*Review* **Biosensing on the Centrifugal Microfluidic Lab-on-a-Disc Platform**

#### **Celina M. Miyazaki 1,\* , Eadaoin Carthy <sup>2</sup> and David J. Kinahan <sup>2</sup>**


Received: 17 September 2020; Accepted: 22 October 2020; Published: 28 October 2020

**Abstract:** Lab-on-a-Disc (LoaD) biosensors are increasingly a promising solution for many biosensing applications. In the search for a perfect match between point-of-care (PoC) microfluidic devices and biosensors, the LoaD platform has the potential to be reliable, sensitive, low-cost, and easy-to-use. The present global pandemic draws attention to the importance of rapid sample-to-answer PoC devices for minimising manual intervention and sample manipulation, thus increasing the safety of the health professional while minimising the chances of sample contamination. A biosensor is defined by its ability to measure an analyte by converting a biological binding event to tangible analytical data. With evolving manufacturing processes for both LoaDs and biosensors, it is becoming more feasible to embed biosensors within the platform and/or to pair the microfluidic cartridges with low-cost detection systems. This review considers the basics of the centrifugal microfluidics and describes recent developments in common biosensing methods and novel technologies for fluidic control and automation. Finally, an overview of current devices on the market is provided. This review will guide scientists who want to initiate research in LoaD PoC devices as well as providing valuable reference material to researchers active in the field.

**Keywords:** biosensors; LoaD platforms; microfluidics; centrifugal microfluidics; PoC devices

#### **1. Introduction**

The development of microfluidic biosensors with rapid, sensitive, and selective response has been the focus of many research laboratories. Biosensors can be focused both on the clinical diagnosis of silent diseases, such as cancer, to guarantee that the patient is referred to as quickly as possible to the appropriate treatment, and in the routine control of chronic diseases such as diabetes, to give a rapid and easy indication of the medication dosage. Biosensors are defined by the International Union of Pure and Applied Chemistry (IUPAC) as "a device that uses specific biochemical reactions mediated by isolated enzymes, immunosystems, tissues, organelles or whole cells to detect chemical compounds usually by electrical, thermal or optical signals" [1]. The biomolecule responsible for the analyte recognition is denoted as the biorecognition element [2], and is the key for the specificity of the biosensor. Since 2010, about 28,000 articles have been published in accordance with ISI of knowledge (Web of Science) with the topic "biosensors" (black bars on Figure 1A). Biosensors exhibit advantages over the traditional analytical methods including low-cost of manufacture, fast response time, easy handling (does not demand trained operators), portability, better batch-to-batch reproducibility, and often have comparable sensitivity and selectivity. Despite this, few biosensors have been commercialised due to the challenges with translating laboratory-based platforms to devices and instruments suitable for widespread application in practical situations.

Sample preparation is a critical step in ensuring that a biosensor can meet the required sensitivity and specificity. Sample preparation is often an arduous and labour-intensive task involving a range of laboratory unit operations (LUOs) such as metering, mixing, separation, resuspension, and elution. The versatility and usefulness of a biosensor is greatly reduced if sample preparation must be performed manually. Therefore, much effort has been made to encapsulate biosensors within microfluidic devices, or Lab-on-a-Chip devices, which can fully automate the required LUOs and therefore offer sample-to-answer performance. Microfluidics is related to the science and technology of systems that process or manipulates small amounts of fluids [3,4]. Biosensors based on microfluidic technology can offer reduced reagent volumes, reduced sample volume, increased automation, and can function without requiring expensive and large instrumentation. These advantages, combined with the potential for mass manufacturing, can lead to low-cost tests and can prevent sample and operator contamination by minimising the user intervention [5]. The combination of "biosensors" and "microfluidics" has been a consistent focus of the research community as shown in Figure 1A (connected blue squares).

**Figure 1.** Web of Science results analysis by publication year using the topic (**A**) "biosensors" (black bars), the combination of "biosensors" and "microfluidics" (connected blue squares), and the terms "lab-on-a-disc" or "lab-on-a-CD" or "bio-disk" or "lab-CD" or "lab-disc" (connected red circles). (**B**) "lab-on-a-disc" results distributed in applications focus. Data from Clarivate Analytics (ISI Web of Knowledge [6]) in 18 May 2020.

Flow control (via pumping and valving) is a critical technology for enabling microfluidic systems. Prior to undertaking the design and fabrication of LoaD technology, it is important to understand non-centrifugally induced fluid flow mechanisms. Many methods of fluid manipulation and delivery have been reported. Passively driven microfluidics is a popular method as it is operated without an external fluid delivery system or actuator. Techniques such as osmosis [7,8], capillary action [9], pressure [10], gravity-driven flow [11], surface tension [12], vacuum-driven pressure [13], and hydrostatic flow [14] techniques can all be applied to achieve adequate fluid flow within microchannels. Applying these methods to microfluidic devices creates a robust, self-sufficient platform which can be utilised in any POC setting. Applying external triggers can further the capabilities of the devices and incorporates a variety of active driven device techniques. Manipulating fluid using an external trigger can allow for more complicated assays to be carried out in situ. Active fluid flow methods include electroosmotic [15], acoustic [16], photo-actuated [17], and pumping [18]. Both passive and active pumping mechanisms are important in understanding the dynamics of microfluidic devices. These methods may also be applied to centrifugal devices and can be amalgamated to create highly comprehensive POC tools. Centrifugal platforms can incorporate both types of fluid flow techniques; however, centrifugal forces play the main role of fluid displacement within the platform architecture. The Lab-on-a-Disc (LoaD) platform uses a typically disc-shaped cartridge and the pumping is via the centrifugal force; the disc-shaped chip is rotated about its axis. The general advantages of polymer chip-based microfluidics are that (i) mass production is possible with non-expensive infrastructure and

common polymers; (ii) low-cost acrylics or thermoplastics have replaced glass for chip production, and, therefore, the cost per unit is significantly reduced compared to other in vitro diagnostic (IVD) tests [19]; aside from those advantages, the LoaD systems present particular advantages, specifically that (i) the fluidic propulsion depends only on a low-cost and controllable spinning motor, without the need of pneumatic interfaces and pumps; (ii) liquid handling is widely independent of the sample properties, i.e., pH or ionic strength (pivotal for electrokinectic-based methods); (iii) possible full integration, automation, and miniaturisation are possible; and (iv) there is possible large-scale parallelisation and multiplexing of bioanalytical assays. Because of the optical transparency of polymer LoaDs and the no-contact nature of the detection set-up, the optical detection has been the most common approach in LoaD biosensors. Expertise developed in manufacture in disc-based storage media (Compact Disc TM and Digital Versatile Disc TM) provides a strong knowledge base to support process development for manufacture of LoaDs. Figure 1A (connected red circles) illustrates the number of publications concerning LoaD platforms. Because different terms have been used in the literature, the search was made using "Lab-on-a-disc" OR "lab-on-a-CD" OR "Bio-disk" OR "Lab-disc" OR "Lab-CD" as topic keywords. The number of publications are generally trending upwards demonstrating an increased interest in the LoaD platform while the variability in the trend likely reflects that LoaD remains a niche platform. The application focus distribution of LoaD publications is shown in Figure 1B.

Recent developments in microfluidic research has allowed for the integration of well-established detection methods, such as optical or electrochemical, to create robust platforms capable of a sample-to-answer within a significantly reduced time [20]. The use of centrifugal microfluidic platforms allows for complex and automated sample handling integrated within the microfluidic chip. Alongside the favourable features listed above, a key advantage of the LoaD platform is the capability to apply centrifugation to the preparation of the sample for the biosensor. This is particularly useful when processing complex sample matrices such as whole blood; here, for example, it can be separated into its constituent components (plasma, red blood cells, and white blood cells) in a single step [21]. Furthermore, the reduced costs of micro-processors has also resulted in the emerging area of electronic LoaDs (eLoaD) where the instrumentation required by the biosensor can be miniaturised into a support instrument and, in some cases, can co-rotate with the LoaD. This allows for detection and analysis of biomolecules with significantly reduced data acquisition times [22]. Contextualising, for the majority of electrochemical-LoaDs, the electrodes are integrated onto the platform to connect the biosensors to the instrumentation while the electrochemical readouts are often made from a stationary disc [23,24]. Recently, measurements during disc rotation were enabled by electrical slip-rings [25] or, where the high background electrical noise associated with some slip-rings may greatly interfere with the electrical readout signal, using integrated micro-controllers with wireless data transfer [26,27].

Lab-on-a-Chip platforms (which are often called micro total analysis systems (µTAS)) that are based on LoaD platform are particularly suitable for point-of-care (PoC) diagnostics in low-resource settings. Application of these platforms in poor and remote areas is termed extreme point-of-care [28]. Guidance from the World Health Organization (WHO) recommends that the development of diagnostic tests to resource-limited settings should follow the ASSURED (Affordable, Sensitive, Specific, User-friendly, Rapid/Robust, Equipment-free, Deliverable) criteria [29,30]. An updated acronym was proposed in 2019 as REASSURED, with the inclusions of R as Real-time connectivity and E as Ease of specimen collection and Environmental friendliness [31]. Researchers have made great efforts to meet all these criteria but challenges still remain [19,28,32]. Certainly, centrifugal microfluidics has features which position it as the ideal platform for PoC diagnostic testing and is, therefore, at the forefront of extreme PoC device development.

The existing academic literature contains a number of well-regarded review papers addressing both the general LoaD platform and specific aspects of this technology platform [5,28,33–39]. In this review, we discuss the recent advances in centrifugal microfluidics focusing on biosensing applications. We tried to concentrate our discussion (but do not limit) in the literature over the past 10 years. We divide this paper into sections concerning the main protocols in biosensors. We aim to provide an

accessible guide for the ample audience of biosensor researchers with interest in leveraging LoaD technology to enhance their existing research while also providing a valuable resource for microfluidic specialists. Section 2 introduces the fluidic control in LoaD biosensors. Section 3 discusses the LoaD manufacture pointing both polymer microfabrication and biorecognition element immobilisation strategies. Section 4 describes the essential processing in LoaD biosensors while Section 5 discusses the existing and emerging LoaD PoC devices. We finalised with future perspectives and our final remarks in Sections 6 and 7, respectively.

#### **2. Fluidic Control in LoaD Biosensors**

The main driven-force in LoaD biosensors is the centrifugal force which acts outwards from the centre of rotation. Other pseudo-forces which play a role are the Coriolis force (applied when a particle/fluid moves in a rotating reference frame) and Euler force (induced by accelerating or decelerating the disc rotation). A combination of channel design, capillary action, and the placement of valves control can control the movement and timing of fluid movement. When a disc spins, the centrifugal force induces fluid flow radially outwards from the centre to the edge of the disc. Considering a fluid of mass density *ρ* on a planar disc rotating at a distance *r* from the central axis at an angular velocity *ω*, this liquid experiences a centrifugal force *f<sup>ω</sup>* (Equation (1)) [34,35,40]

$$f\_{\omega} = \rho r \omega^2 \tag{1}$$

a Euler force *f<sup>E</sup>* given by Equation (2) [34,35]

$$f\_{\rm E} = \rho r \frac{d\omega}{dt} \tag{2}$$

with a rotational acceleration *<sup>d</sup><sup>ω</sup> dt* , and a Coriolis force *f<sup>C</sup>* (Equation (3)) [34,35,40]

$$f\_{\mathbb{C}} = 2\rho\omega v \tag{3}$$

where *υ* is the fluid velocity in the plane. These three forces are represented in Figure 2, and can be controlled by the frequency of the rotation *ω* [35]. The centrifugal force induces the fluid to flow radially outward from the centre of the disc to the outer circumference [40]. The Coriolis force depends on the direction of rotation. If the direction changes, the direction of the Coriolis force changes [41]. It can be applied to mixing of samples, for flow switching, or directing of sample in specific channels [40]. In the "shake-mode" with continuous change in the spin speed, the angular momentum caused by the acceleration and deceleration induces Euler forces, resulting in a layer inversion of the fluid in the microfluidic chamber [34].

**Figure 2.** Forces acting in centrifugal microfluidics. The centrifugal force *f<sup>ω</sup>* acts radially outward, the Coriolis force *f<sup>C</sup>* acts perpendicular to *ω* and fluid speed, and the Euler force *f<sup>E</sup>* is proportional to the angular acceleration.

The centrifugal flow rate depends on the rotational speed, radial location of the fluid reservoirs, channel geometry, and fluid properties such as viscosity, density, etc. [36]. Physicochemical properties as pH and ionic strength have no significant influence on the centrifugal flow rate, making it possible to pump many different fluids and integrate various processes on the same disc.

In a non-rotating reference frame under gravity (i.e., on the surface of the earth), the hydrostatic pressure is defined by the simple equation ∆*P* = *ρgh* where *g* is the acceleration due to gravity and *h* is the height of the fluid column. Analogously, the centrifugally induced hydrostatic pressure is dependent on the radially inward and radially outward locations of a liquid columm on the disc (defined by *r*<sup>1</sup> and *r*<sup>2</sup> respectively in Figure 2). The centrifugally induced pressure, ∆*P<sup>c</sup>* is defined as

$$
\Delta P\_{\varepsilon} = \rho \omega^2 (\frac{r\_2^2 - r\_1^2}{2}) \tag{4}
$$

In literature, this equation is commonly represented in an alternative form

$$
\Delta P\_{\mathbb{C}} = \rho \omega^2 \Delta r \overline{r} \tag{5}
$$

where ∆*r* is the radial height of the liquid column (∆*r* = *r*<sup>2</sup> − *r*1) and *r*¯ is the radial center of the liquid column (*r*¯ = (*r*2−*r*<sup>1</sup> ) 2 ) [5].

While rotating, the centrifugal force acts on all liquids present on the disc and, therefore, to automate common assays, valves are required. Valves act just like faucets that control if and when the channels will deliver the fluid. Different types of valves have been proposed in the past years ranging from simple designs to highly sophisticated implementations. On the LoaD valves are typically "normally closed" (NC) such that in their initial state, they are closed and will open with some stimulus. Valves can also be "normally open" (NO), but these are less common and useful on LoaD. Most valves are classified as "single use" (i.e., changing from closed state to open state), but others can be changed between open and closed states. Valves are also classified as rationally-actuated or instrument-actuated [42]. Rotationally-actuated valves (also called passive valves) are controlled by changes in the disc spin rate (centrifugal force) while instrument-actuated valves (also called active valves) use an external actuator or source of energy (e.g., a laser, a pneumatic pump, a robotic arm) to control on-disc valves. The different types of valves and flow control techniques have recently been thoroughly reviewed [34,43]. However, here, we will introduce some of the most common and relevant valves to demonstrate how they are key components in the integration and parallelisation of bio-analytical processes on LoaD platform.

