High-throughput screen design
The nutrient screen was designed to provide a miniaturised, automated, high-throughput platform for rapid low-cost optimisation of nutrient conditions. The 96-well microwell plate format was chosen as a basis for the design as it provides future flexibility to expand sample scaling (for example, 384-, 96-, 48-, 24-, 12- and 6-well) as well as array scalability (for example, from the current eighteen 96-well plates = 1,728 wells) while achieving acceptable errors associated with miniaturisation.
Optical density (OD750), which is a measure of light scattering, was used as a proxy for biomass and to determine microalgal growth rates [29]. It was chosen as it is a standard measure of growth kinetics and is highly correlated with biomass yield. Furthermore, it provides the required precision and accuracy for this broad screen and is cheap, simple and suited for automation. The use of OD750 eliminates the effects of varying chlorophyll content of the cells. Although OD750 solely measures light scattering and so does not differentiate between algae, bacterial or fungal biomass, detritic compounds or algal exudates, the use of axenic algae cell cultures eliminated most of these complications. Subsequent further precision testing of high-performance conditions identified during the screen can be conducted to deliver higher precision if required. However, as OD750 is based on light scattering, it is influenced by cell size. For the purposes of this broad screen, it is therefore necessary to assume that cell size does not vary significantly over the duration of the experiment. This is clearly an approximation. However, the screens were species specific (measuring intra-species variance related to nutrient effects) and the average cell size of most algae species do not vary more than two- to threefold in diameter throughout the growth cycle. Unless synchronised, this is the case for most populations containing a mixture of cell sizes [30]. Cultivation was conducted for 75 h or less to minimise evaporation effects. A second-generation system design could potentially incorporate additional checks for accuracy of biomass estimation.
Automated media preparation
A Tecan robot (Freedom Evo 150, Tecan Group Ltd., Männedorf, Switzerland) (Figure 1a) equipped with a liquid handling arm (1) was used to accurately dispense the nutrient screen matrix stock solutions into 1,728 wells. The liquid handling arm (1) dispenses the stock solutions from a set of 100-mL troughs (2) into eighteen 96-well plates placed on two platforms (3) to generate the nutrient blend matrices required for growth trials. A large trough (4) located between the troughs (2) and platform (3) was used to wash the tips between the dispensing steps of different nutrients. The dispensed nutrient plates were then gamma sterilised at a dose of 2 k-Gy. Microwell plates containing the media were wrapped in cling film and were stored at −20°C until used. A filter sterilised vitamins B1 and B12 solution (Acrodisc 0.22-μm filter) was added to the gamma sterilised media together with the microalgae inoculum (see Additional file 1: Figure S2, S3 and Additional file 1: Table S2).
Automated growth chamber
System layout
A second Tecan system (Tecan Freedom EVO 150 robotic workstation, Tecan Group Ltd., Männedorf, Switzerland) was configured and further developed into an automated microalgae growth chamber (Figure 1b). Specifically, it was fitted with three orbital shakers (IKA KS 130 Control microwell plate shakers, IKA Werke GmbH & Co., KG, Staufen, Germany), each holding six 96-well microwell plates (5), enabling the use of a total of eighteen 96-well plates (1,728 samples). The system was operated at room temperature and during the experiment remained within a range of 23°C ± 0.5°C.
Illumination
Controlled top illumination (Figure 1b (6)) and bottom illumination (7) have been integrated into the system, with capacity for both continuous illumination and day/night cycling. The top illumination system was designed to closely match the visible part of the solar spectrum. It consists of alternating fluorescent lights (12 Cool white Phillips PL-L55W/840 Cool White, Philips International B.V., Amsterdam, Netherland, and 11 Philips PL-L55W/830 Warm White lights, Phillips International B.V., Amsterdam, Netherland). The fluorescent light sources extend beyond the whole cultivation area and were positioned over approximately 1.5 m above the microwell plates, to ensure even illumination. Uniformity of illumination across the full cultivation area was confirmed through detailed light meter measurements and achieved a maximum light intensity of 450 μmol photons m−2 s−1 at the microwell plate level.
