The first literature data on microfluidic approaches is a paper by Lew and Fung (1969). They were the first to outline the concept of simulation theory in the attempt to mimic the fluidic dynamics of the blood flow circulation and bronchioles air flow simulations within lung tissue. They asserted that the flow generated by the uniform axial velocity is relevant for blood flow in branching blood vessels and for the motion of plasma in the capillary blood vessels. This theoretic microtube system, with a diameter equal or less to 100 μm, is generated by complex algorithms governing the fluid dynamics. These Authors were the first to introduce the notion of inlet/outlet in the liquid flow simulations for live systems such as blood circulation and lung bronchioles. This fluidic model has been repurposed by an independent laboratory 20 years later (Xue and Fung, 1989). The importance of this first simulation study lies on the notion that the main role of microfluidic devices is to recapitulate the multicellular living systems in their multifaceted aspects, including biochemical cues behind the cell-cell communication. This key milestone could not be reached without the LC systems. LCs proved decisive for a better knowledge of cell-cell signaling, mimicking specific inner events including those relative to immune cells.

The first use of a LC for investigative purposes on living (eukaryotic) cells was made by Esch et al. (2001a, b) for the detection of Cryptosporidium Parvum, a waterborne pathogen. They used a method relying on the ability of these bacteria to release some typical and exclusive DNA sequences that are PCR-amplified within ad hoc fabricated microfluidic chips. These studies demonstrated that microdevices can be used to investigate the effects of chemicals and proteins released from bacteria and cells. For these reasons, the majority of scientists coined the term LC to evidence how these devices can be conceived as actual miniaturized laboratories. Such mini-labs are focused at investigating specific substances such as chemicals or proteins, but are not suitable for living cells.

In 2012, Farmer and co-Workers generated a LC microfluidic system to recapitulate a red blood cell. They were the first to mimic a cell model, even though not a living cell, with microfluidic approaches (Farmer et al., 1988). This first study in a microfluidic cartridge has certain been facilitated by the non-living nature of the system. This work can be considered as one of the first attempts to obtain a primitive cell-mimicking system. The first microfluidic system to immobilize particular types of proteins came on the 1990s, with a study from Johnsson et al. (1991). This research describes how microfluidic systems can be applied in the field of sensors engineering. On the other hand, it represents one of the first studies reporting a complete microfluidic chip fabrication process as an important part of device design.

A relevant giant step on the creation of LCs as sensors useful for cancer prevention comes from an elegant paper of Bahavarnia and co-Workers. Here a paper-based immunosensor has been generated, coupled to a fluidic chamber to propagate the fluid toward the sensor area in which the CA125 protein will be detected by an ELISA-based assay (Bahavarnia et al., 2019). Similar approaches based on the immunosensors have been described in literature, useful to study apoptotic activity (Dang et al., 2019) or soluble immune checkpoint inhibitors (Reza et al., 2019) in biological fluids or clinical samples. These studies represent interesting examples of LCs to be applied for oncoimmunology applications.

Lab-on-Chip are generally appropriate to study specific mechanical activities or fluidic phenomena, often to be addressed to specific living cell mimicry events. With the development of more advanced microfluidic fabrication techniques and with the advent of sophisticated biotechnology tools, LC platforms have evolved into complex systems to investigate on some features of single cells. For instance, an LC platform was developed with the aim to identify the disparate antigen specific antibodies secreted by single B-lymphocytes during the development of innate immune responses (Gérard et al., 2020). Fabrication technologies have also allowed implementation of LC into efficient and affordable biosensors. LC biosensors have been recently developed and optimized to be applied in oncoimmunology. A clear example is represented by LC biosensors to detect cancer biomarkers, where the main LC biomaterial is represented by paper and nano-inks (Bahavarnia et al., 2019). Despite the discovery of disparate materials well suited for biological studies, the use of Polydimethylsiloxane (PDMS)-based techniques still remains the elected methodology to fabricate all microfluidic devices, including LC biosensors. Next generation chips will be represented by microfluidic platforms where the use of the PDMS may be supplemented, rather than superseded, by the use of other biomaterials needed for chip optimization.

While in vitro Two dimensional (2D) and Three dimensional (3D) culture systems are useful to study specific cell features, the advent of LC equipped investigators with versatile tools to monitor cell behaviors in real time. This began the initiation of the CC era.