Capillary valves (see Figure 3A) are based on the balance between the capillary pressure and the centrifugally induced pressure [36]. In a common configuration, the capillary burst valve is connected to a reservoir through a straight channel, and when the disk is at rest or in low rotation speed, the liquid flows to fill the hydrophilic channel. However, it stops at the inlet of a suddenly expanded valve due to the capillary pressure [44], so the liquid will pass through a capillary valve only when the centrifugal force is high enough to break the capillary barrier pressure. In the hydrophobic valve (Figure 3B), hydrophobic regions in the microchannel prevent the fluidic movement. The valve opens when the centrifugal force overcomes a certain critical value [34,36]. In a siphon valve (Figure 3C), the disc is rotating in a high speed and the centrifugal force keeps prevents priming (via capillary pumping) of the siphon. With the decrease of the rotation speed, the siphon's hydrophilic channel is primed by the capillary force (acting against the direction of the centrifugal force). The liquid in the siphon is pumped by the capillary force over the siphon crest and radially outwards until the liquid meniscus passes the radial inward location of the liquid column (i.e., location *r*<sup>1</sup> described in Figure 2). At this point, centrifugal pumping far exceeds capillary pumping and the siphon valve opens when emptying of the reservoir can be sped up by increasing the disc spin rate/centrifugal pumping force [36]. Variations of siphon valves including centrifugo-pneumatic siphon valves [45], sequential siphon valves [46,47], and interruptible siphon valves [48] can also be used for additional control.

**Figure 3.** Centrifugal microfluidics valving methods: (**A**) capillary valve on a hydrophilic microchannel; (**B**) hydrophobic and (**C**) siphon valves. Adapted from [36] with permission from The Royal Society of Chemistry. (**D**) Laser irradiated ferrowax microvalves: at the top, from a closed to open state, the laser beam is focused at the ferrowax that melts and flows to an assistant valve chamber (AVC); at the bottom, to block an open channel, the laser is focused at the pre-loaded ferrowax located adjacent to the main channel. The molten ferrowax bursts into the main channel and solidifies, blocking it. Adapted from [49] with permission from The Royal Society of Chemistry; and (**E**) event-triggered valve based on dissolvable film (DF): (**a**) sample is loaded and the two DF tabs create a closed pneumatic chamber. The counteracting gas pressure keeps preventing the liquid to enter the pneumatic chamber; (**b**) the ancillary liquid enters and dissolves the control film (CF) and the pneumatic chamber is vented, allowing the main liquid to wet the load film (LF), and (**c**) sending liquid to the outlet chamber. Reproduced from [42] with permission from The Royal Society of Chemistry.

Sacrificial valves are another common valving technology. These valves involve the integration of some material which acts as a destructible barrier layer to allow liquid release. The release can be triggered externally, for example, by laser irradiation [49] (shown in Figure 3D). Laser irradiated ferrowax (iron oxide nanoparticles dispersed in paraffin) microvalves goes from a closed to open state by laser irradiating the ferrowax, which melts and flows to an assistant valve chamber (AVC). From open to closed state, the laser is focused at the pre-loaded ferrowax located adjacent to the main channel. The molten ferrowax bursts into the main channel and solidifies, blocking it [49]. Dissolvable films (DFs)-based valves [50–52] allow the passage of the fluid when its surface is wet, and can be used to implement an event-triggered system (Figure 3E) [42] which act analagous to a (single-use) electrical relay or transistor. DF valves are schematically positioned at specific sites on the disc, the arrival of liquid at this point triggers the release of liquid at another point [42]. In Figure 3E, the sample (main liquid) is loaded and the two DF tabs (named control film (CF) and load film (LF)) create a closed pneumatic chamber (dark green in (a)). The counteracting gas pressure keeps preventing the liquid to enter the pneumatic chamber, and thus wet the DF. At an appropriate timing, an ancillary liquid enters the control chamber and dissolves the CF, thus venting the pneumatic chamber (b), allowing the main liquid to wet the LF and finally, sending the liquid to the outlet chamber (c). This is particularly useful for sequential delivery steps when the arriving of a liquid in the waste chamber triggers the delivery of the next reagent/buffer. Many novel valving strategies have been described in the literature in the last years, always aiming to facilitate and automate disc protocols, as new passive [53–55] and active (magnetically induced [56], pinch-based valves [57], reversible [58–61], among others) valves. Some of them will be discussed in the next sections as they appear in the control of biosensors processes.

#### **3. LoaD Fabrication**

#### *3.1. Polymer Microfabrication*

At the end of the 1990s decade, polymers started to be used for microfabrication [62,63]. Opposite to silicon and glass, which are fragile and expensive, requiring time-consuming and complex processing, polymers are low-cost, of easy and scalable processing, and are available with a range of chemical and physical properties. LoaD fabrication techniques have followed the same strategies than conventional polymer-based chips. The choice of the material accounts on the cost, but especially on the desired properties, e.g., on-disc colorimetric measurement demands optically transparent polymers, some types of valves demands flexible structures, intricate design demands polymers which can be processed by high resolution techniques. As many specialised reviews have dedicated on microfabrication techniques [64–66], this review does not intend to go deep in such subject but displays the main LoaD fabrication approaches and offer the address to find out more about it.

A LoaD biosensor shelters various LUOs and commonly a net of 3D channels and reservoirs are needed, which is obtained by stacking of independently processed layers bonded to each other. The LoaD architecture is created in a computer-aided design (CAD) software, and each layer is previously processed by techniques as replica moulding (soft-lithography), xurography, laser ablation, etc. Figure 4A illustrate (a) 1.5 mm thick-poly(methyl methacrylate) (PMMA) plates machined using a CO<sup>2</sup> laser ablation, and (b) 86 µm thick-pressure sensitive adhesive (PSA) processed by a knife-cutter. (c) After removing the protective covers, the LoaD is assembled by stacking the PMMA and PSA layers using an alignment jig and were rolled using a laminating machine. The side view is shown in (d). Although the cutting process is automated and scalable, the assembly stage is more difficult to be automated while the manual assembly of layers is time-consuming and laborious. Yet, researchers widely use this approach at the proof-of-concept stage because the prototyping is fast and without the need for mould fabrication.

**Figure 4.** (**A**) Photographs of steps of layers assembly during the LoaD manufacture: (**a**) laser ablation cut PMMA discs, (**b**) knife-cut PSA layers, (**c**) stacking and alignment of layers, and (**d**) side view during the mounting of layers. (**B**) Schematics of PDMS moulding to fabricate the disc: Pinch valves-based disc fabricated by PDMS moulding. Reproduced from [57] with permission from Elsevier. (**C**) Hot embossing fabricated LoaD biosensor: (**a**) photograph of the LoaD, (**b**) fabrication of the PDMS mold using a milled PMMA master, and (**c**) demoulding of the hot embossed COP foil. Reproduced from [67] with permission from The Royal Society of Chemistry.

Polydimethylsiloxane (PDMS) is usually patterned by replica moulding (soft-lithography) in which the PDMS reagents (base + curing agent) is poured over the master mould and cured to solidify. Commonly, the master mould is fabricated using UV photolithography of SU-8 (negative photoresist) exposing the pattern through a mask. The unexposed parts are dissolved while the cured photoresist remains on the substrate defining the pattern. Figure 4B illustrates a schematics of fabrication of a microfluidic disc by PDMS casting over the SU-8 master mould. At the design suggested by the authors to create pinch-valves, metal balls were placed over the first cured PDMS layer, and covered with a new layer of PDMS. The PDMS replica containing the metal balls are peeled off and united to a PDMS-coated and plasma-treated PMMA disc [57]. The active valves are controlled by spring plungers which push the balls deforming the flexible PDMS to block the flow [57]. Alternative low-cost techniques for master fabrication, as micromilling, laser engraving, cutting plotters, are discussed in [66]. PDMS is optically transparent in the visible range, biocompatible, flexible, and can be easily sealed to glass or another PDMS part (no need of adhesive layers). Soft-lithography provides high-resolution replicas (nm-scale), but pattern deformations and defects in the moulded product can occur in the peeling process [64]. Other disadvantages concern the cost, relatively high when compared to common thermoplastics [65], and the difficult to implement mass production [66]. Details in PDMS micromolding can be found in [68,69].

Hot embossing and injection moulding of thermoplastic materials are remarkable concerning cost-efficient mass production [70–72]. At hot embossing processing, the temperature is elevated above the polymer glass transition temperature (*Tg*), and the mould is pressed against the polymer substrate transferring the pattern [73]. Nanometre range can be achieved by hot embossing [71]. As polymer pattern resolution depends on the mould resolution, delicate patterns usually demand lithography/etching processed silicon moulds [70,74] accompanied by time-consuming and expensive fabrication [64]. In addition, the masters have limited cycles of life due to cracking and warping from the stress and temperature conditions [75]. To overcome these drawbacks, µ-scale PDMS patterns were produced by photolithography to act as hot embossing mould [75,76]. PDMS moulds are relatively inexpensive, rapid to fabricate and reusable for many replications, and possess thermal stability that enables to stamp common thermoplastic polymers [75,76]. Figure 4C present (a) a photograph of a cyclic olefin polymer (COP) foil disk, fabricated by hot embossing and sealed with PSA, and representations of (b) demoulding of the PDMS mould from the milled PMMA master, and (c) demoulding of patterned COP foil after hot embossing, cooling, venting, and opening of the chamber [67].

3D printing is an emerging technology in microfluidics [77]. Yet, it has some limitations concerning resolution. Usually, supporting material is needed to fill the void spaces during the printing process. Later, its removal is manually made which is time-consuming and difficult to do in small channels. The solvent assisted removing is mostly hampered because the chemical composition of supporting and construction materials are often similar [78]. The produced rough surfaces may also cause dead volumes and irregular surface modifications [78]. Concerning LoaD applications, 3D printing has been used to fabricated specific parts, as valves [61] or valve actuating discs [57,59], or in the manufacture of peripheral structures [25].

#### *3.2. Immobilisation of the Biorecognition Element*

The immobilisation of the biorecognition element is always a concern in the construction of biosensors because the sensitivity depends on the total activity of proteins. LoaD platforms, like most microfluidic devices, are usually built in polymer materials, and the direct passive adsorption of proteins as antibodies and enzymes onto these surfaces is mainly driven by hydrophobic interactions. The adsorption-induced conformation changes can lead to protein denaturation and reduce protein activity by as much as 90% [79,80]. Since hydrophobic surfaces induces higher denaturation than hydrophilic ones [81], some surface modification strategies have been proposed to increase the hydrophilicity of conventional polymers for microfabrication. These strategies include plasma treatment [82], coverage with layers of hydrophilic polymers [83,84] and graphene oxide [85]. Poly(ethyleneimine)—PEI introduces hydrophilic amine groups and acts as a spacer

(increases the space between the biomolecules and the hydrophobic surface) and has been applied in PMMA [80,84,85] and cycloolefin [83] surfaces. PEI coating [80], PEI coating following oxygen plasma treatment [84], and nanostructured layer-by-layer film of PEI and graphene oxide [85] have all demonstrated capacity to improve ELISA performance on PMMA surfaces. Functionalisation of metal surfaces is also crucial for electrochemical and reflectivity-based approaches. Conventional methods based on thiol functionalisation using cysteamine and alkanethiols followed by NHS-EDC chemistry have also been applied to the LoaD [86,87].

The anchoring of bio-active beads within the LoaD to act for the biorecognition element anchoring is an alternative to direct immobilisation of biomarkers onto the flat polymer surface. Beads can be easily functionalised with the biorecognition element *a priori* off-chip and then be loaded to the microfluidic platform. This is particularly useful when the protein coating may cause difficulty sealing microfluidic chips when surfaces have been functionalised [88]. One of the most important advantages using beads is the higher surface to volume ratio (1 g of microbeads with a 0.1 µm-diameter has a surface area of 60 m<sup>2</sup> ) [89]. This can be achieved, for example, by a 3D column filled by the functionalised beads. Additionally, analytes and bioreagents can be easily transported in the fluidic system when attached to beads, by the centrifugal flow or by external forces as pressure-driven flow or especially, by the magnetic force [89,90]. Additionally, some protein-functionalised beads are commercially available or have well-established protein anchoring protocols. Immunoassays applications usually encompass the capture antibody anchoring via carbodiimide chemistry on carboxylated-beads [91,92], or by biotinylated antibodies on streptavidin-coated particles [93].

#### **4. Essential Processes in LoaD Biosensing**

#### *4.1. Reagents and Sample Storage and Supply*

At the R & D stage, the input of reagents in a LoaD platform is usually by manual loading, via pipetting, of the reagents and samples. However, long-term storage strategies, which minimises operator handling, reduces cold-chain requirements, reduces the need for specialised user-training, and which avoids contamination of both user and sample, are of critical importance. A number of different technologies have been demonstrated including stick-packages [94,95], glass ampoules [96], and elastic-membrane micro-dispensers [97].

Pre-storage of reagents in glass ampoules placed on LoaD platforms for DNA extraction was demonstrated with no loss of ethanol and water for 300 days at room temperature. Frozen storage was also possible without ampoule rupture. The release of liquids (buffers and ethanol) is made by a mechanical force (fingertip pressure) through the elastic lid of the cartridge. While the liquid is centrifugally displaced, a filter prevents the glass shivers to go forward [96]. van Oordt et al. [94] have developed stick-packs of aluminium/polyethylene composite foil aiming at long-term storage of reagents. The transverse frangible seal was fabricated by ultrasonic welding and it is adjustable for a specific burst pressure. The hermetically sealed stick-packs can store both liquid or dry reagents, with the first frangible seal (that separates solid and liquid) bursting at lower rotation frequency. After mixing, the second frangible seal bursts to release the mixture [94]. These stick-packs were further applied in a LoaD for nucleic acid-based detection of respiratory pathogens [95].

A long-term storage micro-dispenser was created by Kazemzadeh et al. [97] and tested in lab-on-a-chip and LoaD platforms. The micro-dispenser comprises a tube with a hole and an elastic membrane covering this hole. When the internal pressure is increased (by the centrifugal force) in enough amount to stretch the membrane, a path is provided for the liquid release. Figure 5A(a) illustrates the micro-dispenser placed on a disc. No liquid release happens while the centrifugal force does not exceed the membrane force. Increasing the internal pressure by the rotation speed, membrane stretches and releases the fluid (b). To implement this technology for blood separation, a micro-dispenser was created with a tube with two apertures, each one covered with different elastic materials (see Figure 5B(a)). The top and the bottom membranes are for plasma and blood cells delivery, respectively. The location of each aperture is well-planned, e.g., the first top aperture is located at a point above the level where the plasma is separated from blood cells. Figure 5B(b) illustrates three identical assays for blood separation on disc and (c) a single assay with a micro-dispenser placed on the disc with a piece of tape and a tacky adhesive. Because of the different elastic properties, each of the membranes releases the liquid at different rotation speeds. In this case, a higher centrifugal force is needed to release the blood cells (bottom membrane). The long-term stability of the micro-dispensers were demonstrated to be two and one year(s), for DI-water and ethanol 70%, respectively [97].

**Figure 5.** (**A**) Micro-dispenser working principle in a LoaD platform. (**a**) When the centrifugal force is lower than the membrane resistance, no liquid is released. (**b**) Increasing the rotation speed, the centrifugal force overcomes the membrane force and the fluid is temporary released. (**B**) Separation of blood on a LoaD platform. (**a**) Micro-dispenser with two apertures covered with two different membranes, C-flex and latex, and (**b**) a schematic of the LoaD platform for 3 identical assays for blood separation. (**c**) Picture of the disc dispensing blood plasma and blood cells to two different chambers at different rotation frequencies due to the difference in the membranes properties. Reproduced from [97] under a Creative Commons Attribution 4.0 International License.

