1. Introduction

In recent times, scientific interest has addressed the development of functional ingredients or foods specifically conceived to improve human health and wellbeing [1]. Some functional foods or ingredients contain probiotics, “living microorganisms that, when administered in a sufficient number, confer a benefit on the health of the host” [2]. Probiotics may exhibit these health benefits through a plethora of processes. Health benefits range from regulating intestinal flora to maintaining the host immune homeostasis and producing specific molecules with antioxidant and antibacterial properties. In addition, they can support the human body in preventing obesity and other dysmetabolic diseases [3]. They can help in maintaining host immune homeostasis by downregulating inflammatory cytokines, inducing immune stimulation, and modulating gut-associated gene expression. They can prevent obesity and related chronic diseases by regulating cholesterol-, lipid-, and glucose-metabolism [4]. Furthermore, probiotics can beneficially affect human health through some metabolic by-products generated in a matrix during the anaerobic fermentation of organic nutrients [5]. The need is to receive enough number of such microorganisms in foods. Still, their amount should not decrease in the gastrointestinal tract (the number of live bacteria should be not less than 106–107 CFU/g to give beneficial effects, where CFU stands for Colony-Forming Unit) to provide their health benefits. Even though some bacteria strains are considered probiotics based on lab tests and medical studies, they are susceptible to certain conditions. Heat, oxygen, pH levels, water activity, stomach enzymes, bile salts, and other stress factors can affect them. One of the most used methods to protect probiotics is represented by microencapsulation, which allows the probiotic to resist stress, reach the colon in optimal conditions, colonize the environment, and begin to exhibit all those positive properties stated above.

Microencapsulation is a process that allows enclosing micron-size particles, living organisms, or food ingredients into a shell or coating exhibiting desirable properties. Alginate polymer is often preferrable as a material for the shell/coating matrix since it is food grade and safe grade, so much that it is also used in the food and pharmaceutical industries to provide the controlled release of the capsules’ content. Introduction of microencapsulation is significantly broadening the number of functional foods [6]. In recent years, there has been an increased interest in developing new encapsulation systems capable of protecting probiotics more and more effectively, ensuring the preservation of their vitality and biological properties [7]. Generally, the gastrointestinal transit of microcapsules containing probiotics can be followed through in vitro systems and, at certain times, similar to those of gastric and pancreatic digestion. When the microcapsules are recovered, the effectiveness of the microencapsulation process can be followed with conventional techniques, commonly used in microbiology, which, although highly valid and always current, require hours, or even days to give a result. A striking example is the treatment of sodium alginate microcapsules with a simulated gastric juice and pancreatic juice and the subsequent treatment, for example, with sodium citrate, which determines the widening of the meshes of the alginate network and the release of the bacterial cells, which, once recovered, are used for microbial counting on specific growth medium plates. Although widely used, the plate counting method for viability assessments comes with several limitations, mainly that culturing bacteria on agar plates is highly time-consuming, mostly due to the preparation of media and long incubation times [8,9]. Hence, for the time being, plate microbial counting could be unacceptable in sectors such as industrial and control ones, which require increasingly more analytical, fast, and sensitive methodologies [10]. Modern methods, such as droplet microfluidics, can allow the high-throughput analysis of bacteria [11]; new nanotechnology methods based on optical techniques have been developed in some cases, especially for monitoring and detecting pathogenic bacteria. The most recently developed strategies to develop sensors for detecting pathogens, such as Escherichia coli, are colorimetric, fluorescence, surface plasmon resonance, surface-enhanced Raman spectroscopy, localized surface plasmon resonance, and chemiluminescence. At present, optical detection of E. coli in smartphone, paper-based, and portable devices are also considered [12].

Here we follow an optical approach, sketched in Fig. 1, for profiling bacteria strains that are candidates to serve as probiotics. We develop a platform relying on a coherent light probe that screens the strains based on their motility properties and rank them based on their attitude to be microencapsulated [13,14], to act as expected during microencapsulation, and to move freely inside the destination buffer after the release from the capsules. The platform investigates the all-optical spatial-temporal signature of motile bacteria and the way these interact with the light probe. It consists of two complementary and distinct modules. The former is a Bio-Speckle Decorrelation Analysis (BSDA) apparatus aimed to test the microencapsulation performance at the cell population level. The latter is a Digital Holographic (DH) microscope that can follow in real time and map the 3D motion of single elements of the strain after these are released by the capsules.

 

Fig. 1. Dual-module platform for screening probiotic bacteria candidates. (a) Ideal behavior of effectively microencapsulated probiotics under simulated gastro-intestinal transit. (b) Proposed test based on BSDA and 3D holographic tracking. The test can discard ineffective microencapsulation and bacteria candidates exhibiting unsuitable motion profiles out of the capsules.

Download Full Size | PDF

Bacteria have been studied with optical approaches both at the single element level and as bacteria colony ensembles. At the single cell level, morphological features of single species (e.g., length and diameter of E. coli) have been measured by inverse light scattering [15] and Fourier transform light scattering [16]. Ensemble approaches do not resolve necessarily the single element but have the advantage of inferring information from statistically relevant populations. Microbial contamination from Pseudomonas aeruginosa, Staphylococcus aureus, and Bacillus subtilis has been monitored by elastic light scattering approaches [17]. Within the framework of smart farming, food production and conservation, hyperspectral imaging approaches, often aided by artificial intelligence, are widely used to detect the presence of heterogeneous mixtures of bacteria strains [18,19].

All the above-mentioned approaches can detect the presence of bacteria and characterize the population. However, they do not provide motility information to be associated to a specific condition. Biospeckle dynamics can be employed instead at this scope. The temporal decorrelation of the biospeckle can be used to assay biological activity [20–22]. Also, the contrast of speckle patterns can be associated to the bacterial load and bioactivity [23,24]. Biospeckle has been used to distinguish fungi from motile bacteria on agar plates [25], to detect the presence of motile E. coli and Bacillus cereus in turkey breast [26], and to assess the performance of microencapsulation [27]. In this framework, AI can be used as a segmentation tool, e.g., to associate various parts of an object under test with the detected bacterial activities [28]. As mentioned before, the sole BSDA is valuable to assess the microencapsulation effectiveness but is not enough to rank the strains based on the single cell motility. In other words, a population could undergo microencapsulation with a high percentage of vital bacterial load out of the capsules. This condition would produce large biospeckle decorrelation. However, the single elements (not resolvable or trackable by BSDA) could be not motile enough to reach the epithelial cells of the intestine to complete the probiotic action. Hence, a method that resolves and track the motion of the single elements should follow the BSDA.

