Label-free bacteria quantification in blood plasma by a bioprinted microarray based interferometric point-of-care device

Existing clinical methods for bacteria detection lack in speed, sensitivity and importantly in Point-ofCare (PoC) applicability. Thus, finding ways to push the sensitivity of clinical PoC biosensing technologies is crucial. Aiming that, we here report a portable PoC device based on Lens-free Interferometric Microscopy (LIM). The device employs high performance nanoplasmonics and custom bioprinted microarrays and is capable of direct label-free bacteria (E. coli) quantification. With only one-step sample handling we offer a sampletodata turnaround time of 40 minutes. Our technology features detection sensitivity of a single bacterial cell both in buffer and diluted blood plasma and is intrinsically limited by the number of cells present in the detection volume. When employed in a hospital setting, the device has enabled accurate categorization of sepsis patients (infectious SIRS) from control groups (healthy individuals and non-infectious SIRS patients) without false positives/negatives. User-friendly on-site bacterial clinical diagnosis can thus become a reality.


Introduction
Fast spreading bacterial infections induce acute infections, critical illnesses as meningitis, as well as life-threatening conditions as sepsis. 1,2 In particular, sepsis with a mortality rate of 30% [3][4][5][6][7] requires fast and accurate diagnosis as the survival chances decreases by 7-8% for every hour that the infection remains untreated. [8][9][10] Rapid, sensitive and quantifiable bacterial detection from patient blood is thus a clinical demand. Blood culture is still the gold standard method used to perform a microbiological diagnosis of sepsis 6,11,12 , but its slow turnaround times delay the delivery of optimal personalized therapy. Furthermore, its low sensitivity results in a positive detection of the pathogen in approximately only 50% of cases. False-negative results can also be the consequence of presence of antibiotics in blood or the inability of the bacteria to grow under the laboratory culture conditions. 13 Recently, molecular diagnostic tests requiring a polymerase chain reaction (PCR) amplification step is becoming popular. Although they are labour-intensive, they have improved the sensitivity in addition to reducing the processing time to 38 h 4,8,14 , but is still not optimal for sepsis treatments.
Outdoor or post-operative patients suspected of bacterial infections need rapid diagnosis and categorization 15 into bacteraemia, sepsis or non-infectious SIRS i.e, Systemic Inflammatory Response Syndrome (a non-infectious process exhibiting similar symptoms to that of sepsis and hence requiring different treatments). Appropriate treatment would thus be possible with sensitive accurate and fast (<1 h, as per hospitals) detection. Bacteria detection with various other technologies like SPR 16 , bimodal waveguide 17 , Raman 18 , as well as fluorescence-based smartphone 19,20 devices have been reported, but most of them have not attempted detection from biofluids like blood or plasma and specifically not addressed sepsis diagnosis. Thus, a rapid sensitive direct and affordable method for bacterial detection with feasibility of hospital on-site testing (i.e., requiring minimal operational expertise) depicted as a clinical point-of-care (PoC) device, is the need of the hour.
Within the framework of a European project (H2020 RAIS Project, http://www.rais-project.eu/) we have developed an integrated portable and stand-alone instrument based on optical interferometry which when employed with specialized nanoplasmonics can directly detect bacterial cells from patient blood plasma. In order to demonstrate its usability as a PoC device, we customized our assay for specifically targeting Escherichia coli (E. coli). In addition, real patient samples were analysed in a hospital setting using a simple one-step process with sampletodata turnaround time of 40 minutes.
E. coli, a gram-negative bacteria, was selected as the target as it is one of the major causative agents for approximately 30% of bacterial infections including sepsis. 13,21,22 Our technology operates with low sample volumes (10 µL) and is a quick (40 minutes) one-step quantification method without the need of multiple expensive laboratory instruments, reagents or skilled technicians, thereby offering a user-friendly fast and sensitive clinical PoC device. Our approach paves the way for modern implementable PoC diagnostics in the clinical settings for pathogen detection.

