On one hand, high permeability of the tumor vessels and a lack of functional lymphatic vessels results in the EPR effect, driving NP extravasation; on the other hand, these phenomena lead to high IFP, limiting NP extravasation. In normal tissues, IFP is around 0 mm Hg; whereas tumors, exhibiting interstitial hypertension, have an IFP almost identical to the microvascular pressure (with a range of 10–40 mm Hg) [66–68]. High IFP limits convection of NP, paradoxically promoting passive diffusion [69]. Diffusion is a much slower transvascular process than convection, especially for the transport of large NP [14]. Moreover, stroma cells compress intratumoral blood and lymphatic vessels, which consequently impairs blood flow, leading to blood stasis, loss of function, and further inhibition of NP penetration [70]. The vascular abnormalities can also cause hypoxia and acidosis. Hypoxia renders tumor cells resistant to both cytotoxic drugs and radiation, while also inducing genetic instability and selecting for more malignant tumor cells with potentially metastatic properties [59, 71]. Finally, because of the steep drop in IFP on the edge of tumors, intratumoral fluid can escape from the tumor periphery into the surrounding tissue, expulsing therapeutic NP, and also excreting growth factors (e.g., VEGF-A, PDGF-C) to facilitate tumor progression [14]. Altogether, the high IFP and abnormal vasculature pose a formidable barrier to both the delivery and efficacy of nanodrugs.

Pericytes are a ubiquitous part of the TME [54, 72]. They were first identified in 1923 and named based on their function as a major constituent of mural cells lining against microvessels [73]. Although no specific molecular marker has been identified specifically to pericytes, alpha smooth muscle actin (α-SMA), PDGFR-β, NG2 proteoglycan, RGS5, and XlacZ4 are commonly used [54, 73, 74]. Signaling pathways implicated in the development of pericytes and their interactions with endothelial cells have recently been reviewed in detail [54]. Briefly, pericytes recruitment involves multiple pathways in a tumor type-specific manner. They are recruited mainly by endothelial cells through PDGF-BB/PDGFR-β signaling. Alternative recruitment signaling includes HB-EGF, pericyte-expressed EGFR and SDF-1α/CXCR4 [75–77]. VEGF-A, a potent mediator of endothelial sprouting and neovascularization, acts as a negative regulator of pericyte function and vessel maturation [78]. Therefore, VEGF-A and PDGF-BB coordinately regulate pericyte coverage. Ablation of pericytes by anti-PDGF antibody or VEGF has been reported to increase vascular tortuosity and tumor growth in low PDGF-BB tumor models [79, 80]. This is paradoxical, since one would expect that increased leakiness of blood vessel with low-pericyte coverage would severely facilitate NP extravasation and inhibition tumor cell proliferation.