Below the microwell plates, a customised diode array lighting system was also fitted (see Figure 1b insert). This illumination system positioned one light-emitting diode (SMD 3020, Epistar, Hsinchu City, Taiwan) below each well of each 96-well plate (LEDs are rated to +/−5%). The maximum illumination intensity is approximately 3,000 μmol photons m−2 s−1 and can be adjusted between 0% and 100% of maximum intensity in 1% increments. This ability to vary light intensity enables ‘dynamic’ day/night cycling. Programs coded in Arduino© (Arduino SA) provide the ability to run: (1) A fixed light cycle, (2) A day/night cycle with light flux changing at manually set time increments (for example, 5% every 30 min to a maximum or minimum level) to simulate outdoor solar conditions and (3) a rapid flashing light cycle to simulate mixing of cells in photobioreactors (maximum cycle speed is 10 ms−1). The top and bottom illumination systems can be used individually or in combination.
CO2 control
The growth chamber was also fitted with an atmospheric CO2 control system (Get Red-y 5 system, Voegtlin Instruments AG, Aesch, Switzerland). Specifically, two thermal mass flow controllers (Red-Y Smart Controllers, Voegtlin Instruments AG, Aesch, Switzerland) were fitted to regulate the mass flow of air and CO2 into the chamber based on the measured CO2 concentration. The CO2 concentration was measured using the CO2 probe (CARBOCAP® GMT 220 CO2 probe, Vaisala, Oyj, Finland) shown in Figure 1b (8). To minimise the use of CO2 required to maintain a stable 1% enriched atmosphere, a specifically designed low wall mounting (9: dimensions: 110 cm × 45 cm × 13 cm) was fitted around the shakers. The volume within it (approximately 65 L) is approximately 11 times less than the total volume of the entire Tecan enclosure (dimensions 115 cm × 130 cm × 50 cm) and, as it does not have a top, does not interfere with the light path from the top lights. A stable 1% ± 0.3% CO2-enriched atmosphere could therefore be maintained much more precisely and with a reduced CO2 requirement by flooding the 1% CO2 mix into the bottom of the enclosure via a looping perforated tube system.
Time course assays
A robotic manipulator arm (10 - Tecan, ROMA) was fitted to transfer the plates to a plate reader after removal of the lid (11 - Tecan Infinite M200 PRO, Tecan Group Ltd., Männedorf, Switzerland) to measure optical density at defined intervals (typically every 3 h).
Algae growth media variations for the screening
The ‘midpoint’ and elemental screen range of the screen was based on an extensive literature search and the average values obtained. In total, 11 different fresh water media (TAP medium [31], HSM medium [32], Johnson medium [33], Bristol medium [34], Botryococcus medium [35], Spirulina medium [35], M4N medium [36], Modified Bold 3 N [35], Del Río medium [37], BG11-1 medium [38] and Modified BG11 medium [39]) were analysed, and their elemental compositions are compared in Additional file 1: Table S3. NaNO3 and NH4Cl were found to be the most common nitrogen sources. In this screen, urea ((NH2)2CO), a common and cheap fertiliser, and ammonium nitrate (NH4NO3), which provides an opportunity for microalgae to dynamically switch N sources during growth, were also tested.
For microelements, the Hutner’s trace formulation [40] was modified by the inclusion of selenium, vanadium and silicon. Other elements have been included because many elements are not essential but beneficial for growth and to make the screening systems applicable to a broad variety of microalgae strains, such as diatoms. The average nutrient concentration based on these 11 media was used as the average values for Screen 1 and the initial mid-value for the Screen 2 system. It was noted that average concentration values derived from the literature search analysis may not be optimal but provided a sensible starting point for optimisation. Solubility constants of each element were examined to ensure that the formulation did not induce precipitation.
