PETR: A novel peristaltic mixed tubular bioreactor simulating human colonic conditions

A novel bioreactor simulating human colonic conditions for in vitro cultivation of intestinal microbiota is presented. The PEristaltic mixed Tubular bioReactor (PETR) is modular designed and periodically kneaded to simulate intestinal peristalsis. The reactor is introduced, characterized from a bioprocess engineer's perspective and discussed in its ability to mimic colon conditions. PETR provides physiological temperature and appropriate anaerobic conditions, simulates intestinal peristalsis, and has a mean residence time of 32.8 ± 0.8 h comparable to the adult human colon. The single‐tube design enables a time‐constant and longitudinally progressive pH gradient from 5.5 to 7.0. Using a dialysis liquid containing high molecular weight polyethylene glycol, the integrated dialysis system efficiently absorbs short chain fatty acids (up to 60%) and water (on average 850 mL d−1). Cultivation of a typical gut bacterium (Bifidobacterium animalis) was performed to demonstrate the applicability for controlled microbiota cultivation. PETR is unique in combining simulation of the entire colon, peristaltic mixing, dialytic water and metabolite absorption, and a progressive pH gradient in a single‐tube design. PETR is a further step to precise replication of colonic conditions in vitro for reliable and reproducible microbiota research, such as studying the effect of food compounds, prebiotics or probiotics, or the development and treatment of infections with enteric pathogens, but also for further medical applications such as drug delivery studies or to study the effect of drugs on and their degradation by the microbiota.

around 4 × 10 13 in total (Sender et al., 2016).In addition to this quantity, the diverse composition with more than 160 different bacterial species (Qin et al., 2010) is a key determinant of microbial functions, making the microbiota an emerging field of research.
However, there are still major challenges in terms of detailed experimental studies.Studies of the human gut microbiota can generally be performed in vivo in humans, in vivo in animals, or in vitro in bioreactors (Williams et al., 2015).Human in vivo studies are the most reliable, but complex and costly and therefore limited in quantity (Marzorati et al., 2011), restricted to stool samples (Payne et al., 2012) which are not representative for spatial variations across the gut (Stearns et al., 2011), fraught with ethical drawbacks, and highly influenced by the uniqueness of the patient-specific microbiota (Eckburg et al., 2005).In vivo studies with animals (e.g., mice) also have their limitations and results are not directly transferrable to humans due to differences in anatomy, physiology, and microbiota (Nguyen et al., 2015).
In vitro systems can overcome these limitations, especially when using defined and reproducible microbiota.They are alternative and versatile research tools as they provide flexible, controllable, and reproducible culture conditions without any ethical constraints (Roupar et al., 2021).However, their limitations lie in their accuracy in mimicking physiological gut conditions and functions (Williams et al., 2015).In vitro systems vary widely in complexity, from simple single-vessel batch approaches over complex multistage and pHregulated continuous cultures to artificial digestive systems, each with different advantages and limitations (Payne et al., 2012).
Although simulation of some important conditions of the natural situation (e.g., physiological temperature, anaerobia, typical pH values) is common to all systems, they differ in the accuracy of other properties and functions (e.g., simulated gut region[s], pH gradient, mixing technique/peristalsis, absorption of water and metabolites).To date, in vitro systems are usually simple in design (e.g., cascades of stirred reactor vessels (Williams et al., 2015)) and often lack important colonic functionalities such as absorption (Roupar et al., 2021) or appropriate luminal gradients (Isenring et al., 2023).
Many bioreactor systems mimicking human colonic conditions in vitro are based on a three-stage continuous culture system, which was first introduced by Gibson et al. (1988) and later adapted and validated by Macfarlane et al. (1998).The basic principle is a cascade of three stirred bioreactors connected in series to replicate the changing conditions along the colon with increasing nutrient depletion and pH value.This three-stage approach has been widely adopted in other in vitro systems (Barroso et al., 2015a;Cinquin et al., 2006;Feria-Gervasio et al., 2014;Mäkivuokko et al., 2005;Molly et al., 1993) to simulate the conditions of the ascending, transverse, and descending colon in separate compartments.Other systems focus on single parts of the colon, mainly the cecum and ascending colon (Zihler Berner et al., 2013;Feria-Gervasio et al., 2011;Minekus et al., 1999) with their high fermentation activity, but sometimes also the distal part (Auchtung et al., 2015;McDonald et al., 2013).While restriction to one colon part has the advantage of reduced system complexity, it fails to simulate the varying conditions of the natural colon environment (Payne et al., 2012).In contrast, multivessel approaches allow simultaneous study of different spatial, nutritional and physicochemical conditions (Payne et al., 2012).However, natural occurring gradients (nutrients, pH, redox potential) are not properly addressed in current systems (Isenring et al., 2023).
Whether single-vessel or multistage concepts, almost all systems are based on stirred bioreactors, which hardly fits the in vivo situation.They oversimplify the natural peristaltic mixing of the intestinal tract, cannot reproduce the physiological residence time distribution (RTD), and have different fluid mechanics and shear forces, affecting, e.g., wall growth (Spratt et al., 2005).A first approach to incorporate natural peristalsis was introduced by Minekus et al. (1995) in form of vessels with flexible walls in the TIM-1 (Netherlands Organization for Applied Scientific Research [TNO] Intestinal Model 1) system simulating stomach and small intestine, and later adopted in the TIM-2 system to mimic conditions in the proximal colon (Minekus et al., 1999).The TIM-2 system also includes a dialysis system that absorbs water and metabolites and maintains short chain fatty acid (SCFA) concentrations at physiological levels (Minekus et al., 1999).Although TIM-2 is a much more comprehensive system, it does not aim to mimic the varying conditions along the colon and is usually operated to simulate only the proximal colon (Venema, 2015).Nevertheless, it can also be operated with a temporal pH gradient by increasing the pH over time (Venema, 2015) as it is also done in other single-vessel systems (Cinquin et al., 2004;Wiese et al., 2018) to simulate the passage through the different colon regions.
The limitations of other systems described above highlight the need for a sophisticated bioreactor system that mimics human colon conditions more comprehensively than previous in vitro systems.
Therefore, the main objective of the present investigations was to develop a novel bioreactor designed to replicate the conditions of the cecum, the colon, consisting of ascending, transverse, and descending colon, as well as the sigmoid colon and rectum.The system does not have to be an exact replica of the anatomy and physiology of the human gut, but it must provide the appropriate conditions that are exerted on the microbiota in a controlled and reproducible manner.
For this purpose, the reactor system needs physiological temperature and must be equipped with sectional pH control to set a pH gradient along the entire axial length, and an appropriate resorption function to gradually remove water from the cultivation broth.To mimic in vivo conditions as closely as possible, the reactor system needs a peristaltic mixing system to simulate intestinal peristalsis and must provide the ability to continuously culture intestinal bacteria under anaerobic conditions.At the same time, easy access for sampling along the entire length of the reactor is an advantage over in vivo capabilities.A novel bioreactor system is presented, which essentially consists of a modular, PEristaltic mixed Tubular bioReactor (PETR) with dimensions comparable to the human colon.In addition, this newly designed bioreactor is characterized from a bioprocess engineer's perspective, compared to other existing systems, and discussed in terms of its ability to mimic human colon conditions.A 7 VORLÄNDER ET AL.
| 1119 days continuous cultivation of a typical gut bacterium (Bifidobacterium animalis) is presented to demonstrate the biological applicability in a first proof of concept.A Bifidobacterium was selected because Bifidobacteria play an essential role in the gut of infants, but their amount becomes less with increasing age (Arboleya et al., 2016).Due to their health promoting characteristics, Bifidobacteria are of great interest as probiotics (Chen et al., 2021), which is one of the topics PETR was developed for.