DH is one of the preferred microscopy modalities to image and track microparticles, cells and living organisms in liquid environment. It is an interferometric technique well assessed and widely applied in diagnostics and medicine for blood counting and screening [29–31], circulating tumor cells detection [32] and drug resistance assays [33], sperm motility analysis [34,35], parasites and waterborne pathogens detection [36], as well as environmental monitoring, also in the form of field portable devices [37,38]. Exploiting the advantageous DH numerical refocusing, 3D-tracking of bacteria freely swimming inside a volume of liquid has been performed [39–41]. Both low density and high-density ensembles [40] have been tracked, and the local interplay between the elements has been investigated using optimization of cost functions for localization [41] or machine learning approaches [42].

To the best of our knowledge, the two methods at the basis of the two platform modules have never been bridged to characterize microencapsulation performance and to screen probiotics. While the first BSDA test is a vitality assay inherently related to microencapsulation, DH characterizes their capability to move free inside the basic environment which is an important feature to reach the epithelial cells of the intestine walls.

2. Working principle

2.1 Methodology

Figure 1 sketches the proposed method to analyze and rank bacteria strains as potential probiotics. Encapsulated bacteria are exposed to liquid at different pH conditions that simulate the gastrointestinal transit [14] and probed in through transmission by coherent laser light. The biospeckle pattern recorded by the system and its evolution over time provides ensemble information of the microencapsulation performance in relation to the liquid environment. BSDA can monitor the evolution of the speckles with high sensitivity to tiny space-time shifts. Motile bacteria populations are expected to decorrelate the speckle pattern over time when released by the capsules. Ideally, this condition should be verified only in the presence of basic environment. Indeed, alginate microcapsules are expected to shrink in the presence of an acid solution (simulating the passage through the human stomach) and to dissolve within a basic environment (e.g., the human intestine), thus releasing their microbial content (Fig. 1(a)). Once tightened, the average size of the alginate meshes is slightly lower than 1 µm; thus, meshes do not allow single micron-sized elements to exit until the different external pH condition is reached.

BSDA access a wide Field of View (FOV) at the cost of lateral resolution. Thus, the information gathered by the BSDA are statistically significant since they account for the ensemble behavior of many bacteria. Despite lateral resolution of the BSDA module is not large enough to spot single bacteria (typically 1 micron-sized), the speckle pattern they project onto the sensor in far field is observable and resolvable in time and space. In this sense, the use of temporally coherent light is convenient. This module acts as a first screener of potential candidates and allows to assess the microencapsulation performance, but cannot yield a single-element characterization of the locomotion profiles of a strain. Once a candidate strain has exhibited a potential ensemble probiotic locomotion profile, the platform allows further analysis at the single cell level (see the sketch of the screening pipeline in Fig. 1(b)). At this scope we use a DH microscopy apparatus in transmission mode. The DH setup provides higher lateral resolution to image each single cell of the colony. A DH microscope encodes the whole complex information of label-free samples into a fringe pattern modulated by the sample. Once reconstructed, the hologram allows accessing the object complex amplitude, i.e. both the amplitude and the phase-contrast map. The reconstructed wavefront can be then back-propagated in post-processing. In other words, each sample can be registered out of focus and then refocused afterwards. The numerical refocusing capability allows to image motile samples while they move in 3D during holographic sequence captures, and to quantitatively estimate the coordinates and 3D motion dynamics. Thus, here we characterize the 3D motion profiles in qualitative and quantitative mode, thus furnishing a ranking between the strains based on their motility profiles.

Due to the high sensitivity of optical systems based on coherent light, both modules suffer from a performance worsening in the presence of non-ideal conditions of the measurement environment. We first calibrate and test the effective use of the platform by analyzing Lacticaseibacillus rhamnosus, a strain that has shown in previous analysis good probiotic properties [27]. Then, we successfully screen and rank five more bacteria strains, and we prove the complementarity and a substantial coherence between the readout of both modules that result in the unambiguous ranking between them. We benchmark the flexibility of the system under non ideal lab-environment conditions (e.g., uncompensated shaking, convective air fluxes, seismic noise) with the aim to demonstrate its applicability out of the optical lab, e.g., in industrial settings. Within this scope, we develop a novel software suite that can cope with the above-mentioned non-ideal conditions. We believe that the label-free microscopy approach, the absence of dyes, tags and chemical analysis, and simplicity of the apparatus outperforms in versatility the existing solutions.