#### *4.2. Samples and Reagents Processing*

#### 4.2.1. Blood Processing

Blood is the most utilised sample for diagnostics, and its separation is usually the first step of many biological assays. Cell isolation and pathogens detection demand an initial step of selective blood separation. Conventional benchtop processes typically use centrifugation. This is relatively simple but often requires large sample volumes and, in case of infectious disease, there are concerns about operator safety. Based on the different densities of blood components, it is quite easy to promote separations simply by the rotation of the disc. Common protocols for plasma extraction on disc use moderate (about 1200 RPM for 3–8 min [98,99]) to fast rotation (2000–5000 RPM for few seconds [100–103]). Due to the difference in density of the plasma and the red and white blood cells, blood cells are driven toward the bottom while the plasma forms the supernatant layer on the top [101]. One of the great advantages of LoaD platforms is that the plasma extraction step can be easily and conveniently integrated with the following metering, mixing, and detection steps. It has been reported for plasma extraction, metering and mixing with reagents for the prothrombin (PT) time tests [101–103], nitrate/nitrite assays [98], cancer cells detection [86,104], virus detection [90] and many other sensing platforms.

Some geometry parameters can be defined to increase the speed of blood separation. Kim et al. [105] found that higher tilt angles and narrower channels promote faster plasma separation. The enhanced sedimentation in inclined channels is known as Boycott effect [106], and it is associated with the increase in the surface area available for the particles settling [105]. The impact of the Coriolis force on blood sedimentation has also been investigated [107]. *Spira mirabilis*-like structures (equiangular spiral) demonstrated an enhanced speed of sedimentation due to the increased Boycott effect, herein applied with a density gradient medium (DGM) [108]. Sedimentation over DGM uses solutions with specific densities, such as Ficoll and Percoll, to separate and sort blood cells into layers according to their inherent density differences [109]. This can facilitate the collection of the separated components, and thus the analysis of multiple blood samples. Figure 6A(a) illustrates the density distribution of human blood cells and (b) a schematic of the separation using Ficoll (density = 1.077 g mL−<sup>1</sup> ), with the dense cells moving to the bottom while the low density-cells remain on the top layer. Efforts have been done to develop DGM-based blood fractionation and separation of specific cells [21,86,104,110–113]. Morijiri et al. [113] proposed an elutriation-based separation in a centrifugal disc. Particles are introduced in the separation chamber with a low-density fluid under centrifugation. While particles experienced a centrifugal force in an outward direction, the fluid force acts in an inward direction. The retention position of each particle depends on the size and density, and are controlled by the balance of the centrifugal force and the fluid drag force. Small and low-density particles are not retained in the separation chamber and are the first to move towards the outlet. By introducing solutions with higher densities, the balance of forces is changed allowing a step-wise elution and recovery of particles [113].

Leukocytes extraction can be particularly important. Abnormal leukocytes counts can be related to anemia, infection, inflammatory conditions, and certain cancers [114]. Furthermore, they can interfere with the detection of circulating tumor cells (CTCs) [104]. CTCs have large size variation and their size can overlap with leukocytes causing unreliable size-based separations. Because CTCs are disseminated from primary tumors or metastatic sites to the bloodstream, LoaD platforms to separate them from whole blood have been developed [115,116]. Park et al. [104] developed a centrifugal platform with the selectivity based on the anti-EpCAM covered microbeads, which specifically bind to the CTCs, and thus make them heavier than other blood cells. This leads to sedimentation through a DGM and thus being isolated. Figure 6B shows the disc designed for sorting CTCs (left). The blood sample and microbeads conjugated with anti-EpCAM were injected via the blood chamber inlet while the DGM (Percoll) was pipetted through the DGM chamber inlet. Blood and microbeads are shown in the blood chamber (a), and after plasma separation, the triangle obstacle structure (TOS) played a role to retard the convection of the separated blood cells in the blood cell zone while the disc is stopped to open valve 1—V1 (to remove plasma to the waste chamber) (b). After the complete removal of plasma (c), V2 is closed to initiate the bead-binding (incubation) process by mixing (d). V3 is opened and the sample was transferred to the DGM chamber, where under centrifugation the microbeads-CTCs complexes were moved to the collection chamber, passing through the DGM layer due to their higher density, while the other cells remain on the top (e). After the complete separation, V4 is closed before the collection to avoid contamination (f) [104]. Here, the valving was based on the laser irradiation ferrowax [49], discussed previously.

Filtration-based strategies have also been successfully applied for blood separation. Sequential and tangential flow filtration using two track-etched polycarbonate membranes were integrated in the chambers by Kim et al. [117]. The pore size of the filters was carefully selected considering the size of platelets and blood cells. Figure 6C(a) depicts a photograph of the platelet isolation disc highlighting the filters. Schematics in Figure 6C(b) shows Filter-I (3 µm) to eliminate both white blood cells (8–15 µm) and red blood cells (6–8 µm). The plasma runs through Filter-II (600 nm) which captures platelets while washing residual plasma contents out to the waste chamber, as extracellular vesicles, proteins, lipids, and cell-free DNA [117]. Platelet analysis revealed high purity (>99%), free of white blood cells contamination, and the yield of the platelets recovery was significantly superior (>fourfold) than that conventional benchtop centrifugation [117]. Refer to the source for more details [117]. Table 1 summarises the blood separation and analysis integrated into LoaD platforms.

**Figure 6.** (**A**) Sorting of blood cells based on the DGM centrifugation. (**a**) The density distribution of human blood cells and (**b**) the movement of blood cells upon centrifugation with blood layer placed on the top of the Ficoll solution. The arrows indicate the movement direction of cells. Reproduced from [109] with permission from Elsevier. (**B**) Disc design for sorting CTCs from whole blood. TOS: triangle obstacle structure; V1, V2, V3, and V4 are on/off operation valves (wax that changes from liquid to solid state) controlled by laser irradiation. (**a**) Cross-sectional view of the blood chamber after injection of blood and microbeads; (**b**) after plasma separation; (**c**) after removal of the plasma with opening V1; (**d**) during incubation to bind CTCs to the antibody-functionalised beads; (**e**) the cross-sectional view of the DGM chamber showing that only CTC-microbead complexes (with higher density than DGM) are moved to the collection chamber; (**f**) V4 is closed to prevent sample contamination during collection. Reprinted with permission from [104]. Copyright 2014 American Chemical Society. (**C**) Filtration-based platelet isolation disc: (**a**) photograph and (**b**) schematics. Filter-I to eliminate both white blood cells and red blood cells from a platelet-rich-plasma (PRP). PRP runs through Filter-II which captures platelets while washing platelet-poor plasma (PPP) out to the waste chamber. Reproduced from [117] with permission from The Royal Society of Chemistry.


**Table 1.** Summary of the blood separation strategies performed in LoaD platforms. WB: whole blood, DGM: density gradient medium, PT: prothrombin time, HBV: hepatitis B virus, PBMC: peripheral blood mononuclear cell, CTC: circulating tumour cells, CRP: C-reactive protein, IL-6: interleukin-6.

#### 4.2.2. Volume Metering and Aliquoting

Accurate metering of reagents and samples is an essential step in sample preparation to guarantee a reproducible and quantitatively reliable response from the biosensor. Although different architectures are found in the literature, the volume metering in LoaD is based on the chamber volume (controlled by its geometric parameters). Typically, an overflow channel is connected to this metering chamber to guarantee that any excess of liquid is sent to a waste/overflow chamber, as illustrated in Figure 7A. The same overflow principle can be applied to implement aliquoting and serial dilutions on disc, which are usually time-consuming and susceptible to errors (specially when working in µL-scale) when performed manually. As depicted in Figure 7B, Mark et al. [119] developed a metering structure whereby a dead-end pneumatic chamber enhanced system performance. In this structure, a single metering chamber is connected upstream and downstream by a feeding/overflow channel. The volume metering is based on centrifugo-pneumatic valve structure in which the metering chamber *V<sup>M</sup>* is connected by a narrow channel to an unvented receiving chamber *V*<sup>0</sup> (a). The volume metering chamber is filled but the receiving chamber, which may, for example, contain lyophilised reagents, remains dry (b). When liquid reaches the narrow connection at the bottom, a meniscus is formed which prevents the air from escaping the unvented receiving chamber (creates a pressurised chamber with pressure *p*1) (c). Aliquoting is performed by continuing the liquid feed. After complete the first *V<sup>M</sup>* (d), the overflow channel deliveries the liquid for the next metering channel and so on, as illustrated in Figure 7C(a). After all the metering channels are completed, an increase in the centrifugal frequency reaches a point (burst frequency) in which the centrifugal force overcomes *p*1, releasing the liquid to the receiving chambers, as shown in Figure 7B(e),C(b).

**Figure 7.** (**A**) Simple metering design based on overflow. (**B**) Metering structure based on the centrifugo-pneumatic valve: (**a**) design of a single metering structure with the volume defined by the metering channel *VM*. (**b**) At a first centrifugal frequency, liquid fills the metering channel while air is displaced. (**c**) The pressure *p*<sup>1</sup> prevents the liquid to enter the receiving chamber *V*0. (**d**) After metering, (**e**) an increase in the centrifugal frequency above the burst frequency (to an amount that overcomes *p*<sup>1</sup> ) releases the liquid. Adapted from [120] with permission from The Royal Society of Chemistry. (**C**) Layout of an aliquoting structure: (**a**) multiple *V<sup>M</sup>* chambers been fed by liquid, and (**b**) aliquots distributed to the receiving chambers. Adapted by permission from Springer Nature [119], Copyright 2011.

Active valving based on individually addressable diaphragm (ID) valves [61] was used for on-disc serial dilutions by Kim et al. [121] (Figure 8A). These valves are assembled from an elastic epoxy diaphragm and a 3D printed actuator (Figure 8B). A simple push-and-twist action is required for closing and opening the channel. This open/close functionality allows the volume of sample and buffer to be metered in the two different zones; thus dictating the degree of dilution. These two zones are detailed in Figure 8C,D. An example of fivefold dilution is presented in Figure 8E (opened and closed valves represented by green and red colours, respectively): (i) at buffer metering zone, valve 1 (V1) is opened while V2, V3, V4, and V5 are closed for metering of total 9 µL-buffer. Simultaneously, at the sample metering zone, V6 and V7 are opened while V8 is closed, totaling 8 µL sample metered; (ii) at the sample metering zone, V6 is closed and the 8 µL-metered sample is sent to R1 (with valves 7, 8, 9, and 10 opened while V11 is closed). From the initial 10 µL sample, 2 µL remains in the mixing chamber. At the buffer metering zone, only the V5 is kept open, sending the buffer excess to the waste chamber; (iii) the 2 µL sample in the mixing chamber is diluted by the delivery of 8 µL buffer

(only V2 and V3 are opened while V4 is kept closed keeping 1 µL buffer retained). Finally, the 5-fold diluted solution is metered (5 µL, by keeping V6 opened while V7 is closed) and sent to R2. The next dilution can be performed to the remained solution in the mixing chamber based on the similar steps (protocols for 2-,5-, and 10-fold dilutions are summarised in the table of Figure 8E). Buffer can be re-loaded by opening V1 [121]. Although the system at first appears complex and unwieldy, the recent miniaturisation and cost-reduction in micro-controllers means this approach can be easily automated and can flexibly address a wide array of biosensor applications.

**Figure 8.** (**A**) Disc design for serial dilution consisting of sample and buffer loading chambers, mixing chamber and buffer waste chamber, two metering zones (for sample and buffer), five collecting reservoirs (R1–R5) for the diluted samples and seventeen ID valves (numbered circles). (**B**) A reversible ID valve composed of a diaphragm (orange) embedded on a top layer and 3D printed valve actuator (gray). Simple push and twist actions open and close the channel. Detailed buffer (**C**) and sample (**D**) metering zones. (**E**) 5-fold dilution steps: (i) in the buffer metering zone, a total of 9 µL is metered keeping V1 opened while V2, V3, V4, and V5 are closed. At the sample metering zone, V6 and V7 are opened while V8 is closed, totaling 8 µL metered; (ii) at the sample metering zone, V6 is closed and the metered liquid within the channel is sent to R1 (V7, V8, V9, and V10 are open while V11 is closed). At the buffer metering zone, V1, V2, V3, and V4 are closed and V5 is opened to send the excess to the waste chamber; (iii) 8 µL buffer is delivered (just V2 and V3 are open while V4 is kept closed keeping 1 µL of buffer retained) into the mixing chamber, where remains 2 µL sample. Thus, a dilution of 2 µL sample to a final 10 µL volume solution was achieved. The table presents the summary of the metered volume of sample and buffer, and the state of the valves to achieve 2-, 5-, and 10-fold dilutions. Reproduced from [121] with permission from Elsevier.

#### *4.3. Mixing and Washing*

Mixing protocols should provide an effective and fast homogenisation of the fluids. However, adaptation of simple bench-top protocols as agitation, vortexing and stirring onto the LoaD can cause unwanted opening of rotationally actuated valves; and the subsequent uncontrolled pumping of liquid. Therefore, systems which require rigorous mixing must be carefully designed. Several studies have been focused on developing disc protocols to allow an integrated and adequate mixing. On-disc mixing can be passive, just based on the rotation control, based on intrinsic forces as Euler or Coriolis [122–124] or by means of control of geometry [125,126]; or can be active based on external perturbations, e.g., compression by external actuators [127] or use of magnetic-beads inside the disc in combination with permanent magnets [128,129]. Several micromixing technologies are discussed in [130].

One of the most utilised mixing protocol is the "shake-mode" mixing [124] comprising periodic changes in the sense of rotation and rapid changes of the spinning frequency. The difference in the angular momentum leads to the appearance of a shear force that drives an advective current within the liquid [124]. The mixing time was down to 3.0 s for 25 µL-volume while a mere diffusion-based mixing took about 7 min. Combining the "shake-mode" mixing with magnetic beads (pre-filled in the mixing chamber) periodically deflected by a set of permanent magnets resting in the lab-frame, an effective homogenisation was attained in only 0.5 s [124].

The use of pneumatic pressure to promote reciprocating flow-based mixing [131] was introduced later, and has been successfully applied in LoaD biosensors [87,132]. This scheme utilises an auxiliary pneumatic chamber, and the sequential decrease and increase in the centrifugal force causes the compression and decompression of air pushing the liquid through different chambers and thus allowing an accelerate mixing [131]. Figure 9A(a) depicts an immunoassay disc. The biorecognition element (antigen) is immobilised on the reaction chamber to specifically capture the primary antibody (sample) while the secondary antibody is to allow the colorimetric detection. The schematics is illustrated in Figure 9B(b). The sequence of images in (c) describes: (i) the sample stored in the loading chamber; (ii) acceleration causing air compression in the pressure chamber; (iii) by deceleration, the air expansion in the pressurised chamber pushes the liquid back to the upper chamber; (ii and iii) the increase and decrease in rotation speed (3200–7000 RPM) are reciprocated to effective and rapid mixing; (iv) the rotation frequency is increased to the maximum (7500 RPM); followed by (v) a rapid decrease to very low value (480 RPM). This increases the liquid level above the crest point, primes the siphon valve and sends liquid to the waste chamber, (vi) leaving an empty reaction chamber [132]. Pneumatic valving and mixing are reviewed in [133]. Another reciprocating mixing technology demands lower rotation frequency (0–1500 RPM) and encompass the micro-balloon pumping [134]. With higher centrifugal force, the liquid is pushed into an air chamber inflating a latex micro-balloon. A reciprocating mixing can be performed increasing (inflating the micro-balloon) and decreasing (flattening the micro-balloon) the rotation frequency [134].