In 1997 Li and co-Workers carried out a glass microfluidic CC system suitable to monitor the mobilization of red blood cells, yeast cells as well as Escherichia Coli bacteria in a device composed by several interconnected channels. This chip constituted one of the first approaches to study living cells by means of a device composed by multiple substructural units (Li and Harrison, 1997). A first study analyzing cell docking and alignment was produced by Yang and co-Workers on 2002 with human hematological HL-60 tumor cell lines (Yang et al., 2002). Another report where cells were monitored in a CC system employed a feochromocytoma derived PC12 cell line (Huang et al., 2004) in the attempt to follow their mobilization features. The development of CC research with tumor cells has been facilitated by the fact that these cells do not require stringent factors for their growth, compared to primary cells (Zielinski et al., 2017). For this reason, tumor cells can be easily studied on chip, in particular for long-term time points. In 2005, primary cells were successfully studied on a CC system in an interesting work focused at recapitulating the cardiac myocyte contraction (Li and Li, 2005). This and other studies (Ho et al., 2006; Kokkinos et al., 2008; Zhao et al., 2012) constitute some representative examples of the first primitive studies carried out in CC systems. All of them utilized a single cell type with simple analyses and evaluations, such as mobilization within microscale-sized spatial units and brief time intervals.

The first attempt to study a specific immune cell population on chip was reported in 2006 by Matsumura and co-Workers, who investigated the ability of macrophages to start and sustain zymosan-induced phagocytosis in fibronectin-coated CC by time-lapse microscopy. This represented a first study in which a functional immune event has been reproduced on chip (Matsumura et al., 2006). Another interesting CC system has been generated by Kang and co-Workers. Their structure is represented by a chamber where bead-labeled blood cells flow upon a micromagnetic field. This field has a key role in capturing rare circulating tumor cells or metastatic cells (Kang et al., 2012). The particularity of this microfluidic chip is the possibility to culture a single cell type by flushing away the other undesired cells, such as circulating leukocytes.

An important phenomenon to be investigated in the field of oncoimmunology is the migratory behavior of immune cells toward the primary tumor site. Thus, it is not surprising that researchers pay particular attention on the study of immune cell migration on chip. A first example of such studies was provided by Molino and co-Workers. They described a CC system properly designed to follow leukemic cell migration toward an oil droplet (Molino et al., 2016). For some aspects, these droplets resemble immune cell moving in the CC microchannel. This work is a clear example of how CC studies inspired experimental implementations to design the more complex OOC systems, in order to study the interaction between immune cells and cancer within ad hoc fabricated microfluidic devices.

The term OOC underlines the attempt to load different cell types residing organs or multicellular complexes inside a microfluidic device, often coupled to sophisticated microscopy systems. In general, such a definition fits also for chips loaded with at minimum two different cell populations, even if these are not able to develop in complex multicellular systems (Sackmann et al., 2014; Reardon, 2015; Chen P. et al., 2020). The simpler the OOC is the easier cells are monitored within it.

A first approach to study immune system on chip has been developed by Whitesides and co-Workers. They created a simple OOC platform to co-culture two different types of macrophage-like cells (Wong et al., 2008). In this report, BAC1.2F5 and LADMAC macrophage cell lines were loaded in separated chambers of the microdevice. Then, the production of the colony stimulating factor-1 (CSF-1), released by LADMAC and required by BAC.2F5 cells for survival, was monitored over time. This simple OOC prototype can also be used to study the mutual crosstalk between immune and cancer cells.