2.2 Experimental setup for BSDA

The speckle decorrelation setup is built on a vibration isolation optical table, which can isolate vibrations from 3 Hz to 50 Hz when working in normal mode. A He-Ne laser (emitting at 632.8 nm wavelength) is chosen to be the light source for ensuring that the coherence is sufficient to produce speckles. Herein, the minimal acceptable coherence length for creating high-contrast speckles is 11.87 mm, and He-Ne laser can meet this need. Laser beam is expanded and filtered by a Beam Expander (BE) setup after its light intensity is adjusted by a neutral density filter. In particular, expansion allows turning the beam diameter from 2 mm to 15 mm. After basic beam shaping, mirrors M₁, M₂, M₃ are used for light path adjustment (beam steering system), and the angle between illumination beam and sample surface is controlled by M₃. The sample is placed on a tip & tilt stage; in a conventional BSDA setup, the angle between the sample surface to be measured and the illumination beam is 45°, as sketched in Fig. 2(a). A Charged Coupled Device (CCD) camera is positioned on the same side as the reflected beam; adjustable camera mount will ensure the reflected beam is orthogonal to the camera screen. Here, XIMEA MD028MU-SY, 1940 × 1460 (pixel size: 4.54 µm) is used to record speckle images; its frame rate is set to 15 fps. However, in order to avoid secondary reflections from the objects in the Petri dish that generate unwanted interference during speckle recording (as sketched in Fig. 2(c)), herein we slightly modified the typical 45° illumination geometry by introducing an angle α as sketched in Fig. 2(d). The two reflective surfaces are the upper surface of the solution and the bottom surface of the petri dish. The beam reflected from the bottom of the Petri dish carries information of targets inside the solution, while the beam reflected by the interface between air and the liquid buffer does not convey useful insight and should be considered as an unwanted interference term. When 45° illumination is performed, the two reflected beams will form interference fringes on the CCD screen; conversely, introducing a small angle, α, with respect to this configuration has the effect of shifting the area of specular reflection and thus it avoids the unwanted interference (Fig. 2(d)). In an ideal state of 45° illumination, the distance of speckle patterns of two beams on CCD screen depends on incident angle, refractive index and thickness of the solution. Therefore, once the height of the solution in the Petri dish is fixed, the incident angle of illumination beam can be adjusted by slightly tilting the sample plane. In our experiments, the angle $\alpha \approx 2^\circ $ was chosen. Measurements are carried out for three minutes for each sample, although one-minute recording is in principle enough to reach a stable saturation level, as it will be discussed in the following. To ensure that obtained data from different experiments are comparable, we choose to implement speckle recording always at the edge of one capsule, which means that in a single FOV there is at least one observable capsule. Once the sequence of speckle patterns is registered, we select a suitable measurement region and calculate the bio-speckle correlation coefficient with respect to the first frame. The region for performing decorrelation calculations is selected as a 1000 × 1000 square pixels at around 100 pixels distance along the x-axis from the edge of the capsule. In the following we describe the signal processing steps we propose to get rid of unwanted disturbance due to a non-ideal measurement environment.

 

Fig. 2. BSDA system. (a) Schematic diagram of optical system. (b) photo of the setup. (c) and (d) shows the introduction of the angle α to avoid unwanted interference.

Download Full Size | PDF

2.3 BSDA calibration

First, BSDA requires the compensation of noise from high frequency airflow disturbance. Since experiments are carried out by removing the cover lid of the Petri, in principle this disturbance due to convective airflow is hard to be avoided in full, and is not compensated by the optical table neither. We developed a specific algorithm to cope with this issue, which is described in the Methods section. In order to calibrate the system and the relative algorithms, we first inserted the test strain (L. rhamnosus) in alginate microcapsules and filled the dish with a Phosphate Buffered Saline (PBS) buffer. In this condition no decorrelation is expected due to the bacteria movement (in PBS the alginate meshes keep narrow and constraint the bacterial load inside the capsules), so any decay of the biospeckle correlation coefficient should be attributed to external factors. Then, we test the behaviour of the capsules with buffers at different pH as previously described. Figure 3 shows an example of application of the Normalized Cross-Correlation (NCC) strategy (described in the methods section) [43] to the case of L. rhamnosus. Different frames are shown in Fig. 3 before (a) and after (b) applying the NCC algorithm to compensate for the FOV motion, Visualization 1 shows the video results. Figure 3(c) shows the corresponding biospeckle correlation coefficient, ρ. This drops down to average values close to ρ∼0.55 (green dotted line), a decay that has to be attributed to environmental motions. After applying the NCC algorithm, the correlation curve gets stabilized (red solid line) with average values ρ∼0.90, which proves the effectiveness of the method in discarding the contribution of unwanted convective airflows. The same algorithm can be applied to all the sequences without the need to check whether the disturbance affected the measurement, since the algorithm does not change substantially the curves obtained in the absence of airflow disturbance (see e.g., Fig. 3(d)). In this way, supervision by the operator is not necessary. Figure 3(e) shows the biospeckle correlation coefficient vs. time for different buffer conditions. In particular, we report the case of capsules in PBS (light blue line), pH 2 buffer (red line), pH 8 buffer (yellow line) and capsules exposed to the pH 8 buffer for 3 h (purple line). The light-blue curve corresponds to a negative control and substantially determines the limit of detection of the system, i.e., correlation decays smaller than the saturation value of the light blue curve cannot be ascribed to bacteria ensemble motion. The red curve shows a substantial stability, which means that most of the bacteria remain inside the capsule in the case of the pH 2 buffer, as expected as a result of an effective microencapsulation process. On the contrary, the decay measured in the case of the yellow curve indicates how in pH 8 the process of enlargement of the alginate meshes and diffusion of the bacterial load occurs in the first 10 s after the contact with the basic buffer. Hence, the biospeckle decorrelation curve reaches values ρ<0.6 after 40s from exposure to the buffer. However, the decay of the purple curve is sharper than the yellow curve. This means that after 3 h from the exposure to the pH 8 buffer, most of the bacteria are outside the capsules with motility. After 60 s the purple curve reaches values ρ<0.5, i.e. the biospeckle is uncorrelated after one minute. These observations demonstrate the effectiveness of the BSDA test and suggest how observing the capsules after 3 h from the exposure to pH 8 buffer is convenient to access an ensemble picture of the status of motility of bacteria after microencapsulation. Thus, we will hereafter use this analysis protocol for comparing the different strains on the basis of the BSDA.

 

Fig. 3. (Visualization 1) Suppression of high-frequency airflow disturbance by NCC algorithm applied to BSDA. (a) Original sequence of speckle patterns subject to FOV motion. (b) Calibrated sequence. The yellow area shown in (a) and (b) are the result of overlaying the corresponding area of the first frame with 50% transparency and related frames. (c) Measurements of microencapsulated L. rhamnosus in the presence of high-frequency vibration, pH 2. Measurements of microencapsulated L. rhamnosus in the absence of high-frequency vibration, pH 8. (e) BSDA of L. rhamnosus after applying the NCC algorithm. Each data curve is calculated based on the average of three sets of measurement with the same time period.

Download Full Size | PDF

In principle, the use of the NCC method should be enough to remove unwanted vibrations due to non-ideal environmental conditions of the laboratory where the platform is operated. However, unwanted and uncompensated interfering signals could occur during an operation of candidates screening in industrially relevant environments. For instance, this was the case for some of the measurements carried out during our experimental campaign, where an unexpected periodic interference signal (with long period T∼21 s) was detected as superimposed to the biospeckle decorrelation signal. A mitigation strategy for unexpected low-frequency interference is described in the Methods section.

