**4. Conclusions**

In this article, the different steps of the conception of a magnetoresistive chip cell-counter were detailed. This detection technique has a grea<sup>t</sup> potential. The production, use and integrability of GMR sensors are easy and the tool allows for the detection of targets one by one. This test was evaluated regarding several essential qualities of diagnostic tools (sensitivity, specificity, reproducibility and duration) on a biological model, murine myeloma cells immunocaptured by commercial magnetic beads of 1 μm in diameter. The reached sensitivity of about 10<sup>4</sup> cells/mL is equivalent to that of an ELISA test realized with the same reagents (NS1 cells, mAbs, buffer . . . ). Our test is simpler to perform than an ELISA test. Indeed, the GMR test can be performed within 2h30 (2 h of labeling as assessed by our kinetic study, briefly described in Appendix A and 30 min/mL of sample for the test) without any washing steps, while the compared ELISA test requires several washing steps. Data treatment can be done in a few minutes for ELISA test and can be integrated in the acquisition chain and done in real time for the GMR test. One can note that both techniques can benefit from large parallelization of tests. Moreover, the time of the GMR test can be further reduced by increasing the flow rate in the channel. The labelling processes strongly depends on the target, the beads and the biological probe and other groups reported times between 30 and 180 min [51,53,57]. This time will have to be optimized on the final system, in the real biological sample.

Flow cytometry, although not optimized to give absolute cell counts, have a sensitivity ten times lower than the GMR test. However, this method is more complicated, with washing steps, causing loss of cells and thus discrepancies in counts. The Figure 7 shows the extrapolated number of cells counted by both techniques in positive samples of 1 mL as a function of the expected counts. The agreemen<sup>t</sup> between the two techniques is remarkably good for all concentrations except at 10<sup>3</sup> NS1/mL. Our technique however, presents the interest that the count of signals can be automated while flow cytometry data treatment requires an expert.

**Figure 7.** Experimental results of flow cytometry (in blue) compared with experimental results of the GMR test (in red). Mean values and standard deviations are represented.

The relatively high limit of detection of some 10<sup>4</sup> cells/mL is due to two main phenomena. First, some specific events are missed. Indeed, some less efficiently labeled cells are flowing high in the channel and cannot be detected specifically. Secondly, the detection count threshold has a high value. This can be explained both by the number of beads aggregates, increasing the average number of non-specific signals and by the variability of experimental parameters, increasing the standard deviation of the number of non-specific signals. These uncertainties rise from the use of 5 distinct batches of functionalized beads for the experiment, the random order in which the samples were passed and the involuntarily fluctuations in channel geometry.

This study shows the importance to take into account the biological parameters (antigen distribution, labeling efficiency, cell survival, matrix effect, etc.) in the test evaluation. The high detection count threshold value demonstrates the crucial importance of having negative controls and to repeat experiments in different conditions several times in order to define correctly performances of such technologies. The development of diagnostic tests are based on these two pillars (physical and biological parameters) and correct definitions of performances of a test should systematically integrate these cross-cutting aspects. Here, the focus was set on a rigorous evaluation of non-specific signals measured by the GMR sensor. The study showed that these non-specific signals were due to the detection of beads aggregates.

To lower the detection threshold without complicating the device, the challenge is to diminish drastically the number and the sizes of the MP aggregates. As a matter of fact, decreasing the number of beads in aggregates would enable grea<sup>t</sup> changes in the chip design. The separation layer, added to reduce the impact of non-specific events, could be thinned and thus less efficiently labeled cells would be easier to detect. A better understanding of these aggregation phenomena and development of solutions to reduce the number of these non-specific events will help to reach a better reproducibility and sensitivity.

The elimination of aggregates can be performed by microfluidics sorting techniques relying on hydrodynamic or magnetodynamic forces [61–65] but this would necessarily waste a certain amount of expensive mAbs-coated beads. Another way to deal with these beads suspensions instabilities would be to address directly the cause by a better design of the magnetic beads, such as adding a PEG coating [66].

The real solution may lie in designing magnetic beads tuned especially for this application and thus, to continue the development of this diagnosis tool, the natural next step should be to add chemistry as a third project pillar.

**Author Contributions:** Conceptualization, M.G., F.-D.D., C.F., C.F.-T., S.S. and G.J.-L.; Data curation, M.G. and C.F.-T.; Formal analysis, M.G., C.F.-T. and G.J.-L.; Funding acquisition, E.E., S.S. and G.J.-L.; Investigation, M.G., F.-D.D., A.W. and C.F.-T.; Methodology, F.-D.D., C.F., C.F.-T., S.S. and G.J.-L.; Project administration, G.J.-L.; Resources, G.C., M.P. and E.P.; Software, P.B., M.T. and C.F.; Supervision, C.F., C.F.-T., S.S. and G.J.-L.; Validation, C.F.-T. and G.J.-L.; Visualization, M.G.; Writing—original draft, M.G.; Writing—review & editing, F.-D.D., A.W., C.F., C.F.-T., S.S. and G.J.-L.

**Funding:** This research was funded by the Direction Générale de l'Armement, the Ile de France region and its DIM-ELICIT program and the CEA transverse programs "Techno-Santé" and "PTC Instrumentation et Détection".

**Acknowledgments:** We are greatly indebted to Bakhos Jneid and Karine Devilliers who produced the recombinant IpaD protein and the monoclonal antibody IpaD315. We would like to thank Gérald Le Goff and Dominique Duet for their regular help with technical issues. We are grateful to Olivier Limousin, Antoine Pallandre, Anne-Marie Haghiri-Gosnet, Auélie Solignac and Hervé Volland for fruitful scientific exchanges.

**Conflicts of Interest:** The authors declare no conflict of interest.