Additionally, mixing can be done with gas bubbles integrating external pumps [135] or with gas generation by internal reactions [136]. Figure 9B illustrates the schematics and images from two separated layers of solutions contained in a vented reservoir (top channel) (a), air pushing through the lower channel creates bubbles that promote the mixing (b), thus creating a homogeneous solution in about 100 ms (c) [135]. To avoid the need of external apparatus, Burger et al. [136] implemented a bubble-based mixing by gas produced through a chemical reaction (Figure 9C(a)). The oxygen bubbles are generated in the reaction chamber by the hydrogen peroxide decomposition, which provokes a drag flow through the liquid. A strong buoyancy causes deformation and rupture of these bubbles inducing mixing flows, as represented in (b). In a DNA extraction experiment, this approach allowed a similar performance compared to the "shake-mode" mixing and can be an alternative approach when mixing has to be performed at a fixed rotational frequency.

**Figure 9.** (**A**) Pneumatic pressure flow-based reciprocating mixing. Schematics of (**a**) the fluidic system and (**b**) the antibody capture assay to take place in the reaction chamber; (**c**) images of the system in operation: (i) sample is loaded; (ii) the increase in the rotation frequency promotes the air compression in the pressure chamber; (iii) the decrease in the rotation speed promotes the air expansion that pumps the liquid towards the center; (ii) and (iii) can be repeated in a reciprocated manner. (iv) The increase of rotation to the highest speed (7500 RPM) followed by (v) a decrease to the lowest rotation frequency (480 RPM, see protocol in (**d**)) allow the liquid to prime the siphon valve and (vi) send the liquid to the waste chamber. Reprinted from [132], with permission of AIP Publishing. (**B**) Bubble-based mixing schematics with an external pump: (**a**) from an unmixed state of two dyes, (**b**) under bubbling mixing, to (**c**) a homogeneous solution. Reproduced from [135] with permission from The Royal Society of Chemistry. (**C**) Buoyancy driven bubble mixer set-up. (**a**) Hydrogen peroxide flows into the reaction chamber and the decomposition reaction produces oxygen which is directed through the gas channel to the mixing chamber. (**b**) The mixing chamber with blood and lysis buffer in separated layers (left) and a bubble detaching and ascending through the liquid (right). Reproduced from [136] with permission from The Royal Society of Chemistry.

Mixing and washing steps are crucial for biosensors concerning especially the incubation of the sample with the biorecognition element, which is fundamental for biochemical reactions to occur. Incubation can be performed at a constant low rotation for some assays (400–600 RPM for few minutes [98,137]), while longer periods are necessary for some complex assays such as nucleic acid amplification, e.g., LAMP assays (about 60 min at low or paused rotation [138–141]), NASBA (about 70 min [142]), and RPA (about 15–30 min [143,144]). In ELISA the incubation protocols have been performed even in constant spinning (at 20 Hz for 15 min [85]) or in mixing modes ("shake-mode" mixing from 15 to 30 Hz for 10 min [99] or by magneto-balloon assisted mixing at 900 RPM for 30 min [60]). Incubation protocols should be developed and optimised individually for an assay. In biosensors based on the formation of a complex of magnetic nanoclusters with antibody- [145] or aptamer- [146], the presence of the target triggers the magnetic particles agglutination, which is quantified with an optomagnetic readout. A two-step protocol promoted reliable results with reduced

incubation time: the first 1 s incubation step occurs in a strong magnetic field to facilitate clusters formation, followed by the second step of 2 s mixing by shaking to break unspecific bindings and to facilitate the reorientation of the beads for the readout [145,146]. A protocol with 180 cycles of incubation and mixing was established to detect the NS1 Dengue biomarker in serum [145].

After any incubation, the washing step has to guarantee that any impurities, as well as unspecific binders are removed. Washing is usually performed in "shake-mode" mixing with proper buffer solutions, and can be performed more than once if necessary. For each LoaD platform, it is important to select the most suitable approach. "Shake-mode" mixing is the simplest way to achieve good mixing in a passive set-up (no need of external devices), but demands to vary the speed rotation for relatively high frequencies. If this is not possible, external forces (e.g., magnetic or pressure) can be introduced, or bubbles can be generated internally, however, this demands a precious disc area (additional storage reservoirs) and disc area can be quite limited since many lab protocols has to be included in a LoaD biosensor.

#### *4.4. Detection Methodologies*

A broad variety of analytical approaches have been used in LoaD biosensors. There is no rule to follow concerning which method should be or should not be used in LoaD platforms, but all of them exhibit advantages and limitations that should be considered. A complete review of detection methods for centrifugal microfluidic platforms is found in [39]. If the analyte is detectable with enough sensitivity and selectivity, the optical colorimetric strategy is the most cost-effective and easy integrated readout method. With technological advances and miniaturisation of detection systems, other approaches raise attention to PoC devices. Special attention to the label-free methods as electrochemical impedance spectroscopy (EIS) and SPR (surface plasmon resonance), which can greatly simplify immunoassays by reducing fabrication-cost and fluid handling. Table 2 lists some detection techniques used in LoaD biosensors comparing the advantages, limitations and solutions, some of them especially thought to adapt the analytical detection set-up to the LoaD platform. Following, this section discusses recent and relevant developments concerning optical and electrochemical detection.

#### 4.4.1. Optical Detection

Optical strategies require three basic components: a light source, a photodetector, and a reaction chemistry, which is analyte concentration-dependent. Particularly to the absorption and fluorescence techniques, considering that the discs can be fabricated in transparent polymers, easy integration of the detection module is possible with the optical transducer placed as a peripheral optical device without the need of electric contacts and wires. Optical strategies for centrifugal microfluidic devices are reviewed previously in [38,39]. Especially for biosensors, optical approaches are widely used because many biological analytes absorb light in a specific wavelength or can be easily marked to do so. In addition, many commercial kits were available for quantitative determination by colorimetric and fluorimetric approaches, which greatly simplifies the R & D stages of LoaD optical biosensors.


#### **Table 2.**Detection strategies in LoaD biosensors, advantages, limitations and solutions.

The Beer–Lambert Law shows the dependence of the light absorption with the absorbing species concentration, which allows a simple approach of sample quantification. A simple set of light emitting diode (LED) with the adequate wavelength emission and a proper photodetector perpendicularly aligned to the disc plane can be used to create a compact and low-cost detection module [154]. The Lambert-Beer Law also states that the absorbance depends on the optical length. Because of that, achieving a reasonable limit of detection in thin discs can be difficult. This can be overcome with thick discs, or by increasing the light path length with an optical beam-guidance set-up based on the total internal reflection (TIR) [147,148].

Various well-established colorimetric lab protocols were successfully implemented in LoaD platforms as loop mediated isothermal amplification (LAMP) and polymerase chain reaction (PCR) as well as the immunoreactions-based fluorescence-linked immuno-sorbent assay (FLISA) and enzyme-linked immuno-sorbent assay (ELISA). LAMP-based LoaD platforms for bacteria [138,139,141] and virus [155–157] detection, as well as PCR for virus [90,95,158] and bacteria [137,159] detection have been published. DNA amplification is a reliable detection method and has existed for decades. The implementation of this technique on a LoaD however is a more difficult task. The LoaD proposed by Li et al. [90] is illustrated in Figure 10, a PoC apparatus for Hepatitis B Virus (HBV) detection which boasts the same diagnostic capabilities as that of a centralised laboratory test. On-board valving allows for serum separation and reagent storage which allow for a fully automated test. A laser diode activates the valves allow for the release of reagents. Centrifugal forces generated by the double rotation axis of the microfluidic platform, which rotate to allow the stationary magnets to displace the internal capture magnetic beads for nucleic extraction. The complex system includes three resistors and thermistors which control the specific temperatures required for the PCR process. Finally, a 470 nm-blue LED served as an excitation source. The fluorescent signal was then detected using two optic fibers when the turntable containing the LoaD was position to the fixed detection region. The only user interaction required is the insertion of 500 µL of whole blood where the systems boasts an LOD of 10<sup>2</sup> copies/mL in just 48 min [90].

As exemplified before, nucleic acid amplification strategies, as LAMP and PCR assays, demand controlled temperature raising. Temperature sensors and heating system can be also integrated into the LoaD platform. Technologies using hot air gun [160] and heater boards [139,155] have been proposed to control temperature profiles. Figure 11A depicts a set-up for a LAMP-based bacterial infection sensing. The heater board with embedded resistive heating elements and thermistors was connected to a printed circuit board (PCB) which allows temperature control and wireless communication with a remote computer [139].

**Figure 10.** (**A**) The optically transparent disc and (**B**) the schematic of the LoaD architecture for HBV DNA detection in whole blood. Sample is inserted into the centre reservoir of device where pre-stored DNA extraction reagents in wash 1, 2, elution ad lysis buffers and magnetic beads. Ferrowax valves control the release and flow of each reagent to initiate the DNA extraction which then mixes with pre-stored primer and PCR reagents for amplification purposes. Reprinted with permission from [90]. Copyright 2019 American Chemical Society.

**Figure 11.** (**A**) LoaD set-up for LAMP assay-based bacterial infection sensor. A belt-driven motor to provide the centrifugal force; power coupling device; wireless data communication module for temperature control and signal detection; light source (blue laser at 488 nm) for the fluorescence excitation; and spectrometer linked to a data acquisition computer. The disc is on an electronic circuit board (PCB) accommodating the heating elements for microfluidic disc to the temperature control. Reproduced from [139] with permission from Elsevier. (**B**) SPR immunosensor set-up: photographs of (**a**) SPR optics integrated to a spin stand, and (**b**) the reflected laser spots from aqueous glycerol pattern solutions (with different refractive indexes) showing the position of the dark lines (caused by the plasmon absorption). Adapted from [87] with permission from Elsevier.

Immunoassays in format of ELISA [99,161] and FLISA [98,162] have also been implemented on LoaD platforms. Conventionally, the detection is made by the absorption (or emission) measurement using a monochromatic light source and an optical sensor, as demonstrated by Thiha and Ibrahim [161] with a PoC sandwich-type ELISA platform for Dengue detection. Oh et al. [163] have applied a similar colorimetric measurement, combined with LAMP, for detection of food-borne pathogens.

Immunoturbidimetry method was used to measure hemoglobin A1c (HbA1c), a glycated hemoglobin recommended by the American Diabetes Association for diabetes diagnosis [164]. The process automated on disc includes the rupture of erythrocytes by a hemolysis reagent, the attaching of the hemoglobin on latex particles, and the specific agglutination of HbA1c, which causes the solution turbidity. The specificity is given by the HbA1c monoclonal antibody added to induce the particles agglutination. A calibration curve reading the absorbance in 660 nm against Hb1Ac concentration was plotted for further quantification of hemoglobin in unknown samples. Fourteen

blood samples were tested giving a rapid response (8 min), a good correlation with laboratory results (ion-exchange chromatography), and a standard deviation of ± 0.36% HbA1c (of order with standard laboratory equipment) [164].

Despite ELISA presenting the advantage of the easy colorimetric detection, it requires labeled proteins. Label-free methods as those based on surface plasmon resonance (SPR) simplify the disc design (less reagents and washings) and reduce the production cost (labeled proteins are expensive). SPR biosensing is based on changes in the refractive index very close to the sensor surface (usually gold or silver) caused when the analyte (in solution) binds to the biorecognition element immobilised on the sensor surface [165]. Grating-based [166] and prism-based [87] SPR approaches have been proposed. We have demonstrated an SPR-LoaD platform to simple multi-analyte immunoassay adaptable to detect specific diseases just by changing the capture antibody loaded onto the gold sensor [87]. The gold sensor was attached to the top of the disc immediately before the assay running exempting the need for special storage and transport cares and generalizing the device manufacture. The SPR optics (Kretschmann configuration) were integrated to a spin stand (Figure 11B(a)), and were tested in glycerol solutions to check on the dependence of the SPR angle position with the refractive index of solution. Pictures of the reflected light spot were recorded with a smartphone camera in a dark room (shown in b), and further image analysis provided a plot of reflected light intensity versus incident angle, with the minimum reflectivity (SPR angle) caused by the plasmon absorption. Shifts in the SPR angle were proportional to the antigen concentration when the biorecognition element (specific antibody) was immobilised on the gold sensor. Details can be found in the source [87]. Although the SPR optics can appear to be complex when integrated into LoaD, automated approaches turn SPR a very sensitive optical detection technology for LoaD biosensors. The company Biosurfit (Lisbon, Portugal) already commercialises SPR-based platforms [167].

Raman [168] and surface enhanced Raman scattering (SERS) [169,170] based discs for the detection of organic compounds have been recently published bringing good perspectives for future LoaD biosensors. Because of the expensive and complex detection set-up, these methods are not appropriate to PoC devices at this point. However, certain exigent applications can be worth due to the single-molecule sensitivity [171]. Optical methodologies are far the most explored approach for LoaD platforms due to the cost-effectiveness, the possibility of miniaturisation and personalising of PoC devices, being the closest to fully satisfy the (RE)ASSURED criteria. There are a huge number of publications on optical detection-based LoaD sensors, some of them are summarised in Table 3 to exemplify the variety of detection approaches, analytes, assay types and processes integrated on LoaD biosensors.


**Table 3.**Summary of some optical detection-based LoaD biosensors. DR: detection range, LOD: limit of detection, WB: whole blood. 


**Table 3.** *Cont.*

#### 4.4.2. Electrochemical Detection

Electrochemical methods are widely used in conventional off-chip sensors because of the fast response, high sensitivity, and relatively low-cost, however, they have been under explored in LoaD devices. This is related to the complexity in integrating the electrochemical signal transducer to the LoaD platform, which demands electrodes included in the disc, and a potentiostat in direct contact with these electrodes. In other words, this usually implicates that the rotation should be stopped to connect the electric contacts to the disc electrodes to perform the measurement. Nwankire et al. [86] implemented gold electrodes functionalised with capture biomolecules where label-free electrochemical detection methods were deployed for measurement. Whole blood was centrifuged on-board, where the cancer cell rich plasma is extracted via a siphon microchannel. Centrifuge-pneumatic DF valves control the realise of reagent for the assay to be carried out. Five identical testing sites exist on the electrochemical-LoaD, allowing for multiple assays being carried out in parallel. The disc was stopped to Electrochemical Impedance Spectroscopy (EIS) measurement for each compartment invidually. The device boasts an 87% capture efficiency over a dynamic range of three orders of magnitude with a lower LOD of 214 cells/mm<sup>2</sup> . The lower LOD equates to just 2% of the total working electrode surface [86]. The detection method of label-free EIS allowed for rapid cancer cell detection. The change in the electrical interfacial properties of the surface due to biological binding events is detected. For label-free detection, the changes in the dielectric constant, resistance and capacitance on the surface due to the capture of the target molecule is measured. The change in the impedance would be an indication of whether an analyte has been captured or not, and depending on the change, whether it is a high or low concentration of the target analyte.