2.4 Setup for holographic 3D-tracking of bacteria

A DH microscopy system has been optimized for the analysis of motility profiles of individual bacterial elements belonging to selected strains as potential probiotic candidates. We implemented a Mach-Zehnder interferometer in off-axis configuration coupled with a CW laser diode (Melles Griot, 5 mW @ 473 nm), as shown in Fig. 4(a). The use of blue laser wavelength is the best option to maximize resolution. The generated light wave is split by a Beam Splitter (BS₁) in two contributions, namely object and reference beams. The object beam is reflected by M₁. After passing through the sample, it is collected by a Microscope Objective (MO₁, Nikon, 50×/NA = 0.45). Simultaneously, the reference beam passes through a Iris Diaphragm (ID) and a second Microscope Objective (MO₂, Newport, 20×) for balancing the phase curvature generated by MO₁. Finally, the two contributions are recombined by a second Beam Splitter (BS2) and the resulting fringe pattern is recoded by a CMOS camera (CP-M UI-3370CP-M-GL Rev.2, 2048 × 2048 pixels, 5.5 µm pixel size). With this setup implementation the spatial resolution, measured using a test resolution target (Newport HIGHRES-1), is 512.0 lp/mm, corresponding to the group 9-1 of USAF 1951 target, as shown in Fig. 4(c). We recorded DH videos with frequencies of capture equal to 5 fps. The obtained holograms are the intensity images of the interference pattern between object (O) and reference (R) beams:

Therefore, as shown in Eq. (1), the off-axis configuration provides a resulting signal consisting in a low frequencies contribution $({|R |^2} + {|O |^2})$, and other two terms, called real (+1 order, ${R^\mathrm{\ast }}O$) and conjugate (-1 order, $R{O^\mathrm{\ast }}$) images, which are of interest since they contain the object, O, as a complex term . Therefore, for suppressing unwanted terms we applied filtering in the Fourier domain which allows separating the aforementioned contributions contained in the holograms [44]. After obtaining the demodulated holograms, we solved the diffraction propagation integral numerically by means of Angular Spectrum method which allows refocusing each single bacterium in a different focus plane than the acquisition plane for each frame, improving the contrast and simplifying the segmentation step. Indeed, label-free bacteria are transparent pure phase objects that show a sharp phase-contrast when they are in the best focus plane. The focus plane is unpredictable a priori due to the bacteria motion in the volume, thus the possibility to refocus a posteriori is an important capability of DH microscopy exploited here.

 

Fig. 4. Digital holographic geometry for 3D bacteria tracking. (a) Schematic view of the DH setup. BS, Beam Splitters; M, Mirrors; MO, Microscope Objectives; ID, Iris Diaphragm. (b) Digital hologram of USAF-1951 resolution target, interference fringes are revealed in zoom in area. (c) Reconstructed amplitude, sectional view of group 9-1 is shown.

Download Full Size | PDF

Recovering the best-focus condition provides a correct estimation of axial and lateral position of the single bacterium; the identification of the best-focus distance is assured using the Tamura Coefficient (TC) optimization criteria [45]. Despite the actual magnification of the DH system is equal to 62×, tracking the bacterial elements is not a trivial process due to their small size (about 1 µm length). Improving the system magnification and lateral resolution would further reduce the available FOV, which would impair visualizing enough bacteria coming out of the capsules. In this regard, we implemented a specific 3D tracking algorithm to tackle this problem without changing the optical configuration, as described in the Methods section. After the 3D tracking of each element is performed, 3D velocity calculation allows establishing a ranking between the strains.

3. Experimental results

We benchmarked the developed platform by considering five more different strains, namely L. acidophilus, Lacticaseibacillus casei Shirota, Lactobacillus gasseri, Lactiplantibacillus plantarum, and a strain derived from fermented Sardinian cheese, hereafter indicated with 37 5 AS. All the tested strains are rod-like shaped with approximately 1 µm length and an average ∼0.8 µm diameter. Microencapsulation performance and loco-motion profiles have been evaluated qualitatively and quantitatively. Thus, as an outcome of the analysis we conducted, we ranked them in terms of their suitability to serve as probiotics.

According to the results of the calibration experiments, microcapsules were left 3 hours inside the basic buffer in order to permit the micro-meshes to enlarge and the bacteria content to be effectively released and detected in the form of ensemble speckle pattern by the BSDA module. In the case of the DH microscopy experiments, we chose to assay the sole capsules with OD = 0.50, in order to maximize the probability to observe single bacteria elements in the reduced FOV of the DH optical system. Data processing and analysis was conducted using the suite of BSDA algorithms described in the Methods section. In the case of the DH analysis, we used conventional routines for reconstructing off-axis holograms by demodulation in the Fourier domain, and backpropagation based on the Angular Spectrum method. The determination of the frames of interest from the DH video sequences and the contextual estimation of the 3D centroid of the cells were carried applying the Fourier Spectra correlation (FSC) method and the full DH-based 3D tracking pipeline described in the Methods section.

Figure 5 shows the trends of the speckle correlation coefficient for all the analyzed strains, for both the ODs, and for both pH conditions of the buffer. All the tested strains exhibited satisfactory microencapsulation performance, suggesting the efficacy of the protocol of preparation of alginate microcapsules and the whole microencapsulation process.

 

Fig. 5. BSDA of the probiotic candidates L. acidophilus, L. plantarum e L. gasseri, L. casei Shirota, 37 5 AS. (Left) OD = 0.50. (Center) OD = 0.35. (Right) Comparison between the two different bacterial loads after 3 hours exposure of microcapsules to pH 8 buffer.

