In this manuscript, we have successfully created an alpha prototype of a novel microfluidics technology that is potentially capable of automated minimally disruptive manipulations of cells and fluids when implanted in vivo. These abilities are also “addressable” meaning that they can be performed independently of each other, at any desired location on a surface of a scaffold in an animal. To that end, this section describes the technology's ability to collect cellular and/or fluidic samples from within the device and transport them to an external assay chip for ex situ analysis (thereby overcoming the need for explantation). The nondisruptive nature of such monitoring could potentially eliminate the reliance on terminal assays; which, in turn, would drastically cheapen the cost of animal testing, ensure data continuity for multi-time point experiments and reduce the number of animals that need to be sacrificed (which is more ethical).

Furthermore, Figs. 4(c)–4(e), show a diagram and photographs of the cross-sectional view of the device, respectively. Specifically, they show that the cylindrical PDMS part of the device (i.e., the various layers and sub-layers from Sec. that have been bonded together) is joined with the scaffold via a small bolt-and-nut. Therefore, there must be a hole in the center of the device through which the bolt is inserted. For example, Fig. 4(d) shows how a 0.5 cm bolt is used to connect the PDMS device and the PLLA scaffold. It should be noted that a flathead bolt is used to support the scaffold from the bottom, to minimize any protrusion into the inner space of the cranium. Furthermore, the PDMS and the PLLA parts of the device in Fig. 4(d) are intentionally separated from each other on the bolt for demonstration purposes. In contrast, Fig. 4(e) shows the two parts of the device fully pressed together with a nut on top of the bolt. This is the final configuration in which it would be implanted into the skull of an animal.

A. Modifying the addressable microfluidic plumbing to encircle the scaffold attachment bolt in a constrained space of the calvarial defect

As mentioned in Sec. , the spatial constraints of the calvarial defect model,55. P. P. Spicer, J. D. Kretlow, S. Young, J. A. Jansen, F. K. Kasper, and A. G. Mikos, Nat. Protoc. 7, 1918–1929 (2012). https://doi.org/10.1038/nprot.2012.113 coupled with its circular shape and with the need to have a bolt in the middle of the device (for connecting it with the biomaterial scaffold below), necessitated changes to our original microfluidics design that consisted of a square addressable ports array in our previous publication.66. A. Tong, Q. L. Pham, V. Shah, A. Naik, P. Abatemarco, and R. Voronov, ACS Biomater. Sci. Eng. 6, 1809–1820 (2020). https://doi.org/10.1021/acsbiomaterials.9b01969 Therefore, in this manuscript, we have modified the addressable plumbing so that it would meet the following requirements: (1) it must fit into a limited space of 12 mm diameter (i.e., the device will be slightly wider than the 8 mm implant and will sit between the rat's skull and the skin covering it); (2) it must curve around the center hole that fits the bolt for attaching the biomaterial scaffold below the device; (3) the various plumbing inlets and outlets must be positioned as far apart from each other as possible to allow space for the numerous tubing insertions [see Fig. 4(a)] in the constrained area of the device. If the latter are placed too close to each other, the needles that penetrate the device can create leakages between the neighboring channels and undesired stresses in the PDMS material.Figure 5(a) shows our updated plumbing design that fits these parameters: the payload/probing plumbing layer (red) is composed of four circularly curved channels, with four addressable ports on each channel; and the control valve layer (blue), which is bonded on top, has four valves (each corresponding to an addressable port) per channel. Also, note that each of the red channels requires an inlet and an outlet (making for a total of eight tubing connections for the layer), but the blue channels only require one inlet/outlet each (i.e., four tubing connections for the Layer). Furthermore, the latter connections have been placed on the alternating sides (i.e., top-bottom-top-bottom) of the device to reduce the spatial constrainment of the needles penetrating the PDMS.Figure 5(b) shows a PDMS prototype of the plumbing design in Fig. 5(a), using food dyes to attain the same color scheme. From this figure, it is apparent that all the channels, the inlets, and the external tubing connection needles fit snugly into the 12 mm diameter circular area of the device. Furthermore, Fig. 6 shows that the in vivo addressable implant is still capable of the same fluid and cell manipulation, as our original ex vivo addressable microfluidics device (compare to Fig. 8 and Video S1 in Ref. 66. A. Tong, Q. L. Pham, V. Shah, A. Naik, P. Abatemarco, and R. Voronov, ACS Biomater. Sci. Eng. 6, 1809–1820 (2020). https://doi.org/10.1021/acsbiomaterials.9b01969).The action of delivering and sampling chemicals within the addressable device is shown in the left and right panes of Fig. 6, respectively. In the former case, a red dye is delivered to the top left corner of the addressable grid of microfluidic ports [see Figs. 6(a) and 6(c)], while in the latter case, the same red dye is withdrawn back via a neighboring port in the second row of the addressable matrix [see Figs. 6(b) and 6(d)]. In the in vivo setting, the picked-up fluids and/or cells would then be sent off to an external sensor for ex situ analysis on the SAAC (as in Fig. 2-right). This eliminates the need for explantation of the scaffold, ensuring continuous monitoring of the biology occurring during bone regeneration within the animal's native tissue environment. Furthermore, this can be done continuously over long periods of time, given that the entire process is automated (and as such, does not require any human involvement).However, the actions of either delivering or withdrawing fluids within the implanted device can result in a pressure imbalance in the hosts head, which can disrupt the experiment by either breaking the biomaterial scaffold that is being analyzed and/or presenting a danger to the animal by upsetting its intracranial equilibrium. Therefore, Sec. presents some preventative measures in our design.