The first attempt to design a microfluidic platform employing different cell types on chip was reported by Shuler and co-Workers on 2009. They adapted a microfluidic chip for simultaneous co-culture of tumor cells (colon cancer cells HCT-116 or Hepatoma cells HepG2, embedded in hydrogels) with liver and bone marrow cells (Sung and Shuler, 2009). This system allowed to elegantly test the cytotoxicity of a specific anticancer agent on tissues when directly delivered on chip. Despite allowing the simultaneous loading of different cells, this OOC system does not allow to study the crosstalk between loaded cells, such as migration of bone-marrow cells toward the tumor in response to the drugs. In a study investigating the crosstalk between immune cells and cancer cells, Businaro and co-Workers successfully attempted to investigate on the mutual behavior of melanoma and immune cells (Businaro et al., 2013). This approach was inspired by the difficulty to study the events behind the immune cell infiltration inside the TME, and represents a new dawn for the oncoimmunology on chip. Previous work in vivo showed that accelerated B16.F10 melanoma progression in non-immunocompetent mice was the result of altered cytokine/chemokine production and impaired infiltration of different murine immune cell subsets in the TME (Mattei et al., 2012). Based on these observations, microfluidic devices were employed as a simple OOC system to evaluate gradient-dependent interactions between B16.F10 melanoma cells and spleen-derived immune cells from naïve immunocompetent vs non-immunocompetent mice. In detail, a PDMS microdevice constituted by a 1,000 μm-sized central chamber connected to two side chambers by two arrays of 12 μm-sized microchannels has been used. The two side chambers were loaded with spleen cells (from immunocompetent or IRF-8 deficient mice) and B16.F10 melanoma cells. Time-lapse video recording was carried out over a 48 h-period in a region comprising the B16 side chambers and the adjacent microchannels. The results evidenced a mutual crosstalk between B16 cells and spleen cells. Specifically, in OOC loaded with splenocytes from immunocompetent mice, these cells displayed a high migratory ability toward the B16 chamber coupled to a slow B16 invasion extent. Conversely, IRF-8 deficient splenocytes displayed poor migratory extent toward the tumor, which reflected a high invasion ability of melanoma cells. This report constitutes the first evidence of mutual interaction between two different cell types in a microfluidic chip. Further analysis on time-lapse videos demonstrated that the migration of immune cells can be quantitatively and qualitatively evaluated by extrapolating single cell tracking profiles of each cell migrating toward melanoma cells (Agliari et al., 2014). From an oncoimmunology view, these tracking profiles represent relevant functional data expressing the overall behavior of immunocompetent vs non-immunocompetent cells faced to melanoma cells into this OOC system.

On the other hand, time-lapse microscopy is useful to evaluate the tumor cell membrane elasticity as a function of cell viability during their permanence in the OOC system. Fully functional tumor cells show high variations in the membrane fluctuation over each image frame (Figure 1). On the contrary, the membrane fluctuations are less evident in dying or stressed tumor cells. This perturbation factor has been quantified by Agliari and co-Workers by targeting time-lapse on single tumor cells inside the OOC. It can be used as a function of the stress status of a cancer cell after migration in the chip and reflects the ability of immune cells to limit cancer cell growth on chip as a consequence of tumor-immune cell interactions (Agliari et al., 2014). This “migrate-to-interact” concept can be adapted for every OOC platform designed to study how cancer cells respond to the presence of immune cells (Figure 1).

Further advancements in OOC development in oncoimmunology have been made by using human hepatitis B virus (HBV)-derived cancer cells and T cells that can be loaded and monitored on chip. Here, a customizable OOC was used to load engineered T cells expressing HBV-specific antibodies in liquid media with tumor cells derived from the same cancer patient. Tumor cells were loaded in a central chamber resuspended in a collagen matrix, thus recapitulating the TME on chip. In this setting, it was possible to follow and quantitate the tumor cell killing activity of T cells with accuracy (Pavesi et al., 2017). Moreover, this study allowed a better knowledge of the suitability of extracellular matrices for OOC systems, such as Matrigel, hydrogels or collagen complexes to better recapitulate the TME. Indeed, extracellular matrices represent a mix of extracellular proteins markedly improving the capability of cells to survive inside the microfluidic devices. In this situation, the cells engage chemical bonds with the components of these matrices then activating intracellular survival signals. Thus, the use of these matrices is particularly relevant for the activation and survival of some primary cell types, such as dendritic cells and hepatocytes (Mehta et al., 2010; Mölzer et al., 2019; Serna-Márquez et al., 2020). On the other hand, it is becoming increasingly evident that the use of gel matrices represents an important step to an optimal development of OOC platforms in oncoimmunology, as evidenced by the key role of the disparate types of matrices employed to reproduce the TME (Narkhede et al., 2017; Hassell et al., 2018; Sontheimer-Phelps et al., 2019).

In a recent study, Fang and co-Workers reported the realization of an OOC system with human tumor spheroids or murine organoids (Fang et al., 2019). They loaded the device with fibroblasts and MCF7 or 4T1.2 breast tumor cells to allow the on-chip generation of tumor organoids, with the fibroblasts spontaneously infiltrating the tumor mass. In this case they used the agarose as main biomaterial of the OOC. In addition, a microfluidic prototype has been developed by Borenstein and co-Workers to study the immune checkpoint inhibitors in human cancer (Beckwith et al., 2019). Specifically, this OOC system allows the trapping of a lung tumor biopsy and a real-time monitoring of the programmed cell death protein-1 (PD-1) expression. The originality of this study is represented by the development of an OOC platform optimized to monitor PD-1 on fresh biopsies in 24 h.

