Selectively cross-linked hydrogel-based cocktail drug delivery micro-chip for colon cancer combinatorial drug screening using AI-CSR platform for precision medicine

Abstract

Cancer, ranked as the second leading cause of global mortality with a prevalence of 1 in 6 deaths, necessitates innovative approaches for effective treatment. Combinatorial drug therapy for cancer treatment targets several key pathways simultaneously and potentially enhances anti-cancer efficacy without intolerable side effects. However, it demands precise and accurate control of drug-dose combinations and their release. In this study, we demonstrated a selectively cross-linked hydrogel-based platform that can quantify and release drugs simultaneously for in-parallel cocktail drug screening. PDMS was used as the flow channel substrate and the poly (ethylene glycol) diacrylate (PEGDA) hydrogel array was formed by UV exposure using the photomask. Employing our platform, cocktails of anticancer drugs are precisely loaded and simultaneously released in-parallel into HCT-116 colon cancer cells, facilitating combinatorial drug screening. The integration of an artificial intelligence-based complex system response (AI-CSR) platform successfully identifies optimal drug-dose combinations from a pool of ten approved drugs. Notably, our cocktail drug chip demonstrates exceptional efficiency, screening 155 drug-dose combinations within a brief two and a half hours, a marked improvement over traditional methods. Furthermore, the device exhibits low drug consumption, requiring a mere 1 μL per patch of chip. Thus, our developed PDMS drug-loaded hydrogel platform presents a novel and expedited approach to quantifying drug concentrations, promising to be a faster, efficient and more precise approach for conducting cocktail drug screening experiments.

---

1. Introduction

Cancer treatment remains a formidable challenge in the medical field, with the effectiveness of chemotherapy limited to approximately one in four patients, or even fewer.¹ Specifically, Colorectal carcinoma (CRC) is the third most common cancer worldwide. The lifetime risk of developing CRC is ≈1 in 10 (13.9%).^2,3 The chemotherapy response rate of colorectal cancer is 25–35% in CRC patients.⁴ Recognizing the need for more targeted and potent therapeutic approaches, combinational drug therapy, often referred to as cocktail drug therapy, has emerged as a crucial strategy in the battle against cancer.⁵ However, the conventional methods of drug screening are marred by their time-consuming, painful nature and side effects, prompting a concerted effort within the scientific community to develop high-throughput and user-friendly platforms for more efficient drug testing.^6–9 Therefore, the need of the hour is to develop alternate, high throughput, cost effective and user friendly laboratory drug screening methods for highly efficient drug testing.

Microfluidic-based platforms have emerged as highly efficient chip-based assays for cell line tests, offering inherent advantages such as minimal reagent consumption, precise control, and scalable high throughput. These platforms provide robust analytical tools for investigating intricate biological processes at the cellular level, thus contributing to advancements in the field.^10–15 The efficacy of anticancer drugs depends on the cancer type and its specific location, with each drug targeting distinct sites. For instance, antimetabolites like 5-fluorouracil and capecitabine, the plant alkaloid irinotecan acting as a topoisomerase inhibitor, chemo-protectant folinic acid inducing folic acid deficiency leading to cell death, alkylating agent oxaliplatin inhibiting cell cycles, genetically modified drug bevacizumab regulating angiogenesis, and protein-based drug cetuximab targeting endothelial growth factor receptor proteins are among the drugs approved by the US drug agency for chemotherapy.^16–18 Despite their individual merits, none of these drugs alone can comprehensively and efficiently eliminate cancer.

Considering the limitation of single-drug approaches, a combination of multiple drugs, akin to a cocktail, presents a potential strategy for achieving enhanced treatment outcomes.¹⁹ The synergistic effect of drug combinations not only allows for lower individual drug dosages but also augments the overall effectiveness against various cancer targets. This approach concurrently diminishes the drug resistance capabilities of cancer cells.²⁰ Combinatory drugs have demonstrated higher efficacy and reduced individual drug dosage in treating various diseases, including cancer.²¹ However, it is crucial to note that the dosage of drugs in combination regimens critically influences both efficacy and toxicity.

The concept of cocktail drug treatment refers to a therapeutic method involving the combination of two or more drugs. This innovative approach was initially introduced by Dr. Ho Dayi in 1996 for the treatment of HIV and has since expanded its application to various diseases, including cancers and the COVID-19 coronavirus, through highly active antiretroviral therapy (HAART). The distinctive advantage of cocktail therapy lies in its ability to target different pathways, thereby inducing virus or cell apoptosis. For instance, in colorectal cancer drug chemotherapy, a diverse array of drugs is quantitatively mixed, including targeted drugs and traditional chemotherapy cocktails. Traditional experimental methods, though foundational, are criticized for their time-consuming nature owing to high volume ratios and the inefficient use of cells and drugs. To address these limitations, various research teams have pioneered solutions that enable the efficient preparation of different drug concentrations or the creation of platforms with drug mixed concentration gradients to streamline drug screening experiments.

Jiao Zhai's team has engineered a wafer based on electro wetting over dielectric (EWOD) to control liquids for drug distribution and mixing.⁸ Simultaneously, they introduced a novel technique relying on the droplet ejection phenomenon triggered by a high-voltage pulse wave driving signal, aptly named a drug dispenser. This technology facilitates the rapid dispensing of drug concentrations across three to four orders of magnitude, utilizing fewer cancer cells and less drug consumption.

Hu's team developed a high-throughput drug testing chip that incorporates 3D cells.⁹ This innovative chip involves incorporating cells into a gelatin methacryloyl (GelMA) light-sensitive hydrogel, forming a 3D hydrogel on the wafer using partial exposure of the mask combined with acoustic force assembly technology. The chip's unique design features two inlets and a dendritic micro-channel network, allowing for the creation of drug concentration gradients and the simultaneous testing of the response of 100 groups of 3D cells to drugs. This approach significantly expedites the drug screening process and represents a noteworthy leap in efficiency.

Zhang's team has developed a chip (μ-TGSM) that can mix three liquids simultaneously, featuring 3 sample inlets and 12 outlets with 18 double-helix micromixers.²² This design enables the rapid establishment of three stable, accurate, and controllable drug concentration gradients. This innovative approach facilitates the study of the individual effects and interactions of two drugs on a single cell under three sets of gradients.