Electrochemical biosensors of this nature have attracted major interest for label-free analysis as they don't require as many assay steps as their labelled counterpart for signal acquisition. This form of biosensing attains sensitivity and simplicity which makes them a reliable, quantifiable diagnostic tool for whole cell capture and detection [176]. Opting for whole cell detection rather than DNA detection also reduces the cost and time of detection, as no cell lysis or additional reagents are necessary. The use of sputter-coated electrodes allows for the platform to be disposed of after a single-use and more cost effective, making this type of PoC device economical. These results give a true insight into the huge potential that exists for electrochemical-LoaDs as they are an efficient prognostic devices where minimal sample preparation is required and can be further developed for clinical applications [86]. The relative changes in impedance (difference between values of the anti-EpCAM coated gold electrodes before and after interaction with different concentrations of SKOV3 cells) varies linearly with the captured cell number [86].

To overcome the drawback concerning the integration of the detection system, recent works have focused on the electrochemical modules to allow measurements during the disc rotation [25,153]. This opening up opportunities of real-time monitoring of bioreactions on disc [25,153]. Andreasen et al. [25] developed a low-noise component (electrical slip-ring) to on-disc electrochemical measurements. Cyclic voltammetry tests performed from 0–600 RPM indicated only a small perturbation of the peak currents with the spin rate. Recently, Rajendran et al. [153] demonstrated a modular lightweight (127 g) and wireless potentiostat-on-a-disc (PoD) able to perform square-wave voltammetry (SWV) and amperometry, controlled by a custom-made software. The current resolution of 200 pA was in agreement with commercial potentiostats. During the experiment, data was transmitted via Bluetooth to a Windows PC, plotted live and displayed in a custom-developed LabVIEW program (Figure 12A). The PoD detection unit was composed by a Qi-based wireless power supply, a core circuit platform and a module (shield) (an exploded view in Figure 12B). Figure 12C(a) presents the cyclic voltammograms (CV) of 1 mM potassium ferricyanide (FiC) collected in stationary-mode using the PoD and a commercial potentiostat and (b) CVs collected in different spinning rates by the PoD. At 0 and 1 Hz, the reaction was diffusion-limited while reaction-limited (due to the mass transfer) at 2 Hz and above [153]. The comparison between PoD and the commercial equipment was extended to SWV: (c) voltammograms of 10 mM ascorbic acid and (d) the respective calibration

curve. Both cyclic voltammetry and SWV measurements obtained by the PoD were comparable to the commercial system. The dependence of the amperometric response with the spin rate was tested in a multichannel shield, shown in (e), with the current response for 500 µM FiC increasing with the rotation frequency, as observed in current *versus* FiC concentration plot, shown in (f). Refer to the source for details [153].

**Figure 12.** (**A**) Photograph of the experimental set-up with the spindle motor, LoaD platform, PoD and computer to control the software interface. (**B**) Exploded view of the PoD and LoaD device. (**C**) PoD results: (**a**) CVs of 1 mM FiC (Au working electrode with area of 50 mm<sup>2</sup> , scan rate of 50 mV/s comparing PoD and commercial potentiostat results, and (**b**) CVs collected in various rotation speeds at scan-rate of 100 mV/s; (**c**) SWVs of 10 mM ascorbic acid (carbon working electrode with 50 mm<sup>2</sup> , potential step 0.004 V, frequency of 10 Hz and amplitude of 0.025 V), and (**d**) calibration curve for ascorbic acid concentrations. (**e**) Amperometric response for 500 µM FiC (pH 7.4 PBS as supporting electrolyte, applied potential of −0.4 V *vs* pseudo-Au reference electrode; working electrode area: 0.69 mm<sup>2</sup> ; Au counter-electrode) in different rotation speeds, and (**f**) the normalised current changes recorded for different FiC concentrations during rotation at 1.0, 1.5, and 2.0 Hz (*n* = 3). Adapted with permission from [153]. Copyright 2019 American Chemical Society.

Sanger et al. [177] report a centrifugal platform capable of capturing and sensing pathogens. The robust LoaD contains an integrated sample pre-treatment system where SWV are used for cell free detection of a secondary metabolite, p-Coumaric acid (pCHA), a biomolecule produced by genetically modified strains of E-Coli bacteria. A common attribute of LoaD platforms is their ability to contain multiple, identical testing sites. A total of eight regions are available for individual biosensing, where a 0.2 µm filter membrane system exists in each section for sample filtration. This filtration and separation occur under centrifugal conditions for 5 min at approximately 12 Hz. The supernatant is metered and then displaced into the detection chamber. Unlike other electrochemical-LoaDs, the sensors are patterned onto an acrylic base of the platform using e-bean evaporation. Contact points connecting both potentiostat and sensor exist through the depth of the LoaD to establish a connection through a slip-ring. Measurements are obtained under static conditions and showed very promising results for

the detection and quantification of the two E. coli pCHA metabolites compared to results obtained from other detection methods such as HPLC [152].

This group also reported a novel, on-board pre-treatment based on supported liquid membrane (SLM) technology. This extraction method is used to purify and enrich analytes directly from complex matrices such urine and saliva [23,49]. Using the SLM to facilitate on-disc separation, the main hindering compound, Tyrosine, is subsequently removed and the pCHA is enriched. The platform enables multiple measuremtns to be obtained at certain stages of the production process when the target analyte is quite low (typically 100 µM). The main biosensing technique deployed were CV and SWV measurements. As the device is deemed as low-cost it was aimed to be a single use device, however it has been shown that the device can be re-used using typical electrode cycling to clean the surfaces. This work poses a significant insight into the exciting advantages and potential of on-disc sample pre-treatment and electrical detection for analytes in a compact, portable testing [152].

Amperometry is another electrochemical technique where the current response due to a redox on the surface of the biosensor is measured over a period of time at a constant potential [178,179]. It is often used as a detection technique when coupled with electrochemical-LoaD platforms. The potential is maintained at the working electrode via the reference electrode where the current is recorded due to oxidation or reduction of electroactive species. In cases where the analyte is not electrochemically active, the target cannot be measured directly but indirect approaches can be used. Screen-printed carbon electrodes (SPCE) which were modified using graphene-polyaniline nanocomposites were incorporated onto a LoaD to enhance electrochemical detection of glucose. The enzyme and glucose solutions, loaded into separate chambers, mix in a serpentine channel under centrifugal conditions creating the hydrogen peroxide through the enzymatic reaction in under 8 min. The electrochemically active hydrogen peroxide was quantified to deduce the amount of glucose present in the sample, with a reported LOD of 0.29 mM [152]. Common interfering compounds ascorbic and uric acid were tested where the sensor remained specific towards glucose. This is a critical attribute of a biosensor, as it is important to determine to working capability of the sensor for real applications.

Li et al. [180] created a whole blood (WB) analysis system with a primary functionality of performing basic metabolic testing. Different concentrations of glucose, uric and lactate acids in WB spiked samples were analysed within minutes [180]. WB is a very difficult medium to work with, as various blood components non-specifically bind to sensor surfaces. Figure 13(A) depicts the design of one of the testing sections comprising a microfluidic layer, an insulating layer an a biosensing layer. The system set-up contains a nano-porous Au plated Si electrode. Carbon nanotubes on the working electrode increases the electrode surface area which increases the rate of immobilisation of Prussian blue (catalyst) and the analytes. Sample sizes of just 16 µL was required to perform each test, where each PDMS device contained eight individual testing sites. Amperometry allowed for rapid detection and LODs of 0.3 mM of glucose, 0.1–0.2 mM of uric acid and 0.7–1.5 mM. Figure 13(B) illustrates (a) the WB sample loaded. The centrifugation of the WB samples separates the blood components and plasma for analyte capture and electrochemical detection, as illustrated in (b). To further verify the capabilities of the device, a comparison study was carried out using a commercial colorimetric assay, where results indicated deviations for glucose, lactate and uric acid of 4.2%, 5.2% and 6.0%, respectively, between both methods [180].

**Figure 13.** (**A**) Conceptual design of one of eight testing sites displaying each consecutive layer of the PDMS electrochemical-LoaD: a microfluidic layer, an insulating layer and a biosensing layer. (**B**) Electrochemical-LoaD whole blood analysis: (**a**) sample loaded ready for rotation, and (**b**) sample centrifuged and the plasma isolated for measurement of electrochemical signal. Reproduced from [180] with permission from The Royal Society of Chemistry.

Amperometry on a LoaD was again reported by Kim et al. [23] where the cardiac marker, C-reactive protein (CRP), was detected in 20 min. The on-board ELISA assay detected using gold electrodes, fabricated using E-beam lithography. 300 Å of chromium was deposited acting as an adhesion layer, followed by deposition of 3000 Å of gold to create the desired electrode pattern for the electrical connection [181]. Ferrowax valves allows for on-board reagent storage which are actuated using a mobile laser under static conditions [37]. The CRP LOD for this system was 4.9 pg/mL, a 17-fold improvement in quantification by optical density (OD) which is the standard ELISA detection method. This device adds to the very small research bracket of microfluidic-electrochemical crossovers for successful detection for very low concentrations of biomarkers.

#### **5. Biosensing on Existing and Emerging Commercial Point-of-Care Lab-on-a-Disc Devices**

Among the first companies to offer a LoaD platform on the market was Abaxis with the Piccolo Xpress [182]. In 1995 they launched the Primary Health Panel (a nine-test reagent disc), and the General Health Panel (a 12-test reagent disc). The Liver Panel Plus arrived the next year. Nowadays, Abaxis offers fast blood analysis (about 12 min) with 31 tests across 16 complete chemistry panels [182]. Thenceforth, other assays became available, as the PCR-based devices from Revogene Platform by Meridian Bioscience [183] (tests for gastrointestinal conditions, pediatric and neonatal conditions, respiratory conditions and enterobacteria infection), and the Simplexa assays to be run in 3M Integrated Cycler [184] (real-time PCR for quantitative and multi-analyte detection of virus infections, as well as to develop extraction and amplification assays using the kits the company offers in its catalog) .

Strohmeier et al. [34] presented a complete list of companies selling LoaD platforms until 2015 in a review paper. On this occasion, Hahn-Schickard [185] was cited as in the development stage. Today, this German company operates as an R & D service provider, developing solutions in microsystem technology, with a wide portfolio of services. Recently, Hahn-Schickard and related company Spindiag GmbH jointly received an investment of 6 million Euros from the German government to establish a SARS-CoV-2 virus rapid test to address COVID-19 [186] based upon their existing platforms. Spindiag [187] is a medical technology specialised start-up founded in 2016 as a spin-off of the Hahn-Schickard. In May 2020, Spindiag announced an additional 16.3 million Euros investment to prepare market launch of the Rhonda mini-lab for SARS-CoV-2 virus. The Rhonda system is a PCR-based device and will provide results in 30–40 min with minimal contact between the operator and sample. Swab samples from the nose and throat are directly inserted into the cartridge, which contains all the required reagents. The system is on an analytical test and it is expected to receive

market approval in Germany and European Union during the third quarter 2020 (information from Spindiag website [187]).

Immunoassays often require multiple steps such as sample incubation, washing and elution steps, blood separation and a final detection output. Gyros ABTM has proven commercial success with a first-generation centrifugal testing platform for immunoassay applications. The platform uses hydrophobic regions to regulate the sequential flow inside the injection moulded device which permits on-board metering and aliquoting of reagents. One device provided by the company, Gyrolab Bioaffy CD [188], has up to 112 channels which allows for high throughput testing of the sample where the users own specific biomarker may be detected. The device contains a series of metering structures, which evenly distributes the capture reagent. Streptavidin coated beads in each region are incubated with the capture biomolecules, where the excess reagent is displaced into another chamber by increasing the spin frequency of the platform. Samples, of 200 nL defined volume, enter the incubation area by capillary action where the analyte is captured. A secondary detection reagent is then passed through the capture column, creating a sandwich assay. This nanolitre immunoassay significantly reduces sample and reagent usage where fluorescence detection from the assay is reportedly obtained within the hour. The unique selling point of the device is that specific immunoassay can be designed for implementation on the device based on the required dynamic range and analyte concentration where the company itself will provide support whilst designing the LoaD assay. Ghosh et al. [189] reported using the commercially available Bioaffy 1000 CD to measure protein drugs capable of neutralises vascular endothelial growth factor to save patient vision with retinal vascular diseases. Cynomolgus ocular final concentrations were determined using an Alexafluor labelling kit and where visualisation and measurements were achieved using laser-induced fluorescent detection at a 647 nm wavelength [189]. Many advantages are offered by this device regarding assay volume sizes being realised with this platform, there are many clear drawbacks with the concept. The company does not provide with ready to use devices, where all biorecognition molecule pairs for the sandwich assay require specific user knowledge. As the assay detects biomolecules, some form of sample preparation is required which cannot be performed on the LoaD [34,190].

The LabGEO IB10 device is commercially available centrifugal platform distributed by Samsung. The device is an integrated system that combines a microfluidic platform and an analyser for cardiac biomarker detection from whole blood. This LoaD only requires 500 µL of a whole blood sample from the user, where all other reagents are pre-loaded. Proteins such CK-MB, Troponin and Myoglobin are extracted from whole blood and tested within 20 min where an optical density ranging over 10 specific wavelengths are obtained by the accompanying centrifuge analyser platform [191]. The platform contains freeze-dried reagents for blood chemistry analysis. The fluid flow is controlled using laser actuated ferrowax microvalves to activate specific reagent steps for the assay. All samples processing is carried out automatically. Other commercial LoaDs have combined both testing and optical density analysis conjoined system [154] however, Samsung appear to be the only competitors on the current market which have been able to provide analysis using whole blood analyte extraction on a PoC system.

Another example of an optical detection system is the ViroTrack® cartridge produced by BluSense Diagnostics (Denmark). Although the chip is not a LoaD by definition, it is a segment which operates using centrifugal microfluidic technology. The ViroTrack® is a single use cartridge which provides results with just one drop of blood in 10 min. The assay technology contained within the chip uses magnetic nanoparticles (MNPs) coated with capture biomolecules that will react and capture the specific target. These MNPs are released which bind with the target biomolecule within the blood. After the incubation period, a strong magnetic field aggregates with MNPs where the density of these are subsequently measured by the BluBox instrument to produce semi-quantitative results. Various versions of the chip are available to test for Dengue, Zika and Chikungunya, Covid-19 viruses [192,193].

Although there are already some LoaD platforms on the market, there is still room for extensive research into new devices. Researchers from University of Freiburg developed a disposable ready-to-use microfluidic cartridges (termed GeneSlices) for PCR assays [194,195]. Four GeneSlice cartridges were fit in a custom made rotor replacing the standard rotor in a commercial Rotor-Gene (Qiagen, Germany) [194,195]. In collaboration with Hahn-Schickard, they created an automated PCR-based disc for bacterial pathogen detection to run on a portable LabDisk-player [137] with a minimal manual step since the reagents and buffers were pre-stored in stick-packs [94]. Pathogens detection via LAMP nucleic acid amplification have been also developed [196,197]. An extreme PoC device for malaria field detection was based on a real-time LAMP assay in a compact analyzer and a disposable microfluidic compact disc. The device was capable to process 4 samples simultaneously within 50 min, with a material cost ∼\$1/test, and achieving a LOD of 0.5 parasites/µL whole blood (sufficient to detect asymptomatic parasite carriers) [196].

