source: 14.33 mmol/gDW/h) with corresponding malate synthesis rates of 13 and 8 mmol/gD/h, respectively. In addition, the malate synthesis rate of the corresponding route constantly increased, resulting in a negative influence on the optimal PMA synthesis rate. When the malate synthesis rates of the reductive TCA route, oxidative TCA route, and glyoxylate shunt reached 120, 37, and 37 mmol/gDW/h, respectively, the PMA synthesis rate and biomass growth rate were reduced to 0 mmol/gDW/h (glucose or sucrose as a carbon source). Furthermore, the PMA synthesis rate decreased gradually as the biomass synthesis rate increased for all simulation results. These results showed that PMA and biomass synthesis showed a competitive relationship. However, in the reductive TCA route, biomass synthesis could be maintained with a relatively high PMA synthesis rate compared with the other two routes, regardless of whether glucose or sucrose was used as the carbon source. In our previous study, we found that high levels of the nitrogen source favored cell growth but decreased PMA synthesis [44]. Therefore, the efficiency of the PMA synthesis route should balance cell growth and PMA formation. In our previous study, in silico analysis of the carbon flux distribution and changes in PMA synthesis rates showed that, in the context of a high PMA synthesis rate, a large amount of carbon flux was transformed into pyruvate and channeled into the reductive TCA route by pyruvate carboxylase [27]. These in silico results showed that the reductive TCA route, as the optimal PMA synthesis route, showed a relatively high optimal PMA synthesis rate and mitigated the competitive relationship between PMA and biomass synthesis. In addition, the simulation results also showed that the reductive TCA route required a relatively high route flux (the reductive TCA malate synthesis rate to 15 mmol/gDW/h), yielding a high PMA synthesis rate (the optimal PMA synthesis rate to 15 mmol/gDW/h). Therefore, pyruvate carboxylase (pyc) in the reductive TCA route was selected as the target to be overexpressed to improve PMA yield and productivity.

Moreover, according to the simulation result from sucrose and glucose as carbon sources, sucrose was not dominant compared with glucose as a carbon source for the optimal PMA synthesis rate and cell growth by A. pullulans strains in the conversion efficiency of metabolic pathways. This could be explained by the observations that the reaction of sucrose hydrolysis exhibited no energy consumption or carbon loss in silico.

Overexpression of the pyc gene in A. pullulans

To evaluate the effects of the pyc gene on PMA biosynthesis, the endogenous pyc gene was amplified using the genomic DNA of A. pullulans as a template (Additional file 3: Figure S2). The PgpdA promoter of Aspergillus nidulans, which directs the constitutively high expression of the glyceraldehyde-3-phosphate dehydrogenase (gpd) gene, can be used to induce high levels of constitutive expression of genes of interest or marker genes and to drive the overexpression of various genes in fungi [45–47]. Therefore, the PgpdA promoter of A. nidulans was used to constitute the pyc cassette by cloning into pBARGPE1 (Additional file 3: Figure S3). In our previous study, we developed a simple and efficient system for genetic transformation of A. pullulans using A. tumefaciens [30]. Therefore, A. tumefaciens carrying the pK2-hyg-pyc binary plasmid was used to transform A. pullulans. Because ATMT-mediated integration is random, five different ATMT-derived clones were tested for PMA production in shake-flask culture. The PMA titers of five clones compared with the wild-type strain were improved from 3.5 to 7.1% (Additional file 3: Figure S4). The differences among mutants were due to random integration, causing interruption of gene function by insertion of T-DNA, and were not caused by intrinsic factors [30]. This result indicated that the increased PMA titer was mainly dependent on pyc gene overexpression. Thus, we selected the highest PMA producer (E10 strain; as called FJ-PYC) for further analysis of the sequences flanking the T-DNA in the genome. As shown in Additional file 3: Figure S5, the hygromycin B resistance cassette and overexpression pyc cassette were inserted into the noncoding regions of g4940.t1 and g4941.t1 on the A. pullulans genome. The insertions did not disrupt the g4940.t1 and g4941.t1 open reading frames. Furthermore, we found that the expression level of the pyc gene in strain E-10 was increased by 9.5-fold compared with that in the wild-type strain (Fig. 5). These results also revealed that the promoter PgpdA was an effective tool for regulating gene expression in A. pullulans.