Chung's team has introduced a high-throughput drug test method, utilizing a unique structure on the substrate to allow hydrogel to fill in the hole. Using PEGDA to coat the drug, the team aimed at the cell culture chips, merging them to allow the drug in the hydrogel to diffuse for 24 hours, subsequently estimating cell viability. In earlier research, the team led by the authors leveraged the concept of hydrophilic and hydrophobic differences to design a drug screening wafer using liquid bead forming technology.¹⁴ This wafer could accurately quantify drug concentrations through simple steps, processing five different drugs simultaneously and saving considerable time in drug screening experiments.

As the landscape of drug discovery and testing evolves, recent advancements in technology have paved the way for more sophisticated and efficient methodologies. One such promising development is the integration of artificial intelligence (AI) into drug screening processes, particularly in the form of the AI-CSR platform.^23–27 With an AI based neural networks approach, the drug-dose inputs are correlated with the phenotypic outputs with a complex system response (CSR) which can be represented by a polynomial type CSR function.²⁸ With a few calibration tests to determine the coefficients of the quadratic algebraic equation governing CSR, CSR dictates the composition and the ratio of a globally optimized drug combination from a very large search space in in vitro tests.²⁹ This innovative approach harnesses the power of AI algorithms to analyse complex phenotypic responses of cells to various drug combinations, enabling researchers to determine the optimum drug dose combinations with unprecedented precision. Traditional drug screening methods often rely on a trial-and-error approach, testing a limited set of drug combinations with predefined concentrations. In contrast, the AI-CSR platform considers a multitude of factors, including genetic variations, signalling pathways, and cellular microenvironments, to generate comprehensive insights into how cells respond to different drug doses.^30–32 Moreover, the AI-CSR platform has the potential to uncover synergistic interactions between drugs that may not be evident through traditional screening methods. By identifying combinations that elicit a more robust and targeted cellular response, researchers can tailor treatment regimens to maximize therapeutic benefits while minimizing adverse effects. This precision in drug combination optimization holds great promise for personalized medicine, where treatment strategies can be tailored to individual patient profiles for enhanced efficacy and reduced toxicity.^24,33

This study introduces a novel cocktail drug delivery chip employing UV cross-linked poly (ethylene glycol) diacrylate (PEGDA) hydrogel arrays on a PDMS substrate, enabling precise release of ten drugs at three different doses for in-parallel testing against colon cancer cells (Fig. 1). The platform is designed for swift quantification of drug concentration in cocktail drug screening experiments. Innovative cocktail drug selection involves exposing specific hydrogel lengths to different drug concentrations through a precisely designed photomask. Material analysis focuses on PEGDA hydrogel and photo initiator LAP concentrations, validated through cytotoxicity tests, demonstrating minimal impact on cells with a UV curing time of 20 seconds. The developed cocktail drug chip for ten drugs exhibits remarkable efficiency, completing the screening of 155 drug-dose combinations based on the AI-CSR platform in just two and a half hours. Additionally, the device boasts low drug consumption, requiring only 1 μL per chip patch. Interestingly, our findings indicate a slightly higher cell survival rate in the traditional method compared to our chip. In summary, our PDMS drug-loaded hydrogel platform offers a novel and expedited approach to quantifying drug concentrations, providing efficiency gains and addressing specific challenges encountered in traditional drug screening methods.

Fig. 1 Schematic illustration of the selectively cross-linked hydrogel-based cocktail drug delivery chip based on AI-CSR platform. (A(i)–(vi)) Process flow of the cocktail drugs delivery chip (B) process flow of the AI-CSR platform for optimization of drug-dose combinations.

2. Materials and methods

2.1. Materials

Polyethylene glycol diacrylate (PEGDA) polymer (Fisher Scientific, Model B2200R-1, Pittsburgh, PA), 1_ PBS buffer, pH 7.4 (gibco®, by life technologies). Importantly, all the chemicals used in the UVGELDRUGCHIP preparation were dissolved and prepared in 1X PBS buffer solution. Human colorectal cancer cell lines (HCT116) was obtained from the American Type Culture Collection (ATCC, Manassas, VA, USA). Cells were cultured in McCoy's 5A medium (Dow Corning Co., Midland, MI, USA) for HCT116, containing 10% heat-inactivated fetal bovine serum (Dow Corning Co., Midland, MI, USA), 100 units (U) mL⁻¹ penicillin/streptomycin (Dow Corning Co., Midland, MI, USA), in a humidified atmosphere of 5% CO₂ at 37 °C. Ten drugs are selected from existed commonly clinical use including five chemotherapy drugs and five target drugs. The chemotherapy drugs were 5-FU (BD Biosciences, San Jose, CA, USA), oxaliplatin (Sigma-Aldrich), gemcitabine (Selleckchem chemicals LLC, TX, USA), irinotecan (Selleckchem chemicals LLC, TX, USA), folinic acid (Sigma-Aldrich) and the target drugs were regorafenib (Selleckchem chemicals LLC, TX, USA), lenvatinib (Selleckchem chemicals LLC, TX, USA), bevacizumab (Selleckchem chemicals LLC, TX, USA), cetuximab (Selleckchem chemicals LLC, TX, USA), panitumumab (Selleckchem chemicals LLC, TX, USA).