| MATERIALS AND METHODS
As the development and final design of the bioreactor system is a substantial part of this study, PETR is described in detail in   SI1.Precultures were performed in 100 mL serum bottles at 37°C, shaking frequency of 80 min −1 (rpm) and 50 mm shaking diameter (Multitron II, Infors AG) for around 24 h.

| Residence time
To determine the RTD in PETR, demineralized water was continuously pumped (Ismatec Reglo ICC, Cole-Parmer Instrument Company) through the reactor at the same flow rate used for continuous cultivation (1.35 mL min −1 ).Conductivity was measured in M4 (Conducell 4UHF Arc 120 connected to Arc View Handheld, Hamilton Bonaduz AG) and the other pH ports were equipped with polyvinyl chloride dummies of the same dimensions as the pH electrodes.PETR was run without dialysis system and the corresponding ports were plugged with rubber stoppers.
PETR was mixed peristaltically and a pulse of 3 mL of 4 M KCl solution was added to M1.

| Mixing time
To determine the mixing time, PETR was set up as for the RTD, but operated in batch mode.In a first approach, the forward mixing time was investigated by adding the tracer (3 mL of 4 M KCl) in M1 and recording the conductivity in M4.In a second setup, the tracer was added in M4 and conductivity was measured in M1 to determine the back-mixing time.

| pH gradient
To validate creation and maintenance of the desired pH gradient, PETR (without dialysis system) was equipped with a pH electrode (405-DPAS-SC-K8S/120, Mettler-Toledo GmbH) in each module and filled with MCM mineral salt solution containing the same minerals as the MCM (Supporting Information S2: Table SI1).Gas phase oxygen was measured with a fiber optical sensor (FireSting + Robust Oxygen Trace Probe, PyroScience GmbH).The incubation chamber was gassed first with 25 L min −1 nitrogen for 4.3 h to rapidly displace oxygen in the gas phase and then with 10 L min −1 to maintain the low oxygen concentration.Exhaust air leaves the incubation chamber through a port on top to a laboratory extraction arm.

| Absorption
To quantify the resorption capacity of water and SCFAs, PETR was equipped with the dialysis system which is based on a dialysis tube (diameter 16 mm, molecular weight cut-off (MWCO) 14,000 Da, Membra-Cel™, Carl Roth GmbH & Co. KG) going through the reactor.
Then, 60 mM acetic acid, 23 mM propionic acid, and 17 mM butyric acid were added to the continuously fed (1.35 mL min −1 ) MCM mineral salt solution to mimic typical SCFA concentrations in vivo (Blaut, 2018;Cummings et al., 1987), and the pH was adjusted to 5.5 using 8 M NaOH.The dialysis liquid was also based on MCM mineral salt solution but contained 10% or 20% (w/w) of high molecular weight polyethylene glycol (PEG 35,000, Clariant AG) by omitting the same amount of water, and the pH was adjusted to 6.25 with HCl.
The mass of the medium, dialysis liquid and dialysis waste (DW) was continuously measured (Kern PCB 10000-1, Kern & Sohn GmbH) and recorded via a Processing script.

| Reactor cultivation
To sterilize PETR in an autoclave, the assembled system was sealed with blind stoppers and equipped with air filters (Midisart 2000, Sartorius Stedim Biotech GmbH).The dialysis membrane and pH electrodes were autoclaved separately in water and 3 M KCl, respectively, and integrated sterile into the autoclaved PETR in a biological safety cabinet.After connecting all peripheral hoses and electrical connections, the incubation chamber was closed and gassed with nitrogen while PETR was filled with 2.7 L medium.Heating was started at least 3 h before inoculation.The preculture was transferred in a syringe through an airlock to the incubation chamber and injected into PETR through the sampling port of M1.After an initial batch phase of around 16 h, continuous operation was initiated by starting pH control and pumps for fresh medium (1.35 mL min −1 ) and dialysis liquid (15 rpm/~1.9g min −1 ).The dialysis medium was supplemented with vitamins, hemin, and cysteine after autoclaving to prevent their loss from the MCM (Supporting Information S2: Table SI1).

| Statistics
Statistical analyses were performed using OriginPro 2023 (OriginLab Corporation).Student's t test was used to test significance between two independent samples, and p < 0.05 were considered as significant.

| Reactor design
The first step of the reactor design was to define the most important colonic conditions and functionalities that have to be implemented.These include physiological temperatures (Pearson et al., 2012), anaerobic conditions (Zheng et al., 2015), and peristaltic mixing (Sarna, 2010), as well as varying conditions across the different colon parts, for example, in terms of nutrient availability and fermentation activity (Macfarlane et al., 1992) or increasing pH in distal direction (Cummings et al., 1987), and especially the absorption of water (Sandle, 1998) and fermentation products (McNeil, 1984).Besides the trend of miniaturizing gut-mimicking systems (Davis Birch et al., 2023;Jin et al., 2022;Makivuokko et al., 2005;Robinson et al., 2014;Wiese et al., 2018), the bioreactor was intentionally developed as a comparatively large system with dimensions comparable to the adult human colon (Kararli, 1995).This ensures differences and gradients over the length and the ability to collect sufficient sample volume without disturbing the running system.
With a single-tube design and dimensions comparable to the adult human colon, PETR provides a novel reactor design intended to overcome some limitations of stirred cascades due to more realistic replication of the colon anatomy.The reactor essentially consists of four stainless steel modules (M1 to M4) and five flexible silicone hoses (Figure 1a).The steel modules (Figure 1b) provide the connections for the medium inlet (only M1), pH measurement and control, and sampling.Each module has a length of 150 mm, an inner diameter of 47.5 mm, 2″ Tri-Clamp connections at both ends and is equipped with a stainless steel cable gland (bg 220VA, Pflitsch GmbH & Co. KG) for mounting the pH electrodes, as well as two or three stainless steel tubes (3 mm inner diameter, 1.5 mm wall thickness).
The flexible segments consist of 190 mm long silicone tubes (45 mm inner diameter, 2 mm wall thickness, J. Lindemann GmbH) equipped with Tri-Clamp fittings (21.5 mm length, 47.5 mm inner diameter, 50.5 mm outer diameter) at both ends to connect them with the steel modules or end caps.The total reactor length of around 165 cm corresponds very well to the mean in vivo length of 160.5 ± 33 cm (Hounnou et al., 2002).Supporting Information S2: Figure SI1 shows a schematic illustration of PETR with the surrounding incubation chamber and periphery for process control.
PETR simulates intestinal peristalsis by periodically squeezing the silicone tubes by servo motor driven kneaders.Each kneading device is composed of two high torque servo motors (HS-645MG, Hitec RCD USA, Inc.) connected to two roller arms to increase deflection and hence the amount of liquid displaced during compression (Figure 1c).The two roller arms of one kneading device squeeze the silicone tube by mirror-symmetrical rotation by a defined angle, causing the tube be squeezed increasingly horizontally to a minimum distance of 15 mm between both rollers before the compression is released and the rollers then move back again (Supporting Information S1: Video 1).The rollers keep a defined distance to ensure integrity of the integrated dialysis membrane.
To replicate the pH gradient, each module is equipped with a pH A dialysis tube made of regenerated cellulose (MWCO of 14,000 Da) passes through PETR and is fixed at the end caps with customized connectors made of stainless steel and polytetrafluoroethylene (Figure 1d).The membrane is attached to the connection tubes with a groove and sealing ring, and they are screwed with cable glands at the end caps.To absorb fermentation products as well as water, the dialysis liquid contains 20% (w/w) PEG-35,000 for osmotic removal of water from the cultivation broth.
A specially designed holding device made of aluminum profiles allows flexible positioning of PETR on a guide rail system (Figure 1a).
The steel modules rest on T-pieces which allows flexible alignment, as well as linear bearing during compression of the silicone tubes.In addition, the reactor holder can be tilted slightly to promote the movement of accumulated fermentation gases to the waste outlet at the distal end.
PETR is surrounded by an incubation chamber (3 m x 0.5 m x 0.5 m) for temperature and oxygen control (Supporting Information S2: Figure SI1).To simplify the reactor design and minimize temperature gradients due to slow mixing, temperature control is achieved by heating the incubation chamber to 37°C.Temperatures are measured at different positions with waterproof digital temperature sensors (DS18B20, AZ-Delivery Vertriebs GmbH) connected to an Arduino Nano microcontroller that regulates chamber heating by controlling the fan speed (P12 PWM PST CO, Arctic GmbH) of waterpowered heat exchangers (360 G2 slim, Magicool Europe).The heat exchangers are connected in series to an external thermostat (RE104, Lauda Dr. R. Wobser GmbH & Co. KG) providing water with 50°C.
An anaerobe atmosphere is achieved by continuously gassing the incubation chamber with nitrogen.Due to their high oxygen permeability (Robb, 1968), the silicone hoses enable deoxygenation of the cultivation medium without direct sparging of the liquid phase.
Particularly with regard to the low shear stress mixing approach, the silicone hoses are also thought to promote biofilm formation, which might be beneficial to simulate colonic conditions more precisely, as bacteria are known to cover food particles with biofilms (van Wey et al., 2011) and are assumed to colonize the mucus layer in form of mucosal biofilms (Sicard et al., 2017).Silicone surfaces are well known to be colonized and covered by biofilms and pieces of silicone hoses in a packed-column biofilm reactor were previously described as suitable carriers for mucosal gut communities (McDonald et al., 2015).The silicone hoses of the flexible parts can be easily replaced, enabling to cut out pieces of the hoses to analyze structure and composition of biofilms adhering to them.
PETR differs in many ways from other in vitro systems published so far.It combines more properties and functionalities of the natural colon environment and offers a unique and more comprehensive in vitro simulation than other systems.Major characteristics are the tubular design enabling a continuous pH gradient, the peristaltic mixing, and the dialytic metabolite and water absorption.A comparison of the most important characteristics and conditions of the human colon and how they are addressed in PETR and in other in vitro cultivation systems is presented in Table 1.