Download Full Size | PDF

In the cases of pH 2, the correlation signal keeps stable over the observation time, as expected, from which we can infer that the microcapsules are able to keep their meshes narrow enough to trap their bacterial content in the case of the acid environment simulating the stomach. This trend is apparent for both ODs, thus suggesting that even the larger OD = 0.50 bacterial load does not affect the efficacy of the alginate meshes in their confining action. This is also a confirmation that the system sensitivity, mostly governed by the pH 2 negative control experiments, does not change with the OD, which allows one making direct comparisons between the measured curves at different bacterial loads. In the case of pH 8 buffer, we observe a signal decorrelating fast over time. In this condition, capsules enlarge enough their meshes (as sketched in Fig. 1(a)) and release their bacterial load inside the buffer. Observations of Fig. 5 suggest that bacteria are vital after microencapsulation and free to move inside the medium. The ensemble effect of the collective motion of the population out of the capsules is a non-negligible decorrelation of the speckle pattern, which is apparent from the plots of Fig. 5. The signal start decreasing over time until it reaches a saturation value, after which it does not change significantly. This also suggests that longer observations are not necessary or useful for the scope of determining the collective behavior of the strains through BSDA.

Besides, we observe a stronger decay of the BS signal for the measurements at OD = 0.5, for the case of the L. acidophilus, L. gasseri, and L. plantarum. Here, a larger number of bacterial elements is diffused inside the medium in the observation FOV, each one playing its role in decorrelating the ensemble biospeckle pattern over time. On the contrary, L. casei Shirota e 37 5 AS, show a weaker decay for OD = 0.5 with respect to OD = 0.35.

This could be explained as it is assumed that homofermentative lactobacilli that are typical of the human host are represented by 3 groups: 1) the L. acidophilus group, involving strains that are recognized today as L. acidophilus, L. gasseri, L. crispatus, and L. johnsonii ; 2) Lactobacillus salivarius ; and 3) the Lactobacillus casei group, involving strains of L. casei, L. paracasei and Lactobacillus rhamnosus [14].

This difference in the behavior is detectable after the first BSDA screen of the platform (module 1) and might denote different attitudes of the strains in surviving microencapsulation, so that a lower load could promote a more effective process for some strains. This effect is interesting from the viewpoint of the encapsulation process design, and worth to be further investigated in the next future. As a result of the BSDA, we notice differences in the collective behavior of the analyzed strains. These are measurable in terms of the saturation value of the biospeckle correlation coefficient, ${\rho _{sat}}$, and allow one making a comparative analysis and a BSDA ranking of the strains as probiotic candidates.

Figure 6 compares the curves obtained for the different strains after exposing the capsules for 3 h to the pH 8 buffer, respectively in the case of (a) OD = 0.35 and (b) OD = 0.50. The direct comparison clearly highlights the differences between the strains. Moreover, the average distance between the curves is significantly larger in the case OD = 0.50, suggesting this bacterial load as the optimal to augment the sensitivity of the of the BSDA in pointing out the differences between strains that are judged as effectively microencapsulated. L. acidophilus exhibits the lowest, ${\rho _{sat}}$ value, thus resulting the most effective among the tested candidates. Similarly, L. plantarum and L. gasseri show significant decorrelation, with ρ<0.8 after the saturation region is reached.

 

Fig. 6. Comparison of the biospeckle correlation coefficient, ρ vs. Time, for the strains under test in pH 8 buffer after microencapsulation with (left) OD = 0.35 and (right) OD = 0.50 bacterial load.

Download Full Size | PDF

Table 1 provides a summary of the quantitative BSDA assay for the selected OD = 0.50. In particular, we report the a and b values of the fitting curve $y = a{x^b}$ and the saturation level ${\rho _{sat}}$, which describe the velocity the extent of biospeckle decorrelation and allow establishing a first ranking between the analyzed strains, to be confirmed by the DH analysis of the second module of the optical platform.

Table 1. Measured BSDA parameters and BSDA ranking

View Table

Then, we carried out an experimental campaign using the DH microscopy module to investigate the 3D motility profiles of single bacteria elements after they exit the capsules and are free to move in the pH 8 buffer. Figure 7 shows the 3D tracking profiles of the analyzed strains, while the corresponding projections in the plane normal to the optical axis are reported in Fig. 8. From the figures we can clearly appreciate the substantially different motility signatures which constitute a sort of fingerprint of each probiotic candidate. It is worth noticing that, although the measurements have been conducted using capsules with the same bacterial load and we observed the holographic FOV for the same time for all the strains, the number of elements detected in there significantly varies with the species. Some of the strains are characterized by high motility in the three dimensions, i.e. they significantly move also along the axial coordinate, as in the case of L. acidophilus e L. plantarum. The non-negligible axial movement justifies in full the use of a DH microscope, since the dynamic motion of the bacteria inside the volume of liquid could not be followed by any system based on a mechanical adjustment of the focus position. Instead, the flexible refocusing capability in the post-processing stage, which is a typical advantage of holographic microscopy, turned out to be pivotal for this analysis. Other strains exhibit poor motility at the single element level. This is the case for L. casei Shirota, whose elements perform tiny shifts per time unit and keep close to the initial centroid in the observation time window. The larger number of live elements found in the FOV could be the reason for the decorrelation observed by BSDA, for which this strain passed the BSDA screening of the first module, although BSDA ranked it as the worst among the candidates. We believe this outcome is a good example that shows the necessity of using both modules for testing the candidates.

 

Fig. 7. 3D holographic tracking profiles of single bacteria out of the alginate microcapsules, pH 8, OD = 0.50. Recording is made at 5 fps for the overall time window, T = 500s. Time instants are reported with different colors.

Download Full Size | PDF

 

Fig. 8. Projections on the x-y plane of the 3D holographic tracking profiles of single bacteria elements out of the alginate microcapsules, pH 8, OD = 0.50. Recording is made at 5 fps for the overall time window T = 500s. Time instants are reported with different colours.