2.2. Chip design

In this study, parallel micro polydimethylsiloxane (PDMS) channels were meticulously crafted through a moulding process, utilizing a computer numerical control (CNC) milled polymethyl methacrylate (PMMA) mould. The subsequent step involved the use of reactive ion etching (RIE) oxygen plasma treatment to render the PDMS channel surfaces hydrophilic (Fig. 1A(i)), ensuring seamless passage for the hydrogel liquid (Fig. 1A(ii)). The cocktail drug selection process was ingeniously executed by exposing specific lengths of the hydrogel to varying drug concentrations using a precisely designed photomask (Fig. 1A(iii)). Since the resolution required is greater than 100 μm, a laser printing technique was employed to fabricate the photomask, which offers significant enhancements in cost, flexibility, speed and customization. Post-exposure, the hydrogel underwent a curing process facilitated by 365 nm UV exposure. Following this, the non-exposed hydrogel, which remained in a liquid state, was carefully removed using tissue papers. This strategic step was crucial for retaining a precise and quantitative drug-loaded hydrogel (Fig. 1A(iv)). To facilitate further experimentation, the chip, now carrying the drug-packed hydrogel, underwent segmentation. The chip was meticulously cut into individual pieces (Fig. 1A(v)). Subsequently, these cut pieces were strategically placed into a 96-well cell culture plate (Fig. 1A(vi)), enabling the controlled release of the various drug combinations into individual wells simultaneously. This final step enhanced the efficiency and accuracy of drug screening experiments, showcasing the versatility and applicability of the developed microfluidic platform.

The configured microfluidic channel possesses a width of 700 μm, a wall width of 150 μm, and a height of 250 μm, as illustrated in Fig. 2. Each patch within the channel has a holding capacity of 1 μL for a single drug (Fig. 2). Notably, at the inlet area, the wall width is intentionally designed wider to prevent unwanted liquid mixing with adjacent channels. To enhance precision during the masking process, barriers have been strategically incorporated within the channels, effectively preventing dislocation of the cured hydrogel. In the course of drug screening experiments, the chips undergo segmentation into smaller pieces, each with a width of 6500 μm. These segmented chips are simultaneously placed in a cell culture dish. To maintain a systematic and organized approach, preventing confusion regarding the sequence of chips containing distinct drug combinations, a set of symbols has been devised. These symbols serve as a numerical identification system for the chips, ensuring a clear and unambiguous correlation between the chips and their respective drug combinations throughout the experimental process. The detailed process of chip fabrication has been detailed in ESI† section S1.

Fig. 2 Design of the cocktail drugs delivery chip.

2.3. Cytotoxicity test of hydrogel chips

Human colorectal cancer cells, specifically HCT-116 were cultured in 96-well plates to assess the cytotoxicity of the material under investigation. Utilizing the designed chip, 10 μL of distinct concentrations of hydrogel and photo initiator were loaded, respectively. After subjecting the loaded chip to a 20-second UV light curing process, it was placed into a cell culture dish containing the relevant cells. The entire setup was then incubated in a controlled environment at 37 °C with 5% CO₂ for 24 hours, 48 hours, and 72 hours. Subsequently, the number of cells was quantified at each designated time point, providing insights into the material's cytotoxic effects over the specified duration.

2.4. Estimation of half maximal effective concentration (EC50) from ten drugs on HCT116 cell line

To determine a suitable cell seeding number for each cell line, different (300, 500, 1000, and 3000) numbers of cells were seeded in 96-well plates. In the pursuit of establishing the EC50 values for ten drugs on HCT-116, a gradient concentration for each drug was meticulously prepared. Cell viability was determined using a real-time viability assay based on engineered luciferase (RealTime-Glo Assay Reagent, Promega, Madison, USA). Luminescence intensity was measured at 4, 8, 16, 24, 36, 48, 60, and 72 h after adding the reagent to obtain the cell growth curve. Small molecule drugs including 5FU, oxaliplatin, gemcitabine, irinotecan, folinic acid, regorafenib, and lenvatinib were formulated at concentrations of 100, 50, 25, 10, 5, 1, 0.1, 0.01, and 0.001 μM respectively. On the other hand, macromolecular drugs such as bevacizumab, cetuximab, and panitumumab were prepared at concentrations of 1, 0.5, 0.25, 0.1, 0.05, 0.01, 0.001, 0.0001, and 0.00001 μM respectively. Cell viability assessments were conducted using a colorimetric assay based on tetrazolium salt cleavage, specifically the CCK-8 assay reagent. Following the addition of drugs, cell viability measurements were obtained after one and four days of cell seeding. After a 2-hour incubation with CCK-8, absorbance at 450 nm was quantified utilizing the GloMax® Explorer Multimode Microplate Reader (Promega, Madison, USA). Relative cell numbers were determined by comparing them to control cells. The EC50 values, crucial indicators of drug efficacy, were calculated employing Graphpad Prism software. This comprehensive approach serves as the foundation for building an optimized drug combination model.

2.5. AI-CSR based drug combination screening and analysis

The design of experiment used for the platform was the orthogonal array composite design, which combined both fractional factorial and orthogonal array designs, as previously described by Xu et al.³⁴ This experimental design allowed for a minimum number of drug combinations to be tested for factor screening and model fitting:^23,28The experimental data points were fit into a second-order quadratic regression model, as described in different previous papers^24,32,35 and second-order polynomial was:²⁵

γ = β₀ + β₁x₁ + ⋯ + β_nx_n + β₁₂x₁x_n + ⋯ + β_mnx_mx_n + β₁₂x₁² + ⋯ + β_nnx_n²

where y represents the desired output, x_n is the nth drug dosage, β₀ is the intercept term, β_n is the single-drug coefficient of the nth drug, β_mn is the interaction coefficient between the mth and nth drugs and β_nn is the quadratic coefficient for the nth drug. After determining the EC50 values of each drug in the cell lines, the effective concentration at which the drug produces 10% of the maximum biological response observed (EC10) and effective concentration at which the drug produces 10% of the maximum biological response observed (EC20) values of each drug were calculated in ten drug search spaces. EC0 is the effective concentration at which there is no observable biological response, i.e. zero concentration of drug is used for testing. Therefore, the drugs were mentioned with respective concentrations. For example, the EC0, EC10 and EC20 of cetuximab were annotated as C0, C1 and C2 respectively. The results of the AI-CSR analysis were correlated into a second-order quadratic series, as previously described by Rashid et al.²⁴ For in vitro experiments, the relative cell viability was used as an experimental data point for fitting the regression mode.

