Materials and methods

Characterization of the food waste

The food waste used in this study was provided by the Portuguese company, Valorsul, S.A., which is responsible for the collection and treatment of food waste produced in hotels and restaurants in the Lisbon metropolitan area. In this facility, after collection, the waste un-dergoes mechanical pretreatment, solids are hydrolyzed and fermented before feeding into an anaerobic digester. The feed for this study was collected before the anaerobic digestion. A preliminary characterization of the fermentate was carried out to evaluate its viability to be used as feed and it is shown in Table 3.1. Owing to the low FP to total COD (CODTOT) ratio, high FP to soluble COD (CODSOL) and high solid content, the waste was centrifuged before being fed to the reactors, removing most of the solids which comprised the non-FP COD, such as proteins and carbohydrates. In the case of the lab-scale study, FFW was collected, centrifuged and stored in the freezer before use. As for the pilot-scale part, FFW was collected and stored in the fridge due to impossibility of storing such a large volume in the freezer.

Experimental setup at lab scale

3.2.2.1 Culture selection

A 2 L SBR was inoculated with activated sludge from the wastewater treatment plant of Mutela in Almada, Portugal and operated with a feast and famine regime to select a PHA-accumulating consortium. Operating conditions are summarized in Table 3.1. The 12h-cycle configuration consisted in periods of 665 min of aeration and 55 min of settling with no aera-tion or stirring: feeding of FFW (10 min), reacaera-tion (630 min), biomass withdrawal (20 min), set-tling (55 min) and mineral medium feeding (5 min). The phases of the cycle were controlled by timers and a software programmed in Labview. The composition of the mineral solution was as follows (mg L^-1): FeCl3•6H2O: 3.00, H3BO3: 0.30, CoCl2•6H2O: 0.30, MnCl2•4H2O: 0.24, ZnSO4•7H2O: 0.24, Na2MoO4•2H2O: 0.12, CuSO4•5H2O: 0.06, KI: 0.06. During the night, biomass was not withdrawn and the reaction phase lasted for 650 min. In the last 20 minutes of the settling phase, effluent was withdrawn according to the selected HRT. Except for the settling phase, the reactor was fully aerated and mixed, keeping the dissolved oxygen (DO) constantly above 2 mg L^-1. pH (7-9.5) and temperature (18-22 °C) were monitored, but uncontrolled. The OLR was set to 3.56 ± 0.07 gCOD L^-1 d^-1 at the beginning of operation and increased gradually

until 7.06 ± 0.12 gCOD L^-1 d^-1, where the reactor was stabilized for at least 3 sludge retention time (SRT) and characterized (Condition L1). Then, OLR was increased until reaching a maxi-mum of 9.42 ± 0.29 gCOD L^-1 d^-1 and the reactor performance was stabilized and characterized (Condition L2).

Table 3.1. Preliminary characterization of the hydrolysate obtained from Valorsul

TC = Total carbon; TOC = Total organic carbon; IC = Inorganic carbon; ProtTOT = Total protein; ProtSOL = Soluble protein; CH = Total carbohydrates; CH = Soluble carbohydrates; TS = Total solids; VS = Total solids

Parameter Value

CODTOT (gCOD L^-1) 119 ± 0.28 CODSOL (gCOD L^-1) 37.5 ± 1.34

TC (g L^-1) 12.4 ± 0.01

TOC (g L^-1) 12.3 ± 0.02

TOC (C-mmol L^-1) 1026 ± 2.04

IC (mg L^-1) 0.04 ± 0.01

ProtTOT (gCOD L^-1) 14.9 ± 0.73 ProtSOL (gCOD L^-1) 1.85 ± 0.04 CHTOT (gCOD L^-1) 7.53 ± 1.02 CHSOL (gCOD L^-1) 1.17 ± 0.38 FP (gCODFP L^-1) 31.9 ± 0.45

FP (C-mmol L^-1) 898 ± 12.7

FP/CODSOL (gCODFP gCOD^-1) 0.85 ± 0.02 TS (gTS L^-1) 62.4 ± 3.94 VS (gVS L^-1) 53.6 ± 3.68 VS/TS (gVS gTS-1) 0.86 ± 0.005

N-NH3 (gN L^-1) 1.11 ± 0.01

N-NO3 (gN L^-1) 0

N-NO2 (gN L^-1) 0

Total Kjeldahl-N (gN L^-1) 3.2 ± 0.1 N-NH3/Kjeldahl-N (gN gN-1, %) 34.2 ± 1.2

P-PO4 (mgP L^-1) 331 ± 3.92

N-mol:P-mol 7.39 ± 0.15

3.2.2.2 PHA accumulation

A 1 L fed-batch reactor was used for the PHA accumulation step to evaluate the storage performance of the culture. This reactor was inoculated with biomass selected in the SBR (0.4 L) and fed with the FFW in pulse-wise mode. Separate fed-batch accumulation tests were car-ried out after SBR stabilization at each OLR (Test L1 and L2). The amount of substrate fed in each pulse was calculated to maintain the food to microorganism ratio equal to the one in the SBR during pseudo-steady state. The reactor was fully aerated (at least 1 vvm) during operation to maintain the dissolved oxygen (DO) above 2 mg L⁻¹, temperature was set to 20 ºC with a heat blanket and pH was monitored and uncontrolled during operation and ranged from 7 to 9.5.