Careful formulation of the microelements was crucial to produce accurate and sensible information from the nutrient screen systems for application to the larger scale systems such as bioreactor and open pond systems. Selenium (0.1 μM) [14], vanadium (0.009 μM) [38], silicon (273 μM) [38], vitamin B1 (52 μM) [15,41] and vitamin B12 (0.1 μM) [15,42] were used as a baseline of both screens in addition to the Hutner’s trace elements [40] and concentrations (Table 1) used for TAP media [31]. In addition, 0.5373 mM Na2-EDTA, pH 8.0 (chelating agent) and 100 mM Tris-HCl (pH 7.4) buffer are added to the formulation (concentrations derived from range finding experiments, data not shown). Extensive preliminary trials were conducted to monitor optical density changes of the screen media over an experimental run period to ensure that no salt precipitation occur that could contribute to increased measured OD. Given this and to maximise the efficiency of statistical design, blank wells were not included in the runs. This is however optional.
To optimise the efficiency of the screen statistically (that is, to maximise the multidimensional search space and minimise sample number), the screen was configured into a two stages process (Screen 1 and Screen 2).
Screen 1 - N and P optimisation
Screen 1 was designed to identify the best N type and concentration tested (Figure 2), and these are based upon the average literature values (Additional file 1: Table S3) and the concentration ranges listed in Table 1. The rationale for this approach is that different algae have different N preferences and that the effects of N and P are so important that without their initial optimisation, the statistical influences of the other elements on algae growth will be masked. For example, ammonium requiring algae would show very low growth in nitrate based media.
The nitrogen (N) source concentration was adjusted to account for the number of N atoms in the source (for example, NaNO3 = 1, NH4NO3 = 2). A chelating agent (0.5373 mM Na2-EDTA, pH 8.0) and a buffer (100 mM Tris-HCl, pH 7.4) were added to the formulation (concentrations derived from range finding experiments - data not shown). It is recognised that such high levels of EDTA and Tris-HCl would not likely be suitable for subsequent scale up cultivations; however, they are required here to ensure pH stability and to prevent precipitation in a miniaturised system that cannot be controlled in an automated fashion as in scaled up photobioreactors.
The full factorial design of Screen 1 investigates the effect of the four different nitrogen sources and one phosphorous source at five and three concentration levels, respectively (Table 1), for each algal strain in the test. In total, Screen 1 consists of 60 different photoautotrophic conditions and 3 positive photoheterotrophic controls (TAP media).
Nutrient Screen 2 uses the best N and P conditions from Screen 1 and is based on the statistical incomplete factorial Box-Behnken design. It is designed to measure the effects of Ca, Mg, Fe, Mn, Cu, Zn, B, Se, V and Si on microalgal growth performance. The elements were tested at three concentration levels coded as −1 (low), 0 (middle) and +1 (high). Other nutrients were supplied at constant concentrations. These consisted of CoCl2, (NH4)6Mo7O24, Na-EDTA (pH 8), Tris-HCl (pH 7.4) and vitamins B1 and B12 (Table 1) which excludes them from being tested variables in the current screen configuration. These elements, though not a complete set of nutrients at this stage, were considered to be the most critical for initial testing of a broad range of species. The low and high concentration levels for each of the nutrient elements were set as a twofold difference from the middle concentration (Table 1). The Box-Behnken experimental design allows the observation of primary effects and nutrient interaction effects on microalgal growth to be determined and presented via response surface analysis [11]. Minitab 15 software (Minitab Inc., State College, PA, USA) was used to design the experiment and generated 180 different media formulations (experiments) (Additional file 1: Table S4 ).
The three-level second-order response surface model for m factors (x1,…, x
m
) in n runs is described by Equation 1 [43].
$$ y=X\beta +\varepsilon $$
(1)
y = the n × 1 response vector
n = number of runs (equals number of concentrations tested)
X = n × p model matrix with n 1 × p row vectors
x = (1, x
1,…, x
m, x
1
x
2,…, x
m−1
x
m
, x
1
2,…, x
m
2)
m = number of factors (here 10)
x
m
= growth rate of factor m
β = p × 1 vector of parameters (to be estimated)
ε = n × 1 vector of errors (with zero mean and covariance matrix I
n
σ
2)
Screen format
A total of 24 species (or 8 species in triplicates) can be analysed in a single Screen 1 run, and a total of 9 species (or 3 species in triplicates) in 180 conditions can be analysed in Screen 2. All nutrient elements were prepared as individual stocks. Both screens include a triplicate photoheterotrophic/mixotrophic growth condition controls in TAP media to compare between runs of the same strain (quality control) as well as to photoautotrophic growth conditions. Microalgal growth rates in media containing acetate as additional carbon source (TAP controls) are expected to be higher than rates in photoautotrophic growth conditions using CO2 as sole carbon source.