3. Results and discussion

3.1. O₂ plasma treatment of the channel surface

Upon the fabrication of the wafer substrate, the inherent hydrophobic nature of the PDMS material becomes apparent, as depicted in Fig. 3A, where the water contact angle measures approximately 99°. To enhance the fluidity of liquid flow through the channels, we subjected the chip to oxygen plasma treatment (O₂ plasma). The treatment parameters were set at O₂ flow rate of 10sccm, RF power of 100 W, and a duration of 180 seconds. Immediately post-treatment, the surface undergoes a remarkable transition, becoming highly hydrophilic with a water contact angle of less than 10°, and after a brief interval of five minutes, the water contact angle stabilizes at around 26° (as shown in Fig. 3B), indicative of a sustained hydrophilic state. This phenomenon persists for approximately 30 minutes before gradually reverting to its original hydrophobic state.

Fig. 3 Water contact angle of the PDMS surface (A) before and (B) after O₂ plasma treatment (C) evaluation of fluid flow using colored water testing through the microchannels.

To validate the effectiveness of the oxygen plasma treatment, we conducted tests using different coloured water solutions within the channels. Fig. 3C visually confirms that, post-treatment, the liquid flows seamlessly through the channels, ensuring the absence of cross-contamination between the ten channels. This demonstrates the successful modification of the PDMS material's hydrophobic nature to facilitate improved fluid dynamics within the microfluidic channels. Subsequently, a selective cross-linking assessment of the hydrogel was conducted utilizing the photomask and UV light exposure as given in ESI† Fig. S1. The outcome revealed that the cross-linked hydrogel remained appropriately positioned following the removal of the mask and elimination of the uncured hydrogel liquid.

3.2. UV curing time optimization for different concentrations of PEGDA and photo initiator

Among the anticipated ten drugs for testing, five are chemical drugs with molecular weights spanning 130–558 g mol⁻¹, two are small-molecule drugs with molecular weights of 426 and 482 g mol⁻¹, and the remaining three are macromolecular protein drugs with molecular weights around 140–150 kDa. To ensure uniform drug release efficiencies despite varying molecular weights, adjustments were made using different concentrations of hydrogels and photo initiators. With the aim of operating all ten drugs simultaneously in subsequent experiments, it was imperative to standardize the UV curing time for all hydrogel and drug mixtures. The UV light machine employed in the experiment had a wavelength of 405 nm and a light intensity of 1600 mJ cm⁻². The results provided in Table 1, revealed that even with the addition of 0.35% LAP to 5% of the hydrogel, complete gelation could not be achieved. Consequently, the minimum usable hydrogel concentration was set at 6%. For subsequent experiments, the parameters selected were 6% PEGDA + 0.3% LAP, 7.5% PEGDA + 0.25% LAP and 10% PEGDA + 0.25% LAP, illuminated for 20 seconds.

Table 1 UV exposure time requirements for various concentrations of PEGDA and LAP. The table outlines the necessary UV exposure times corresponding to different concentrations of PEGDA and LAP. The subsequent experiment will adopt a fixed exposure time of 20 seconds, with the minimum PEGDA concentration set at 6%

PEGDA concentration	5%		6%		7.5%		10%
LAP concentration	0.3%	0.35%	0.25%	0.3%	0.2%	0.25%	0.2%	0.25%
Time needed for gelation	—	—	25 s	20s	25 s	20s	22 s	18 s

---

The choice to refrain from using a higher concentration of LAP was deliberate, aiming to avoid a faster gel formation rate due to the manual operation required for switching the optical machine. It is important to note that higher concentrations of photo initiator can lead to increased cell damage, emphasizing the need for a balanced approach in the experimental parameters.

3.3. Drug releasing test of different concentrations of hydrogel

Initially, standard solutions for ten drugs were meticulously prepared, and their absorption values were measured using a UV/Vis spectrometer to establish a quantitative standard curve. Subsequently, the release efficiency of drugs with diverse molecular weights was examined, employing different hydrogel and photo initiator concentrations while maintaining a consistent illumination time of 20 seconds at the 24-hour mark.

The drug was first formulated into a 200 μM hydrogel solution, effectively solubilized and loaded successfully into the hydrogel to achieve a standard drug screening concentration in the final solution. The drug was then loaded into the hydrogel and 10 μL of the resulting hydrogel-drug solution loaded onto a wafer and cured for 20 seconds. (Taking 5-FU drug as an example, the molecular weight is 130.077 g mol⁻¹. We take 2.61 g of 5-FU powder and add 100 ml of deionized water to prepare a 200 μM 5-FU solution). The cured sample was then released in a 200 μL PBS solution and kept in a 37 °C environment for 24 hours. Absorbance measurements were taken to estimate the release efficiency, with the detailed results presented in Table 2. Initially, it was evident that the release efficiency of these drugs after 24 hours' ranges from approximately 74% to 86%. Despite reaching 67.2% (time constant) within the initial 150 minutes, the decision to extend the release duration to 24 hours was made to minimize drug wastage.

Table 2 Release rate of ten drugs with different molecular weights. These release rates were taken into consideration for subsequent experiments

Chemo drugs	5-FU	Oxaliplatin	Gemcitabine	Irinotecan	Folinic acid
Molecular weight	130.77 g mol^−1	397.29 g mol^−1	263.19 g mol^−1	558.64 g mol^−1	473.44 g mol^−1
PEGDA concentration	10%	7.5%	10%	7.5%	7.5%
LAP concentration	0.25%	0.25%	0.25%	0.25%	0.25%
Release rate (24 h)	86.7%	85.2%	84.2%	78.6%	80.3%

---

Target drugs	Regorafenib	Lenvatinib	Bevacizumab	Cetuximab	Panitumumab
Molecular weight	482.8 g mol^−1	426.9 g mol^−1	149 kDa	146 kDa	147 kDa
PEGDA concentration	7.5%	7.5%	6%	6%	6%
LAP concentration	0.25%	0.25%	0.3%	0.3%	0.3%
Release rate (24 h)	80.1%	82.7%	75.5%	74.23%	76.0%

---

3.4. Cytotoxicity assessment of PEGDA and photo initiator LAP

To evaluate the cytotoxicity of the material, HCT-116 colorectal cancer cells were co-cultured with the substance in 96-well plates using RealTime-Glo™ for real-time cytotoxicity measurement. The three groups selected from previous experiments were examined: 6% PEGDA + 0.3% LAP, 7.5% PEGDA + 0.25% LAP, and 10% PEGDA + 0.25% LAP. Additionally, a comparison was made between the hydrogel solution cured with 20 seconds of UV light and the one without UV light curing.