All these studies facilitated the ad hoc development of OOC systems and led to a new generation of more complex OOCs. Another representative example in using diverse cell populations in OOC research is the development of multi-organ units on chip, outlined by a recent work depicting a platform aimed at generating and connecting liver and gut on chip (Sung, 2020). The Supplementary Table 1 illustrates the technical and analytic details of several OOC systems and their potential use for oncoimmunology studies. This list suggests that OOC can be designed by lithographic techniques and 3D bioprinting methods, with or without gel matrices. As expected, the structural composition of the microdevices largely changes in function of their specific functional and analytic requirements.

An OncoImmuno chip defines a microdevice in which the TME’s main components are isolated, reconstituted on-chip and coupled to time-lapse and image analysis systems to monitor immune and tumor cells overtime, for a maximum of 72 h (Biselli et al., 2017; Mencattini et al., 2020a; Torres-Simón et al., 2020; Figure 2).

There are several important advantages to track immune cell migratory behavior toward cancer cells. For example, in an experiment where immune and drug pre-treated cancer cells are loaded into two separate compartments, it will be possible to acquire, frame-by-frame (FBF), a set of single cell tracks throughout the entire duration of the time-lapse (Figure 3). This FBF acquisition of immune cells yields an array of cell tracks, namely the immune cell tracking profile, whose behavior is strictly dependent on the anti-tumor effects of an administered drug (Figure 3). When each immune cell reaches the compartment containing drug-treated cancer cells, it interacts with them for a variable time. The computation of the duration of such an interaction is then fundamental for the quantification of therapeutic efficacy. Such a post-migration interaction represents a functional parameter indicative of the ability of the immune cell to “sense” cancer cells inside the device upon drug exposure (Figure 3).

There are some typical classes of trajectories relative to immune cell motion that are highly indicative of the response to attractant or repellent substances emanated from tumor cells (Biselli et al., 2017; Comes et al., 2020b). These distinct track patterns have functional effect on cancer cell fate (Figure 4) and can be represented by several kinematic and non-kinematic factors. The first, directionality ratio (Gorelik and Gautreau, 2014), expresses the ability of an immune cell to reach its target cancer cell by taking the shortest route possible.

The mean step length of a track is another kinematic entity intrinsically associated to the track of an immune cell (Figure 4). It expresses the mean walking displacement of the immune cells between two consecutive points of the time-lapse (Figures 3, 4). Assuming that a single step length, s, is yielded by the formula A delineated in Figure 5 (Caplan, 1972; Simmons et al., 2010). Here, Δx and Δy represent the projections of the cell step length s on the X and Y axes in a Cartesian XY domain (Figure 5). By denoting as (x₁, y₁) and (x₂, y₂) the starting and the ending points of a generic displacement, i.e., △x = x₂−x₁ and △y = y₂−y₁, then the single step length of an immune cell can be also obtained by the formula B in Figure 5. The mean step length is calculated as the average s value among all the steps of the immune cell along its track and directly depends on the speed of that cell (Perner and Perner, 2009; Dewan et al., 2011; Biselli et al., 2017).

Immune cell speed returns the step length s from a starting (x₁, y₁) point to an ending point (x₂, y₂) divided by the time required to move from the two points (Figures 4, 5).

The track angle represents another important quantity of the immune cell migration path strictly associated to the directionality of the track. It is expressed by the angle α (in degrees) the current step vector s forms with the X reference axis (Figure 5). The angular deviation can be yielded by the difference between the two α values of the immune cell at two consecutive points.

Assuming that the step length is an s vector composed by its projection on X and Y axes, as mentioned in Figure 5, the angular deviation a constitutes the angle between the s vector and its projection on X axis. Such an angular deviation α of the immune cell can then be expressed with a generalized formula C in Figure 5, derived from the Pythagorean theorem, and is also a function of the attractive forces exerted by the adjacent cancer cell. As delineated by the two formula D and E in Figure 5, Δy and Δx represent the projections of the step lengths on Y and X axes, respectively. The two catheti Δy, Δx and the hypotenuse s constitute a right-angled triangle (Figure 5), where the angular variation α a is easily computable by the inverse tangent formula C in Figure 5 (Dummer et al., 2016; Lin, 2019; Cankaya et al., 2020; Zhang et al., 2020). Of note, each step length s of the immune cell track is indissolubly linked to its relative α value conventionally referred to the X axis, as evidenced in Figure 5.