Concept and novelty of the technology
To attempt bacterial detection in hospital settings, we here present an in-house built portable optical device based on lens-free interferometric microscopy (LIM). Figure 1 shows an overview (left) of the developed PoC device (middle) and its working principle (right) for E. coli detection. The LIM device, recently reported 23 , measures the optical phase shift due to the accumulated mass (contribution of biomolecular height and its surface density) of the designed assay, providing an optical path difference (OPD) value, translating into E. coli quantification in cells/mL. The sensor substrate is based on extraordinary transmission characteristics of the fabricated plasmonic gold nanohole substrate 24 with holes of 200 nm diameter and 600 nm period and an area of 1 cm 2 sensor substrate (shown in inset with substrate holder). When the nanohole sensor substrate is inserted into the LIM device along with the sample holder it forms a sandwiched configuration between two polarizers P1 and P2, and savart plates SP1 and SP2, pairs. A collimated light emitting diode (LED) source (660 nm) is split by SP1 into two orthogonally polarized beams sheared with respect to each other (shear distance is 50 μm). After passing though the microarray biofunctionalized nanoplasmonic substrate the two sheared beams recombine through SP2 and are further interfered by P2, which is orthogonal to P1. The optical phase shift is thus recorded onto the CMOS image sensor (~25 mm 2 ). The LEDs are chosen to spectrally overlap with the extraordinary transmission peak positions of the gold nanohole substrate in air at 660 nm. The nanohole sensor substrate upon customized microarray based biofunctionalization (shown in circular inset) will be capable of specifically capturing E. coli from patient blood plasma. The captured target physically changes the topography of the surface, which is measured with the lens-free interferometric microscope (LIM) PoC device as shown in figure 1 middle.
The device incorporates a LED light source and a low-cost CMOS detector in contrast to lasers and CCD cameras, making it cost-effective. Additionally, to minimize operational costs, we utilize a waferscale nanoplasmonic gold nanohole substrate fabrication technology. Finally, cost is also reduced by handling minimal reagents at the testing site as our approach of simple one-step direct label-free detection eliminates usage of secondary antibodies or nanoparticles and thus allows for testing in PoC settings without the need of highly trained laboratory technicians. Furthermore, we attempt to obtain fast (<1 h) detection from a low blood plasma volume. The device features dimensions of 201423 cm and weighs approximately 3-4 kg which provides both compactness and portability and meets the crucial requirements for a clinical PoC device. With such combination of features the technology presents itself as an ideal candidate for PoC medical diagnostics. an ultrasensitive axial topographic sensitivity and enables the simultaneous imaging of thousands of discrete small sensing areas (µm size) with a high lateral resolution. These features emphasize the potential of using bioprinted microarrays as customized sensor substrates rather than uniform surface functionalization techniques. Hence, we opted to explore printing biomolecular microarrays via dippen nanolithography (DPN) techniques 25 , which offers high versatility (i.e. by varying the printing ink biomolecular composition with proteins, peptides, DNA or antibodies, etc.) to suit our needs 26 . For our application, specific antibodies with the capability of recognizing surface antigens of the target E. coli were chosen as the capture bio-element. In particular, polyclonal antibodies were selected as they could recognize a variety of different epitopes on a single cell. To maximize the recognition potential of the antibodies, its appropriate orientation is crucial, especially in a label-free microarray based biofunctionalization approach, where area and amount of recognition elements might be a limiting factor to reach desirable sensitivity. Therefore, instead of bioprinting antibodies directly onto the substrate, it was essential to employ another biomolecule which could steer the antibody to a preferable orientation for optimal target detection. We achieved this by employing protein G, which in its native form has high affinity for binding to both the Fc and Fab fragments of an antibody. We especially used a recombinant protein G which lacks binding sites to Fab regions, thus ensuring only the capture through the Fc region and maximizing the desired tail-on orientation 27 . We thus bioprinted protein G via the high-resolution patterning technique of dip pen nanolithography as shown in figure 2A. A series of experiments (details in Supporting Information section 1, figure S1) were performed in order to optimize printing conditions and were chosen to obtain maximum OPD signal. With the optimised conditions, the printing of 8×8 array with 250 µm of spacing between individual micro-spots was carried out at a protein G concentration of 500 µg/mL with 5% glycerol in PBS with a contact time of 3 s per micro-spot in a controlled humidity chamber (75-80%) resulting in an individual micro-spot of size 50×55 µm. Four such micro-arrays were printed onto a 1cm 2 nanoplasmonic sensor substrate, as shown in figure 2B, which was followed by 2 h incubation at room temperature and rinsing and drying steps to anchor the protein G molecules to the gold substrate, as depicted in figure 2C. The SEM images of the bioprinted microarray are shown in figure 2D(a). Figure 2D