As illustrated in Fig. 4, the cured hydrogel exhibited minimal impact on cell growth after 72 hours of co-culturing. Contrastingly, the addition of uncured PEGDA and photo initiator LAP hindered cell growth, because photo initiator is not fully polymerized with in un-cured PEGDAI like in UV cured PEGDA. So, free radicles from un-polymerized photo initiator in un-cured PEGDA may directly deteriorate cell viability. Reactive monomers formed during polymerization can suppress formation of free radicles, which leads to higher cell viability. These results underscore that 6% PEGDA + 0.3% LAP, 7.5% PEGDA + 0.25% LAP, and 10% PEGDA + 0.25% LAP, when cured by UV for 20 seconds, do not inhibit cell growth, indicating their biocompatibility and suitability for further experimentation.

Fig. 4 Cell viability comparison between cured and uncured hydrogel. Addition of cured PEGDA showed no significant impact on cell growth.

3.5. Determination of EC50 values of HCT116 for drugs

Prior to determining the EC50 values for each drug, the time-dependent growth of the cell lines was assessed by seeding varying cell numbers per well over a 72-hour period. Specifically, based on the growth curve characteristics, an initial cell concentration of 1000 cells per well was employed for HCT116 (ESI† Fig. S2). Building upon the earlier assessment of drug release efficiency, subsequent experiments will consider the released drug concentrations from the hydrogel in conjunction with the water gel preparation. Taking 5-FU as an example, if the original water gel solution was prepared with a concentration of 100 μM, and considering the 24-hour drug release rate in the hydrogel is 86.7%, the drug concentration in the hydrogel solution would be adjusted to 115.4 μM to attain the desired concentration post-release. In the ensuing experiments, cells co-cultured with an unloaded hydrogel chip were utilized as the control group, mitigating any potential influence from the hydrogel and the chip.

Before embarking on the drug combination experiment, it is imperative to establish the EC50 value for the cell line concerning different drugs released from the hydrogel. We went up to a resolution of 1 nM for small molecular drugs such as 5-FU (U), oxaliplatin (O), gemcitabine (G), irinotecan (I), folinic acid (F), regorafenib (R), and lenvatinib (L) and a resolution of 0.01 nM for macromolecular drugs such as bevacizumab (B), cetuximab (C), and panitumumab (P). The cell viability of HCT116 across various concentrations of 10 drugs are illustrated in Fig. 5, while the EC50 values for HCT116 for these drugs are provided in ESI† Table S1. A comparative analysis with the results of traditional drug screening experiments published by our group¹³ reveals a similar trend, indicating that the drug retains its efficacy even after being combined with hydrogel and photo initiator under UV light.

Fig. 5 Drug dose–response curves of ten drugs at different concentrations for HCT116 where drugs were released from hydrogel and provided a side-by-side comparison between EC50 curves of the drugs directly added to the well plate (from our group's previously published paper).¹³

3.6. AI-CSR-based drug combination screening on HCT-116

Building upon the establishment of EC50 values and releasing rate data for all 10 drugs, we proceeded to test the 155 drug combinations provided by the AI-CSR against HCT116 cells. The drug combinatorial arrays were derived from the orthogonal array composite design as previously outlined.¹⁹ This approach ensures the utilization of the minimum number of combinations necessary to comprehensively explore the designated search space. In the context of cell line experiments, 155 combinations were systematically tested within a ten-drug search space, spanning three different concentrations (EC0, EC10, and EC20). Table S2† showed different concentrations of the ten drugs considered for the cocktail drug screening. Each code (0, 1 and 2) on the top of the tables corresponds to the different levels of concentration of the all the drugs. Herein, code 0, code 1, code 2 represents the EC0, EC10 and EC20 values of the individual drugs. In vitro experiments employed the relative viability of cells as a crucial data point for fitting. CCK-8 was employed as the reagent to obtain the cell viability for each group. AI-CSR, based on this data, generates a ranked list encompassing all possible permutations, ranging from top efficacy to the least effective.

Twenty experimental runs with different permutation combinations (from the 155 drug combinations generated using AI-CSR) of cocktail drug concentrations were conducted to test our hydrogel-based cocktail drug chip on HCT116 cells. The values represented the cell viability for each cocktail testing result and has been listed in the ascending order of their cell viabilities in Table 3. At the same time, we also compared the cell viability obtained by a conventional technique and our technique (Table 3). The readout (conventional) represents the viability of cells treated with drugs alone in conventional micro-titer plates via a direct drug pipetting process and served as reference data. Our results are very much analogous to those obtained using the conventional method and even the cell viability of most combinations using the hydrogel chip are slightly lower than that of the traditional method. A t-test statistical analysis was performed using MATLAB to compare cell viabilities between two methods and t-statistic value obtained was −1.039 and P-value 0.306. These results indicate that there is no statistically significant difference in cell viability between the hydrogel chip method and the conventional method (since the p-value is greater than the common alpha level of 0.05). This slight difference in cell viability could be because in the traditional method, all drugs need to be mixed and formulated first, then be added to the cells one by one, which may cause the drugs to have some reactions with each other during the process, causing some drugs to lose their original effects. In contrast, the hydrogel chip method initiates drug interactions after the chips are placed in the cell culture plate, circumventing issues encountered in the traditional approach.