Not less relevant, one of the most important parameter used to describe the type of motion for an immune cell is the mean square displacement (MSD)(Gorelik and Gautreau, 2014). Early studies that attempted to mathematically model cell migration in isotropic medium assumed that a cell moves like a Brownian particle (Dunn, 1983). Namely, an immune cell was assumed to undertake a persistent random walk, in which it moves directionally at short time intervals, but it loses its persistence at longer time intervals. The time to cross from the persistent regime to the random one is the directional persistence time, whose estimation requires a fitting of the MSD curve (Gorelik and Gautreau, 2014). The MSD is a crucial factor since it embeds global as well as local kinematic characteristics of the immune cell track at different time lag, thus evidencing cell behavior at shorter time but also the overall trend at larger time. In an oncoimmunology context, the MSD is strictly dependent on how cancer cell drug treatment can affect the immune cell trajectory on chip.

In conclusion, cell tracking analysis returns several hidden kinematic and non-kinematic values, that are an intrinsic feature of the FBF analysis in a time-lapse video. This can be easily extrapolated in an OncoImmuno chip, by analyzing the time-lapse video and single image frames with a specific image analysis software. In this regard, the open source software ImageJ is equipped with a number of plug-ins (i.e., Manual Tracking or Automated Tracking plug-ins) that allow to automatically process all the aforementioned entities, then facilitating the extrapolation of all kinematic factors associated to each single tracked immune cell. The use of ImageJ with the integrated Automated Tracking plug-in is a clear example of image-assisted analysis of cell migration tracking. A list of the most commonly used cell migration tracking software are depicted in Supplementary Table 2 (Emami et al., 2021).

Overall, the use of tracking analysis in OncoImmuno chips can recapitulate the interactions between cancer and immune cells in TME in the steady-state or following a defined therapy by exploiting hidden data acquired exclusively through these platforms and not accessible by in vivo TME methodologies.

The massive advent of sophisticated biotechnologies has been accompanied by a parallel evolution of instruments in which these technologies were computationally implemented. These analytical methods are recently exploiting the beginning of the Artificial Intelligence framework that is revolutionizing the field of biomedicine with its diverse applicative approaches (Ramesh et al., 2004; Buch et al., 2018; Balkanyi and Cornet, 2019; Holzinger et al., 2019; Mintz and Brodie, 2019; Kaul et al., 2020). In the area of machine learning (Rajkomar et al., 2019; Sidey-Gibbons and Sidey-Gibbons, 2019; Alsuliman et al., 2020), a novel kind of algorithm-based methodologies have recently gained a profound interest. These advanced methods belong to a sub-branch of machine learning, the deep-learning (Cao et al., 2018; Wainberg et al., 2018; Bock et al., 2021), and, in particular, often refer to the so called Generative Adversarial Networks (GAN). GANs can have disparate practices, covering social sciences, image analysis and biological problem solving. To this purpose, GANs have been applied for prediction of human actions (Liu et al., 2017), image processing applications (Nie et al., 2018; Li et al., 2019c; Xue et al., 2020; Yang et al., 2020), and computational analysis of natural environments (Saleemi et al., 2009; Negin and Brémond, 2019; Comes et al., 2020a, b). GAN are based on a game scenario where the generator network must compete with an adversary, the discriminator network. The two network models are trained together until the discriminator network is fooled about half the time from the image created by the generative network, meaning that it is generating plausible examples (Sahoo and Narayanan, 2019; Xu et al., 2020).