Growth rate determination
Assuming that the specific growth rate μ (h−1) represents the average growth rate of all cells present in the culture, it defines the fraction of increase in biomass over a unit of time and is proportional to the biomass of the cells during exponential growth phase (Equation 2). OD750 was used as the basis for maximum specific growth rates μ
max determination of each condition. These rates were used to compare different conditions within the nutrient screens for each algae strain. In general, batch culture growth phases can be divided into lag, exponential, linear and stationary phase with μ
max occurring in exponential phase.
$$ \mu =\left(\mathbf{ln}\;\mathbf{O}{\mathbf{D}}_{750\left(\mathbf{t}2\right)}\hbox{--} \mathbf{ln}\;\mathbf{O}{\mathbf{D}}_{750\left(\mathbf{t}1\right)}\right)\;/\;\left({t}_2-{t}_1\right) $$
(2)
μ = specific growth rate
OD750(t1) = OD750 at time
OD750(t2) = OD750 at time
t
1 = time 1 (h)
t
2 = time 2 (h)
High-throughput data processing requires a form of curve fitting that appropriately compensates for irregularities, such as circadian rhythm or scattering effects, to ensure a high comparability of different growth conditions (see Equation 3). Under optimal growth conditions, the microalgae growth curve from lag phase to stationary phase can be usefully described by a sigmoidal curve. Non-linear regression was used to normalise (curve fit) the recorded 3-h OD750 data points to a simple sigmoidal model (variable sigmoidal curve fit (GraphPad Prism, GraphPad Prism Inc., La Jolla, USA)) (Equation 3). The sigmoidal curve fit approach was selected because it describes the usual physiological behaviour of the system where reasonable growth occurs. In non-ideal growth conditions where specific growth rates are low, a sigmoidal fit cannot be achieved (for example, linear growth with no plateau). Strong circadian rhythms in some algae can also interfere with curve fitting (Figure 6). Under these conditions, the growth rates were excluded from the screen.
$$ \mathrm{Y}=\mathbf{k}{\mathbf{t}}_{\mathbf{o}}+\left(\mathbf{k}{\mathbf{t}}_{\mathbf{n}}\mathbf{\hbox{--}}\mathbf{k}{\mathbf{t}}_{\mathbf{o}}\right)/\left(\mathbf{1}+\mathbf{1}{\mathbf{0}}^{\left(\mathbf{log}\ {t}_{\frac{1}{2}}\mathbf{\hbox{--}}t\right)*\mathbf{Hill}\;\mathbf{slope}}\right) $$
(3)
Y = normalised OD750 data xpoint,
kto = raw OD750 at time 0,
ktn = raw OD750 at time n,
log t
½ = log10 of time when the OD750 is between t
0 and t
n,
Hill slope = the steepness of the curve at t½.
Specific growth rates were then determined using sigmoidal fitted 3-h OD750 data points and calculating the slope of two consecutive data points (Equation 2). The highest slope value represents μ
max of a condition. Good growth conditions were selected by comparing all μ
max values for each strain.
The quality of the fit was assessed using R-square (where a value more than 0.85 was chosen to indicate good quality) and absolute sum of squares (value less than 0.1 was chosen to indicate good quality) (Figure 6). A data cutoff limit based on the R-square (R
2) value of the normalised growth plots was applied. This was designed to screen and remove fitted growth curves with R
2 smaller than 0.85. Growth curves that can be fitted accurately to the regression model have smaller sum of square regression (SSreg) than sum of square total (SStot). The curve fitting process first generates a number of possible curve fits to the raw data and then identifies the model yielding the highest R
2 and the least sum of square. In the situation in which only limited data can be fitted or the chosen model is too complicated, the Not Converged or Ambiguous remarks respectively are generated by the GraphPad Prism software [44].