Table 3 Comparison of cell viability from experimental run of 20 drug-dose combination (from AI-CSR platform) test between traditional methods and using selectively crosslinking hydrogel cocktail chip

1. 5-FU (U)	Run	Drugs combinations	Cocktail chip cell viability (%)	(Conventional) cell viability (%)
2. Oxaliplatin (O)	1	U2R2L2B2P2	14.723	16.872
	2	I2G2F2B2C2P2	15.753	14.220
3. Irinotecan (I)	3	I2F2B2C2P2	17.017	17.819
	4	I2B2P2	18.152	20.557
4. Gemcitabine (G)	5	O2I2F2R2L2B2C2P2	20.434	19.913
	6	U2I2G2R2L2B2P2	20.661	22.290
5. Folinic acid (F)	7	I2G2L2B2C2	21.057	20.208
	8	U2O2I2R2B2C2P2	21.126	23.761
6. Regorafenib (R)	9	U2I2G2F2R2L2B2C2P2	21.505	23.111
	10	I2G2P2	22.023	23.730
7. Lenvatinib (L)	11	U2O2I2G2B2P2	22.964	22.621
	12	U2O2I2G2R2C2P2	23.942	26.052
8. Bevacizumab (B)	13	G2R2B2C2P2	24.336	23.952
	14	I2F2R2L2B2	24.805	25.621
9. Centuximab (C)	15	U2F2L2B2P2	24.908	26.525
	16	U2I2F2L2B2P2	24.944	24.162
10. Panitumumab (P)	17	G2L2B2P2	25.136	27.556
	18	U2O2I2R2L2	25.163	29.513
EC10 (1)	19	U2O2I2F2L2B2C2	25.216	28.630
EC20 (2)	20	U2G2F2R2B2P2	25.839	27.015

---

The comparison of our developed hydrogel based cocktail drug platform in different aspects over the traditional well plate method and other similar works are listed in ESI† Table S3: (1) while the traditional method¹³ requires approximately four hours to add each drug into the well plate, label and mix the drugs in desired combinations, the experiment utilizing the hydrogel chip was completed in just about two and a half hours. This confirms the time-saving potential of the designed chip for drug combination screening experiments. There was also the risk of drug–drug reaction in case of traditional method which may reduce drug effectiveness, while there was no risk of drug–drug reaction observed in our cocktail drug chip (2) our chip offers enhanced customization of drug loading through selective UV exposure, allowing for precise control over the drug concentrations in each of the ten parallel chambers. This level of customization is superior to some of the previously published works.^36,37 (3) Significantly, less wastage of drugs during the drug testing as compared to traditional methods, as our hydrogel-based cocktail drug chip hold minimum capacity of just 1 μL drug per patch of the chip that can be loaded in a single patch of the cocktail drug chip, which can specifically impact the cell viability, (4) the new chip supports parallel screening of ten different drug combinations, significantly increasing throughput compared to the previous studies,^37–39 (5) application of AI-CSR platform to find the optimum drug-dose combinations for the best cell survival rates conducting less experiments comparatively and (6) our method simplifies the process of creating combinatorial drug environments, reducing the complexity, time consuming nature involved in microfluidic printing⁴⁰ and enhancing customization of photomask and user-friendliness compared to some complex microfluidic platforms.^41,42

While robotic liquid handling systems like Opentrons can indeed prepare drug combinations rapidly, not all laboratories have access to such advanced infrastructure. Robotic liquid handling systems may face challenges in completely avoiding cross-contamination between different drug combinations, especially when dealing with very small volumes and high-throughput setups.

4. Conclusions

A PDMS drug-loaded hydrogel wafer chip was introduced for the rapid and precise quantification of drug concentrations for cocktail drug screening experiments. The hydrogel exhibited precise selective cross-linking capabilities using a photomask for different drug doses. Material analysis validated the chosen concentrations of polyethylene glycol diacrylate (PEGDA) hydrogel and the photo initiator LAP (6% PEGDA + 0.3% LAP, 7.5% PEGDA + 0.25% LAP, 10% PEGDA + 0.25% LAP) through cytotoxicity tests, ensuring minimal impact on cells. A UV curing time of 20 seconds was employed for experimental purposes. In drug release efficiency experiments, varying hydrogel concentrations were selected based on different drugs. In contrast to the traditional drug combination experiment method, which takes approximately four hours for the preparation and testing of 155 drug combinations, the proposed hydrogel chip streamlined the process, completed the experiment in just about two and a half hours. This chip significantly reduces the time required for drug combination screening experiments and mitigates errors associated with manual addition of drug. Upon comparison of the results, it was observed that the cell survival rate in the traditional method was slightly higher than that of hydrogel based chip. Additionally, the chip demonstrated an improvement in addressing the issue of early reactions between drugs, showcasing its potential to enhance the precision and reliability of drug screening experiments. The chip's ability to handle drug combinations involving ten drugs and multiple dosages aligns well with the goals of personalized medicine. The design and functionality of the chip open up new avenues to explore complex drug interactions, study the effects of drug combinations on different cell types, and investigate the mechanisms of drug resistance for personalized medicine.

Data availability

The data supporting this article have been included as part of the ESI.†

Author contributions

Kiran Kaladharan: writing – original draft & editing, data curation, formal analysis. Chih-Hsuan Ouyang: methodology, data curation, investigation. Hsin-Yu Yang: data curation, formal analysis. Fan-Gang Tseng: conceptualization, writing – review & editing, supervision, funding acquisition.

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

Authors would like to thank the Shekelton project (MOST108-2638-E-007-001-MY2), NSTC personal project (MOST110-2221-E-007-017-MY3), VGH and Taiwan University system cooperation project (VGHUST112-G6-1-1) and 3R project/NSTC (NSTC 113-2321-B-007-002) for their support towards this research.