One of the most promising LoaD devices is in the field of circulating tumor cells (CTC). In 2014, Professor Cho's group from Ulsan National Institute of Science and Technology (UNIST) suggested a LoaD platform for CTC isolation by mechanical filtration. The size-selective CTC detection system used a commercially available track-etched polycarbonate membrane with 8 µm-pore-size, specially chosen based on the size of the CTC. The entire process of CTC isolation, staining and detection was integrated on disc by a programmable operating system. When compared to the commercial ScreenCell, the caption efficiency for the MCF-7 cells was slightly lower (56 ± 2% compared to 69 ± 6%) (*n* = 3), however, the contamination by white blood cells was 20-fold lower with the disc system [198]. In 2017, they suggested the fluid-assisted separation technology (FAST) for CTC isolation using the same filter but now with the pores filled with a stably held liquid during the entire filtration process, achieving an impressive recovery rate of 95.9 ± 3.1% [199]. While in a non-FAST mode, the filtration left the majority of particles located in the outer rim of the filter (only a small part of the filter is used), the membrane fully wet before and during the filtration process allows the liquid transport at a pressure lower than the capillary pressure, leading to a uniform filtration throughout the entire membrane [116]. The FAST device was successfully validated in a pre-clinical test in unprocessed whole blood of 40 patients with non-small cell lung cancer [116]. FAST disc was patented, licensed to Clinomics [200] (Ulsan, Korea). A pilot study using the FAST device was performed in whole blood of 13 patients with ovarian cancer, for isolation and enumeration of CTC, and to assess the correlations among CTC, cancer antigen-125 levels and the clinical course of the disease [201]. The FAST disc presents great potential to provide minimally invasive way to detect metastasis and for monitoring treatment response in patients.

Through technological advances, the current wireless technology and Bluetooth connectivity can allow the complete automation of processes on disc. Besides the control of spinning protocols, important experimental parameters, as valve action and temperature control [202], as well as real-time data acquisition [141,153,154,203–205] can be performed by Bluetooth connection. To avoid stopping the disc and/or off-disc measurements, an electrified-LoaD (eLoaD) platform [203,204] was created to allow continuous measurement while the disc is spinning with data transmission via Bluetooth, which has special importance for time-dependent reactions and biochemical reaction kinetics. The platform is assembled on a modular set-up, with an interchangeable and non-disposable "application disc" specifically designed for the intended application, sandwiched between the LoaD microfluidic disc (top) and the eLoaD platform (bottom). Authors highlighted the versatility by adapting the "application disc" to the intended protocol, for instance, manufacturing an "application disc" for optical detection by housing proper set of LEDs and photodetectors, or by housing a set of temperature sensors and heaters to automate the temperature control, etc. [204]. The wireless power is transferred during the disc rotation from a 5 W-Qi Standard transmitter, commonly used to charge smartphones. An Arduino microcontroller, a Bluetooth communication module and an SD-card module are settled in the platform [204]. The portability and the easy control by Tablets and Smartphones can be a key feature for extreme PoC devices, and for daily diagnostics for a rapid data transfer to health care professionals by remote consultation.

Technological advances in materials processing and digital innovations (mobile-health and internet of things) have contributed to create diagnostic tests to satisfy (or nearly satisfy) the REASSURED criteria for PoC devices [31].


#### **6. Future Perspective of Biosensing Platforms**

Biosensing LoaDs performing bio-assays often requires complex sample processing with an increased level of integration, making them robust and effective tools in PoC testing. Advancements have been made in LoaD manufacturing and fabrication strategies in the past decade, however significant challenges are prevalent for the successful integration of microfluidic PoC systems into routine clinical diagnosis [206]. One of the main challenges for centrifugal microfluidics to become commercially desirable is extending the shelf life of the PoC device. For biosensing and assay performance, multiple reagents are necessary for testing. For these platform to remain highly attractive products on the market, the LoaDs must be semi/fully automated. This attribute is dependent on multiple factors, including;


#### **7. Final Remarks**

In search of complementarity between microfluidics and biosensing, the centrifugal microfluidic platform has emerged as an preeminent platform of significant interest to academic researchers and commercial enterprises. The key advantage is the capability of the LoaD to enable transfer of LUOs (pipetting, metering, mixing etc.) directly from the conventional lab "wet bench" to "on disc" while requiring minimial protocol changes or optimisation. This ease of automation results in faster and more robust assay development compared with other emerging PoC platforms, such as paper microfluidics [221], where advantages such as greater deployability (for example not requiring a spindle motor to function) are counterbalanced by a greater upfront burden of assay development or biosensor optimisation. Furthermore, the LoaD platform is continuously evolving and has been shown to be a platform which can be adapted to leverage the latest emerging technologies (such as the Arduino controllers and wireless mobile-phone charging technology described above). Every year new LoaD automation and biosensing technologies have been developed to demonstrate sample-to-answer PoC biodetection with minimal user interaction. Even with devices already commercialised, there is plenty of space for researchers to develop the LoaD with new technologies to address new challenges. We hope this review paper has provided researchers interested in working in this field a comprehensive overview of the recent advances on LoaD biosensors.

**Author Contributions:** Writing—original draft preparation, C.M.M., E.C., and D.J.K.; writing—review and editing, C.M.M., E.C., and D.J.K. All authors have read and agreed to the published version of the manuscript.

**Funding:** Sao Paulo Research Foundation grants 2014/15093-7 and 2015/16311-0 and partly funded by Enterprise Ireland under Grant No. CF/2019/1080.

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

#### **Abbreviations**

The following abbreviations are used in this manuscript:



#### **References**


**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

### *Article* **Controlling Nanoparticle Formulation: A Low-Budget Prototype for the Automation of a Microfluidic Platform**

**Dominik M. Loy 1,\* , Rafał Krzyszto ´n 2,3 , Ulrich Lächelt <sup>1</sup> , Joachim O. Rädler 2,3 and Ernst Wagner <sup>1</sup>**


Geschwister-Scholl-Platz 1, 80539 Munich, Bavaria, Germany **\*** Correspondence: dominik.loy@cup.uni-muenchen.de

**Abstract:** Active pharmaceutical ingredients (API) with suboptimal pharmacokinetic properties may require formulation into nanoparticles. In addition to the quality of the excipients, production parameters are crucial for producing nanoparticles which reliably deliver APIs to their target. Microfluidic platforms promise increased control over the formulation process due to the decreased degrees of freedom at the micro- and nanoscale. Publications about these platforms usually provide only limited information about the soft- and hardware required to integrate the microfluidic chip seamlessly into an experimental set-up. We describe a modular, low-budget prototype for microfluidic mixing in detail. The prototype consists of four modules. The control module is a raspberry pi executing customizable python scripts to control the syringe pumps and the fraction collector. The feeding module consists of up to three commercially available, programable syringe pumps. The formulation module can be any macro- or microfluidic chip connectable to syringe pumps. The collection module is a custom-built fraction collector. We describe each feature of the working prototype and demonstrate its power with polyplexes formulated from siRNA and two different oligomers that are fed to the chip at two different stages during the assembly of the nanoparticles.

**Keywords:** nanoparticle; lipoplex; polyplex; raspberry pi; siRNA; python; microfluidics

#### **1. Introduction**

Packaging active pharmaceutical ingredients (API) into nanoparticles can alter the pharmacokinetic properties of drugs fundamentally [1–3]. It is a well-established strategy to improve the biodistribution of small molecule drugs like paclitaxel [4] as well as larger, oligomeric drugs like nucleic acids [5,6].

Many methods have been developed to produce nanoparticles containing the target API either with the top-down [7–11] or the bottom-up approach [12]. Mixing cationic oligo- or polymeric excipients with negatively charged nucleic acids to induce nanoparticle formation, for example, is an established method from the bottom-up approach [13]. These nanoparticles are named polyplexes [14].

The properties of polyplexes and ultimately their effectiveness is determined by their individual components and the formulation process parameters [15].

On the one hand, polyplex properties can be controlled by the application of solidphase supported synthesis (SPSS) [16] to precisely design the chemical components of the polyplex. This technique enables the synthesis of oligomers with defined structures, for example, sequence defined oligo(ethanamino)amides [17]. These oligoamides form the backbone of more elaborate structures with additional functional elements integrated at exactly defined places in the oligomer sequence [18,19].

**Citation:** Loy, D.M.; Krzyszto ´n, R.; Lächelt, U.; Rädler, J.O.; Wagner, E. Controlling Nanoparticle Formulation: A Low-Budget Prototype for the Automation of a Microfluidic Platform. *Processes* **2021**, *9*, 129. https://doi.org/10.3390/ pr9010129

Received: 29 November 2020 Accepted: 5 January 2021 Published: 8 January 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

<sup>80539</sup> Munich, Bavaria, Germany; rafal.krzyszton@stonybrook.edu (R.K.); raedler@lmu.de (J.O.R.) <sup>3</sup> Graduate School of Quantitative Biosciences (QBM), Ludwig-Maximilians-Universität München,

On the other hand, the formulation process as well exerts a decisive influence on the polyplex characteristics. The polyelectrolyte complex formation is thermodynamically favorable due to a high entropy gain from counter ion released during complexation [20]. Nevertheless, formulation parameters determine the polyplex properties due to kinetic control over the complex formation process [21].

These considerations indicate that both the production process and the defined polyplex components are important for the therapeutic success of the final formulation. One approach to reduce unintended variation of process parameters is the automation of the process. For example, particle properties can be improved measurably by controlling the educt feeding rates to a T-junction to produce lipoplexes [22] or polyplexes [23].

Additionally, transferring the mixing process to the micrometer scale reduces the degrees of freedom of the system [24]. At this scale, forces from interfaces greatly surpass inertial forces that dominate the macro scale [25]. These effects are exploited in microfluidic devices and potentially lead to an increased control over the mixing process itself. Many studies have demonstrated that methods based on microfluidics can improve the physicochemical properties of nanoparticles [26–29]. In our previous publication [30], we demonstrated that the application of a microfluidic chip increased control over the sequential polyplex formulation process by exploiting the advantages of solvent exchange in combination with flow focusing inside the microchannel. The educts and the production method are depicted in Figure 1. Sequence-defined formulation components were a lipo-oligomer for nanoparticle core formation [31], a lipid anchor-polyethylene glycol (PEG)-ligand for coating [30], and siRNA. Schematics of the utilized microfluidic chips (Supplemental Figures S1 and S2) are presented in the Supplemental Information. The resulting multi-component polyplexes were well-defined due to the increased level of control over the formulation process and allowed the establishment of structure-function relationships between the PEG-ligand length and the siRNA transfection efficiency [30].

**Figure 1. Production methods for polyplexes from oligomers and siRNA.** Formulation components used are depicted with the id label of their corresponding syringe (S1–S4). Two different channels were used to produce nanoparticles during solvent exchange, a single meander channel (left) and a double meander channel (right). In the single meander channel, pre-assembled core particles (S1) were mixed with lipid anchor polyethylene glycol (PEG)-ligand oligomer (S2). In the double meander channel, the polyplex was assembled by microfluidics in two subsequent steps from its starting components (S3, S4, S2). PEG: polyethylene glycol. HBG: HEPES buffered glucose pH 7.4. Core oligomer (S3): cationic lipo-oligomer. Lipid anchor-PEG-ligand (S2): oligomer containing fatty acids to hydrophobically adsorb to the core oligomer, a PEG chain and folic acid as ligand for the folate receptor. Core polyplex (S1): polyplex from siRNA (S4) + core oligomer (S3). Reproduced with modification from Loy et al., 2019 [30], https://doi.org/10.7717/peerj-matsci.1/fig-1. under copyright permission from PEERJ.

Exploiting the benefits of automated nanoparticle production systems is often associated with significant investments in hardware, software, and development work to integrate any microfluidic platform into a typical lab environment. To ameliorate this problem, we have developed a modular, low budget system around the microfluidic chips from our previous publication [30]. The system can easily integrate most microfluidic platforms driven by syringe pumps into its setup. Here, we describe the set-up of the system and its application in detail using educts from our previous publication as an example.

#### **2. Results**

The complete nanoparticle production system is depicted in Figure 2. It consists of four modules that can be used independently: the feeding module—up to three programmable syringe pumps—is responsible for supplying educts to the formulation module, which can be any macro or microfluidic chip. The collection module—a custom-built fraction collector—is responsible for collecting the final product into standardized well plates. The control module is a remotely accessible raspberry pi which controls the syringe pumps via a Recommended Standard 232 (RS232) interface and the fraction collector via the generalpurpose input/output (GPIO) pins. The design of the fraction collector as well as the python program code are published together with this paper on GitHub [32]. This setup allows the employment of most microfluidic chips while additionally providing the ability to sample the product from the chip directly into standardized well plates. We describe all modules in detail in the following sections.

**Figure 2. Overview over the nanoparticle production system.** The system consists of four modules that can be used independently. The control module (green) is a raspberry pi which controls the collection module (blue), a custom-built fraction collector, via its GPIO pins. The raspberry pi controls the feeding module (red) via the RS232 interface. It is assembled from up to three syringe pumps. The formulation module can be any macro or microfluidic chip.

#### *2.1. Control Module*

The control module is a raspberry pi model 3B. It is a small and inexpensive singleboard computer with open hardware, e.g., general-purpose input/output (GPIO) pins. We used the open source operating system (OS) Raspbian version 9 (Stretch) with the open source programming language Python version 3.7.3 for this system. The python code for controlling the feeding module and the collection module is available on our GitHub repository [32].

We chose this computer because of several features. First, its size together with the ability to access it remotely over the network increases the mobility of the complete device. Second, its various interfaces enable the communication with the feeding module via a USB to RS232 interface as well as the communication of the collection module via the GPIO pins. Third, it can be easily replaced by any other computer from the raspberry pi family since it is cheap and the OS together with all data, e.g., the logs of each experiment, are stored on a micro SD card. Therefore, changing the control module, e.g., because it was damaged or more computing power is needed, is only a matter of switching the SD card into the new device.

#### *2.2. Feeding Module*

The feeding module consists of up to three syringe pumps that are daisy-chained to the raspberry pi via a R232 to USB interface. Here, we used LA120, LA122, and LA160 from Landgraf Laborsysteme HLL GmbH. LA120 and LA160 are standard syringe pumps with two and six channels, respectively. LA122 is microfluidic syringe pump with two channels which is especially suited for dispensing smaller volumes due to its higher precision.

In principle, any syringe pump can be integrated into the system if it satisfies the following prerequisites: First, the pumps must have an interface that can be connected to the control module, e.g., the RS232 serial interface. Second, the pump must be programmable. In order to reduce the risk of interferences during particle production, the complete program is written to the pumps in advance and the pumps execute the production program independently. If a pump with a different command structure is integrated into the system, however, commands sent to the pump must be adjusted. A detailed description on changing commands sent to the pumps can be found in the Supplemental Information (3.2.3. Module: Module\_pumps.py).