Although the screen was validated through triplicate runs, each screen is designed to be conducted without replicates to broaden the screen by maximising the number of conditions and algae strains assayed per run. The screen is not intended to be highly analytically precise but to identify optimal regions of nutrient search space which can be analysed more precisely using conventional assays, while excluding the vast majority of conditions. Validation of well-to-well (repeatability test) and run-to-run reliability (reproducibility test) indicated excellent internal data consistency between replicate experiments (see Additional file 1: Figures S4, S5 and S6), in particular for high growth rate samples.
Statistical analysis on microalgal growth rate using response surface method
The Main and Interaction Effects analyses (response surface method (RSM)) were used to identify specific effects and statistical interactions between the nutrients as well as to determine the significance of nutrients that can improve microalgal growth.
Main and Interaction Effects analysis
The Main Effects analysis identifies the statistical significance of individual nutrients on the microalgae growth rate (Figure 4). The Interaction Effects analysis determines significant statistical interactions between multiple nutrient factors and their effects on microalgal growth rate. When the Main and Interaction Effects exhibited significance (p ≤ 0.05), the nutrient factor involved in these cases should be fine-tuned for growth performance improvement. The analysis is based on the average value of the growth rate of specific nutrient concentration level (−1, 0 and 1) in the changing background of 180 experiments conducted within the Box-Behnken matrix. In principle, important conditions will significantly influence the relative growth rates within these changing backgrounds.
The Interaction Effects analysis determines the synergistic or antagonistic effects of two nutrient elements on microalgal growth rates. Nutrient elements that exhibited statistical significant interaction effects could be subsequently evaluated or optimised on a rational basis to increase the microalgae growth rate. The analysis was based on the average value of the growth rate of specific nutrient concentration levels (−1, 0 and 1) from 180 experiments (Additional file 1: Table S4).
Algae strains and culture conditions
The microalgae strains M. inermum (18-1), A. gracilis (18-2), R. complanata (SF-150), C. sorokiniana (21), M. convolutum (9-FW), C. pyrenoidosa (22), M. reisseri (13), P. falcate (4A-1) were isolated in the vicinity of Brisbane, Queensland, Australia. Identification consisted of morphological investigation (Olympus BX42 and Nikon Ti-U, × 200 and × 400 magnification) and molecular classification by rDNA analysis (see Additional file 1: Figure S1 and Table S3). The amplification of 18S rDNA and its sequencing was outsourced to the Australian Genome Research Facility (AGRF). Sequences were aligned using nucleotide BLAST (NCBI, http://blast.ncbi.nlm.nih.gov/Blast.cgi) against the ‘nucleotide collection (nr/nt)’ database.
Microalgae cells from agar plates (TAP + 0.3% yeast extract + 1.5% agar) grown at 23°C in 50 μmol photons m−2 s−1 were used to build up inoculation cultures grown in 150-mL flasks with TAP medium [31] (23°C, 120 μmol photons m−2 s−1) on an orbital shaker (approximately 120 rpm). Algae strains that did not tolerate acetate were grown in tris phosphate (TP) media only. Algal strains originating from brackish water were supplemented with 250 mM NaCl. Cell densities were determined using optical density measurements at 750 nm (OD750) using a microwell plate reader (Infinite M200 PRO, Tecan Group Ltd., Männedorf, Switzerland). Algal cells during log phase growth were collected by centrifugation (500 g, 10 min, 25°C using Hettich Zentrifugen Universal 320R, Hettich Instrument Inc., Beverly, USA) and washed once before resuspending in 100 mM TRIS buffer (pH 7.4). The cells were inoculated into sterile 96-well plates, each well having an individual media composition using a starting OD750 of 0.1 using the microwell plate reader. All algae strains were grown in 150 μL in 96-well plates (5-mm culture depth) on an orbital shaker (580 rpm) under continuous light using top illumination (120 μmol photons m−2 s−1) at 23°C ± 0.5°C and 1% CO2 atmosphere (±0.3% CO2).