Protein MRI contrast agent with unprecedented metal selectivity and sensitivity for liver cancer imaging

Significance

Primary and metastatic liver cancers that are associated with high mortality rates and poor treatment responses are only diagnosed at late stages, due to the lack of highly sensitive contrast agents and robust imaging methodologies. We have developed a protein MRI contrast agent (ProCA32) by engineering high-affinity Gd³⁺-binding pockets in rat and human α-parvalbumin. ProCA32 can function as both a T₁- and T₂-weighted contrast agent, which enables noninvasive detection of early-stage micrometastatic liver tumors with sizes as small as 0.24 mm using T₁- and T₂-weighted or T₂/T₁ ratio MRI. Our protein-based MRI contrast agents and imaging methodology are expected to provide robust results for the early detection of liver cancer as well as other liver diseases.

Abstract

With available MRI techniques, primary and metastatic liver cancers that are associated with high mortality rates and poor treatment responses are only diagnosed at late stages, due to the lack of highly sensitive contrast agents without Gd³⁺ toxicity. We have developed a protein contrast agent (ProCA32) that exhibits high stability for Gd³⁺ and a 10¹¹-fold greater selectivity for Gd³⁺ over Zn²⁺ compared with existing contrast agents. ProCA32, modified from parvalbumin, possesses high relaxivities (r₁/r₂: 66.8 mmol⁻¹⋅s⁻¹/89.2 mmol⁻¹⋅s⁻¹ per particle). Using T₁- and T₂-weighted, as well as T₂/T₁ ratio imaging, we have achieved, for the first time (to our knowledge), robust MRI detection of early liver metastases as small as ∼0.24 mm in diameter, much smaller than the current detection limit of 10–20 mm. Furthermore, ProCA32 exhibits appropriate in vivo preference for liver sinusoidal spaces and pharmacokinetics for high-quality imaging. ProCA32 will be invaluable for noninvasive early detection of primary and metastatic liver cancers as well as for monitoring treatment and guiding therapeutic interventions, including drug delivery.

Tumor metastasis is the main cause of nearly all human cancer-related deaths. Liver is a common site for metastases of a variety of cancers, including melanoma, breast, pancreatic, and colon cancers (1, 2). For example, uveal melanoma, the most common primary intraocular tumor, has a 40% risk of metastasizing to the liver within 10 y of diagnosis of the primary tumor. Hepatic metastases, which occur in 95% of patients with uveal melanoma metastasis, result in death in almost all cases. This high death rate is related to the recognition of liver metastasis at a late clinical stage (at stage II or later in the TNM system) in which the metastatic uveal melanoma is resistant to currently available systemic chemotherapies (3, 4). The liver may also give rise to primary tumors such as hepatocellular carcinoma (HCC), which is the most common primary malignancy worldwide (5). However, currently there is no reliable way to detect primary liver cancer and hepatic metastases at early stages with high sensitivity and specificity.

MRI is a widely used clinical imaging modality that provides exquisite soft-tissue contrast without using ionizing radiation (6, 7). More than 35% of MRI scans use MRI contrast agents, particularly paramagnetic gadolinium (Gd³⁺)-based contrast agents, which shorten T₁ and lead to an increase in MRI intensity (8). All clinically approved Gd³⁺-containing contrast agents are based on small chelators with relaxivity (r₁ and r₂) values around 4–6 mM⁻¹⋅s⁻¹. Their low relaxivities severely limit the sensitivity of current MR imaging methods with respect to detection of lesions and treatment planning and monitoring. Repeat MRI scans are frequently requested for patients with ambiguous small lesions and are used as a routine follow-up modality for high-risk patients, but they are clinically suboptimal and cost ineffective and may lead to mistreatment and metal toxicity related to the high injection doses required. Imaging of liver lesions is particularly challenging due to rapid liver excretion of contrast agents. Detection of small liver lesions at early disease stages requires contrast agents that have a significantly improved sensitivity and a large dynamic range with reduced background water signal in tissue, as well as appropriate liver distributions and retention times for high-quality imaging. The most commonly used clinically approved T₁-weighted liver contrast agents, gadolinium–ethoxybenzyl–diethylenetriamine pentaacetic acid (Gd-EOB-DTPA) (Eovist, United States; Primovist, Europe; Bayer) and gadobenate dimeglumine (Gd-BOPTA) (Multihance; Bracco), have 50% and 5% hepatocyte uptake, respectively (9). They cannot detect early-stage small liver tumors and metastases (size, <0.5 mm), differentiate dysplastic nodules from HCC, or discriminate tumor thrombosis from platelet-fibrin thrombosis. Thus, there are pressing unmet needs for innovative MRI contrast agents with markedly improved sensitivity, metal stability, and pharmacokinetics, as well as more robust imaging methodologies to enable the earlier detection of small tumors, thereby leading to earlier diagnosis and more efficacious treatments as well as to monitor disease progression of high-risk patients.