Download Full Size | PDF

The differences in the motility profiles can be quantified in synthetic motion parameters that facilitate a ranking based on the DH test. Plots in Fig. 9 show (a) the average velocities of 3D movement and (b) the width of the overall axial displacement for each strain. As expected, L. acidophilus and L. plantarum show the highest velocities and the widest axial motion dynamics. L. casei Shirota is confirmed to be the worst in terms of motility at the single element level. An important result of the analysis we conducted is the substantial agreement between the BSDA and DH microscopy rankings, which suggests the robustness of the screening platform as a whole. Indeed, L. acidophilus and L. plantarum show the lowest ${\rho _{sat}}$ in Table 1 and are ranked in first and second position according to the BSDA module. Similarly, L. casei Shirota is the last in both rankings, as discussed. On the contrary, the DH analysis indicates the strain extracted from the Sardinian cheese (namely, 37 5 AS) as preferrable to the L. gasseri in terms of motility of the single elements, although the ensemble BSDA would suggest the latter to be more effectively incapsulated. From a biological viewpoint, the different behaviors of the analyzed elements can be ascribed to the intrinsic interspecies heterogeneity. Although the microorganisms used in our work can be all associated to the Lactobacillaceae family, they belong to different species. Hence, alongside common interspecies genetic characteristics, they are characterized by genetic peculiarities that translate into a more or less different biochemical pathway and biological properties. These peculiarities are also expressed according to the external environment surrounding them (culture medium, presence of gastric juices, bile salts, lysozyme, etc.), and according to the energy source that is administered to them.

 

Fig. 9. Quantitative DH analysis of the 3D motion and ranking among the analysed strains as probiotic bacteria candidates, OD = 0.50, pH 8. (a) Average velocity of the 3D motion. (b) Width of the overall axial displacement.

Download Full Size | PDF

4. Conclusions and discussion

We have developed a platform and a testing pipeline for screening bacteria strains that are candidates to serve as probiotics in alginate microcapsules. Two optical modules both based on coherent light are part of the platform. The BSDA module probes the sample at the population level. It sacrifices spatial resolution in order to access a wide FOV. Although lateral resolution is not enough to look at the single bacterium, the related biospeckle pattern of the population captured by the camera is very sensitive to tiny variations in space and time, thus allowing detecting their presence and vitality out of the capsules. BSDA has shown its efficacy in assessing the effectiveness of microencapsulation. The second module is an off-axis DH interferometer in transmission mode that tests the motility of single elements of the species released by the capsules in the presence of an environment simulating the human intestine. In order to facilitate the tracking of multiple elements, in-line DH arrangements providing wide FOV have been typically preferred so far, since accessing a wide FOV facilitates numerical processing for 3D tracking elements with large lateral displacements [39–41]. The wide FOV of in-line DH solutions is paid in terms of lateral resolution and, above all, visibility of interferometric fringes. The poor scattering efficiency of in-line DH can limit the accuracy in axial localization. Differently, in off-axis DH a separate reference beam introduces a spatial carrier modulated by the object in the hologram encoding, whose visibility can be tuned by properly balancing power and polarization of object and reference beams. Thus, the poor scattering efficiency of single bacteria is a less severe problem in off-axis DH. Nevertheless, so far this configuration has not been preferred for the abovementioned difficulties in 3D tracking, which we overcome here by designing the ad-hoc 3D tracking pipeline (described in the Methods section) to follow the 3D locomotion over an ample volume ($12,4 \times {10^6}\; \mu {m^3} = 0,012\; \mu l$)). We 3D tracked bacteria and inferred motility information in order to rank the candidates based on their locomotion profiles. Among the tested strains, L. acidophilus and L. plantarum have shown the best performance in terms of both effective microencapsulation and free-swimming motility in the pH 8 environment. On the contrary, our measurements have shown that L. casei Shirota can be effectively microencapsulated but exhibits poor motility out of the capsules. The sensitivity of the system in appreciating differences between the behaviours of the strains is enhanced using capsules with high bacterial load (OD = 0.50 was preferable to OD = 0.35 in our experiments).

We made several steps forward with respect to our previous proof-of-concept [27] on this topic. i) We augmented significantly the number of independent experiments, by testing multiple times a larger number of strains (six in this work) in different conditions of pH, times of exposure to acid environment, and bacterial loads. ii) We coupled BSDA to a holographic module in order to infer information on motility profiles out of the capsules at the single cell level. iii) We developed a novel suite of 3D tracking algorithms to cope with bacteria motion with large axial displacements and poor scattering efficiency. iv) We developed a signal processing pipeline to make the platform resilient to non-ideal recording conditions, e.g., the presence of high frequency noise, convective airflows, or low frequency disturbances due to seismic noise and/or nearby equipment in industrially relevant environment. All these interference terms are not negligible for a system with high sensitivity and should get rid of in order to use the platform for screening and ranking probiotics in food and pharmaceutic industry. It is worth mentioning that for time-varying speckle, temperature and humidity do not have a significant impact on its representation. Therefore, for the BSDA method itself, the variations of temperature and humidity are not expected to influence the optical apparatus. However, temperature could be a potential factor affecting bacterial activity. For this reason, all the experiments were carried out at room temperature set by the room AC system (establishing a temperature approximately ranging in the interval 20°−23°).

During the experimental design stage, we have considered the sensitivity of bacteria to visible laser irradiation. Indeed previous works report that visible wavelength blue laser illumination has the capability to inactivate bacteria under certain conditions [46,47]. Therein, the sensitivity of bacteria to blue laser is demonstrated under sufficient irradiation dose conditions, i.e. 21 mW cm⁻² for > 50 mins or 350 mW cm⁻² for > 10 mins; the effect on cell activity appears as inhibition. For these reasons, the maximum holographic recording time performed in our experiments was set to 500 s and the heat flux density of the bacteria was set to 5 mW cm⁻², to sidestep inhibition by avoiding the use of excessive optical power and long exposure times. Pulsed illumination has been avoided as well.

We expect this platform and the quantitative analysis thereof, will be useful to investigate the environmental conditions that could promote or impair motility of various organisms other than bacteria, e.g. waterborne pathogens, parasites, sperm cells, or motile marine organisms studied as bioprobes of marine pollution levels. In the framework of food industry and pharmaceutics, the platform could serve as a toolkit to test and boost the efficacy of probiotics in production lines for the market and research needs. Moreover, the test proposed in this work will allow to compare the locomotion patterns of the same strain in the presence of different matrixes, e.g. glucose, fructoligosaccharides, or prebiotics substances such as inulin or pectin. In other words, the same platform could be used as a tester for prebiotics in the presence of well-characterized probiotics. The same system could be used to monitor the status of conservation of solid and liquid foods, detecting the presence of bacteria colonies, and identifying the concurring species within a heterogeneous mixture.