| Peristaltic mixing technique and mixing time
PETR is characterized by its peristaltic mixing technique, simulating intestinal peristalsis more realistic than commonly used stirred reactors.Based on their spatiotemporal characteristics (frequency, amplitude, duration, propagation), gut contractions propel luminal contents, produce back and forth movements mixing the content, or both (Sarna, 2010).Whereas some strong ultra-propulsive contractions cause mass movements like during defecation, non-propagating contractions produce bidirectional movements and mixing of the chyme.Especially in the colon, spatially disorganized contractions closing the lumen completely or only partially occur, resulting in intense mixing with only slight propulsion (Sarna, 2010), and this contraction pattern is addressed by the regularly compressions of PETR's silicone hoses.
Therefore, kneading devices (cf. Figure 1c) squeeze the flexible silicone hoses regularly one after the other.The movements of the kneading devices are controlled by an Arduino nano microcontroller, with adjustable squeezing speed and compressions per time.Starting from the proximal side, the roller arms of the first kneading device move to the right (in distal direction) and back to the left.After a defined time delay, this movement is performed on the next hose in distal direction and starts again at the proximal side after squeezing the fourth hose (Supporting Information S1: Video 1).By this, each silicone module is squeezed at a frequency of 7 min −1 .Comparable short-duration rhythmic phasic contractions occur at a similar frequency of 3-12 min −1 in the human colon (Sarna, 2010).The compression of the silicone hoses leads to fluid displacement and motion and therefore mixing of the reactor content.The rollers of the kneading system have a minimum distance of around 15 mm during compression to ensure integrity of the dialysis tube inside.This correlates to around 70% occlusion, which is comparable with the occlusion found in the stimulated human colon (59 ± 18%) (Stamatopoulos et al., 2020).Due to the partial closure, the liquid is displaced in both proximal and distal direction and causes mixing in both directions, as it is a common contraction pattern of the human colon (Sarna, 2010).Only four of the five silicone modules are equipped with a kneading device to reduce mixing at the distal end.In the human colon, water absorption increases the viscosity, which decreases mixing in distal direction and therefore needs to be considered for adequate in vitro replication (Macfarlane & Macfarlane, 2007;Venema & van den Abbeele, 2013).However, as PETR has not achieved in vivo like solid contents/viscosities so far, mixing is reduced by omitting the kneading device at the fifth silicone module for closer simulation of deceasing mixing.
Figure 2 shows PETR's mixing times in water as Newtonian fluid with a dynamic viscosity of ~1 mPa s in batch mode.The forward mixing time was determined by adding the tracer in M1 and measuring the conductivity in M4, whereas the setup was vice versa for the backward mixing time.Mixing times of 11.1 ± 0.8 h and 9.6 ± 0.4 h were obtained for forward and backward mixing, respectively (average of 10.4 ± 0.5 h for both).Although homogeneity increased significantly (p < 0.01) faster in forward mixing during the first 2.5 h, indicating greater propulsive fluid motion from proximal to distal, backward mixing was overall slightly but significantly (p < 0.05) faster.Nevertheless, similar mixing times in both directions highlight that, in contrast to the classical approach of stirred cascades, mixing takes place over the entire length.PETR's mixing times are far from typical bioreactor mixing times in the range of several seconds, but this is essential to create and maintain a pH gradient.If mixing is too intense, the desired pH gradient and also other gradients (nutrient availability, microbiota composition) would not be properly created.
On the other hand, mixing intensity must ensure local homogenization for adequate pH control, and prevent cell sedimentation.As PETR does not reach in vivo like solid contents/viscosities so far, water was used as model fluid.While this is representative for current reactor cultivations, it is limited in its ability to accurately reflect the in vivo situation.Viscosities in vivo are much higher and T A B L E 1 Overview of selected in vitro systems to cultivate colonic microbiota under controlled conditions.