References

1. H. Sung, J. Ferlay, R. L. Siegel, M. Laversanne, I. Soerjomataram and A. Jemal, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries, Ca-Cancer J. Clin., 2021, 71, 209–249,  DOI:10.3322/caac.21660.
2. R. L. Siegel, K. D. Miller and A. Jemal, Cancer statistics, 2019, Ca-Cancer J. Clin., 2019, 69, 7–34,  DOI:10.3322/caac.21551.
3. R. L. Siegel, K. D. Miller, A. Goding Sauer, S. A. Fedewa, L. F. Butterly and J. C. Anderson, et al. Colorectal cancer statistics, 2020, Ca-Cancer J. Clin., 2020, 70, 145–164,  DOI:10.3322/caac.21601.
4. C. Grávalos, J. Cassinello, I. Fernández-Rañada and E. Holgado, Role of tyrosine kinase inhibitors in the treatment of advanced colorectal cancer, Clin. Colorectal Cancer, 2007, 6, 691–699,  DOI:10.3816/CCC.2007.n.038.
5. J. Zhai, C. Li, H. Li, S. Yi, N. Yang and K. Miao, et al. Cancer drug screening with an on-chip multi-drug dispenser in digital microfluidics, Lab Chip, 2021, 21, 4749–4759,  10.1039/d1lc00895a.
6. X. Hu, X. Hu, S. Zhao, S. Zhao, Z. Luo and Y. Zuo, et al. On-chip hydrogel arrays individually encapsulating acoustic formed multicellular aggregates for high throughput drug testing, Lab Chip, 2020, 20, 2228–2236,  10.1039/d0lc00255k.
7. J. Holch, S. Stintzing and V. Heinemann, Treatment of metastatic colorectal cancer: Standard of care and future perspectives, Visc. Med., 2016, 32, 178–183,  DOI:10.1159/000446052.
8. J. Zhai, C. Li, H. Li, Y. Jia and R. P. Martins, et al. Cancer drug screening with an on-chip multi-drug dispenser in digital microfluidics, Lab Chip, 2021, 21(24), 4749–4759,  10.1039/D1LC00895A.
9. B. N. Yalamandala, P. H. Chen, T. Moorthy, T. M. H. Huynh, W. H. Chiang and S. H. Hu, Programmed Catalytic Therapy-Mediated ROS Generation and T-Cell Infiltration in Lung Metastasis by a Dual Metal-Organic Framework (MOF) Nanoagent, Pharmaceutics, 2022, 14(3), 527,  DOI:10.3390/pharmaceutics14030527.
10. P. S. Dittrich, K. Tachikawa and A. Manz, Micro total analysis systems. Latest advancements and trends, Anal. Chem., 2006, 78, 3887–3907,  DOI:10.1021/ac0605602.
11. P. A. Auroux, D. Iossifidis, D. R. Reyes and A. Manz, Micro total analysis systems. 2. Analytical standard operations and applications, Anal. Chem., 2002, 74, 2637–2652,  DOI:10.1021/ac020239t.
12. P. S. Dittrich and A. Manz, Lab-on-a-chip: Microfluidics in drug discovery, Nat. Rev. Drug Discovery, 2006, 5, 210–218,  DOI:10.1038/nrd1985.
13. H.-Y. Yang, V. S. Goudar, Y.-C. Hung, C.-H. Ouyang, M. B. M. A. Rashid, L.-Y. Juo, J.-K. Wu, Y.-W. Cheng, P.-L. Wei, H.-E. Tzeng, J.-K. Jiang, E. K.-H. Chow, C.-Y. Yang, C.-M. Ho, Y. Yen and F.-G. Tseng, In Vitro Study on AI-PRS Enabled Precision Cocktail Drugs Design for Treating Human Colorectal Carcinoma, Adv. Ther., 2023, 6(6), 2200298,  DOI:10.1002/adtp.202200298.
14. S. J. Bryant, C. R. Nuttelman and K. S. Anseth, Cytocompatibility of UV and visible light photoinitiating systems on cultured NIH/3T3 fibroblasts in vitro, J. Biomater. Sci., Polym. Ed., 2000, 11, 439–457,  DOI:10.1163/156856200743805.
15. T. Moorthy, B. M. Hathim, C. H. M. NagaMahesh, G. Anburaj, S. S. Ahmed and V. Gopinath, et al. Controlled release of kaempferol from porous scaffolds augments in-vitro osteogenesis in human osteoblasts, J. Drug Delivery Sci. Technol., 2023, 83, 104396,  DOI:10.1016/j.jddst.2023.104396.
16. M. Peeters, T. J. Price, A. Cervantes, A. F. Sobrero, M. Ducreux and Y. Hotko, et al. Randomized phase III study of panitumumab with fluorouracil, leucovorin, and irinotecan (FOLFIRI) compared with FOLFIRI alone as second-line treatment in patients with metastatic colorectal cancer, J. Clin. Oncol., 2010, 28, 4706–4713,  DOI:10.1200/JCO.2009.27.6055.
17. J. Y. Douillard, S. Siena, J. Cassidy, J. Tabernero, R. Burkes and M. Barugel, et al. Randomized, Phase III trial of panitumumab with infusional fluorouracil, leucovorin, and oxaliplatin (FOLFOX4) Versus FOLFOX4 alone as first-line treatment in patients with previously untreated metastatic colorectal cancer: The PRIME study, J. Clin. Oncol., 2010, 28, 4697–4705,  DOI:10.1200/JCO.2009.27.4860.
18. C. Heidelberger, N. Chaudhuri and P. Danneberg, et al., Fluorinated Pyrimidines, A New Class of Tumour-Inhibitory Compounds, Nature, 1957, 179, 663–666,  DOI:10.1038/179663a0.
19. H. Podder, S. M. Stepkowski, K. L. Napoli, J. Clark, R. R. Verani and T. C. Chou, et al. Pharmacokinetic interactions augment toxicities of sirolimus/cyclosporine combinations, J. Am. Soc. Nephrol., 2001, 12, 1059–1071,  DOI:10.1681/asn.v1251059.
20. P. M. LoRusso, R. Canetta, J. A. Wagner, E. P. Balogh, S. J. Nass and S. A. Boerner, et al. Accelerating cancer therapy development: The importance of combination strategies and collaboration. Summary of an Institute of Medicine Workshop, Clin Cancer Res, 2012, 18, 6101–6109,  DOI:10.1158/1078-0432.CCR-12-2455.
21. D. O'Hagan, C. Schaffrath, S. L. Cobb, J. T. G. Hamilton and C. D. Murphy, Biosynthesis of an organofluorine molecule, Nature, 2002, 416, 279,  DOI:10.1038/416279a.