The gradually increased use of time-lapse microscopy together with the use of microfluidic platforms provided oncoimmunologists with potent tools to perform long-term live cell imaging and high-throughput analysis of on chip platforms (Comes et al., 2019, 2020a; Mencattini et al., 2020b). In this context, GANs constitute potent machine learning tools to predict how a cell migratory path will evolve starting from the real cell tracking captured by a time-lapse video. Specifically, an updated version of the GAN, social GAN (Comes et al., 2020b), can be useful to predict the post-migratory behavior of an immune cell when it is in proximity of a cancer cell in a microfluidic device (Figure 6). To this purpose, all the single immune cells are identified and tracked starting from the original time-lapse video sequence. To do so, GAN algorithm need to be “trained” by inputting a number of initial immune cell positions, namely the Input Tracks in Figure 6. After the training procedure, GAN is able to return high-fidelity immune cell trajectories by extrapolating them from GAN-dependent hidden entities (Fernando et al., 2019; Comes et al., 2020b). This approach is particularly useful to predict the behavior of poorly identifiable immune cells, or simply to accelerate the uptake of the experiment and to avoid phototoxic effects on the cells (Figure 6). GAN can provide a meaningful estimation of the probability for each immune cell to physically interact with the cancer cell in the vicinity (Figure 6).

In conclusion, a GAN learning system can be helpful for the envision of the so called in silico OOC experiments (Swayden et al., 2019). The main goal of such a system is the evaluation of a real OOC experiment without performing until the end reducing time and experimental costs. Moreover, this process can also generate an in silico OOC simulation potentially useful to plan new optimized OOC experiments.

One of the first studies aimed to assess the effects of pharmacological treatments for cancer therapy by using an on chip approach is represented by a study addressing the role of formyl peptide receptor 1 (FPR1)/annexin a1 (Anxa1) axis in anti-tumor response to anthracycline-based chemotherapy (Vacchelli et al., 2015). In this report, OOC experiments in humans and mouse settings were crucial to demonstrate the importance of the integrity of the FPR1/Anxa1 axis to allow immune cell migration and interaction with cancer cells undergoing chemotherapy-induced immunogenic cell death. Indeed, cancer cells treated with anthracyclines (i.e., Doxorubicin or Mitoxantrone) release Anxa1, which acts as a danger signal for immune cells. Single cell tracking profiles extrapolated by OOC system displayed that FPR1 expressed by immune cells, particularly dendritic cells (DCs), is required to “sense” Anxa1 emanated from dying tumor cells and to migrate toward them engaging in stable interactions. Interestingly, in the OOC setting only DC with an intact FPR1 successfully migrate toward anthracycline-killed tumor cells and capture apoptotic bodies released from these dying cells. Post-analysis of time-lapse video recordings in this OOC system allowed the extrapolation of high numbers of single cell tracks associated to high interaction times between immune cells (including DCs) and cancer cells, showing that immune cells with intact FPR1 perform biased random walks toward anthracycline-treated cancer cells, while those with mutated FPR1 show uncorrelated random walks (Biselli et al., 2017). Thus, numerical descriptors and statistical physics applied to OOC systems provide an additional tool for a quantitative description of the immune response to cancer during chemotherapy.

Organs-On-Chip systems can undergo further step-by-step implementations in their structural complexity which do not compromise their state-of-simplicity. In this regard, Nguyen and co-Workers have successfully visualized and quantified the effects of the anticancer drug Trastuzumab in a TME reconstituted on chip, composed by cancer-associated fibroblasts (CAF), an HER2⁺ breast cancer cell line (bearing a HER2-receptor gene amplification), and immune cells (Nguyen et al., 2018). This study demonstrates that an OOC system hold potential for further implementation by adding other TME key cell components, such as cancer associated fibroblasts (CAFs). Addition of CAFs in this system still contributes to maintain a simplified and smart OOC platform. Here, the OOC system has been utilized to dissect the role of Trastuzumab, a reference anti-cancer for HER2+ breast cancer patients, on such a tumor-on-chip complemented with CAFs plus hydrogels to reproduce a central endothelium compartment. The microdevice is constituted by two side chambers, containing the immune cells (embedded in a collagen matrix), cancer cells and CAFs, and a central compartment with endothelial cells. These chips where monitored with or without the presence of Trastuzumab. The extrapolation of interaction times, cancer cell speed and tracking profiles by time-lapse demonstrated that Trastuzumab inhibited breast cancer cell growth with a concomitant induction of cell death. In addition, CAFs hindered the interactions between cancer and immune cells. Overall, these on-chip observations suggests that this tumor-on-chip effectively recapitulates the in vivo scenario referred to the effects of Trastuzumab in human breast cancer. A similar concept was applied in a study illustrating the crosstalk between lung cancer cells and epithelial cells in an OOC system mimicking the lung alveoli. Here, H1975 non-small-cell lung cancer (NSCLC) cells were loaded in a microfluidic device to evaluate the effects of Erlotinib and Rociletinib, two representative third generation Tyrosine Kinase inhibitors. The authors designed a chip composed by two side compressible chambers for air flow, mimicking respiratory processes, and a central compartment further subdivided in two separated chambers, containing cancer cells and endothelial cells. This design had the purpose to recapitulate the alveoli of lungs, under normal or cancer settings. These studies not only showed the feasibility of this lung-on-chip system to study the anti-tumoral effects of Tyrosine Kinase inhibitors, but also revealed an unexpected role of breathing in contrasting cancer growth (Hassell et al., 2018).