5. Methods

5.1 Microencapsulation of bacteria strains in alginate matrix

Different strains of lactic acid bacteria (LAB), Lactobacillus acidophilus, Lacticaseibacillus casei Shirota (LcS), Lactobacillus gasseri LG050, Lactiplantibacillus plantarum 299 V, Lacticaseibacillus rhamnosus GG, obtained from commercial formulation available in a local pharmacy, and a strain derived from fermented Sardinian cheese, named 37 5SA, were used for the experiments of microencapsulation. Cultures were grown for 16 h at 37°C in De Man–Rogosa–Sharpe (MRS) broth. The growth was read at λ = 600 nm (Cary 50Bio, Varian, Palo Alto, CA, USA). Cells were then recovered by centrifugation (7,000 x g at 4°C for 15 minutes using a Biofuge centrifuge (Beckman Coulter Italia, Cassina De’Pecchi, Milan, Italy), and were twice washed with a sterile cold solution of NaCl (8.5 g/l). Microencapsulation was performed following the experimental protocol described by Nazzaro et al. [13], under a laminar flow hood, using a sterile solution of 4% sodium alginate (Sigma Aldrich, Milano, Italy) in deionized water. 1 ml of sodium alginate solution was mixed with 1 ml of the microbial culture. After appropriate mixing, the resulting solution was dropped from a height of 10 cm with a 5 ml syringe into a gelling solution consisting of sterile 0.05 M CaCl2. The capsules, with an average diameter of approximately 2.0 ± 0.3 mm, were gently shaken for 30 minutes, isolated under aseptic conditions using filter paper (Whatman No. 1, Maidstone, United Kingdom), washed and stored cold in a sterile physiological solution. The choice to use alginate as an encapsulating material was taken, due to its chemical-physical characteristics. Such material is capable of tightening the mesh, protecting the bacteria from low pH values, for example, pH 2, like that of gastric juice, which we used to treat capsules for 3 hours. At an alkaline pH (e.g., pH 8, used to treat capsules for 3 hour), such as what they might typically encounter in the intestine, and only at those pH values, the alginate meshes must widen, thus allowing the bacteria to exit from the microcapsules, and maintain a reasonable level of motility in the intestine, necessary to reach the epithelial cells to carry out the probiotic action. Furthermore, we used two final concentrations of microencapsulated bacteria, with λ 600 = 0.35 and λ 600 = 0.50 (corresponding to 2.89 ± 0.13 and 3.49 ± 0.04 CFU, respectively), which define the two optical densities (OD) tested. For the BSDA experiments, we inserted 15 capsules in 50 mm diameter Petri dishes. Capsules were dipped into 3.5 mL buffer.

5.2 Noise suppression strategies for BSDA

5.2.1 Suppression of noise from high frequency airflow disturbance

The bio-speckle signal and thus its correlation coefficient in time are very sensitive to vibrations occurring during the recording process. This high sensitivity to small perturbations, not resolvable by the optical system due to lack of lateral resolution but observable through the speckle signature, is the main reason behind the choice of this recording modality. In most cases, a vibration isolation device is required to minimize system disturbances. During our experiments, the optical table is able to isolate environmental vibrations from 3 Hz to 50 Hz, but for samples in open Petri dishes (the lid of the Petri dish needs to be removed during recording) the buffer can be slightly moved due to room convective airflows (e.g., created by air conditioners, or in the case of setups placed in open-space labs in industrial environment). This type of disturbance is different from the change in speckle caused by bacterial movement, and it appears as an overall shaking of the FOV; thereby, it can be suppressed by FOV matching based on the NCC algorithm [43]. For any continuous speckle pattern recording, we take the first frame as a reference for identifying the area of interest, then for each subsequent frame that needs to be analyzed, the NCC algorithm is used to calculate the mismatch between the FOV in terms of pixel shifts. Here, the range of matching space is set to 5% of the size of FOV. After calculating the offset of each subsequent frame relative to the initial frame in x-axis and y-axis separately, we uniformly crop all frames with same common area to suppress the disturbance by airflow.

5.2.2 Suppression of low-frequency interfering tones

Figure 10 shows examples of biospeckle correlation coefficient signals affected by the presence of low-frequency interference tones. In particular, Fig. 10(a) shows the case of L. plantarum, OD = 0.5, in the pH 2 buffer. In this case, the correlation signal should keep stable in time, but the superposition of the tone provokes unexpected oscillations around an average value higher than 0.8. Another example are the two measurements of Figs. 10(b,c), corresponding to the case of L. acidophilus, OD = 0.35, pH 8. In these cases, we can infer a substantial decorrelation of the biospeckle pattern, although the oscillations impair a correct quantitative evaluation. The estimated frequency of this tone, as detected by the biospeckle correlation signal, was around 0.05 Hz, which cannot not be discarded by using the optical table. The strategy adopted to deal with this issue is to increase the data recording time from the original 60 s to 180 s. Then, in data processing, the frames in 0.0 s and 100.0 s are selected as the reference frames for cross-correlation calculations, and speckle decorrelation calculations are performed for following frames under one minute after these two. Thus, decorrelation curves can be obtained, which generates decorrelation signals compatible and in very good agreement, where the low-frequency interference tone is highly decorrelated and can be reduced by averaging the signals. The averaging process is a first useful way to reduce the interference and improve the subsequent analysis step. Instead of comparing the performance of the strains on the basis of the biospeckle correlation signal, we decided to analyze a fitting curve of the averaged correlation coefficient. The same processing is applied to all the measurements of the experimental campaign, independently on the presence or not of the interference tone. This approach allowed us creating an analysis framework robust against the presence of interference signals that could occur only for a subset of measurements, with the aim to make all the signals comparable. After analyzing all the measured sequences, we found that curves of the type $y = a{x^b}$ are the most suitable to fit the data while discarding the residual interference signal. Figure 10(d) shows an example of the application of the analysis pipeline described above, in particular for the case of L. acidophilus, OD = 0.35. The green solid line shows the signal in the presence of pH 2 buffer. The purple solid line shows the case of pH 8 buffer after the time sequence correlation averaging, while the dashed grey line is the best-fit curve.