22. S. Shen, X. Zhang, F. Zhang, D. Wang, D. Long and Y. Niu, Three-gradient constructions in a flow-rate insensitive microfluidic system for drug screening towards personalized treatment, Talanta, 2020, 208, 120477,  DOI:10.1016/j.talanta.2019.120477.
23. X. Ding, V. H. S. Chang, Y. Li, X. Li, H. Xu and C. Ho, et al. Harnessing an Artificial Intelligence Platform to Dynamically Individualize Combination Therapy for Treating Colorectal Carcinoma in a Rat Model, Adv. Ther., 2020, 3(4), 1900127,  DOI:10.1002/adtp.201900127.
24. M. B. M. A. Rashid, T. B. Toh, L. Hooi, A. Silva, Y. Zhang and P. F. Tan, et al. Optimizing drug combinations against multiple myeloma using a quadratic phenotypic optimization platform (QPOP), Sci. Transl. Med., 2018, 10(453), eaan0941,  DOI:10.1126/scitranslmed.aan0941.
25. X. Ding, W. Liu, A. Weiss, Y. Li, I. Wong and A. W. Griffioen, et al. Discovery of a low order drug-cell response surface for applications in personalized medicine, Phys. Biol., 2014, 11(6), 065003,  DOI:10.1088/1478-3975/11/6/065003.
26. X. Ding, H. Xu, C. Hopper, J. Yang and C. M. Ho, Use of fractional factorial designs in antiviral drug studies, Qual. Reliab. Eng. Int., 2013, 29, 299–304,  DOI:10.1002/qre.1308.
27. A. Weiss, M. Le Roux-Bourdieu, M. Zoetemelk, G. M. Ramzy, M. Rausch and D. Harry, et al. Identification of a synergistic multi-drug combination active in cancer cells via the prevention of spindle pole clustering, Cancers, 2019, 11, 1–8,  DOI:10.3390/cancers11101612.
28. I. Al-Shyoukh, F. Yu, J. Feng, K. Yan, S. Dubinett and C. M. Ho, et al. Systematic quantitative characterization of cellular responses induced by multiple signals, BMC Syst. Biol., 2011, 5, 88,  DOI:10.1186/1752-0509-5-88.
29. A. Silva, B. Y. Lee, D. L. Clemens, T. Kee, X. Ding and C. M. Ho, et al. Output-driven feedback system control platform optimizes combinatorial therapy of tuberculosis using a macrophage cell culture model, Proc. Natl. Acad. Sci. U. S. A., 2016, 113, E2172–E2179,  DOI:10.1073/pnas.1600812113.
30. J. Jaynes, X. Ding, H. Xu, W. K. Wong and C. M. Ho, Application of fractional factorial designs to study drug combinations, Stat. Med., 2013, 32, 307–318,  DOI:10.1002/sim.5526.
31. A. J. Pantuck, D. K. Lee, T. Kee, P. Wang, S. Lakhotia and M. H. Silverman, et al. Modulating BET Bromodomain Inhibitor ZEN-3694 and Enzalutamide Combination Dosing in a Metastatic Prostate Cancer Patient Using CURATE.AI, an Artificial Intelligence Platform, Adv. Ther., 2018, 1, 1–9,  DOI:10.1002/adtp.201800104.
32. X. Jia, Y. Li, A. Sharma, Y. Li, G. Xie and G. Wang, et al. Application of sequential factorial design and orthogonal array composite design (OACD) to study combination of 5 prostate cancer drugs, Comput. Biol. Chem., 2017, 67, 234–243,  DOI:10.1016/j.compbiolchem.2017.01.010.
33. T. Zhang, A. Sharma, Y. Li, Y. Zhou and X. Ding, Orthogonal Array composite design to study and optimize antioxidant combinations in the prevention of UVB-induced HSF damage, J. Photochem. Photobiol., B, 2018, 178, 568–576,  DOI:10.1016/j.jphotobiol.2017.12.007.
34. H. Xu, J. Jaynes and X. Ding, Combining two-level and three-level orthogonal arrays for factor screening and response surface exploration, Stat. Sin., 2014, 24, 269–289,  DOI:10.5705/ss.2012.210.
35. D. Y. Wang, H. Liu, L. P. Tu and G. Y. Ding, An orthogonal electric fish optimization algorithm with quantization for global numerical optimization, Soft Comput., 2023, 27, 7259–7283,  DOI:10.1007/s00500-023-07930-6.
36. G. S. Du, J. Z. Pan, S. P. Zhao, Y. Zhu, J. M. J. Den Toonder and Q. Fang, Cell-based drug combination screening with a microfluidic droplet array system, Anal. Chem., 2013, 85, 6740–6747,  DOI:10.1021/ac400688f.
37. Y. T. Chen, V. S. Goudar, R. G. Wu, H. Y. Hsieh, C. S. Yang and H. Y. Chang, et al. A UV-sensitive hydrogel based combinatory drug delivery chip (UV gel-Drug Chip) for cancer cocktail drug screening, RSC Adv., 2016, 6, 44425–44434,  10.1039/c6ra01733a.
38. G. Quon, Y. Ajena, K. S. Lam and T. Pan, A Plug-and-Play, Drug-on-Pillar Platform for Combination Drug Screening Implemented by Microfluidic Adaptive Printing, Anal. Chem., 2018, 90(23), 13969–13977,  DOI:10.1021/acs.analchem.8b03456.
39. L. Li, Y. Chen, H. Wang, G. An, H. Wu and W. Huang, A high-throughput, open-space and reusable microfluidic chip for combinational drug screening on tumor spheroids, Lab Chip, 2021, 21, 3924–3932,  10.1039/d1lc00525a.
40. Y. Ding, J. Li, W. Xiao, K. Xiao, J. Lee and U. Bhardwaj, et al. Microfluidic-Enabled Print-to-Screen Platform for High-Throughput Screening of Combinatorial Chemotherapy, Anal. Chem., 2015, 87(20), 10166–10171,  DOI:10.1021/acs.analchem.5b00826.
41. F. Eduati, R. Utharala, D. Madhavan, U. P. Neumann, T. Longerich and T. Cramer, et al. A microfluidics platform for combinatorial drug screening on cancer biopsies, Nat. Commun., 2018, 9(1), 2434,  DOI:10.1038/s41467-018-04919-w.
42. Z. Zhang, Y. C. Chen, S. Urs, L. Chen, D. M. Simeone and E. Yoon, Scalable Multiplexed Drug-Combination Screening Platforms Using 3D Microtumor Model for Precision Medicine, Small, 2018, 14(42), e1703617,  DOI:10.1002/smll.201703617.

---

Footnote

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d4lc00520a

---