Subsequently, the engineered strain E10 was tested with glucose as the carbon source in a 5-L stirred-tank fermentor. As shown in Fig. 6, about 36.2 g/L of PMA was produced via batch fermentation, with a PMA productivity of 0.53 g/L h, which was increased by 15.1 and 12.7% compared with the control, respectively. This result indicated that overexpression of the pyc gene could improve the production of PMA, which was consistent with the results of in silico simulation showing that strengthening of the reductive TCA route resulted in increased carbon flux toward PMA synthesis.

Batch and fed-batch fermentation with sugarcane molasses

Sugarcane molasses is a low-cost by-product of sugar production and mainly contains sucrose, with some fructose and glucose. To develop an economical fermentation process, sugarcane molasses without any pretreatment or nutrient supplementation was employed for PMA production by FJ-PYC strain in a 5-L stirred-tank fermentor. As shown in Fig. 7a, during the early fermentation stage, sucrose was rapidly consumed and transformed into glucose and fructose, and glucose was then used for cell growth and PMA synthesis compared with fructose. This phenomenon, called glucose-mediated carbon catabolite repression (CCR), is widely observed in bacteria and yeasts [48–50]. About 31.5-g/L PMA (36.5-g/L MA after hydrolysis) was produced from ~ 95-g/L mixed sugar (61.8-g/L sucrose, 18.5-g/L glucose, and 15.3-g/L fructose from sugarcane molasses) in 60 h via batch fermentation, with an MA yield of 0.44 g/g. Moreover, without adding any other media, the productivity of PMA (0.53 g/L h) was the same as that from glucose fermentation. In fed-batch fermentation (Fig. 7b), a high final PMA titer of 81.5 g/L (94.2-g/L MA after hydrolysis) was achieved in 140 h with a corresponding MA yield of 0.62 g/g and productivity of 0.67 g/L h. As shown in Table 1, compared with other renewable feedstocks, PMA fermentation from sugarcane molasses by strain FJ-PYC resulted in the highest titer and productivity. Nevertheless, the titer and productivity could be further increased by continuous fed operation, because the PMA synthesis rate maintained a high speed. In addition, the fed-batch fermentation with FJ-PYC resulted in an overall MA yield of 0.62 g/g from sugarcane molasses without nitrogen supplementation, which was the same as the highest yield from sugarcane juice [49]. However, sugarcane juice contained only ~ 15% reducing sugars and relatively fewer inhibitors, requiring it to be condensed for high-sugar PMA fermentation and dramatically decreasing its economic efficiency. These results indicated that the strain FJ-PYC exhibited extremely high adaptation and tolerance against sugarcane molasses.

Microorganism	Substrates	Nitrogen sources	Operating mode	PMAa (g/L)	Malic acid (g/L)	Productivitya (g/L h)	Yielda (g/g)	References
NRRL 50383	Corn fiber	Peptone and yeast extract	Batch	10.1	11.7b	0.07	–	[13]
	Wheat straw	Peptone and yeast extract	Batch	23.5	27.1b	0.16	–
ZX-10	Soybean hull hydrolysate	Corn steep liquor	Fed-batch	27.2	31.3	0.48	0.42	[12]
	Soy molasses	–	Fed-batch	62.6	71.9	0.29	0.69
	Sugarcane juice	–	Batch	52.6	60.8	0.32	0.62	[49]
CCTCCM2012223	Hydrolysate of raw sweet potato	NH4NO3	Batch	29.6	33.6	0.28	0.31	[9]
	Hydrolysate of raw sweet potato	NH4NO3	Fed-batch	44	49.9	0.31	0.22
YJ 6-11	Corncob hydrolysate	NH4NO3	Batch	28.6	32.4	0.45	0.41	[33]
FJ-PYC	Sugarcane molasses	–	Batch	31.5	36.5	0.61	0.44	This study
		–	Fed-batch	81.5	94.2	0.67	0.62

-, none or not reported

^aTo facilitate comparisons, PMA yield and productivity were based on the malic acid that can be released from PMA after hydrolysis, PMA (g/L) = 0.87 malic acid (g/L)

^bCalculated from data in this study