The control program of the feeding modules consists of six modules that are described in detail in the Supplemental Information (3.2. Description of the python modules) together with a Unified Modeling Language (UML) class diagram to illustrate the dependencies between the classes of the modules (Figure S3). The main module that calls the required functions from the respective modules to execute a certain pumping program is called 'main.py.' We provide a library with different 'main[ . . . ].py' modules. If the module 'main.py' is executed, the user will be asked to input all parameters during runtime, e.g., flow rates and volumes. If one of the modules 'main\_[ . . . ]\_automated.py' is executed, the parameters defined in the module will be used to run the pre-defined pumping program without requiring any user input. These modules serve as examples of how to define target variables and how to customize the main module. The code of the 'main.py' module is described in the Supplemental Information (3.2.7. Modules main[ . . . ].py).

Several features are implemented in the control program to simplify the employment of different formulation modules, to document experiments, and to save educts: first, formulation module specifications are loaded into the program during runtime from a simple text file. In order to employ a new channel, an updated text file needs to be supplied to the program. A detailed description on adding new formulation module specifications can be found in the Supplemental Information (3.2.1 Module: channels.py). Second, a logging function was integrated into the program, which writes every event and its timestamp to a text file stored on the control module. This log can be used for documenting and for troubleshooting purposes. Third, the implementation of ramping and purging capabilities reduces the waste of educts to a minimum. When large flow rate

changes occur (e.g., when a pump is started), the system needs some time to adapt to the increased pressure. This can lead to the retardation of educts due to the elasticity of the system. Bringing educts efficiently (i.e., without wasting time or educts) to the mixing zone without involuntarily changing the volume ratios is challenging especially at the beginning of a new run. The easiest solution would be to use the flow rates of the first experiment to pump all educts to the mixing zone. Applying this strategy, however, increases waste of time and educts in relation to the flow rate differences between the educts. Ramping all educts to the mixing zone without changing the mean flow rate alleviates this problem and prevents unnecessary waste.

Additionally, employing the ramping protocol can reduce the backflow from the syringe pumps. During transition from preparations to formulation, the ramping protocol ensures a smooth transition between flow rate changes and keeps the overall flow rate constant, minimizing pressured changes that can provoke backflows. Moreover, when flow rates need to be changed during the formulation of the product, the program automatically inserts an overlap volume between those two fractions to allow some time for the flow to stabilize again. The overlap volume can be adjusted according to the magnitude of the flow rate changes. If large flow rate changes take place (e.g., when flow rates between slow and fast pumping pumps are interchanged), the modularity of the program allows another execution of the ramping protocol. Furthermore, fractions affected by backpressure instabilities can automatically be excluded using the collection module.

The ramping program is described in the Supplemental Information (3.2.5. Module: ramping\_class.py).

The purging functions enable the user to choose the least expensive reagent to purge the product from the channel after the experiment. A detailed description of this function can be found in the Supplemental Information (3.2.6. Module: mixing\_class.py).

A flowchart describing the workflow from starting the system to collecting the final product(s) and resetting the system to its original state is shown in the Supplemental Information (Figure S4).

#### *2.3. Formulation Module*

The formulation module can be any micro- or macrofluidic chip that is connectable to syringe pumps. In our prototype, we employed two different microfluidic chips that are based on the design from Krzyszto ´n et al. [29]. These chips are made from polydimethylsiloxane (PDMS) bonded to glass slides. Both chips exploit the advantages of solvent exchange in combination with flow-focusing inside the microchannel to produce polyplexes from siRNA and polycationic oligomers. The layout of both chips together with the utilized educts is shown in Figure 1. The single meander channel (SMC) employs the design of a Y-junction followed by a long meandering channel section while the double meander channel (DMC) features two successive Y-junctions followed by their respective meandering section which allows the assembly of polyplexes in two consecutive steps. Detailed schematics of both channels are shown in the Supplemental Information (SMC: Figure S1, DMC: Figure S2). The chips were made from polydimethylsiloxane (PDMS) bonded to glass slides. Wang et al. [33] have made suggestions to increase durability of these chips. We used both chips in our previous publication to produce well-defined, multicomponent polyplexes that allowed the establishment of structure-function relationships between PEG-ligand length and transfection efficiency due to the increased level of control over the formulation process [30]. Here, we only show data produced with the DMC to highlight the potential of the device to produce sophisticated formulations. For a comparison of the core formulation (siRNA and CO) prepared by the SMC, at a T-junction, or by rapid pipetting, please see Figure 2 in our previous publication [30].

#### *2.4. Collection Module*

The design is based on previously published work [34]. It was optimized for greater robustness and user safety, especially by choosing aluminum to decrease wear, increase resistance to common solvents (except acids), and to increase the accuracy of fit of the machine. Increased user safety was realized by including stop switches into the design.

The fraction collector is controlled by a raspberry pi 3, model B running Raspbian GNU/Linux 9 (stretch). The raspberry pi controls the fraction collector via input/output (GPIO) pins. Figure 3 shows an overview of the complete fraction collector (Figure 3A), the wiring of each component (Figure 3B,D), and the GPIO pin assignment (Figure 3C).

**Figure 3. Overview over the fraction collector.** (**A**) Overview fraction collector. (**B**) Wiring of the end switches. The end switches are supplied with 5V power from the pins of the raspberry pi, and the signal is sent from the switches to GPIO 17 or 27 (green wire). (**C**) GPIO pin assignment. Schematics of the GPIO pins of the raspberry pi. Saturated colors and bold script indicate utilized pins. (**D**) Wiring of the H-bridge. Each H-bridge controls one stepper motor. Power is supplied by a 12 V, 1 A switching power supply and routed to each stepper motor by four output wires (coil one: black and green wires; coil two: red and blue wires). Power distribution is controlled by the GPIO pins. GPIOs 18 and 23 or 05 and 06 (orange wires) control the direction of coil one, while GPIOs 24 and 25 or 13 and 26 (light green wires) control the direction of coil two.

The control program for fraction collector is an independent piece of software. This approach allows the integration of the collector control software into the pumping program but enables usage of this device with other, non-automated processes, as well. It

consists of three modules which are described in detail in the Supplemental Information (4.2. Description of python modules). The main module is called 'main.py.' It calls the required functions from the respective modules to execute a certain collection program. The module serves as example how to define target variables and how to customize the collection program. A video documenting the execution of the 'main.py' module can be found on GitHub [32]. The code of the 'main.py' module is described in the Supplemental Information (4.2.3. Module: main.py). A complete list of all classes and functions of the modules can be found on GitHub, as well [32]. The dependencies between the classes of the program are depicted in a UML class diagram in the Supplemental Information (Figure S3).

#### *2.5. Application for Polyplex Formation*

In the following, we highlight the importance of precisely defined process parameters and show the formulation of three component polyplexes with the automated nanoparticle production system.

Figure 4 demonstrates the influence of formulation parameters on polyplexes formulated from two components by rapid pipetting. By adjusting the volume ratios and the mixing order of the oligomer and siRNA solutions, significant changes in size (hydrodynamic diameter) and polydispersity index (PDI) can be achieved.

**Figure 4. Influence of formulation conditions on manually prepared polyplexes (oligomer CO + siRNA).** Dynamic light scattering (DLS) data are represented as the mean of three measurements. Color and shape encode the mixing order. The volume parts of both educts are denoted on the x-axis. Total volume was 70 µL for each solution. That means, for example, that 64.2 µL of a diluted oligomer solution (oligo) was pipetted to 5.8 µL of a concentrated siRNA solution (light red dot, volume parts 11/1). Blue square: siRNA was pipetted into an oligomer solution. Red dot: The oligomer solution was pipetted into a siRNA solution. (**A**) Mean hydrodynamic diameter (z-average). (**B**) Mean polydispersity index (PDI). Statistics: Error bars correspond to 95% confidence intervals. N = 3.

For this experiment we produced polyplexes using siRNA and a core oligomer (CO) [30,31]. Light red dots present data obtained after the oligomer solution was pipetted into the siRNA solution, and blue squares represent the result after pipetting the siRNA into the oligomer solution. The number written on the x-axis denotes the volume parts of the two solutions in the final solution. The final volume of each solution was always 70 µL. For example, data resulting in the third light red dot (11/1 on the x-axis) were obtained from 64.2 µL oligomer solution that was pipetted to 5.8 µL siRNA solution.

Mixtures of equal volumes of educts solutions produced comparable hydrodynamic diameters (Figure 4A, circle: 82.5 ± 2.7 nm, square: 84.4 ± 3.1 nm) and PDIs (Figure 4B, circle: 0.151 ± 0.058, square: 0.136 ± 0.052) independent of the mixing order. If unequal volumes were mixed, however, the mixing order influenced particle characteristics significantly. Pipetting a smaller volume of the oligomer solution into a larger volume of the siRNA solution (1/11 on the x-axis) produced larger polyplexes (141.0 ± 3.2 nm) with a smaller PDI (0.109 ± 0.030), while mixing siRNA solution to an oligomer solution produced smaller particles (108.0 ± 1.4 nm) with a larger PDI (0.180 ± 0.048).

Pipetting a larger volume of the oligomer solution to a smaller volume of the siRNA solution (11/1 on the x-axis) produced very small particles (Figure 4A, 52.5 ± 9.7 nm) with a larger PDI (Figure 4B, 0.249 ± 0.144). Mixing diluted siRNA solution to concentrated oligomer solution produced slightly larger particles (104.0 ± 6.5 nm) with a comparable PDI (0.131 ± 0.063) in comparison to polyplexes from mixtures of equal volumes. The 95% confidence intervals from the z-average as well as from the PDI, however, were very large, indicating the presence of particles from different size classes.

Nanoparticles formulated from more than two components usually require increased control over the production process. Figure 5 highlights this critical issue. The formulation (siRNA/CO; 1/1 on the x-axis, light red dot) described in Figure 4 was further modified with a third oligomer that contributes shielding and targeting features to the nanoparticle. It consists of a lipid anchor for integrating into the core particle, a PEG12 chain for shielding purposes and an azide moiety that allows the simple addition of further shielding and targeting ligands via strain-promoted azide-alkyne click chemistry [35]. Here, we utilize the two simplest versions of the lipid anchor oligomer with a free azide moiety and with (LPOE) or without (LPO) two additional glutamic acids (E). The sequences of all oligomers are depicted in the Supplemental Information (Figure S7). Results from in vitro experiments with these three component polyplexes with PEG-Folic acid ligands with 12 to 60 ethylene oxide repetitions can be found in our previous publication [30]. In the same publication, we compared the influence of the production method on the biological activity of two component polyplexes. We were able to demonstrate comparable biological activity in vitro [30].

In Figure 5A,B, equal volumes of the three educts were mixed sequentially by rapid pipetting according to the order of appearance denoted on the x-axis. Color and shape indicate if LPO or LPOE was used. Manual production of three component polyplexes from CO, siRNA, and LPO yielded polyplexes with suboptimal hydrodynamic diameters and PDIs regardless of mixing order (CO + siRNA + LPO: d<sup>Z</sup> = 416.2 ± 62.5 nm, PDI = 0.711 ± 0.233; CO + LPO + siRNA: d<sup>Z</sup> = 466.7 ± 33.1 nm, PDI = 0.792 ± 0.068). When LPO was replaced with LPOE, the mean hydrodynamic diameter of the polyplexes was reduced to acceptable levels, but the mean PDI was still too large (CO + siRNA + LPOE: d<sup>Z</sup> = 128.2 ± 22.4 nm, PDI = 0.468 ± 0.101; CO + LPOE + siRNA: d<sup>Z</sup> = 114.5 ± 1.5 nm, PDI = 0.428 ± 0.058).

**Figure 5. Manual or automated formulation of three component siRNA polyplexes.** DLS data are represented as the mean of three measurements. Color and shape encode either the difference in the sequence of lipid anchored PEG12 oligomers (LPO, with or without E (glutamic acid), panel **A**,**B**) or the difference in formulation conditions (oligomer CO dissolved in HBG with or without 50% acetone, panel C,D). (**A**,**B**): polyplexes were formulated manually by mixing all educts with pipettes. The mixing order is denoted on the x-axis. Orange triangle: the sequence of the LPO contains two additional glutamic acids. Green diamond: no additional glutamic acids. (**C**,**D**): polyplexes were formulated automatically inside the double meander channel (DMC, Figure 1) by the nanoparticle production system. Flow rates: siRNA 900 µL/h (S4), CO 100 µL/h (S3). LPO(E) 50 µL/h (S2, two syringes). Total flow rate: 1100 µL/h. The educts are denoted on the x-axis. Red dots: CO was dissolved in HBG only. Blue squares: CO was dissolved in HBG with 50% acetone. (**A**,**C**): mean hydrodynamic diameter (z-average). (**B**,**D**): Mean polydispersity index (PDI). Error bars correspond to 95% confidence intervals. N = 3. Raw data were selected from our previous publication [30], here presented in a new format.

> In Figure 5C,D, polyplexes from the three educts were produced automatically inside the double meander channel (DMC) by the nanoparticle production system. In the first mixing zone, CO and siRNA were mixed. The color and the shape of the data points indicate if CO

was dissolved in HBG with or without 50% acetone. In the second mixing zone, LPO or LPOE was added to the mixture. Polyplexes prepared from CO dissolved in HBG only showed slightly higher hydrodynamic diameters d<sup>Z</sup> and PDI (CO + siRNA + LPO: d<sup>Z</sup> = 153.0 ± 12.7 nm, PDI = 0.210 ± 0.062; CO + siRNA + LPOE: d<sup>Z</sup> = 148.2 ± 8.7 nm, PDI = 0.306 ± 0.003) in comparison to polyplexes prepared from CO dissolved in HBG with 50% acetone (CO + siRNA + LPO: d<sup>Z</sup> = 114.7 ± 1.5 nm, PDI = 0.137 ± 0.045; CO + siRNA + LPOE: d<sup>Z</sup> = 141.9 ± 4.7 nm, PDI = 0.230 ± 0.022).

Automated production of three component polyplexes (CO + siRNA + LPO) generated nanoparticles with smaller hydrodynamic diameters and PDIs compared to manually prepared polyplexes. Incorporating glutamic acid into the structure of the lipid anchor PEG12 oligomer facilitated the production of polyplexes with comparable mean hydrodynamic diameters regardless of production method. Nevertheless, the PDI of manually prepared polyplexes was still larger than the PDI of polyplexes prepared with the nanoparticle production system.

#### **3. Discussion**

A detailed description of the automated nanoparticle production system, its hardware, and its software is provided. The control module is inexpensive, and its parts are readily available. The feeding module integrates syringe pumps as commonly applied in microfluidic systems such as in Liu et al. (PHD 2000, Harvard Apparatus) [36], Debus et al. (Aladdin, World precision Instruments) [37], Lim et al. (model unspecified, Harvard Apparatus) [38], Karnik et al. (SP220I, World Precision Instruments and PHD 22/2000, Harvard Apparatus) [39], and Belliveau et al. (KD200, KD Scientific) [27]. The schematics of the collection module are published together with this paper on GitHub [32], enabling the replication of this module in any workshop. Additionally, building the collection module with additive fabrication methods, e.g., 3-D printing, might also be feasible.