The relationship between pericyte coverage and NP extravasation has been investigated in detail by various groups [1, 81–83]. Different from the initial assumption that all tumor microvessels have low and loose pericyte coverage [54], emerging evidence has demonstrated heterogeneous pericyte coverage within in one single tumor or with regards to different tumor types. By defining pericytes as α-SMA positive cells attached to endothelial cells, Kano et al. has classified malignant tumors into high pericyte coverage and low-pericyte coverage subtypes [81, 82]. They further established that more coverage relates to a worse prognosis and more fibrotic interstitium for pancreatic, diffuse-type gastric cancer, clear cell renal cell carcinomas, and glioblastoma (60–70 % coverage), when compared to low-coverage cancers with a better prognosis such as colon cancer and ovarian cancer (10–20 % coverage) [73]. Disparate intratumoral NP transport in response to cytokine-mediated modification of pericyte coverage has been observed in these two types of tumors in a series of work by Kano et al. [73, 81–83]. Murine colon cancer CT26 is an example of low-pericyte coverage tumor, while the BxPC3 pancreatic model has been characterized by hypovasculature with more than 70 % pericyte (Fig. 3). They compared the effects of three types of kinase inhibitors, including TGF-β inhibitor (LY364947), PDGF-B signaling inhibitor (imatinib), and VEGF inhibitor (Sorafenib) on extravasation of a modeled NP, 2 MDa dextran, and a liposomal formulation, Doxil, on CT26 and BxPC3. By using the BxPC3 model, they are able to show that the TGF-β inhibitor can improve 2 MDa dextran and Doxil penetration, leading to enhanced tumor inhibition (Fig. 3). This is due to the fact that low-dose TGF-β inhibitor can block pericyte proliferation without affecting the function of endothelial cells and tumor cells. Consistent with this finding, various types of TGFβ inhibitors, including small kinase inhibitor and its nanoformulation (Fig. 4), siRNA and antibodies have been shown to decrease pericyte coverage. These inhibitors increase vessel leakiness and improve the intratumoral penetration of sub-100 nm NP, including PEI-PEG-coated MSNP, liposome, and polymeric micelle in other high pericyte coverage tumors, such as diffuse-type gastric cancer and 4T1 breast cancer models [82–86]. This finding enabled delivery optimization of various contrasting media. Two recent studies have indicated that TGF-β knockdown can improve MRI contrast [87, 88]. In contrast to the finding in the BxPC3 model, TGF-β inhibitors cannot increase particle penetration and improve therapeutic outcome in CT26, since the pericyte coverage was too low to achieve any additional effect. Interestingly, the VEGF inhibitor, Sorafenib, increased extravasation of 2 MDa dextran in the CT26 model (Fig. 3). Inhibition of VEGF-A can efficiently diminish nonfunctional microvessels with low-pericyte coverage while increase pericyte coverage in the functional microvessels. Thus, the tumor vasculature is “normalized.” The increased pericyte coverage and tumor vasculature normalization were also observed by McDonald’s group [89]. Recent research by Jain’s group indicated that normalization of tumor blood vessels not only improves small molecule-based chemotherapy, but also facilitates the delivery of nanomedicine with smaller sizes [90]. Therefore, tumor vessel normalization is one explanation for enhanced NP penetration after treatment with a VEGF inhibitor in a CT26 tumor model. The combination of a VEGF inhibitor with Doxil can synergistically inhibit CT26 tumor growth. Thus, the relationship between pericyte coverage and NP extravasation varies with regard to the original pericyte coverage, blood vessel stabilization, and extracellular content. Pericyte coverage is an indispensable factor for vessel stabilization and maturation. Neither leaky, unmatured blood vessels with little coverage, nor over-matured vessels with abundant pericyte coverage are suitable for NP delivery. Moreover, pericyte coverage is just one of many factors that influence the intratumoral transport of NP. We need to take the other barriers into consideration when proposing strategies for the improvement of NP delivery.

The BM is a specialized form of ECM that functions as a scaffold for endothelial and mural cells. The main components of the BM are laminin and type IV collagen, which form distinct sheet-like dispositions linked together by nidogen and heparin sulfates [56, 60]. In normal tissues, more than 99 % of blood vessels are covered with a thin layer of BM. It supports the architecture of the blood vessels and regulates vessel development through gradual secretion of pro-angiogenesis and proinflammatory cytokines, such as TGF-β and TNFα [36]. In contrast to normal blood vessel BM, the BM of tumor microvessels is primarily continuous but conspicuously abnormal [36]. Heterogeneous BM morphologies exist in different regions of the same tumor or different tumor types. The first type of BM is characterized by a loose association with endothelial cells. Spontaneous pancreatic islet cell tumors in RIP-Tag2 mice, MCa-IV breast carcinomas, and Lewis Lung carcinomas belong to this type of tumors [57]. Murine lung cancer 3LL and pancreatic cancer BxPC3 are marked by a second type of BM, with a distribution of brighter collagen nodules condensely overlapped with the capillary. The third type of BM can be observed in the 4T1 model. This model has a larger collagen content, which is completely dissociated from blood vessels. Different from the interstitial matrix, another category of ECM, BM does not induce elevated interstitial pressure, yet functions as a sieve to modulate extravasation of free drug and NP from capillaries into the TME.