B. Maintaining cranial fluid equilibrium and preventing contamination from the surrounding tissues via sampling chambers

An additional implication of implanting a microfluidics device in the cranium of a living animal that needs to be considered is its potential interaction with the host's body, which is not desired. For example, some envisioned problems are: (1) fluid payloads delivered to the device could leak out into the host's surrounding tissues (potentially upsetting the intracranial fluid pressure in the animal's head and potentially hurting or even killing it); (2) a reverse situation could occur where a mix of the host's bodily fluid/cells from the areas surrounding the implant could contaminate the probings collected from the sampling chambers of the device; and (3) if excess vacuum or pressure are created in the device, they could cause the scaffold to collapse or break (thereby jeopardizing the experiment and endangering the animal's well-being). Therefore, it is desirable to: (a) isolate the microfluidics portion of the device from the host's physiology and (b) maintain a fluid equilibrium inside of its chambers such that no excess pressure or vacuum is created by the performed plumbing manipulations. In other words, if any fluid is flown into the device, an equal amount must be removed from it somewhere else simultaneously (and vice versa).

To take care of these issues, we have included a sampling chamber layer, situated between the PDMS part of the device and the biomaterial scaffold [see Layer #4 in Fig. 3(b), and also see Fig. 7(a)]. It serves two purposes: (1) the thick circular band, along the circumference of the layer, serves as a “wall” that sits on top of the skull bone (attached with fibrin glue) and isolates the device's microfluidic plumbing from the interstitial fluids of the host's body [see purple color in Fig. 7(a)]; and (2) the area inside of the circle is broken up into eight identical “pizza slice”-shaped chambers [see Fig. 7(b)], which effectively divide the surface of the implanted scaffold into separate sampling regions (green color). The reason why this is important is because it: (a) reduces diffusion between the neighboring regions, thereby increasing the spatial accuracy (i.e., “resolution”) of the probing; and (b) isolates two addressable microfluidic ports per each “pizza slice”-shaped chamber [see Fig. 7(c)]. The latter is a fluid pressure balancing measure: essentially, while one port is active (i.e., delivering or sampling fluid), the other port in the chamber is kept open to prevent any pressure or vacuum build-up that could potentially lead to the collapsing of the PDMS device; or to the deformation/cracking of the PLLA scaffold and/or causing brain damage in the mouse, due to the strong pressure/vacuum forces transmitted through the PLLA's pores. For example, when one port is open to a vacuum line to collect a sample, the neighboring port would be open to a fresh media source to quench the negative pressure in the same chamber [see Figs. 7(a) and 7(c)].