The software that controls the feeding and the collection module enhances the functionality of any formulation module. With a specific, customized main module for each individual experiment, reproducibility is increased since every production cycle follows the same commands. Additionally, logs of each experiment are available to document the intended execution of the program. With each main module tailored to the specific needs of any experiment, repeating an experiment is done by simply executing the program again. Additional benefits of employing the software to control the feeding module are the ramping and purging functionalities that reduce the waste of educts to a minimum. These functionalities are cumbersome at best to program into each pump manually, but readily available in our software. The ramping functions ensure that all educts reach the mixing zone at the same time and the purging functions enable the user to choose the least expensive reagent to purge the product from the channel after the experiment.

Although volumes and flow rates of each experiment can be taken from its log, the components and concentrations must still be recorded manually. We have demonstrated the importance of detailed experiment descriptions with polyplexes prepared manually from two components. Changing the volume ratios of the educts and the mixing order varied the hydrodynamic diameter and PDI of the resulting polyplexes from 52.5 ± 9.7 nm to 141.0 ± 3.2 nm and 0.249 ± 0.144 to 0.109 ± 0.030, respectively. The effect of volume ratios on particle sizes is probably due to turbulences of varying intensity, which usually promote faster mixing of the educts. This effect is especially important during the polyplex complexation process since charge neutralization occurs in around 50 ms [21]. Additionally, we demonstrated that some formulations might be impossible to be produced with pipettes and require a formulation module—especially formulations from three or more components seem to benefit from the increased control of a microfluidic setup. Krzyszto ´n et al., for example, improved the efficiency of their mNALP (monomolecular nucleic acid/lipid particles) formulation by microfluidic mixing on the same hydrodynamic flow-focusing chip without our device [29]. We have prepared three component polyplexes manually and automatically. Polyplexes prepared by rapid pipetting showed hydrodynamic diameters

and PDIs in suboptimal ranges. With the automated nanoparticle production system, polyplexes with d<sup>Z</sup> = 114.7 ± 1.5 nm and PDI = 0.137 ± 0.045 could be produced. This finding and the application of the automated nanoparticle production system enabled the establishment of structure-function relationships from three component polyplexes in our previous publication [30].

Sizes and PDIs in a desired range, however, do not automatically guarantee superior biological activity of target nanoparticles in vitro or in vivo. On the one hand, nanoparticles produced with controlled methods might show improved formulation characteristics and equal (but not better) biological activity. Members from our lab, for example, demonstrated the reproducible production of polyplexes from pDNA and LPEI (linear polyethylene imine) with an up-scaled micro-mixer. Compared with manually formulated polyplexes, both formulations showed comparable biological activity in vitro [23]. On the other hand, formulations with larger PDIs and sizes might show apparently better transfection efficiencies in vitro. This is usually due to large particles literally "dropping" on the cells fixed to the bottom of the cell culture flask. A formulation with these properties, however, might fail in vivo.

All in all, the application of this versatile software enables the creation and automated execution of a sophisticated program consisting of many individual steps in order to increase control over the formulation process of nanoparticles and foster reproducibility, which will be most relevant for pharmaceutical production.

The next step on the course to automation is the integration of a fraction collector. The device developed here was designed to work with any standard well plate to realize product collection and separation. It is independent of the previously mentioned setup, which makes it suitable for a wide range of applications. It can be integrated into the target automated process, but it can also be used to gather products produced manually. Overall, it is a versatile addition to any product formulation setup relieving the user of additional manual labor.

A well-known disadvantage of microfluidic systems is the scalability problem [24]. Due to the utilization of fluid phenomena—for example, laminar flow—which are only present under certain conditions, the throughput of one microfluidic chip cannot be escalated indefinitely [25]. The obvious solution to employ parallelization is a valid suggestion, but product output does only scale linearly in relation to dedicated resources at the current development stage of the system due to the many individual steps involved in setting up the device. However, a possible solution is already designed in the system. Since the setup is modular, any part of it can easily be replaced by a more efficient one [40].

Placing the complete system into a laminar flow cabinet is a next obvious step for pharmaceutical applications. In the described work, nanoparticles were formulated outside of the cabinet and subsequently transferred inside for in vitro transfections of cells. With the complete system inside the cabinet, direct application of the product to the target cells could be achieved, which would decrease the influences of external factors and human interactions even further.

#### **4. Materials and Methods**

#### *4.1. Materials*

The materials and the software used for the control module are listed in the Supplemental Information (2.1. Materials, Table S2; 2.2. Software, Table S3).

The materials used for building the feeding module are listed in the Supplemental Information (3.1. Materials, Table S4).

Materials used for building the formulation modules used in this publication are listed in the Supplemental Information (1.1. Materials, Table S1).

The design and the manufacturing procedure of the microfluidic chips is described in Krzyszto ´n et al., 2017 [29].

In brief, the design of the microfluidic channels was realized on a silica wafer with soft lithographic methods. The finished wafer was covered with polydimethylsiloxane (PDMS) mixed with 10% (*w*/*w*) crosslinker, degassed, and cured (75 ◦C, 4 h). Subsequently, the solid PDMS was cut and removed from the wafer. Inlets and outlets were pierced with a biopsy puncher and the channel was bonded to a glass slide with an oxygen plasma cleaner (Diener Electronic; 10 W high frequency generator power, 12 s, Pico Model E). Polyethylene tubes (length: 110 mm, inner diameter 0.38 mm) were fitted to the holes and the complete chip was covered in another layer of PDMS to increase resistance to pressure. Wang et al., 2014 investigated the tubing of PDMS channels and offer an improved protocol to prevent channel leakage [33]. Each new channel was tested before application with a standard formulation. The size and PDI measured by DLS were compared to the results from the same formulation produced with the previous channel. A to-scale model of the channels can be found in the Supplemental Information (1.2. Schematics, Figures S1 and S2). We calculated the Reynold's number (Re), Dean's number (De), and backpressure (∆P) for the SMC and DMC at a total flow rate of 1500 µL/h: SMC: Re ≈ 3, De ≈ 1.23, ∆P = 1249.4 mbar; DMC: Re ≈ 2.5, De ≈ 1.25, ∆P = 2498.8 mbar. Solvents used in this paper are classified as low-solubility solvents which are compatible with microfluidic systems fabricated in PDMS by Lee et al. Therefore, they are unlikely to cause considerable changes to the channel geometry due to swelling [41].

Materials used for building the collection module are listed in the Supplemental Information (4.1. Materials, Table S5).

A prototype of the collection module (fraction collector) was built according to the design published at GitHub (https://github.com/Dominikmloy/fraction-collector.git). All parts were cut from aluminum, except the parts noted below. The prototype was built by the workshop of the LMU Munich.

Materials for the formulation of polyplexes and DLS measurements: The synthesis of the oligomers (CO: id = 991, LPO: id = 1203, LPOE: id = 1223) and the required materials are described in detail in [30,31]. The sequences of all oligomers are depicted in the Supplemental Information (5. Oligomers, Figure S7). Solvents: Purified water (produced with Ultra Clear GP UV UF, Evoqua Water Technologies GmbH, Günzburg, Germany), acetone HPLC grade (VWR international GmbH, Darmstadt, Germany). Chemicals: 4-(2-hydroxyethyl)-1 piperazineethanesulfonic acid ultra-pure (HEPES, Biomol GmbH, Hamburg, Germany), D(+)glucose monohydrate DAB (Loewe Biochemica GmbH, Sauerlach, Germany), NaOH pellets puriss. (VWR international GmbH, Darmstadt, Germany), NaOH 1M standard solution (Thermo Fisher Scientific GmbH, Schwerte, Germany), HCl 1M standard solution (VWR international GmbH, Darmstadt, Germany). Nucleic acids: siGFP sense: 5′ - AuAucAuGGccGAcAAGcAdTsdT-3′ , antisense: 5′ -UGCUUGUCGGCcAUGAuAUdTsdT-3 ′ (Axolabs, Kulmbach, Germany). Small letters: 2′ methoxy; s: phosphorothioate.

#### *4.2. Methods*

#### 4.2.1. Polyplex Preparation

Polyplexes were prepared with a final siRNA concentration of 0.025 mg/mL. A nitrogen to phosphate (N/P) ratio of 12 was used to determine the amount of core oligomer CO relative to the amount of siRNA. The N/P ratio relates the number of positive charges from the primary and secondary amines in the oligomer's backbone to the number of negative charges from the phosphates in the siRNA's backbone. The manual method of polyplex preparation was done with pipettes and rapid mixing in a batch wise process. The solvent—if not noted differently—was HEPES buffer pH 7.4 with 5% glucose (HBG). This buffer was used because it does not rely on salts to be isotonic, since polyplex formation relies on charge interactions that could be hampered by ions.

Manual polyplex preparation: CO solution (0.504 mg/mL) was added quickly to a siRNA solution (0.05 mg/mL) of equal volume and mixed by rapid pipetting, achieving a final siRNA concentration of 0.025 mg/mL. Subsequently, the formulation was incubated for 45 min. Concentrations and volumes for mixing polyplexes from unequal volumes were adjusted accordingly: 5.8 µL of CO at 3.023 mg/mL, or 64.2 µL of CO at 0.275 mg/mL. 64.2 µL of siRNA at 0.027 mg/mL, or 5.8 µL of siRNA at 0.300 mg/mL.

For the manual formulation of three component polyplexes, equal volumes (27.7 µL) of CO solution (0.637 mg/mL) and siRNA solution (0.063 mg/mL) were used. The amount of LPO and LPOE was set to 20 mol% relative to CO. Concentrations were set to 0.207 mg/mL LPO, 0.224 mg/mL LPOE, volumes were 14.6 µL. Solutions were mixed sequentially by rapid pipetting. When siRNA was used in the first step, a ten-minute break was taken after the two components were mixed to allow the polyplex to stabilize. After the addition of the third component, the formulation was incubated for 45 min before DLS was measured.

Automated polyplex preparation: The formulation module with the double meander channel (DMC) was used without any additional surface treatment (Figure 1). Before each usage, the channel was washed and primed with the same solvents that were used to produce the polyplexes. Details about the washing/priming process can be found in the Supplemental Information (3.2.4 Module: setup.py). siRNA in HBG (0.033 mg/mL) was loaded into S4 (FR = 900 µL/h) and CO (3.025 mg/mL) in HBG or HBG with 50% acetone to retard siRNA compaction was loaded into S3 (FR = 100 µL/h). LPO or LPOE in HBG with 50% acetone to facilitate solvent exchange were loaded into S2. The flow rate of each syringe S2 was 50 µL/h at a total flow rate of 1100 µL/h, resulting in a flow rate ratio of lipid anchor oligomer to core polyplex of 1:11. The final product was diluted with HBG to 0.025 mg siRNA/mL.

Stability of the formulation presented in this paper has been investigated previously. Troiber et al. have found particles assembled from the same class of oligomers by rapid pipetting to be stable over three weeks [42]. In our previous paper, we have investigated the changes in size, PDI, and zeta potential of our core formulation (siRNA and CO) over 90 min. The core formulation was assembled in the single meander channel (SMC) [30]. We saw no changes in size and PDI. However, changes in the zeta potential of the particles up to the 40 min mark were the reason why formulations were always used after 45 min incubation time.

#### 4.2.2. DLS Measurement

Samples used for dynamic light scattering (DLS) measurements were prepared to contain 1.5 µg siRNA in 60 µL HEPES buffered glucose pH 7.4 (HBG) at 25 ◦C and the corresponding amount of oligomer. Refractive index and viscosity of the solution were calculated using the solvent builder integrated into the software (Zetasizer family software update v7.12). Viscosities and refractive indices (RI) are reported in Figure S6 in the Supplemental Information. RI of all particles was estimated to be 1.45. Scattered light used to determine the hydrodynamic diameter of polyplexes was measured at a 173◦ angle (backscatter) with a flexible attenuator with a Zetasizer Nano ZS ZEN 3600 (Malvern Panalytical Ltd., Malvern, UK) in DTS1070 micro cuvettes (Malvern Panalytical Ltd., Malvern, UK). Samples were measured three times with 12–15 sub runs each. The mean z-average in nm of those three runs is reported with error bars corresponding to the 95% confidence interval of the three runs.

#### 4.2.3. Standardization of the System

The following steps were taken to ensure standardization of the system.


#### 4.2.4. Data Analysis

Data was analyzed with R [43] and RStudio [44]. We always report means with 95% confidence intervals. R code and raw data are made available here: DOI: 10.6084/m9.figshare.13285577.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/2227-9 717/9/1/129/s1: Figure S1: Channel design of the single meander channel, Figure S2: Channel design of the double meander channel, Figure S3: UML class diagram of the control software of the syringe pumps, Figure S4: Flowchart describing the automation process of polyplex formulations, Figure S5: UML class diagram of the fraction collector's control software, Figure S6: Flowchart describing the workflow of the fraction collector, Figure S7: Chemical structures of CO, LPO, LPOE. Table S1: Materials formulation module, Table S2: Materials control module, Table S3: Software control module, Table S4: Materials feeding module, Table S5. Materials collection module, Table S6. Solvents used for DLS measurements. Code 1: Excerpt from the '\_set\_from\_spec\_file()' function, Code 2: Excerpt from the initialization of the logger function, Code 4: 'GlobalPhaseNumber()' class, Code 5: Excerpt from the 'Setup()' class, Code 6: Excerpt from the 'check\_connections()' function, Code 7: Excerpt from the 'rate()' function, Code 8: Excerpt from the 'ramping\_calc()' function, Code 9: Excerpt from the 'overlap\_calc()' function, Code 10: 'main.py' module, Code 11: Excerpt from the 'Initialize()' class, Code 12: Excerpt from the 'Move()' class, Code 13: The 'main.py' module of the fraction collector.

**Author Contributions:** Conceptualization, D.M.L., R.K., U.L., J.O.R., and E.W.; data curation, D.M.L.; formal analysis, D.M.L.; funding acquisition, J.O.R. and E.W.; investigation, D.M.L.; methodology, D.M.L., R.K., U.L., J.O.R., and E.W.; project administration, D.M.L.; resources, D.M.L., R.K., U.L., J.O.R., and E.W.; software, D.M.L.; supervision, E.W.; validation, D.M.L., U.L. and E.W.; visualization, D.M.L.; writing—original draft, D.M.L.; writing—review and editing, D.M.L., R.K., and E.W. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the Deutsche Forschungsgemeinschaft (DFG), 201269156-SFB 1032 (projects B1 Rädler and B4 Wagner), the Munich Center for NanoScience (CeNS), and the Cluster of Excellence Nanosystems Initiative Munich (NIM). Rafał Krzyszto ´n was supported by German Research Foundation (DFG) through the Graduate School of Quantitative Biosciences Munich (QBM) (GSC 1006). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** The data presented in this study are openly available in FigShare at DOI: 10.6084/m9.figshare.13285577 [45].

**Acknowledgments:** We thank Wolfgang Rödl for technical support and we thank the workshop of the LMU for their enormous help in building the collection module. We are grateful to Philipp Klein for synthesis of the CO oligomer. We thank Thomas Unterlinner for his patience and proficiency in answering all our questions about stepper motors.

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

#### **References**