Design and validation of the assay
Bioprinting protein G at 500 μg/mL directly onto cleaned and hydrophilized gold nanohole surfaces constituted the initial step. The protein G biomolecules were then left to physically adsorb onto the gold surface at room temperature (RT) for 2 h which was sufficient for binding via a contribution of electrostatic forces and the amine groups of the protein. A thorough rinse was done at each stage to ensure removal of excess of unanchored biomolecules. Further treatments include blocking non-specific adsorptions onto the non-printed gold surface, as well as anchoring the antibodies to spotted protein G, as schematically depicted in figure 3A. Blocking of the non-occupied gold surface was necessary for minimizing non-specific attachment of blood plasma proteins or other components when analyzing patient blood samples. We employed bovine serum albumin (BSA) as a blocking agent which anchors to the gold via electrostatic interactions. Thiolated blocking agents having strong affinity for gold surfaces were strategically avoided to minimize displacement of the protein G biomolecules. The blocking step also aided in higher loading of antibodies onto the protein G (spotted area), noted as an increase of the OPD signal (see figure S1 in the Supporting Information). This is apprehended to be a combined effect of the inherent higher affinity of antibody to protein G (spotted area) and steric repulsion between antibodies and BSA molecules (unspotted area). A 0.5 % (w/v) of BSA in PBS buffer was used which correlated with employing an excess of a theoretical monolayer of BSA required to cover the unoccupied non-spotted area of 1 cm 2 of the gold nanohole substrate. This was followed by incubating the substrate with E. coli specific polyclonal antibodies at a concentration of 500 µg/mL. Once the array was rinsed and dried, it was then ready to be incubated with E. coli for quantification. Bacterial incubation for 30 minutes in a static mode was employed, as bacterial quantification in hospitals for categorizing infections on-site demands short sample-to-results time. Figure 3B shows the interferometric images with false color scale of the different stages of the biofunctionalization strategy after performing the measurement in air (dry conditions) and figure 3C depicts their respective OPD values. The arrays generated after bioprinting the protein G appear visible, observing OPD values around 30, confirming its binding to the substrate. Both the biomolecule height and its surface density i.e., accumulated mass, is reflected in the OPD signal 23,24 . The blocking step results in OPD signal decrease as the BSA anchors onto the gold surface effectively reducing the spot height (Protein G height  BSA height, where height depends on orientation of the biomolecules) and finally increases significantly (75) due to antibody attachment (both increase in height and surface density) onto the protein G micro-spots. The bioprinting design exhibits high reproducibility, evident from figure S2 in supporting information, in intra-array, inter-array (microarrays within a 1cm 2 substrate) and inter-substrate (microarrays printed on different substrates). Experiments, as explained in the supporting information figure S3, suggests a high bacterial detection sensitivity with an antibody concentration of 500 µg/mL (as compared to 250 µg/mL) and larger spot size of 55 µm (in contrast to 25 µm) and hence a concentration of 500 µg/mL of the antibody onto the protein G micro-spots of 55 µm was finally selected as the optimal one. Considering that a total of 181 different serotypes of E. coli have been classified and different serotypes of E. coli have been associated with bacterial infections, especially with sepsis, 21 polyclonal antibodies specific for most serotypes of E. coli were preferred. Three different polyclonal antibodies pAb (ab31499 rabbit pAb, rPAB80 rabbit pAb and mPAB1:13.5 mouse pAb) were evaluated for performance (see figure S4 in the Supporting Information). The antibody immobilization (OPD antibody  OPDBSA) was similar for rPAb80 and ab31499 while resulting in lower OPD signals for mPAb1:13.5.
Furthermore, we observed a significantly higher E. coli detection sensitivity with ab31499 than the others. Thus, ab31499 was selected for the assay development for E. coli detection.
Being a whole-cell quantification method, it was crucial to evaluate two relevant factors. Firstly, as we aimed at detecting discreet entities of bacteria, the volume of the sample was highly critical especially at low concentrations (i.e., a concentration of 1000 cells/mL results in the presence of 10 cells when the sample volume is as low as 10 µL and 500 cells for a volume of 0.5 mL). Secondly, determining the OPD signal enhancement resulting from averaging OPD signal of an array with several spots (i.e. 8×8 = 64 micro-spots) could introduce a high variability, as at lower range of concentrations the bacteria may not be uniformly distributed on all the micro-spots (see figure 4A for a representative image, where the OPD signal increases (red) in certain spots as compared to others (greenish), depicting high spot-to-spot variability). Therefore, the most efficient approach to account for this variability is to calculate a total OPD enhancement OPDtotal (as shown below) which is the summation of the OPD of each micro-spot i (considering only those micro-spots with ΔOPDi>0, due to E. coli capture) of the entire microarray with n micro-spots i.e., i = 1 to n (here, n = 64).
Introducing this calculation approach a calibration curve in PBS was obtained as shown in figure   4B. The samples were analyzed in triplicates and a mean OPDtotal value was obtained with an interarray standard deviation. Importantly, we were also able to estimate the bacterial load over several log orders, 10 2 10 6 cells/mL, with merely 10 µL sample volumes. When employing 10 µL volume, a  figure 4D). ELISA experiments confirmed this signal came from the slight crossreactivity of the antibody for this bacterium. As B. cereus is not a sepsis-causing primary bacterium, this slight signal should not be a major problem. Moreover, other bacteria tested with an ELISA, such as P. aeruginosa, which is indeed more relevant in sepsis, showed a much lower cross-reactivity with these specific antibodies, even at higher bacteria concentration (data not shown). Additional tests confirmed the lack of non-specific binding of E. coli, when (i) no antibody is present or (ii) a different control antibody is used (no specificity towards E. coli) as reflected in the black and blue curves respectively. The accuracy of the developed assay was evaluated by analyzing several blind spiked samples and correlating it with the calibration curve (see figure 4D). We obtained a linear relationship between the real and the calculated concentrations of the blind samples with a high correlation coefficient. As compared to the reports for bacterial detection with biosensors 17,28-32 , the experimental detection limit of 8 cells/mL in PBS in just 40 minutes with high specificity in a label-free optical biosensor technology is a giant step ahead in this domain. As our main goal was to provide PoC medical diagnostics, we investigated its applicability for the evaluation of patient samples in a hospital setting.