Organs-On-Chip systems have also been exploited to investigate the role of epithelial-mesenchymal transition (EMT). For instance, an interesting OOC model has been generated to study these phenomena. Here, EMT has been recreated on chip with human lung tumor cell spheroids (NCI-H460 or HCC827 cells) resuspended in Matrigel and human umbilical vein endothelial cells (HUVECs). The microdevice is composed of a nylon mesh matrix which recapitulates the separation system between cancer cells and HUVECs. In this microdevice tumor cells and HUVECs are separated by the mesh membrane. When recombinant TGF-β was added in this OOC system tumor spheroids displayed an activated expression of the EMT factors Snail, and Akt, paralleled by decreased levels of E-cadherin. In addition, tumor cells acquired the ability to migrate toward HUVECs. This OOC system may represent an optimal model to study the effects of anti-cancer drugs in terms of control of processes involved in EMT (Li et al., 2019a). Another representative example of on chip anticancer agent evaluation is the development of a metastasis-on-chip model of study (Wang et al., 2020). This OOC is assembled with methyl methacrylate for structural and spatial components, whereas PDMS is used to allow gas and liquid exchange between cells. The PDMS is used to create sheets to which a 10 μm sized resin membrane is sliced. This PDMS plus resin system is then inserted inside the OOC and used for cell loading. Such device was employed for the evaluation of anti-tumor activity of 5-Fluorouracyl (5-FU) on Caki-1 kidney cancer cells loaded with immortalized HepLL hepatocytes. Of note, this OOC system was integrated with a programmable electronic pump to feed cells with nutrients in the chip. This study can inspire the generation of implemented metastasis-on-chip platforms suitable for other metastatic tumors.

The described OOC platforms represent a starting point for implemented systems specially designed to evaluate the efficacy of emerging immunotherapy strategies in the presence of immune cells. Recently, OOC devices were successfully employed to culture murine- and patient-derived organotypic tumor spheroids preserving the immune compartment of the TME and to evaluate the sensitivity and resistance to PD-1 blockade (Jenkins et al., 2018). This system accurately recapitulated the in vivo scenario, was compatible for long-term culture (up to 1 month) and allowed to evaluate multiple parameters of response, such as cell death and growth of tumor organoids from mice or patient-derived organoids loaded inside the microdevice. The microdevice consists in two side chambers utilized to load the conditioned medium (i.e., the anti-PD-1 antibodies) and a central compartment employed to load human tumor spheroids or tumor cells from mouse colon cancer lesions which develop organoids several days after loading (Aref et al., 2018). This platform can be adapted to evaluate other immune checkpoints (i.e., CTLA-4) expressed on cancer cells. OOCs have also revealed as useful systems to study the effect of folic acid in ovarian cancer. Folic acid is the elected anti-tumoral drug to cure this and other solid malignancies. Here, Wimalachandra and co-Workers have proposed an OOC in which the main chamber was rounded by HUVEC cells and filled with OVCAR-3 tumor cells. This microfluidic device has been used to evaluate the migration of DCs and T cells, in presence or absence of folic acid-loaded nanoparticles (Wimalachandra et al., 2019). These encouraging studies suggest not only that nanoparticles are a useful to assess anti-cancer drugs on chip, but can also inspire the use of OOC platforms for the study of tumor cell functions by nanoparticle-encapsulating drugs.