Extravasation of 1 nm doxorubicin (DOX), 50 nm macromolecule FITC-tagged dextran and 80 nm PEGylated liposomes were evaluated on the BM/vessel overlapped 3LL model and the BM/vessel dissociated model 4T1. Results indicated that the extravasation pattern of small molecules (including DOX and dextran) were comparable, suggesting that vascular collagen could only modulate the transport of small molecules to a limited extent [57]. However, the extravasation of liposomes was significantly different between these two types of tumors. Extravasation only occurred from the vessels that were not tightly covered by collagen type IV. An in vitro collagen sleeve model was further developed to mimic collagen surrounding capillaries [57, 60]. Collagen sleeve thickness, which was modeled by changing the number of collagen fiber layers and the size of collagen mesh (with 50–200 nm openings) were evaluated to study their effects on passive diffusion behavior. Since the molecular size of DOX is substantially smaller than any opening in the mesh, both thickness and mesh size failed to provide any resistance. However, diffusion of particles with size larger than 100 nm could be severely impeded by mesh size and thickness. Therefore, the collagen type IV density, mesh size, number of layers, and association with vessels by itself could potentially be a biophysical barrier for limiting drug extravasation and therapeutic efficacy.

Angiogenesis of blood vessels requires degradation of collagen IV by recruiting metalloproteases MMP2 and MMP9, providing a transient niche with a leaky tumor vasculature, loose and thin BM, that is beneficial for NP delivery [91, 92]. However, this transient disruption of contact between endothelial cells and vessel BM leads to endothelial apoptosis and the formation of collagen fragments that antagonize angiogenesis. On the other hand, the residual nondegraded collagen IV accumulates during repeated remodeling, resulting in multiple distinct layers of BM. During the formation of multiple layers, Collagen density increases while mesh size decreases, generating a more resistant pattern for later NP penetration. Furthermore, the underlying cells can regenerate along the surviving BM that acts as a template or scaffold for generating axons, which renders NP diffusion even harder [36].

In conclusion, BM remodeling is a complicated and controversial procedure controlled by angiogenesis. In addition to digesting the BM via intravenous dosing of collagen IV degradation enzyme, closely monitoring angiogenesis process and dosing NP at the optimal interval is required to improve NP extravasation.

The EPR effect improves NP accumulation in tumor microvessels, whereas the tortuous vessel structure, compressed diameter, deficient function, high interstitial pressure, abnormal pericytes, and BM coverage limit NP extravasation from blood vessels into the interstitial space. Therefore, tumor blood vessels are considered as potential targets to improve the therapeutic potential of nano-formulations. One strategy is to remodel tumor blood vessels to a leakier state by decreasing pericyte coverage, BM thickness, or by reducing interstitial pressure to facilitate convection. TGF-β receptor antagonists (including small molecule kinase inhibitors and siRNA) were the most frequently used therapeutic agents to inhibit pericyte recruitment and BM activation [109]. The combination of TGF-β antagonists with NP has shown enhanced particle diffusion and promising therapeutic outcome. TGF-β inhibitors were also found to lower interstitial hypertension. Decrease in IFP instantly increases vessel permeation. In addition to TGF-β inhibitors, VEGF inhibitors such as bevacizumab or anti-VEGF antibody can significantly decrease IFP. A monoclonal antibody against VEGF reduced glioblastoma IFP by more than 70 % [58, 110]. Similar results have been demonstrated elsewhere in other types of cancer [111, 112]. Tumor IFP can also be lowered by using PDGF antagonists [58]. Tumors suitable for this type of treatment, such as pancreatic cancer (APC) and 3LL murine lung cancer, usually have hypovasculature, compressed vessels, extremely high IFP, thick pericyte coverage, and BM coating.

Another strategy is the so-called normalization of blood vessel (Fig. 7) [34, 71, 90]. This treatment type is suitable for tumors marked by hypervasculature with tortuous structure and less pericyte and ECM coverage. VEGF inhibitors decrease IFP and facilitate particle perfusion. Moreover, VEGF blockage (for e.g., using of VEGF receptor-2 blocking antibody DC101) prunes immature vessels, facilitates the recruitment of pericytes, decreases vessel density and diameter, and remodels the vasculature to more closely resemble the structure of normal vessels (Fig. 7a) [34, 90]. Agents with indirect anti-angiogenic effects, such as trastuzumab, can also lead to vascular normalization. A recent study indicated that transient vessel normalization can improve the performance of small anticancer molecule reagents [71]. Vessel normalization might compromise the transvascular transport of large NP (>100 nm) due to the decrease in pore size, but recent research indicates that it can improve the permeability of small hard NP (12 nm quantum dots) and soft NP (50 nm dextran) (Fig. 7b) [82, 90]. Strategies for blood vessel remodeling have to be adapted, based on tumor structure and particle size. In addition, radiotherapy and hyperthermia conditioning can also lead to transient leakiness of blood vessel and thus improve the intratumoral delivery of NP [113, 114].

