**4. Discussion**

Porous silicon membranes are promising biosensors: with their flow-through operation, they overcome the lack of sensitivity of flow-over PSi-based biosensors and are characterized by short response times [34–38]. Moreover, they do not require any stratagems to transport and concentrate the analyte on the transducer. In this work, we demonstrate once

again the potential of these detection platforms and focus on their use for the detection of bacteria. PSiMs were fabricated using standard microfabrication techniques and electrochemical etching. The full fabrication process took less than a week, and dozens of samples could be produced in one attempt. No functionalization was applied to the sensors, as the specificity was based on the use of endolysins. Endolysins are phage-encoded enzymes that induce bacterial lysis for certain targeted strains. In this work, the PlyB221 endolysin was used, targeting *B. cereus*, whose efficiency is demonstrated in [42]. The physical effect of PlyB221 endolysin on the bacteria was observed using SEM. While large clusters of bacterial lysate remained, they were surrounded by nanoscale-sized clusters. It is assumed that these small clusters are able to penetrate the porous membrane and enable a detection.

Combining PSiMs and the use of endolysins produced an innovative biosensing platform that was able to detect *B. cereus* lysate, with an initial concentration of 10<sup>6</sup> CFU/mL, in less than 10 min. In flow-over PSi-based biosensors, similar concentrations could not be detected. Two negative control tests were carried out: one in PBS and one with the PlyB221 endolysin only. The PBS control induced a slight decrease of relative EOT. This can be explained by the slow oxidation and dissolution that PSi undergoes in aqueous media [43,44]. This effect can be minimized by chemically modifying the pore surface using either hydrosilylation [45], thermal hydrocarbonization [46] or atomic layer deposition (ALD) of oxides [47]. Adequate passivation may also help reducing the noise level of the optical sensing, as demonstrated by Rasson et al. with the use of ALD [47]. The second control test, which consisted in flowing a PlyB221 endolysin suspension through the sensor for 1 h, resulted in a slight increase of the relative EOT. While it was expected that the endolysins pass through the membrane, this increase indicates that they were partly retained. We believe two hypotheses might explain this effect: (1) a minor size exclusion effect, which is understandable since there is a large pore size distribution visible in Figure 3 and (2) the binding of proteins to the pore walls.

In order to determine the detection limit, the initial concentration of bacteria was reduced. The sensor was able to detect concentrations as low as 10<sup>5</sup> CFU/mL after 1 h. While concentration of 10<sup>4</sup> CFU/mL induce a significant increase in signal compared to the PBS control test, this increase was however not significant with respect to the endolysin control test. Once again, adequate passivation may help to reduce the deviation between measurements and enable the lowering of the detection limit to 10<sup>4</sup> CFU/mL or lower.

While the specificity of the PlyB221 endolysin has already been demonstrated [42], the specificity of the sensor was still exemplified by adding *S. epidermidis* to the PlyB221 endolysin suspension. The observed shift in relative EOT is similar to the one observed with the PlyB221 endolysin only. This confirms that bacteria, when not lysed, are not able to penetrate the membrane, thus demonstrating the specificity of the sensing platform. A second test was performed adding both *B. cereus* and *S. epidermidis* to the endolysins and the results showed a significant increase in relative EOT, but lower than the one measured for *B. cereus* lysate only. We believe this might be explained by the accumulation of *S. epidermidis* on top of the sensor, blocking part of the pore and preventing the penetration of *B. cereus* lysate. This accumulation of intact bacteria may also explain the larger deviation between measurements when in the presence of *S. epidermidis*: part of the light is scattered by the bacteria, therefore decreasing the optical signal and adding deviations during the fit of the EOT. A solution to minimise this effect is the addition of gold nanoparticles to the porous matrix, enhancing the optical signal by increasing surface reflectivity and reducing EOT fitting deviations [30].

While most PSiM-based sensors rely on functionalization to capture the analyte [34,37,38], the biosensor presented in this work bases the capture on a size exclusion effect. This novelty has several benefits: the lack of functionalization considerably shortens the production time of the sensors and puts no requirements on the storage and detection conditions. This choice of design was made with future industrial specifications in mind: without the constraints of functionalization, a silicon-based device such as the one presented, which is compatible with all standard fabrication techniques used in microelectronic cleanrooms

today, could be easily mass produced and packaged, for a very low cost. The lack of functionalization also enables the use of sensors from the same batch for the detection of different strains of bacteria, as long as a selective endolysin is available. This versatility was illustrated by the detection of *S. epidermidis* using lysostaphin as selective lytic agent.

Besides specificity, the performance of our optical biosensor can be discussed in other terms: response time, sensitivity and limit of detection [10]. The response time for this study is comparable to functionalized PSiMs [34,37,38] and since no rinsing is required to remove unbound species, the total detection time (which includes both lysis and optical monitoring) can be reduced to less than 2 h. Another major advantage is the reduced volume of analyte that is needed: only 1 mL is required for the detection. The approximated theoretical sensitivity of 5745.8 ± 847.7 nm·RIU−<sup>1</sup> is high, but due to the noisy signal, the theoretical limit of detection is also quite high; this value was extrapolated as the RIU value at which the relative EOT shift is equal to the noise level 3σ and amounts to 1.5 × 10−<sup>2</sup> RIU [30]. In terms of bacterial concentration, a limit of detection of 10<sup>5</sup> CFU/mL remains high compared to other PSi bacteria sensors [25,26,48,49], ye<sup>t</sup> there is much room for improvement. Further studies should therefore aim to lower the detection limit to competitive levels (<10<sup>3</sup> CFU/mL) by improving the stability of the sensor and reducing the noise level of the optical signal. Once an optimized prototype is available, real samples may be analysed. A complex microfluidic integration may also enable the optical monitoring of several membranes in parallel, adding the possibility for a blank control test (with the analysed sample only) to exclude false positives, as well as for the simultaneous screening of several lytic agents.