Single agent-based strategies are often insufficient for a successful and complete tumor eradication in cancer patients. Current strategies aim at combining therapies with a second (or even third) drug to amplify anti-tumor responses and/or to broaden the spectrum of responding patients (Bao et al., 2020; Crunkhorn, 2020; GajdÁcs et al., 2020; Iratni and Ayoub, 2020; Jonnalagadda et al., 2020; Karimi et al., 2020; Kawachi et al., 2020; Kim et al., 2020; Mollica et al., 2020; Qian et al., 2020; Song et al., 2020; Yamashita et al., 2020; Zhu et al., 2020). The development of microfluidic devices for oncoimmunology applications based on the use of drug combinations is only beginning to emerge. One of the major challenges to deal with is the precise architectural optimization of the chip devoted to these purposes. Indeed, the evaluation of multiple factors on a single chip implies an adjunctive difficult level, not only in the chip fabrication, but also in choosing the bio-architectural components of the chip. For instance, if one plans to construct a complex 3D chip this should not compromise the control easiness of the cells within it. In this regard, an example comes from an elegant report where Caco-2 colorectal adenocarcinoma cell lines were co-cultured on an OOC platform with hepatic HepG2 cells acting as an artificial liver producing metabolites. This microfluidic system has been successfully employed to evaluate and quantitate the effects of Irinotecan and Temozolomide, alone or in combination (Jie et al., 2017). Noticeably, a superior ability of the double condition to induce apoptotic cell death of tumor cells has been showed. This encouraging study suggests that a chip-based system, when properly designed and loaded with the right cell subsets can be effectively exploited to follow unexpected drug combination effects. Other similar findings come from a simple CC system optimized for use of two anticancer agents at different drug concentrations. Here, PC3 tumor cells have been loaded with Doxorubicin or Mitoxantrone with TRAIL (Kim et al., 2012; Desyatnik et al., 2019).

A significant step forward is the optimization of OOC for evaluating drug combinations by taking into account the role of immune cells to provide a more complete and faithful scenario. OOC platforms have been developed employing immune cells in toto and cancer cells resuspended in Matrigel adapted for the exposure to two different drugs acting alone or in synergy. This OOC design has allowed to demonstrate the superior anticancer activity of the DNA demethylating agent Decitabine (DAC) when administered with Interferon-α (IFN-α) against melanoma in a competitive assay (Lucarini et al., 2017). In detail, this device is composed by a central chamber used to load immune cells (PBMC from healthy donors) and two side channels for loading Matrigel-resuspended human A375 metastatic melanoma cells with or without DAC and IFN-α, alone or in combination. Fluorescence microscopy evaluation and time-lapse experiments have shown that immune cells preferentially migrate toward the A375 cells containing both drugs rather than each single agent (DAC or IFN-α), or no drugs. This approached mirrored in vivo studies performed in melanoma-bearing mice, where combined DAC and IFN treatment promoted immune cell infiltration, limiting tumor progression (Lucarini et al., 2017). The same OOC device was employed to evaluate the behavior of specific immune cell subsets, i.e., dendritic cells (DC) toward tumor cells undergoing combined therapy. Here, IFN-α-conditioned DC (IFN-DC) were monitored within 3D tumor spaces for their ability to infiltrate and engulf collagen-embedded SW620 colorectal cancer cells treated with a combination of drugs (Romidepsin and IFN-α) with respect to untreated (Parlato et al., 2017). We hypothesize an OOC scenario in which DAC induces cancer cells to undergo apoptosis, with a concomitant release of apoptosis-dependent chemokines or via the activation of pathways strictly related to cell death such as the FPR1/Anxa1 axis (Vacchelli et al., 2015). In parallel, IFN activates immune cells as they infiltrate the matrigel microenvironment. This, in turn, favors immune cells to interact with A375 cells and to complete their tumor cell killing activity, leading to further release of apoptotic factors. On the other hand, in the opposite side chamber containing the single drug or saline, immune cells fail to infiltrate the melanoma microenvironment. Overall, this system can be regarded as an OOC provided with two functional TMEs, each responding to local environmental stimuli represented by direct effects of the drugs and by subsequent infiltration of immune cells (Figure 7). Of note, this dual competitive condition chip can is an useful tool to evaluate how these differently shaping TMEs will evolve in function of the immune cells behavior. Indeed, the preferential migration of the immune cells toward one of the two conditions is the peculiar effect noted in this particular device, and is strictly dependent on type of drugs cancer cells are exposed to. To this extent, tumor cells can be elegantly monitored by using this specific dual condition chip to study the cellular dynamics of some drugs endowed with cell death-induction activity, compared to an internal control (Figure 7).