Validation of the designed assay in blood plasma
In order to proceed for clinical testing, we firstly evaluated the effect of blood plasma on the

Clinical testing
We evaluated the clinical utility of the biosensor platform using patient blood plasma samples  figure 5C) providing a quantification of E. coli (cells/mL) in 25% diluted plasma and thereafter a concentration in undiluted 100% plasma sample was calculated by quadrupling the above values. The mean and standard deviation data (between the duplicates) has been plotted in figure   6. The step-wise calculations have been built-into an in-house developed software requiring only few minutes. When employing 10 µL volume of diluted 25% plasma, anything below 1 bacterial cell equivalent to 100 cells/mL in 25% plasma ( 400 cells/mL in 100% undiluted plasma) is undetectable and hence quantification obtained below 400 cells/mL (LOD as marked in figure 6) was considered to be the sole effect of the blood plasma components anchoring to the microarray. As shown in the figure   6, all the healthy controls (section 6a) and non-infectious (no presence of bacteria) SIRS patients (section 6b) provided a concentration below the LOD, whereas a concentration above the LOD was obtained from patients diagnosed with sepsis positive to E. coli (section 6c  In regard to the absolute quantification of E. coli, two major factors need to be considered for the interpretation of the values obtained. Firstly, our system being a label-free optical technique suffers from the limitation of variability of the optical properties of the patient plasma itself, which neither can be overlooked nor can be completely accounted for as it inherently varies from patient-to-patient. Although this gives rise to the high signals obtained from control group (healthy and SIRS patients (fig 6a+b), we here have proven that we can still distinguish between control patients and sepsis patients (multiple patient plasma studied in both cases), justifying the robustness of our PoC technology. Additionally, it is important to note that the detection of the assay is based on the use of specific antibodies targeting bacterial surface antigens (i.e, the antibody used is reactive with all O and K antigenic serotypes). According to the commercial source, the immunogen used to produce the antibodies consisted of a mixture of intact and lysed (denatured) bacteria. Therefore, it is likely that the antibody is able to detect both viable bacteria (which can grow in the appropriate media, such as in blood culture) and non-viable bacteria (dead, lysed etc. which remain undetected in blood culture).
In fact, our group has previously reported 17 that the employed polyclonal antibody ab31499 shows a 6% better sensitivity for lysed bacteria than whole bacteria. We have further verified this by ELISA where a bacteria sample was measured before and after a heating-cooling cycle, which produced lysed bacteria. Despite the decrease in the viable bacteria determined after colony counting, the signal was significantly higher, confirming the employed antibody ab31499 could recognize both viable and lysed bacterial cells, with even higher affinity (i.e. higher signal) when lysed cells are present (data not shown).
Hence, our quantification outcome cannot be strictly correlated with blood culture quantifications. On contrary, molecular tests developed for bacterial identification are carried out in a culture-independent manner. For nucleic acid methods such as PCR results are reported as genome copies (GC) a value that integrates quantification of both viable and non-viable bacterial cells, therefore acknowledging higher bacterial load of 10 3 -10 4 GC/mL 31,33,35 . Overall, the successful and accurate bacterial infection categorization provided by the developed assay with the novel PoC device, obtained within minutes and without the need of highly qualified personnel in a simple one-step analysis (i.e. no secondary antibody or nanoparticle enhancement required), makes it an ideal candidate for clinical PoC diagnostics.

Conclusions
We  context of an H2020 research project; accordingly, all data was protected in accordance with the EU Data Protection Directive 95/46/EC "on the protection of individuals with regard to the processing of personal data and on the free movement of such data". Specifically, all samples and related clinical data were anonymized, and no personally identifiable data has ever been included in any dataset transmitted outside of VHIR. Patient samples (aliquots of blood, plasma and serum) were recovered from the Sepsis Bank. We specifically used EDTA plasma aliquots that were collected from patients for our present study.

Additional information
Supplementary information is available online and includes plots related to optimization of printing conditions (printing buffer, spot size and antibody concentration), reproducibility of the assay protocol and selection of the specific and sensitive antibody for the assay. The table comprises the information of the hospital patient samples.