Herein, we report the development of a protein-based MRI contrast agent, ProCA32, which has both unprecedented metal selectivity and relaxivities in r₁ and r₂, and appropriate liver retention time and distribution. It enables us to noninvasively detect micrometastatic liver tumors at early stages using T₁-weighted, T₂-weighted, or T₂/T₁ imaging as confirmed with histological analysis. Our developed protein MRI contrast agents and imaging methodology are expected to be robust in the early detection of primary liver cancer, liver metastasis, and other liver diseases, such as cirrhosis and liver fibrosis, and in guiding the treatment of these diseases.

Results

Design of ProCA32 with High Gd³⁺ Affinity and Metal Selectivity for Liver Imaging.

One of the prerequisites for in vivo contrast agents is to have required metal stability, especially kinetic stability with high selectivity for Gd³⁺ over physiological metal ions. We previously reported that designing a Gd³⁺-binding site into a scaffold protein CD2, named ProCA1, significantly improves relaxivity by increasing water number q (q = 1.8) and tuning correlation time. Although it enables molecular imaging of biomarkers such as HER2, ProCA1 has relatively low Gd³⁺ binding affinity (K_d = 8.7 × 10⁻¹³ M), compared with clinical MRI contrast agents [1.9 × 10⁻²¹ M for DTPA and 1.4 × 10⁻¹⁷ M for DTPA–bis(methylamide) (DTPA-BMA)], even though its metal selectivities for Gd³⁺ over Mg²⁺/Ca²⁺/Zn²⁺ are better than DTPA-BMA and Gd-DTPA (10–13). Based on our extensive studies on metal binding and the selectivity of various types of proteins with high coordination properties (10, 14), we hypothesized that strong Gd³⁺ binding affinity and metal selectivity for Gd³⁺ over physiological metal ions can be achieved by modifying the electrostatic interaction of well-coupled EF-hand calcium-binding proteins. We further reasoned that improved in vivo sensitivity of contrast agents can be achieved by increasing both r₁ and r₂. Proteins have sufficient secondary and outer hydration shells around the metal-binding site, which contribute significantly to relaxivity. It is possible to increase both r₁ and r₂ by controlling the rotational correlation time, τ_R, at around 2–6 ns at clinical field strengths (Fig. S1). One important benefit of combining T₁- and T₂-weighted imaging is improvement in contrast within the desired tissue or organ as a result of background signal suppression, leading to increased detection sensitivity (Fig. 1A and Fig. S2). Such an imaging strategy is not practical with clinical MRI contrast agents due to their low r₂. In fact, although the currently used clinical contrast agents have the desired r₂/r₁ ratio of 1.2, their low r₂ of 5 mM⁻¹⋅s⁻¹ cannot provide significant T₂ contrast in vivo at physiologically permitted injection doses and feasible echo times (Fig. S2).

Fig. 1.