System/ references
Gut region  increase along the colon, so PETR's mixing performance must be reevaluated when it reaches comparable viscosities.However, in vivo viscosity data are limited and viscosities are also very individual as seen in studies on ileostomy effluents (Dalhamn et al., 1978) and stool samples (Woolley et al., 2014).
As mixing performance of other in vitro cultivation systems is usually not described in detail, a comparison is only feasible regarding the mixing technique, but not on the level of concrete numerical values (mixing times).In addition, mixing in PETR's single-tube design is hardly comparable with mixing in different vessels of a cascade.These vessels are typically stirred magnetically (Barroso et al., 2015a;Cinquin et al., 2006;Macfarlane et al., 1998;Molly et al., 1993) or with conventional agitators (Blanquet-Diot et al., 2012;Feria-Gervasio et al., 2011;Feria-Gervasio et al., 2014;Haindl et al., 2021).In miniaturized reactors, mixing was also conducted with nitrogen sparging (Davis Birch et al., 2023;Jin et al., 2022).Other in vitro systems also incorporate simulation of intestinal peristalsis.For example, TIM-2 simulates the proximal colon and provides a more appropriate mixing technique based on a water pressuredriven peristalsis system (Minekus et al., 1999).Peristaltic mixing was more frequently applied in systems mimicking gastric conditions (Li & Kong, 2022;Sensoy, 2021), but remains the exception for systems simulating colonic conditions (Table 1).
An accurate in vitro replication of long propagating contractions was previously described in the Dynamic Colon Model (DCM), a system for in vitro drug dissolution testing under appropriate hydrodynamic conditions.The DCM consists of a half-filled 20 cmlong tube (5 cm diameter) consisting of 10 segments that can be individually compressed or expanded by an hydraulic system, causing a propagating contraction over the entire length (Stamatopoulos et al., 2016).The contraction frequency was adjusted to mimic those observed in vivo (i.e., two waves per minute) (Stamatopoulos et al., 2020).However, the DCM was developed for drug dissolution testing and not for the cultivation of intestinal microbiota.Accordingly, it does not fulfill the other requirements for an appropriate bioreactor such as anaerobic conditions or pH control, and it is unclear if the device can be sterilized.As in PETR, the contractions in the DCM does not completely occlude the lumen, resulting in back and forth movements in a similar range, which corresponds to the similar mixing times in both directions in PETR.Moreover, the particle velocity was higher in proximal than in distal direction (Stamatopoulos et al., 2020), which could explain the slightly shorter backward mixing time found in PETR.Interestingly, the recent comparison of different contraction pattern (i.e., long propagating contractions in both directions as well as short propagating and single segment contractions) showed best drug dissolution (i.e., mixing) when using periodic local (non or only slight propagating) contractions with high occlusion (50%-75%) (O'Farrell et al., 2022) as simulated with the periodic compressions of PETR's silicon modules.

| RTD
The colonic residence time is very individual and can vary widely (range 1.5-43.7 h) (Haase et al., 2016), but has great impact on colonic conditions and the microbiota, since it directly affects the availability of water and nutrients and determines the luminal wash out rate (Müller et al., 2018).To determine PETR's RTD, the dialysis system was omitted to prevent absorption of the tracer.The (1) is close to typical colonic transit times, which are reported to be, for example, 35 ± 2.1 h (Metcalf et al., 1987) or 35.74 h (Southwell et al., 2009).From three replicate measurements (Figure 3 was calculated to be 32.8 ± 0.9 h, which is in good agreement with hydraulic residence time τ.In principle, the mean residence time t̅ can be easily adjusted by varying the feed rate. At this point, it should be mentioned that the actual culture volume is about 300 mL lower (2400 mL) due to volume displacement by the dialysis membrane, which reduces the (hydraulic) residence time.On the other hand, dialytic water absorption leads to an increasing residence time, which therefore differs during cultivation and depends on the water removal rate.Nevertheless, the course of the RTD gives an additional impression of the reactor characteristics.
After passing the maximum, the RTD follows an exponential decline as expected for a homogenous mixed system.PETR clearly does not behave like a plug flow reactor.Instead, a mixture of the RTDs expected for plug flow and ideal mixed reactors occurs, which can be explained by the bidirectional mixing and slow feed rate (cf. Figure 2).
The natural human colon also behaves like a mixture of both ideal reactor types.Whereas the cecum and proximal colon can be considered as a mixing chamber with characteristics similar to continuous cultures, the mixing decreases in distal direction (Macfarlane & Macfarlane, 2007) leading to a behavior more comparable to a plug flow reactor.The closest in vivo simulation would therefore be a mixture of intensive mixing in the proximal region but reduced mixing in the distal part due to increasing solid content and viscosity as a result of progressive water absorption.
Currently, mixing in the distal region is decreased by omitting the kneading device at the fifth silicone tube, as the water absorption does not reach the very high in vivo rates so far (see 3.2.4).
In fact, with the slow exponential decrease after reaching the modal value, PETR's RTD follows a similar course as reported in vivo (Wiggins & Cummings, 1976).However, the modal value is reached significantly earlier, which might be due to the fact that PETR consists of only one segment, as it is not intended to mimic the whole gastrointestinal tract.In addition, the viscosity increase is thought to reduce mixing effects in the distal colon, resulting in a more sequential passage (Wiggins & Cummings, 1976), whereas in this study, the determination was performed in water, resulting in equal mixability over the entire length.In addition, viscosities in vivo are generally much higher.Nevertheless, PETR's RTD shows a comparable pattern to that observed in vivo.
In other in vitro systems, colonic residence times differ as widely (Table 1) as it was described for in vivo conditions.Some examples from systems mimicking the whole colon are 13 h in a three-stage system mimicking the colon of infants (Cinquin et al., 2006), 36 h in the ARCOL (ARtificial COLon) system (Thévenot et al., 2013) or 76 h in the Simulator of the Human Intestinal Microbial Ecosystem (SHIME) (Molly et al., 1993).As in PETR, many systems provide fresh medium as continuous feed (Zihler Berner et al., 2013;Feria-Gervasio et al., 2014;Minekus et al., 1995;Robinson et al., 2014), whereas other follow a different approach by adding fresh medium or transferring contents sequentially in a semicontinuous manner (Barroso et al., 2015a;Blanquet-Diot et al., 2012;Jin et al., 2022;Makivuokko et al., 2005;Miller & Wolin, 1981) or a combination thereof as it is done in the SHIME (Molly et al., 1993).

| pH regulation
The pH value of the chyme varies significantly in the course of the gastrointestinal tract.Although it is reasonably low in the stomach with pH values between 1 and 3.5, it increases along the small intestine to around 7 in the ileum (Gruber et al., 1987).However, after passing the ileocecal valve and entering the colon, the pH drops to around 5.4-5.9 in the proximal colon, before it raises again to values of around 6.1-6.4 in the transverse and 6.1-6.9 in the distal colon, respectively (Payne et al., 2012).The sharp pH decrease in the cecum and proximal colon results from the bacterial fermentation of carbohydrates entering the colon to SCFAs (Nugent, 2001).The following pH increase along the colon correlates with decreasing SCFA concentrations in distal direction, from 130 mM in the cecum to 80 mM in the descending colon (Cummings et al., 1987).In addition, secretion of bicarbonate to the lumen and the formation of ammonia from bacterial degradation of proteins, amino acids and urea contributes to the pH gradient as well (Nugent, 2001).Protein fermentation occurs primarily in the distal colon (Macfarlane et al., 1992).
Based on the natural values described above, a linear pH gradient of 5.5-7.0 was targeted for PETR to simulate conditions from the cecum to the sigmoid colon and rectum by using pH set points of 5.5, 6.0, 6.5, and 7.0 for the different modules.Continuous cultivation of intestinal bacteria in PETR is preceded by a batch phase, as typically performed in other in vitro systems (Barroso et al., 2015a;Feria-Gervasio et al., 2011;Macfarlane et al., 1998;Probert & Gibson, 2004) as summarized in Table 1.In this phase, pH is measured but not controlled, and decreases due to microbial growth and the release of organic acids.To develop the pH control and validate creation and maintenance of the desired gradient, microbial acid formation was simulated by continuously adding 0.1 M acetic acid to the different modules.Addition was higher in the first two modules to simulate higher metabolic activity in the proximal colon due to higher nutrient availability (Macfarlane et al., 1992).
Figure 4 shows a typical pH course during gradient generation.
Starting from a pH of around 4.5, the set points were achieved in about 1 h and then held fairly constant.In principle, only the dosage of base (1 M NaOH) should be required due to the organic acid formation and the low pH values typically achieved in batch experiments (Supporting Information S2: Figure SI5).However, the addition of NaOH to M2-M4 leads to an increasing pH in M1 due to PETR's bidirectional mixing (cf. Figure 2), and 0.5 M HCl is added to M1 after 1 h to maintain pH 5.5.
As PETR's mixing times are considerable long with 10.4 ± 0.5 h, added acid or base will not immediately affect the measured pH value.To prevent overregulation due to this time delay and to prevent damage of the membrane because of locally low or high pH values, the pH control was programmed with a rather slow supply and acid and base had relatively low concentrations.Therefore, initial gradient formation takes some time, but this is neglectable regarding typical cultivation times of several days or even weeks (Davis Birch et al., 2023;Cinquin et al., 2006;McDonald et al., 2013;Molly et al., 1993) as shown in Table 1.Once the gradient was generated, acid and base were added almost linearly to maintain the set points.The average flows were 7.8 mL h −1 (M1), 3.2 mL h −1 (M2), 0.8 mL h −1 (M3), and 3.7 mL h −1 (M4).Only slight amounts were pumped into M3 after initial gradient generation, which was even more pronounced during the more representative cultivation of B. animalis with simultaneous dialysis (see Section 3.3).Thus, the single-tube design allows an indeed progressive pH gradient, which should simulate in vivo conditions more closely than 3-4 distinct levels in cascaded approaches (Isenring et al., 2023).
The pH values used in other in vitro cultivation systems are summarized in Table 1 and range from pH 5.5 for the proximal to 7.0 for the distal colon.PETR's pH set points of 5.5, 6.0, 6.5, and 7.0 were also used in the four-stage EnteroMix colon simulator (Mäkeläinen et al., 2007;Makivuokko et al., 2005).In a three-stage colon model based on tubular dialysis membranes, pH was also maintained between 5.5 and 7.0 by dialyzing against pH controlled PEG solutions (Spratt et al., 2005).Whereas simultaneous different pH values are only feasible in multistage systems, temporal pH shifts allow to simulate the passage through the different colon regions in single-vessel systems like the TIM-2 (Venema, 2015) or the Copenhagen Mini Gut (Wiese et al., 2018).However, this is only batch-wise and does not allow stabilization of the microbiota.