Ultimately, as the probings are collected, they are sent off to a SAAC, which is discussed in the section.

C. Storing and analyzing collected in vivo samples on an ex vivo assay chip

As mentioned in Sec. , as the probings are collected within the microfluidic implant, the pressure balance within it will need to be maintained. For this reason, the implant will be connected to a media replacement storage (see Fig. 8-left), which will provide the fluids necessary to substitute the volumes subtracted by the probings. In other words, the microfluidic device will draw the replenishment media from this storage as it is collecting samples from the “pizza slice”-shaped sampling chambers (see Fig. 8-center), to maintain the net amount of fluid in the device constant. Once the probings are obtained, they will be flowed out of the implant and into the ex vivo SAAC (see Fig. 8-right), thereby overcoming the need for sample explantation.When on the chip, the collected probings can be either stored to be subsequently removed by an operator for off-line analysis; or on-chip assays can be performed directly in the storage wells. For the latter option, some potential chemical tests that are appropriate for the calvarial defect bone tissue engineering application are discussed in Sec. , while this part of the manuscript is limited to the design of the chip.The chip's plumbing is basically a microfluidic analog of a conventional well plate (e.g., 96 well plate), with each of the storage compartments being individually addressable (see Fig. 9). It has been previously published17,1917. Y. Gao, J. Tian, J. Wu, W. Cao, B. Zhou, R. Shen, and W. Wen, RSC Adv. 6, 101760–101769 (2016). https://doi.org/10.1039/C6RA17633J19. H.-Y. Wang, N. Bao, and C. Lu, Biosens. Bioelectron. 24, 613–617 (2008). https://doi.org/10.1016/j.bios.2008.06.005 and requires no modifications, aside from selecting an appropriate number of rows and columns for the well matrix: given that the dimensions of the chip are not limited by the in vivo size constraints, it is important to note here that the chip can be basically made to be any size, because as many (or as few) probings as needed can be processed by it at any desired frequency. However, here, we have designed it with four rows of wells to match the number of payload/probing plumbing channel rows in Fig. 5/6, because this reduces the plumbing complexity. Meanwhile, the chip's number of columns remains a free parameter, which would be determined based on the number of the desired timepoints, locations and/or signals that the user wants to measure. This is shown in Fig. 9-right.For example, a single time point measurement at all 16 addressable port locations in the implant would require four columns (i.e., a 4 × 4 matrix of wells), while storing two timepoints would require eight columns (i.e., a 4 × 8 matrix of wells), and so on. Alternatively, if an entire “pizza slice”-shaped sampling chamber is considered to be a single “location” [since the addressable ports work in tandem pairs, as was shown in Fig. 7(c)], then the number of the wells per timepoint would be halved (i.e., 2 chamber rows × 4 chamber columns → a total of 8 storage wells would be required). Conversely, if one wanted to measure multiple biological signals per timepoint at each location, then the number of the total wells from the examples above would be multiplied by, however, many signals are being detected. For example, in Sec. , we propose to measure four different biological signals per a sampling chamber. This would translate to a chip with a total of 32 wells. So, for demonstration purposes, we have designed a SAAC with a 4 × 8 addressable grid of storage wells [see Fig. 9(a)]. Fig. 9(b) shows a working prototype of the envisioned array of the micro-wells.In Sec. , we discuss how the external hardware requirement (which is ultimately related to system price and experiment setup difficulty) scales with the increasing SAAC size as well as how to reduce that requirement at a “cost” of an additional microfluidic plumbing complexity.