Design and stability studies of ProCA32 and ProCA32-P40. (A) Cartoon illustration of liver tumor detection by ProCA32-P40. ProCA32-P40 has high r₁ and r₂ relaxivities. Less than 1 h after injection of ProCA32-P40, it is mainly distributed in the liver sinusoidal space and very little ProCA32-P40 are accumulated in the tumor. T₁-, T₂-weighted, and T₂/T₁ ratio imaging by MRI were used for imaging liver tumor. (B) The model structure of ProCA32-P40. ProCA32 was generated from parvalbumin by incorporating the S56D and F103W mutations. ProCA32-P40 was further developed by covalently linking PEG to ProCA32. The model structure of ProCA32-P40 is generated based on the structure of α-parvalbumin (Protein Data Bank ID code 1RWY). (C) Determination of Tb³⁺ (black; n = 3), Ca²⁺ (blue; n = 3), and Mg²⁺ (red; n = 3) affinities for parvalbumin mutants (○, ProCA32; ●, parvalbumin with F103W single mutation) using fluorescence spectroscopy. (Table) Metal stabilities and selectivities of DTPA (*from National Institute of Standards and Technology Standard Reference Database 46), ProCA30 (n = 3), and ProCA32 (n = 3).

Fig. 1B shows the model structure of our developed protein contrast agent (ProCA32), which was created by introducing one more negatively charged ligand S56D in the EF-hand motif 1 of rat or human α-parvalbumin to tune Gd³⁺ binding affinity and metal selectivity. Parvalbumin (ProCA30) was chosen because of its strong calcium affinity (K_d = 8.35 ± 0.29 × 10⁻⁹ M; Fig. 1) and stability. An F103W mutation was introduced to allow determination of metal binding affinity by luminescence resonance energy transfer (LRET) and water number by Tb³⁺ luminescence life time decay. PEGylation of ProCA32 by TMS(PEG)12 (named as ProCA32-P40) allows us to increase blood retention time, liver preference, solubility, and stability of the designed protein (15).

Unprecedented Metal Selectivity, Stability, and Low Toxicity of ProCA32.

Nephrogenic systemic fibrosis (NSF) associated with free Gd³⁺ toxicity of clinically approved contrast agents is reported to correlate with their transmetallation (16). We first examined metal binding affinity of bacterially expressed, purified, and PEGylated ProCA32 using several spectroscopic methods (SI Materials and Methods; Fig. 1C and Fig. S3 A–E). A Tb³⁺-DTPA buffer system (Fig. S3 A and B) was used to control free Tb³⁺ concentration. The dissociation constant (K_d1 = 8.9 × 10⁻²² and K_d2 = 1.4 × 10⁻²¹ M by the Adair equation) between Tb³⁺ and ProCA32 was then determined by Tb³⁺-LRET. The Gd³⁺ binding affinity (2.79 × 10⁻²² M) determined using competition assays (Fig. S3C) is comparable to that of DTPA (1.86 × 10⁻²¹ M) (Fig. 1). Importantly, ProCA32 exhibited 10¹¹- to 10²-fold greater metal selectivity than DTPA for physiological metal ions, such as Zn²⁺, Ca²⁺, and Mg²⁺ (Fig. 1 and Fig. S3 D and E). Additional transmetallation experiments further supported the conclusion that ProCA32 has significantly higher kinetic stability and metal selectivity for Gd³⁺ over Zn²⁺ than those of Omniscan, Eovist, and Magnevist (Fig. S3F). To our knowledge, ProCA32 has the greatest metal selectivity [log (K_Gd/K_Zn)] among all reported Gd³⁺-based contrast agents (Fig. 1). Furthermore, ProCA32 also has a strong proteolytic stability, as revealed by its structural integrity for greater than 12 d of incubation in mouse serum (Fig. S3G). In addition, ProCA32 did not exhibit any cytotoxicity (Fig. S4A), acute toxicity to organs (Fig. S4B and Table S1), and had reduced free Gd³⁺ accumulation in mouse tissues compared with Gd-DTPA at its clinical injection dosage or GdCl₃ 14 d after i.v. injection (Fig. S4C).