| Water and metabolite absorption
A major function of the colon is the absorption of water and SCFAs, but so far absorption has rarely been implemented in in vitro systems simulating colonic conditions (Roupar et al., 2021).PETR contains a dialysis system for water and metabolite absorption.A tubular dialysis membrane passes through the entire reactor.With a diameter of 16 mm and an active length of approximately 160 cm, it has an absorption surface of 0.08 m 2 .The dialysis liquid equals the composition of the cultivation medium, but without nutrients (i.e., glucose and all complex compounds like yeast extract).It contains all minerals and additives added to the medium (vitamins and so on) to prevent their removal from the cultivation broth.In addition, the dialysis liquid contains 20% (w/w) of high molecular weight PEG (PEG 35,000) to create an osmotic removal of water from the fermentation broth.
The dialysis liquid flows in counterflow to the medium, so concentration gradients and driving forces are strongest in the distal part, supporting the development of both increasing dry matter content (Cummings & Macfarlane, 1991) and decreasing SCFA concentrations (Cummings et al., 1987) in distal direction.In addition, this prevents uptake of nutrients in the proximal region and their release at the end and is therefore preferred to simulate the progressive nutrient depletion along the colon (Macfarlane et al., 1992).Moreover, a higher absorption capacity of organic acids in the distal part contributes to the generation of the desired pH gradient.
To determine the absorption capacity of the dialysis system, PETR was continuously fed with MCM mineral salt solution (Supporting Information S2: Table SI1) containing acetate (60 mM), propionate (23 mM), and butyrate (17 mM).The concentrations (Cummings et al., 1987), molar ratio (Blaut, 2018), and daily amount ( Zihler Berner et al., 2013) were similar to the natural situation in the human colon.Two PEG concentrations (10% and 20% [w/w]) were used with two different flow rates each.The upper limit for the PEG concentration is given by increasing viscosity, since the liquid must still be pumpable.In addition, pressure in the dialysis membrane must not be too high to ensure its integrity and tightness, which limits not only the PEG concentration but also the volume flow.
Figure 5a shows the water removal rate for the four different settings calculated from the difference between the mass flows of fresh dialysis liquid and DW (Supporting Information S2: Figure SI2).
Water removal rate reached up to 1.2 L d −1 and specific water absorption was between 110 and 750 mL kg −1 (Supporting Information S2: Figure SI2D).Higher absorption rates were achieved by increasing the PEG concentration to 20% because of the higher osmotic strength.A higher inflow flushes the loaded dialysis liquid (diluted PEG) out of the system faster, also leading to an increased water removal, but only to a lesser extent.This is also reflected in decreasing specific absorption rates, indicating that the time passing through the membrane is too short to exploit the full absorption potential.Thus, increasing the flow rate has only minor impact compared to the PEG concentration.Considering the absorption surface of 0.08 m 2 , the highest removal of 1.2 L d −1 corresponds to an area-specific flow rate of 620 mL m −2 h −1 .
During these experiments, samples were taken from the different reactor modules as well as from the dialysis and reactor waste (RW) for SCFA and pH analysis (Figure 5b,c and Supporting Information S2: Figure S13).In each setup, reduction of SCFA concentrations is evident, which was higher at higher flow rate.After ~1 day, the concentrations remained largely constant indicating that an equilibrium was reached.
Although the differences are not very large, SCFA concentrations were in principle higher in the proximal modules, similar to the situation in vivo (Cummings et al., 1987).This concentration gradient is due to the inflow of SCFAs in M1 mimicking the high metabolic activity in the cecum and ascending colon (Macfarlane et al., 1992) and dialyzing in counterflow.However, as PETR is almost completely mixed within 10.4 ± 0.5 h, the concentrations differ only slightly.
However, dialysis resulted not only in lower SCFA concentrations and pH values, but also created a small but evident pH gradient along the reactor (Figure 5c and Supporting Information S2: Figure SI3).This was also described for other dialysis membrane reactor systems, where the pH gradient was a result of both absorption of SCFAs into the dialysis liquid and diffusion of alkali molecules into the culture broth (Probert & Gibson, 2004;Spratt et al., 2005).
The lowest SCFA concentrations were achieved using 5.0 kg d −1 (30 rpm) 10% PEG (Supporting Information S2: Figure SI3).However, this does not mean that most SCFAs were absorbed with this setting, because it does not consider the concomitant water removal.Thus, the absorption proportion (AP) is calculated based on a mass balance, which corresponds to the fraction of the amount in the DW in relation to the total amount.As the total amount is unknown during cultivation, AP is calculated based on the concentration ratio between DW and RW c c and their corresponding mean mass flows V ̇D and V ̇R leaving the reactor (Equation [3]).
AP decreased slightly but not significant from acetic acid to propionic and butyric acid (Figure 6).This is due to increasing molecular mass and was also observed in TIM-2, using a hollow fiber module (Minekus et al., 1999).While SCFA concentrations were lowest using 5.0 kg d −1 (30 rpm) 10% PEG, AP was highest using 2.6 kg d −1 (15 rpm) 20% PEG.However, luminal concentrations remained comparably high, because the water removal was simultaneously also remarkably high and concentrated the residual acids. ( F I G U R E 5 Water and metabolite absorption of the PEristaltic mixed Tubular bioReactor (PETR) under different dialysis conditions.Concentration of polyethylene glycol (PEG) (10% and 20% [w/w]) and pumping rate of dialysis liquid (7.5 rpm = 0.9 kg d −1 , 15 rpm = 2.6 kg d −1 , 30 rpm = 5.0 kg d −1 ) were varied and water and metabolite absorption were determined.(a) Water absorption rate calculated from the difference of the mass flows of fresh dialysis liquid and dialysis waste (Supporting Information S2: Figure S12).As water makes the largest quantity of the absorbed mass, the mass increase is assumed to equal the absorbed water volume and volumetric removal is used within the text.(b, c) Exemplary course of acetic acid concentrations and pH values for 20% PEG.Complete short chain fatty acid (SCFA) absorption data are shown in Supporting Information S2: Figure S13.This is comparable with the natural situation, were 90%-95% absorption is assumed (Zihler Berner et al., 2013), but total SCFA concentration decreases only from about 130 mM in the cecum to 80 mM in the descending colon (Cummings et al., 1987).2.6 kg d −1 of 20% PEG (15 rpm) performed best for both water removal and SCFA absorption and enabled removal of 1.2 L d −1 (63%) water and almost 80% of SCFAs.
From the 1.5-2 L water entering the colon daily, about 90% is absorbed by the colon (Sandle, 1998).With an absorption capacity of up to 1.2 L d −1 , PETR achieves 60%-80% of the physiological water removal rate in vivo.It has to be mentioned, that these absorption rates were obtained in a simplified setup and the composition of the cultivation broth can affect these rates.In addition, biofilm growth on the dialysis membrane could alter the absorption efficiency, but, however, this did not affected the removal of water in a previously described three-stage membrane based in vitro system (Spratt et al., 2005).The colon microbiota is estimated to produce 100-450 mmol SCFAs per day (Cummings & Macfarlane, 1991) but only 7-20 mmol are excreted (Cummings, 1981), leading to an estimated absorption of 90%-95%.With a feed of 1.9 L d −1 , 190 mM of SCFAs were added to the reactor, mimicking their release from the intestinal bacteria in relevant amounts.At the same time, 64%-78% of the SCFAs could be absorbed through PETRs dialysis system, corresponding to around 70%-85% of the in vivo AP.
Neither water nor metabolite absorption reaches the very high absorption rates of the human colon.One limiting factor might be the much smaller absorptive surface of the dialysis membrane (0.08 m 2 ) compared with the mucosal surface of 2 m 2 of the human colon (Helander & Fändriks, 2014), which will be addressed in further developments of the systems.However, with around 60%-80% of the physiological water removal rate and around 70%-85% of the physiological SCFA AP, PETR's dialysis system already offers substantial absorption capacities and is definitely a major advantage over the many systems that lack such a feature (Table 1).
The few in vitro systems with integrated dialysis use hollow fiber modules or as PETR tubular dialysis membranes, and the latter also used PEG for osmotic water removal (Table 1).TIM-2 simulates conditions of the proximal colon and is equipped with hollow fiber membranes with a MWCO of 50,000 Da for removal of water and metabolites (Minekus et al., 1999).The dialysis liquid equals the cultivation medium omitting the nutrients.The system showed efficient removal of the main SCFAs acetic, propionic and butyric acid with 40% removal during 1 h starting with 20 mM of each acid.
Thus, SCFA concentrations are kept within the physiological range and the withdrawal of water from the dialysis circuit leads to an increasing dry matter content.Water absorption through the dialysis system is controlled via a level sensor to maintain a constant fluid level (Minekus et al., 1999).The ARCOL system uses hollow fibers with a MWCO of 30,000 Da to maintain appropriate electrolyte and metabolite concentrations and the operating volume (Blanquet-Diot et al., 2012), but this absorption is not further specified with numbers.
Probert and Gibson ( 2004) developed a bioreactor system where an 8% (w/v) PEG 10,000 solution maintained at pH 7 circulates through a glass jacket surrounding a tubular dialysis membrane (29 mm diameter, MWCO 1000 Da).This enabled not only absorption of water and SCFAs, but also created a pH gradient from 5 to 6 along the reactor (Probert & Gibson, 2004).A similar approach was described by Spratt et al. (2005), who developed a three-stage tubular dialysis membrane reactor.Three membrane (MWCO 1000 Da, 15.3 mm diameter, 500 mm length) modules were connected in series and PEG solutions (3350 Da, 200 g L −1 ) with different pH values were circulated though their respective shells.With this arrangement, SCFA concentrations decreased over the length, a pH gradient was created, and 64% of the inflow was absorbed (Spratt et al., 2005), which is very similar to PETR's removal of up to 63%.In contrast to these two systems, PETR's dialysis tube is located within the reactor instead of forming the reactor lumen itself.This design results in a smaller mass transfer surface, but is essential for simultaneous replication of intestinal peristalsis.
The molecular weight of the PEG in PETR was with 35,000 Da considerable higher than used in other studies (Probert & Gibson, 2004;Spratt et al., 2005) due to the higher MWCO of the membrane.
A lower molecular weight would be beneficial as it increases the osmotic driving force at same concentration (w/v).By reducing the molecular mass from 10,000 to 3350 Da, Spratt et al. (2005) increased the water removal from 3.3 to 4.38 mL h −1 resulting in an area-specific flow rate of 182 mL m −2 h −1 .However, specific flow rate of PETR was much higher, reaching up to 620 mL m −2 h −1 .This might be explained by the higher MWCO that allows easier diffusion also for small molecules and water.
With the integrated dialysis system, PETR efficiently absorbs water and generated SCFAs from the reactor lumen.The use of an inner membrane facilitates peristaltic mixing and promotes the F I G U R E 6 Absorption proportion (AP) of major short chain fatty acids acetic, propionic, and butyric acid for different dialysis conditions.AP was calculated by the concentration ratio of dialysis waste (DW) and reactor waste (RW), taking into account the mass flows of DW and RW (Equation [3]).Depicted are mean and SD of the different sampling points for each condition.
generation of a longitudinal pH gradient.Even if not reaching the impressive absorption rates in vivo, the dialysis is a major advantage over many other in vitro systems (Table 1) and provides a more realistic replication of colonic conditions.Further developments will focus on increasing the absorption efficiency, e.g. by increasing the mass transfer area or increasing the (utilizable) osmotic strength of the dialysis liquid by using PEG of smaller molecular weight and reducing the dialysis system pressure.