Experimental setup at pilot scale

3.2.3.1 Culture selection

A 60 L SBR was inoculated with a PHA-accumulating culture previously selected from waste activated sludge. Because of the covid pandemic, collection of sludge in wastewater treatment plant was impossible and biomass from a similar sequential batch reactor was col-lected and used as inoculum for this experiment. Operating conditions are summarized in Table 3.2. In condition P1, the 12h-cycle configuration consisted in periods of 665 min of aeration and 55 min of settling with no aeration or stirring: feeding of FFW (10 min), reaction (630 min), biomass withdrawal (20 min), settling (55 min) and mineral medium feeding (5 min). In condi-tion P2, a settling phase was introduced after the feast phase, and the 12 h-cycle was as follows:

feeding of FFW (10 min), reaction (90-180 min), settling (55 min), mineral medium feeding (5 min), reaction (390-480 min), biomass withdrawal (20 min), settling (55 min) and mineral me-dium feeding (5 min). The phases of the cycle were controlled by timers and a software pro-grammed in Labview. The same mineral medium composition as in lab-scale was used. In the last 10 minutes of each settling phase, effluent was withdrawn to adjust for the HRT. Except for the settling phase, reactor was fully aerated and mixed, keeping the DO above 2 mg L^-1. pH (7-9.5) and temperature (18-22) were monitored, but uncontrolled. Antifoam (silicone antifoam with 30% active compound) was supplied at regular intervals throughout the SBR cycle and its dilution factor was adjusted periodically to control foaming. The HRT of the reactor was set to 1 d in Condition P1 and changed to 0.67 d in Condition P2.

3.2.3.2 PHA accumulation

A 1 L fed-batch reactor and a 60 L fed-batch reactor were used for the PHA accumulation step to evaluate the storage performance of the culture. The 1 L reactor was used to charac-terize condition P1 (Test P1), while in Condition P2 both were used. In Condition P2, a test similar to the one carried out in P1 was done (Test P2), as well as another one in which solids from the waste were removed by centrifugation (Test P2S) and a third one at pilot-scale with the 60 L reactor, which was used for the calculation of the overall yield of the process. The 1 L reactor was inoculated with 0.4 L of biomass selected in the SBR, while the 60 L reactor was inoculated with 10 L of biomass. Both experiences were carried out similarly and biomass was fed with the FFW in pulse-wise mode. The amount of substrate fed in each pulse was calculated to maintain the food to microorganism ratio equal to the one in the SBR during pseudo-steady state. The reactor was fully aerated (at least 1 vvm) during operation to maintain the DO above 2 mg L^-1, temperature was set to 20 ºC with a heat blanket, pH was monitored and uncontrolled during operation.

Table 3.2. Summary of the SBR operating conditions

PHA accumulation

Cell dry weight was determined by total suspended solids (TSS) measurements as de-scribed in standard methods [18], COD was assessed using LCK Hach Lange kits (Hach) and Total Kjeldahl Nitrogen (TKN) was determined through a cuvette kit test LCV 909 (Hach). Fer-mentation products (FP; acetate, ethanol, propionate, lactate, succinate, butyrate, iso-butyrate, valerate, iso-valerate, caproate, heptanoate and octanoate) were quantified by high

Parameter Lab Pilot

Condition L1 L2 P1 P2

Working volume (L) 1.8 1.8 40 40

OLR (gCOD L^-1 d^-1) 7.06 ± 0.12 9.42 ± 0.29 9 9

HRT (d) 1 1 1 0.67

SRT (d) 4 4 4 4

Inoculum WAS -

PHA-accumulat-ing consortium -

performance liquid chromatography (HPLC) using a VWR Hitachi Chromaster according to [19].

Ammonia, nitrite, nitrate and phosphate were measured using a colorimetric method by a seg-mented flow analyzer (Skalar San++ System). Gas chromatography with flame ionization de-tector was used to determine the concentration of PHA as described by [12]. Proteins were quantified according to Lowry method [20] and sugars according to the Dubois method [21].

Total organic carbon (TOC) was analyzed using a Shimadzu TOC-VCSH (Shimadzu, Japan). PHA extraction was carried out as described by Cruz et al. with a few modifications [22]: suspended biomass was centrifuged (8000 rpm, 15 min) between washes (twice with deionized water and once acetone), suspended in chloroform (1 g of dry biomass per 50 mL CHCl3) and mixed at 30 ºC for 48 h, filtered and precipitated into cold ethanol (1:10). Molecular weight was determined by size exclusion chromatography according to [23] and thermal properties by dynamic scan-ning thermography according to [23], except for the temperature, which ranged between -90 ºC and 200 ºC.