 

Fig. 10. Appearance of low frequency interference and its compensation. (a) L. plantarum, OD = 0.5, pH 2. (b) L. acidophilus, OD = 0.35, pH 8, measurement #1. (c) L. acidophilus, OD = 0.35, pH 8, measurement #2. (d) L. acidophilus, OD = 0.35, pH 2 (green solid line), pH 8 after time sequence correlation averaging (purple solid line), fitting curve (dashed grey line).

Download Full Size | PDF

5.3 Resolution in BSDA

The essence of the BSDA method is based on speckle decorrelation, which is the measurement of time-varying speckle [48]. Time-varying speckle was initially applied to blood flow measurement [49,50]; the fluctuations of speckles are dependent on the velocity of the targets motion, so that it is possible to obtain information about the motion from the temporal statistics of the speckle [51]. During the measurement of time-varying speckle, the optical feature of a specific target is not needed, but the speckle pattern of the entire area of interest should be recorded, so resolving each single bacterium is not necessary in our BSDA setup. Since BSDA is used to reveal bacterial movement information, the definition of its resolution will focus on velocity feedback. Herein, the targets’ velocities v and the decorrelation time t are inversely proportional and dependent on the wavelength $\lambda $ of illumination beam [52,53]. Once t is fixed, the maximum measurable speed will depend on the recording wavelength $\lambda $, which is $v = \lambda /2\pi t$. Therefore, the velocity resolution of BSDA for bacterial motion depends on the camera exposure time and illumination light wavelength. In the BSDA we performed, the exposure time was set as 0.6 ms. Therefore, BSDA can be used to characterize the activity of bacteria provided that their velocity is lower than 167.9 µm/s, i.e. the maximum measurable speed.

5.4 Holographic 3D tracking

5.4.1 Bacteria detection from holographic sequences by the Fourier spectra correlation method

In the holographic tracking experiment, since a 50× microscope objective is used to achieve 62× magnification imaging, the FOV is limited to 176 × 176 µm². For bacteria in motion, this observation window is undoubtedly small, so that not all frames in recorded holographic videos have trackable bacteria. A single bacterium resolved by the optical system can be tracked when the time it takes to pass through the observation FOV is higher than 0.6 s, i.e. its x-y plane projection speed is lower than 293µm/s. Since a single bacterium only occupies 1∼16 pixels in reconstructed phase, the conventional thresholding-based filtering method cannot be used to detect frames in which bacteria appear. For this reason, we designed an effective frame discrimination method based on cross-correlation for +1 order spectrum to achieve rapid retrieval frames with bacterial occurrences (FSC method). Firstly, we extract frames from a holographic video at certain intervals (as a general rule of thumb, we extract one frame out of 15 from the full sequence) and perform Fast Fourier Transformation (FFT) of them to obtain their spectra. Then, after extracting the amplitude Fourier spectra, we average them; the +1 order of this spectral amplitude mapping is set as a reference during the following cross-correlation operation. This reference spectrum is important to estimate the typical background signal in the Fourier domain in the absence of bacteria occurrences. Finally, for each frame of the original full fps time-series, the FFT is performed to acquire the amplitude spectrum; the +1 order is used to evaluate the cross-correlation with the reference +1 order previously calculated. The method is summarized in Fig. 11, where we report the obtained background amplitude spectrum, which is compared to each amplitude spectrum of the frame under test (Fig. 11(a)). The correlation coefficient is measured over the area denoted by yellow circles in figure. The plot of the correlation coefficient vs. time shows relative minima in correspondence with frames where bacteria appear in the FOV (Fig. 11(b)).

 

Fig. 11. Full DH-based 3D tracking pipeline. An example of application of the FSC method to detect the frames of interest automatically from a long holographic sequence. (a) Time series of Fourier amplitude spectra and synthesis of the background spectrum. (b) correlation coefficient between each spectrum and the background spectrum shows local minima for frame intervals where bacteria occur in the FOV. (c) Estimating the (x,y) coordinates of the centroid of a bacterium in the selected frame by correlation recognition with respect to a simulated target. (d) Numerical back-propagation of the holographic diffraction and estimation of the z coordinate by maximizing the contrast of the phase map. (c) and (d) share the same grayscale bar as shown in (c).

Download Full Size | PDF

5.4.2 3D localization of a single bacterium

As mentioned before, a single bacterium under 62× magnification in the proposed holographic setup occupies 1∼16 pixels in the image plane. For holographic particle tracking in off-axis geometry, this size is too small to apply conventional tracking methods [54]. Therefore, as shown in Figs. 11(c,d), we used an improved holographic 3D-tracking method based on image correlation recognition. The basic process of this method was introduced in Ref. [55,56]. Here, we first randomly reconstruct several frames in which bacteria are present to obtain the characteristic size of individual bacteria; then this characteristic value is used as the basis for identification by evaluating the correlation with respect to the simulated reference target; the coordinates of the centroid corresponding to a correlation coefficient larger than 0.7 are estimated as the (x,y) coordinates of bacteria in each single frame (Fig. 11(c)). After the coordinates in the $({x - y} )$ plane are retrieved, a numerical refocusing process based on Tamura coefficient optimization is performed to obtain the z-axis coordinates (Fig. 11(d)) [45]. Here we set the scanning space of the z-axis from 0 to 400 µm. For the calibration of axis z location, the Tamura coefficient within the 150 pixels area around has been evaluated as a contrast metrics, i.e. as the judgment criterion for each bacterium tracked in 3D. If no Tamura peak value occurs within the propagation range, the upper and lower limits of the numerical propagation distance are synchronously increased until the peak value appears. The minimum numerical propagation step is fixed at 1 µm in all reconstructions. In 3D tracking of bacteria in consecutive frames, the starting position of each frame of scanning is adjusted based on the previous frame, and the scanning range is reduced accordingly. Although the calculation efficiency of this correlation 3D holographic tracking method is poor, this is the optimal solution for targets with small imaging size and big 3D positions displacements.