| Temperature control and anaerobization
As mixing in PETR is rather slow, even heating over the entire reactor surface is beneficial to avoid temperature gradients.Therefore, the temperature of the surrounding incubation chamber is controlled instead of the reactor itself (Supporting Information S2: Figure SI1), also providing more flexibility during reactor and process development.As a normal intestinal temperature of 36.96 ± 0.21°C was previously described (Pearson et al., 2012), the target temperature of the incubation chamber was set to 37°C.After an initial heat-up phase of approximately 3 h, temperature is kept extremely constant in both chamber and reactor (Supporting Information S2: Figure SI4A).
Other intestinal tract models often use glass vessels as reactor compartments that are usually heated to 37°C by circulating water from a thermostat through their double jackets (Cinquin et al., 2006;Feria-Gervasio et al., 2014;Gibson et al., 1988;Molly et al., 1993).
As most bacteria in the colon are anaerobes (Mackie et al., 1999), anaerobic conditions in the lumen can be assumed.Starting with a sterile gut, infants are initially colonized by facultative organisms that consume the luminal oxygen, facilitating the colonization of obligate anaerobes (Mackie et al., 1999).In addition, oxygen is consumed through chemical reactions (i.e., lipid oxidation), leading to low oxygen levels even in germ-free mice (Friedman et al., 2018).
However, only limited data are available describing the exact conditions in humans.A low oxygen partial pressure of 1.5 kPa (corresponding to ~11 mmHg or 7% air-sat.) was reported in the human rectal lumen (Lind Due et al., 2003).Electron paramagnetic resonance imaging with activated charcoal revealed oxygen tensions of 11 and 3 mmHg (corresponding to 2% air-sat.) in the regions of the mid colon and distal sigmoid colon of mice, respectively (He et al., 1999).By phosphorescence oximetry measurements in mice, even lower oxygen concentrations <1 mmHg (corresponding to 0.6% airsat.)were reported in the cecum lumen (Albenberg et al., 2014).
Due to PETR's horizontal design, direct gassing is not applicable as it would lead to a gaseous headspace and a substantial displacement of liquid.Instead, deoxygenation is performed via the oxygen permeable silicone hoses (Robb, 1968) by gassing the surrounding incubation chamber with nitrogen.Assuming a luminal oxygen tension of 2 mmHg based on the literature described above, the dissolved oxygen must be <1.3%air-sat.This depletion is achieved ~14 h after starting the gassing of the incubation chamber (Supporting Information S2: Figure SI4B), indicating the minimum lead time before inoculation.However, even lower residual oxygen concentrations can be expected during cultivation because the medium contains cysteine as reducing agent, which chemically binds residual oxygen (Frei & Hall, 1931).In addition, facultative bacteria can consume residual oxygen (Mackie et al., 1999).
Although not addressed in this study, PETR's design allows to mimic radial luminal oxygen gradients present in vivo (Albenberg et al., 2014) by controlling the oxygen inflow by setting a defined oxygen partial pressure in the incubation chamber.This is very promising for future investigations to replicate the characteristic conditions of mucosal biofilms with microaerobic conditions at the base and anaerobic conditions at the luminal site (Marzorati et al., 2011).