D. Reducing the external equipment requirement via a combinatorial multiplexer chip

Given that the addressable microfluidic plumbing requires a considerable amount of external hardware equipment (e.g., pneumatic solenoid valves, pressure regulators, gauges, etc.)2020. K. Brower, R. Puccinelli, C. J. Markin, T. C. Shimko, S. A. Longwell, B. Cruz, R. Gomez-Sjoberg, and P. M. Fordyce, HardwareX 3, 117–134 (2018). https://doi.org/10.1016/j.ohx.2017.10.001 to run, managing two such chips (i.e., the addressable implant and the SAAC) can become costly and complicated. Furthermore, if many locations, timepoints and/or biological signals need to be measured between the SAAC replacements, then the growing size of the latter will increase that requirement even further. To put it into perspective, the 4 × 4 implantable device described in Sec. uses a manifold of eight solenoids (costs ∼ $600)— the costliest components in the system—to actuate its four control valve channels and four flow channels.Moreover, the cost accumulates further for the hardware needed to run the SAAC: although its rows can be run by the same solenoids as those actuating the implant's payload/probing plumbing channels (because they are basically shared between the two chips), its columns do require extra solenoids to operate. That is, for each additional eight well columns, the cost goes up another $600. So, for a typical 3-day experiment, with a 4-hour probing frequency (i.e., 6 times per day), the number of the columns in the SAAC that detects four biological signals per implant (as in Fig. 9) would have to be increased from 8 to 3 × 6 × 8 = 144. That is, the solenoid cost of running a SAAC of this size would go up by a factor of 18 (which translates to an equipment cost >$10 000 and to an unnecessary tubing/wiring complexity).Luckily, a Combinatorial Multiplexer concept in microfluidics offers a way to reduce the external hardware requirements (albeit at a penalty of increasing the complexity of the on-chip's plumbing).2121. J. Melin and S. R. Quake, Annu. Rev. Biophys. Biomol. Struct. 36, 213–231 (2007). https://doi.org/10.1146/annurev.biophys.36.040306.132646 In particular, due to its N!/(N/2)!22. O. J. Wouters, M. McKee, and J. Luyten, Jama 323, 844–853 (2020). https://doi.org/10.1001/jama.2020.1166 scaling, up to 252 lines can be actuated with just 10 control lines (i.e., with only 10 solenoids). This means that by simply adding the multiplexer to our flow system, the equipment cost of running the 144 column SAAC in the example above can be reduced by a factor of 10. Thus, it is worthwhile to include the combinatorial multiplexer into our design, to make the system costs and setup be more manageable.To that end, Fig. 10 shows our integrated system, with all three of the microfluidics chips connected: the addressable implant, the SAAC and the combinatorial multiplexer. For the latter, we have opted to use the 8 control lines version [multiplexer rows in Fig. 10(a) corresponding to yellow tubing in Fig. 10(b)], since it can actuate up to a total 70 flow lines [multiplexer columns prior to Junctions in Fig. 10(a)], while our system described in Fig. 7(b) and in Fig. 9 requires less than that to run. In fact, multiplexing just 40 flow lines is sufficient for operating the addressable implant and the SAAC described in Secs. and , respectively. Therefore, the remaining 30 unused flow lines of the multiplexer are not fabricated (and are, thus, not shown in Fig. 10).As far as the 40 flow lines that are used, they are merged into pairs [see the 20 numbered Junctions in Fig. 10(a)]. For example, the red Junctions #1–4 are dedicated for multiplexing the inputs of the payload/probing plumbing on the addressable implant. Each of the junctions allows the system to alternate between two different fluid and/or pressure source inputs, while combining them into a single multiplexed output. For example, one of the two lines belonging to the Junction #1 is shown to be active (bright red), while the other is not (dim red). The multiplexer can alternate between them to flow fluids containing different drugs, media, etc., through the payload channel resulting from Junction #1. Likewise, it can also alternate between pressurized and non-pressurized blue lines that belong to control valve Junctions #5–8, to either inflate or deflate the blue O-shaped valves on the addressable implant. Similarly, the green Junctions #9–16 in Fig. 10 are used to either inflate or deflate the addressable O-shaped valves on the SAAC. Last, the red Junctions #17–20 in the same figure are used in tandem with the first four junctions to flow the payloads/samples through the four rows in the addressable implant and SAACs. However, in the case of the former junctions, the multiplexer alternates between non-pressurized and vacuum lines. The negative pressure, in turn, helps to pull the payloads and/or samples through the two chips.An example of the multiplexer's operation is also shown in Fig. 10(a): by actuating its control lines #4 and 6–8, the multiplexer allows for flow to only go through the right arm of the red multiplexed Junction #1, while the rest of the junctions are fully blocked. Similarly, it can activate or disable the inflatable O-shaped valves on the addressable implant and the SAAC to achieve the addressable manipulation at the targeted microfluidics ports or wells, respectively. To that end, Fig. 10(b) shows an actual system prototype using the multiplexer to fill the top row of the SAAC with a red dye.Next, Table I shows how the choice of a different-sized multiplexer affects the total number of time points and/or biological signals (see the last column in Table I) that can be stored/analyzed on the SAAC. For these calculations, we assume that the implantable device's matrix of addressable ports remains at the fixed size of 4 × 4 presented in this manuscript, which takes eight payload/sampling lines [red arms of the Junctions #1–4 in Fig. 10(a)], eight control valve lines [i.e., blue arms of the Junctions #5–8 in Fig. 10(a)] to operate, and eight vacuum lines to operate [i.e., red arms of the Junctions #17–20 in Fig. 10(a)].