Calculations

3.2.5.1 Reactor performance

Feast to famine ratio (FF ratio) was calculated as the ratio between the length of feast phase and the famine phase. Maximum volumetric PHA production rate (rPHAMAX), maximum volumetric substrate uptake rate (rFPMAX), maximum volumetric nitrogen uptake rate (rNMAX) were calculated from the slope of the linear regression of the concentrations of PHA, FP and N, respectively. The specific rates (qPHA,qFP andqN) were calculated by dividing the volumetric rates by the active biomass concentration. In the SBR, storage yield (YPHA/FP) was calculated by dividing qPHA by qFP and growth yields (YXA/FPFEAST and YXA/PHAFAMINE) were calculated by the ratio between biomass produced over FP and PHA, respectively. In the accumulation tests, YPHA/FP

was calculated by dividing qPHA by -qFP of the first pulse and YPHA/FPGLOBAL is the ratio between all the PHA produced and the FP consumed. YXA/FPwas calculated by dividing qXA by qFP of the first pulse. The volumetric PHA productivity (PPHA) was calculated by dividing the amount of PHA produced during the experiment by the length of the experiment and by the final volume of the reactor.

3.2.5.2 Carbon flow in PHA metabolism

A carbon mass balance was carried out for all 8 SBR monitoring data sets and 13 accu-mulation tests in order to estimate monomer production (in mmol) from consumed FP (in

mmol) in each test. The fraction of carbon considered for maintenance and growth was equal to 1-YPHA/FP. For each test, 18 stoichiometric reactions were considered (Table 3.3) and weights were attributed to each reaction to predict each monomers production. Solver from Microsoft Excel was used to adjust the weights of each reaction, using the sum of the percentual error of the monomers as the objective to minimize.

Table 3.3. Reactions considered in the carbon balance of the FP

3.2.5.3 Development of PLS models for prediction of PHA

Projection to latent structures modelling was used to correlate the FP (acetate, ethanol, propionate, lactate, butyrate, iso-butyrate, succinate, valerate, iso-valerate, caproate,

# Reaction Stoichiometry Ref.

1 Succinate → HB C4H6O4 → C4H6O2

2 Lactate → HB 2 C3H6O3 → C4H6O2 + 2 CO2 [24]

3 Lactate + Acetate → HV C3H6O3 + C2H4O2 → C5H8O2

4 Acetate → HB 2 C2H4O2 → C4H6O2 [12]

5 Propionate → Acetate C3H6O2 → C2H4O2 + CO2 [12]

6 Propionate + Acetate → HV C3H6O2 + C2H4O2 → C5H8O2 [12]

7 Ethanol → HB 2 C2H6O → C4H6O2

8 Iso-butyrate → HB C4H8O2 → C4H6O2

9 Butyrate → HB C4H8O2 → C4H6O2 [12]

10 Iso-valerate → HV C5H10O2 → C5H8O2

11 Valerate → HV C5H10O2 → C5H8O2 [12]

12 Caproate → HHx C6H12O2 → C6H10O2 [12]

13 Caproate → HB + Acetate C6H12O2 → C4H6O2 + C2H4O2 [12]

14 Heptanoate → HV + Acetate C7H14O2 → C5H8O2 + C2H4O2

15 Octanoate → HO C8H16O2 → C8H14O2

16 Octanoate → HHx + Acetate C8H16O2 → C6H10O2 + C2H4O2

17 Octanoate → HB C8H16O2 → 2 C4H6O2

18 Acetate → CO2 C2H4O2 → 2 CO2

heptanoate and octanoate) with the monomers of PHA produced (HB, HV, 3-hydroxyhexano-ate (HHx), 3-hydroxyoctano3-hydroxyhexano-ate (HO)). The inputs used in the PLS model were the amount of FP in the feed divided by the initial bioreactor volume (in mmol L^-1). The outputs of the PLS model were the monomers of PHA produced (final less initial) divided by the final volume of the bio-reactor (in mmol L^-1). A total of 21 experiments (including SBR and accumulation at lab and pilot-scale) were used to develop the correlation. Another PLS model was also developed using the same inputs to estimate the total PHA.

The root mean square error of cross-validation (RMSECV) was calculated based on all outputs for the 21 samples through leave-one-out, and the parameter used to select the num-ber of latent variables (LV) of the models (minimum RMSECV). The R² of data fitting and the root mean square error (RMSE) of the final models were used to assess the quality of the cor-relational models.

PLS models were implemented in GNU OCTAVE using the npls algorithm from the n-way toolbox [25]. All input and output data were standardized prior to the PLS models develop-ment. Standardization was performed by subtracting to each value the average and dividing by the standard deviation calculated for each parameter. This procedure ensures similar weight within inputs and within outputs in the modelling process.