| Cultivation of B. animalis YL2
To verify PETR's biological applicability and long-term functionality, a 7 days cultivation of a typical gut bacterium, B. animalis YL2, a member of the Oligo-Mouse-Microbiota OMM 12 (Brugiroux et al., 2016), was performed.Continuous cultivation of intestinal microbiota in vitro is often preceded by a batch phase (Barroso et al., 2015a;Feria-Gervasio et al., 2014;Macfarlane et al., 1998) and this procedure was adopted for PETR. Figure 7 shows PETR's pH course during 16 h initial batch and following 7 days continuous cultivation.
The pH decrease during batch was very similar to that observed in serum bottles (Supporting Information S2: Figure SI5), indicating that PETR provides comparable conditions (i.e., appropriate temperature, anaerobia, and mixing).The pH value dropped to 4.8 during 6 h, which is in good agreement with previously reported inhibition of B.
Continuous cultivation was initiated by starting the addition of fresh medium and dialysis liquid as well as the pH control.Starting from an identical pH value in all modules, the desired gradient was created in about 2.5 h.After initial generation, the longitudinal gradient was maintained fairly stable until stopping the cultivation after 7 days, demonstrating successful replication of the colonic pH gradient during continuous cultivation.To maintain the different set points, acid and base were pumped almost linearly in all modules, but mostly into M1 (201 mL d −1 ) and M4 (264 mL d −1 ) with the lowest and highest set point, respectively.Only slight corrections were necessary in M2 (44 mL d −1 ) and M3 (32 mL d −1 ) as indicated by the small amounts of NaOH added to this modules and can be explained by concomitant resorption of organic acids through the dialysis system supporting the pH gradient together with the tubular design (cf. Figure 5c).These data indicate that PETR is capable to generate a real progressive pH gradient instead of few different levels as in cascaded stirred tank systems (Table 1).
Figure 8 depicts the course of OD as well as concentrations of several organic acids in the reactor modules and the DW.The OD increased slightly from 3.2 after the batch phase to 4.2 after 4 days and then remained constant.Total organic acid concentration of 42 mM was measured after the batch phase, and acetic acid, lactic acid, succinic acid and formic acid were detected in significant amounts as previously confirmed (Scardovi & Trovatelli, 1974;van der Meulen et al., 2006).All organic acids were detected in the DW F I G U E 8 Course of optical densities and main organic acids during continuous cultivation of B. animalis YL2 in the PEristaltic mixed Tubular bioReactor (PETR).All modules were sampled and organic acids were also quantified in the dialysis waste.
as well.However, although considerable amounts were absorbed (Supporting Information S2: Figure SI7), luminal concentrations decreased only slightly.From the main SCFAs, neither propionic nor butyric acid was detectable, which are basically not produced by B. animalis (Scardovi & Trovatelli, 1974).
The mean water absorption was determined to 840 ± 50 mL d −1 (Supporting Information S2: Figure SI6), which was lower than in the abiotic characterization experiments and might be explained by the higher osmolality of the culture broth, reducing the osmotic gradient to the dialysis liquid.This can be clearly seen at the decreased specific water absorption of 310 instead of 460 mL L −1 .In addition, the removal rate decreases slightly over time due to a decreasing flow rate of fresh dialysis liquid, which is a common issue with longterm operated peristaltic pumps (Woodgate, 2018).However, the dialysis system enabled an OD increase by 30%.Although the water removal was lower during the cultivation, it was still remarkable and is definitely a benefit compared to systems lacking a dialysis system.
In contrast to hollow fiber membranes (Minekus et al., 1995), the integrated dialysis has advantages regarding membrane fouling, as dialysis could be performed for 1 week with only slightly decreasing water removal rates.Even formation of biofilms on the membrane should not affect the water absorption considerably (Spratt et al., 2005).
Besides absorption of water, substantial amounts of organic acids were detected in the DW, reaching 44% to 72% of the concentration in the reactor lumen (Supporting Information S2: Figure SI7A).APs for each organic acid ranged from 48% to 60% and decreased again with increasing molecular mass (Supporting Information S2: Figure SI7B).This decrease was significant (p < 0.001) from formic to acetic acid and from lactic to succinic acid, but not significant for the decrease from acetic to lactic acid.APs were lower than in the characterization experiments, which might be due to different media composition, lower water resorption rates as well as addition of acid and base for pH adjustment increasing the waste flow.These are starting points for further improvements at the functional level, e.g. by decreasing the volumes of added acid and base, or increasing the absorption efficiency as previously discussed (Section 3.2.4).Compared with typical SCFA concentrations in the human colon, the sum of organic acid concentrations was rather low and two of the principle SCFAs (propionate and butyrate) were completely absent.However, this was a first feasibility study and neither a single strain nor the cultivation medium used can properly simulate the situation in vivo.Further experiments with intestinal microbiota will provide more insights into the applicability of PETR for microbiota studies and further validation of this system.Sources for these microbiota can be fresh or frozen (or preserved by any other technique) stool samples from a single donor or pooled from multiple donors (Isenring et al., 2023).Especially for basic research, the application of defined microbiota with known composition has several advantages regarding reproducibility and comparability and reduced systems complexity.Such defined consortia are commonly used in gnotobiotic mice models, but some were also used in vitro in bioreactors (van Leeuwen et al., 2023).Some examples are the long and widely used Altered Schaedler Flora ASF (Wymore Brand et al., 2015), the more recently established Oligo-Mouse Microbiota OMM 12 (Stecher, 2021), or the extended simplified human microbiota SIHUMIx, which showed already good stability in vitro (Krause et al., 2020;Schäpe et al., 2019).Sophisticated cultivation media like the simulated ileal efflux medium mimicking the food residues entering the colon (Venema, 2015) could further enhance the preciseness of in vivo simulation in PETR.