The ECM, particularly the collagen and glycosaminoglycan content, limits NP diffusion. To improve drug penetration, a common strategy is to degrade these components and increase the accessibility of the diffusing particles. In addition to hyaluronidase and collagenase mentioned in previous sections, matrix MMP-1 and MMP-8 are proteases frequently used to decrease the level of tumor glycoaminoglycans and improve convection [14, 115].

Besides remodeling of the TME, particle size also plays an important role to enable high-level NP penetration into tumor elements. The smaller the particles the better the transport. Notably, free drugs with smaller sizes can diffuse more rapidly than NP. However, small molecules not only distribute to normal tissue inducing adverse effects, but also fail to be trapped in the tumor tissue for optimized efficacy. Therefore, the size of NP needs to be optimized for each tumor and its metastasis sites Using dextran of various molecular weights in a FaDu tumor model, variable distribution relative to molecular weight has been demonstrated [73, 116]. In this study, 3.3 kDa dextran resembling small molecule drugs entered all tumor tissues quickly. 70 kDa dextran gradually extravasated the blood vessels into the ECM, while 2 MDa dextran remained in the vascular lumen. Polymeric micelles are one kind of NP used widely to delivery hydrophobic chemotherapy drugs. In another study, Cabral and Kataoka et al. prepared a series of micellar nanomedicines (micelle DACHPt) with a diameter ranging from 30 to 100 nm. They found that penetration of NP decreased significantly upon increasing the particle size. Only small particles (30 nm) could penetrate the poorly permeable pancreatic cancer model, BxPC3, and caused promising therapeutic effect (Fig. 8) [85]. In addition, Pain’s work using PEGylated quantum dots further inferred that diffusion of NP with smaller sizes (10–20 nm) can be increased after vasculature normalization, similar to free drugs. However, particles around 100 nm cannot achieve a similar effect [90]. These observations emphasize the importance of tailoring the diameter of NP products, even those with a diameter less than 100 nm.

In addition to particle size, the shape and surface charge of therapeutic NP also plays a key role in extravasation and interstitial transport. For example, cationic particles are more likely to target endothelial cells and exhibit a higher vascular permeability compared to its anionic and neutral counterpart [117]. While extravasated into the interstitial space, cationic particles can aggregate with negatively charged hyaluronan, and anionic particles can aggregate with positively charged collagen. Therefore, neutral particles diffuse faster and distributed more homogeneously inside the tumor interstitial place [118]. As for the influence of particle shape, it is reported that linear, semi-flexible macromolecules can diffuse more rapidly in the ECM than spherical particles with similar size [119]

Though condensed ECM functions as a barrier for NP diffusion, it can also be taken advantage of to improve the efficacy of nanomedicine. Based on the acidic pH, reduced oxygen pressure, and enzyme-rich properties of the TME, NP can be constructed to the advantage of these properties. For example, a multistage quantum dots embedded gelatin NP was engineered to degrade gradually and release 10 nm small quantum dots in response to MMPs, zinc-dependent endopeptidases that are abundant in the ECM [120]. Drug-polymer conjugates have also been designed with a cleavable linker, which is the substrate of MMP and fibroblast activation protein (FAP), a gelatinase that is expressed on TAFs [121–124]. Upon penetration of the ECM, free drug is released upon linker cleavage and diffuses more rapidly than the NP in the interstitial space for better therapeutic outcome. Based on this concept, pH sensitive particles have been designed to trigger the release of free drug from the cargo within tumor elements [125, 126]. In addition, external stimulants, such as electric pulses, magnetic field, ultrasound, heat, and light can also be used to improve NP penetration and free drug release [127–131].