TABLE I. Scaling capacity of the multiplexer plumbing assuming a 4 × 4 addressable implant. Note: The design for the addressable implant optimized for a fixed number of 22 multiplexed channels will be used for the implant payload channels, the implant control channels, and the sample chip vacuum channels. Also, note that the number of control lines can only be an even number for a combinatorial multiplexer.

No. of Control LinesTotal possible multiplexed channelsNo. of remaining possible multiplexed channelsRemaining possible multiplexed junctions (n)Total possible SAAC microwells (4 × n array)Total possible time points and/or signals8704623925 (11)1025222811445628129249004501 800112143 4322 41012056 8204261612 87012 848642425 6961606To explain how Table I is calculated, we provide an example for its first row. Here, we are examining an 8-control line version of the combinatorial multiplexer. Its theoretical maximum is the actuation of 70 flow lines, as per the formula: N!/(N/2)!,22. O. J. Wouters, M. McKee, and J. Luyten, Jama 323, 844–853 (2020). https://doi.org/10.1001/jama.2020.1166 where N = 8. Out of these 70 flow lines, 24 must be dedicated to operating the addressable implant chip. So:

70 (total possible multiplexed flow lines) – 8 (implant payload channels) – 8 (implant control channels) – 8 (sample chip vacuum channels) = 46 (remaining possible multiplexed channels).

Next, we must account for the fact that takes two flow lines to operate a single addressable microwell column on the SAAC, because pressures and non-pressurized lines must be merged together into a single control valve channel (see the formation of green Junctions #9–16 in Fig. 10). Therefore, the number of total possible addressable microwell columns on the SAAC is equal to half of the remaining possible multiplexed channels:

46 (remaining possible multiplexed channels)/2 (multiplexed channels per junction) = 23 (remaining possible multiplexed junctions) → 23 (total possible microwell columns).

Next, we must translate the number of columns on the SAAC translates to the number of microwells. Staying with the assumption that the addressable implant chip has four payload/sample flow channel rows, this means that the number of SAAC wells is quadruple of its columns (because the row channels are shared between the two chips):

4 (implant channel rows) * 23 (total possible microwell columns) = 92 (total possible SAAC microwells).

Last, assuming that the user wants to store/analyze every location in the 4 × 4 addressable matrix, each time point or biological signal measured would have to consume 16 (or 8, if each sampling chamber is treated as a “location”) SAAC microwells:

92 (total possible SAAC Microwells)/16 (implant probing locations) = 5.75 (time points and/or signals).

Given that the final answer is not a whole number, we round it down to the nearest integer to get that 5 (or 11, if each sampling chamber is treated as a “location”) total timepoints and/or biological signals that could be stored/analyzed on the SAAC.