| Limitations and perspectives
PETR was shown to achieve the most important conditions and functions of the natural human colon.However, as in any model of a very complex system, there are some simplifications and limitations in mimicking the natural functions and conditions.For example, the size of the large intestine is highly variable (range 80-313 cm), but PETR (~165 cm) is designed to match average dimensions (160.5 ± 33 cm) (Hounnou et al., 2002).Nevertheless, the length of the silicone hoses can be adjusted (in certain areas) and the whole reactor is on a flexible, slidable reactor mount for easy readjustment of all positions.
PETR has a constant inner diameter of 45-47.5 mm, reflecting the mean diameter of 50 mm (Kararli, 1995), but the diameter is very individual as well and becomes smaller in the descending colon (Mark et al., 2021).Although the geometric dimensions meet those of the colon, the colonic volumes differ significantly.Whereas PETR has a working volume of 2.4 L, a mean colonic volume of 630 mL was reported for healthy humans, which was again strongly individual and ranged from 400 to 1300 mL (Klinge et al., 2022).This discrepancy might result from the flexures and more complex structure of the colon (e.g., semilunar folds and haustra) that differs from a simple tube with flat surface.In addition, a complete filling without considering accumulated fermentation gases was assumed.Other simplifications include the relatively nonselective absorption or that pH control is achieved by only four measuring points, which however seems to be sufficient for a gradual shift from 5.5 to 7.0.However, such limitations and simplifications apply not only to PETR, but to all systems that mimic human colon conditions in vivo, and PETR matches the colonic architecture much better than many other systems (Table 1).
PETR combines already more conditions and functionalities of the natural colon environment and provides a more comprehensive in vitro simulation than other previously described systems.Nevertheless, there are still some topics for further improvements, like increasing the absorption efficiency as discussed before.This will also result in more realistic solid contents and viscosities.The current peristaltic mixing only addresses local contractions to produce sufficient mixing, but a holistic simulation would require a combination of all different spatiotemporal contraction pattern, that is, both local and propagating contractions and both of different occlusion degrees (Sarna, 2010).A similar technique as used in the DCM could further enhance the accuracy of replicating colonic motion (O'Farrell et al., 2022).As the majority of other systems (Table 1), PETR does not take into account the complex host-microbiota interactions so far.However, the system could be easily coupled with external devices such as the Host-Microbiota-Interaction Module (Marzorati et al., 2014).
So far, the biological application was shown with only one exemplary gut bacterium, and in vitro cultivation of intestinal microbiota is necessary to compare, evaluate and validate the system further.

| CONCLUSIONS
A novel and sophisticated bioreactor simulating human colonic conditions for in vitro cultivation of intestinal microbiota was developed and characterized.PETR does not aim to simulate the complex interactions between microbiota and human host, but is a Section 4.1 (Results and Discussion).For better understanding of the following procedures and the assignment of the different positions (Modules M1-M4) mentioned below, the reader is referred to Figure 1a, showing the design of PETR.F I G U R E 1 Overview of the PEristaltic mixed Tubular bioReactor (PETR) as three-dimensional CAD model.(a) Entire reactor on the flexible guide rail system with indicated main fluid streams, target pH values, and assignment of the different colon parts.(b) Detailed view of the first stainless steel module (M1).(c) Detailed view of one kneading module.(d) Fixation of the integrated dialysis tube with specially designed connection tubes at the ends of the reactor.The membrane is fixed and sealed with a groove and sealing ring.

2. 1 |
Bacterial strain and cultivation medium B. animalis YL2 (DSM 26074) was chosen as model intestinal bacterium.Cultivations were performed in pre-reduced mixed culture medium (MCM) containing several complex ingredients (yeast extract, meat extract, peptones), mucin, glucose, minerals and vitamins.The exact composition and preparation procedure are described in Supporting Information S2: Table This solution was used to perform long-lasting abiotic experiments (pH control and absorption experiments) in unsterile conditions without the risk of growth of contaminants.The pH was adjusted to 5.5 and the solution was fed at 1.35 mL min −1 .Acetic acid (0.1 M) was added at 0.3 (M1), 0.3 (M2), 0.1 (M3), and 0.1 mL min −1 (M4) through the sampling ports to simulate acetic acid formation observed during cultivation of B. animalis YL2 in German Collection of Microorganisms and Cell Cultures GmbH medium 58 in serum bottles (22 mM acetic acid during 12 h, data not shown).

2. 5 |
Temperature and oxygen control PETR (without dialysis system) was filled with demineralized water and equipped with an oxygen electrode (Visiferm DO225, Hamilton Bonaduz AG) in M3.The reactor was positioned in an in-house fabricated incubation chamber (3 m × 0.5 m × 0.5 m) specially designed to fit PETR's peripheral requirements.One of the three temperature sensors of the incubation chamber heating system was plugged into the pH port of M4.
electrode and acid (0.5 M HCl) or base (1 M NaOH) can be added through a stainless steel tube.The pH values of the different modules are measured with an Arduino Nano microcontroller and sent around every 10 s to a computer.A Processing script compares actual and target values and controls the dosage of either acid (M1) or base (M2-4) to each module through a computer-controlled four-channel peristaltic pump with independent channel control (Ismatec Reglo ICC, Cole-Parmer Instrument Company).Process data are logged in a text file and graphically displayed using Python for online process monitoring.
Mixing time determination in the PEristaltic mixed Tubular bioReactor (PETR).(a) The tracer was added in M1 and detected in M4 to determine the forward mixing time and (b) vice versa for the backward mixing time.Both experiments were conducted as triplicates and their respective means (black line) and SDs (gray area) are indicated to achieve 95% homogeneity.

F I U R E 3
Residence time distribution of three replicates with indicated mean (black vertical line) of the mean residence time of the replicates.

F
I G U R E 4 Establishment and maintenance of the pH gradient in the PEristaltic mixed Tubular bioReactor (PETR).The cultivationassociated formation of organic acids was simulated by adding 0.1 M acetic acid, distributed over the four modules, to the reactor.It took about 1 h for the generation of the desired gradient which was then kept relatively constant.The solid lines represent the measured pH values and the dashed lines the volume of added HCl (M1) or NaOH (M2-M4).The horizontal dotted lines indicate the respective set points.

F
I G U R E 7 pH measurement and control in the PEristaltic mixed Tubular bioReactor (PETR) during cultivation of B. animalis YL2.The gray background indicates the initial batch phase where the pH was measured but not controlled.After 16 h, continuous operation (pH control, feed of medium and dialysis liquid) was started.The solid lines represent the online measured pH values and the dashed lines show the volume of supplied acid (M1) or base (M2-M4).The set points are indicated by horizontal black dotted lines.The pH of the dialysis waste was measured offline during sampling (red circles).
promising bioreactor system providing a controllable and reproducible simulated colon environment.The single-tube design mimics in vivo anatomy and conditions more accurately (e.g., gradients of nutrients, pH and microorganisms) than the often used cascades of stirred bioreactors with only limited distinct levels, and periodic squeezing of silicone hoses provides more natural fluid motion than a stirred bioreactor.PETR combines more properties and functionalities of the natural colon environment and offers a unique and more comprehensive in vitro simulation than other previously described systems.This is a further step to precise replication of colonic conditions in vitro for reliable and reproducible microbiota research such as testing of drugs, different nutrition or food components, prebiotics and probiotics, investigations of interactions on the microbiota level or colonization by pathogens and how to treat them.As flexible and versatile in vitro system, PETR offers unique possibilities without the limitations of human in vivo studies.