Unique High r₁ and r₂ Relaxivities of ProCA32.

We measured longitudinal and transverse relaxivities (r₁ and r₂) of the designed protein (Fig. 2). ProCA32, with or without PEGylation, exhibited similar per Gd r₁ relaxivities of 33.4 mM⁻¹⋅s⁻¹ at 37 °C and 1.4 T, which were ∼10-fold greater than that of clinically approved contrast agents at 1.4 T (Fig. 2C). The per-Gd r₂ of ProCA32 was 44.6 mM⁻¹⋅s⁻¹ at 37 °C and 1.4 T, which was at least five times higher than that of clinically approved MRI contrast agents, such as Eovist. Remarkably, ProCA32 retains r₁ of 21.9 mM⁻¹⋅s⁻¹ at room temperature and 4.7 T, and r₂ is 56.9 mM⁻¹⋅s⁻¹; r₁ is 18 mM⁻¹⋅s⁻¹ at 7 T (Table S2), which is not predicted by the usual relaxation equations (17, 18). In addition, 5 µM ProCA32 is able to generate a level of contrast similar to 50 µM Eovist with T₁-weighted spin echo and at 4.7 T (Fig. S5A). ProCA32 is able to create a significant T₂-weighted contrast at 75 µM, whereas 150 µM Eovist failed to produce any T₂ effect (Fig. S5B).

Fig. 2.

Relaxivities (r₁ and r₂) and water number of protein-based MRI contrast agents. (A) Changes in R₁ relaxation rate over various concentrations of Gd³⁺ in ProCA32 and Eovist at 1.5 T and 37 °C. (B) Changes in R₂ relaxation rate over various concentrations of Gd³⁺ in ProCA32 and Eovist at 1.5 T and 37 °C (the images in A and B are representative of three independent experiments). (C) Per-Gd r₁ and r₂ relaxivities of Gd-DTPA (n = 3), Eovist (n = 3), ProCA32 (n = 3), and ProCA32-P40 (n = 3) at 1.5 T and 37 °C. Data are expressed as mean ± SD. (D) Water number of Gd-DTPA, Eovist, ProCA30 (n = 3), ProCA32 (n = 3), and ProCA32-P40 (n = 3). Data are expressed as mean ± SD. (E) NMRD profile of ProCA32-P40 measured at 37 °C and 25 °C. ν ¹H stands for proton Larmor frequency. (F) Plot of ln(1/T_2r) versus 1,000/T for water exchange of ProCA32-P40 measured by variable-temperature ¹⁷O NMR. Conditions: [ProCA32-P40] = 10 mM loaded with 20 mM Gd³⁺; pH 7.0; curve fit: ΔH^‡ = 25.6 kJ⋅mol⁻¹, A = 2.3 × 10⁹, k_ex²⁹⁸ = 5.4 × 10⁶ s⁻¹.

Luminescent lifetime measurements of the number of first coordination water molecules on Tb³⁺ yields a value of 0.5 per Gd (Fig. 2D) with two Gd³⁺ per protein. This result is consistent with the addition of a chelating residue in the S56D mutation saturating the first coordination sphere of one Gd³⁺ site (EF-hand 1). Based on differential metal-binding capabilities of the two metal-binding sites (K_d1 = 8.9 × 10⁻²² and K_d2 = 1.4 × 10⁻²¹ M for Tb³⁺) and their strong cooperativity as well as similar pattern of nuclear magnetic relaxation dispersion (NMRD) profiles for metal complex ratios of 1:1 and 1:2, we concluded that Gd³⁺ initially binds to both sites partially with a strong preference for EF-hand 1 due to its stronger metal binding affinity.

The relaxation dispersion profiles shown in Fig. 2E are characteristic of Gd³⁺ coordinated to a protein. The rise in the relaxation rate constant derives from the magnetic field dependence of the electron spin-lattice relaxation rate, which decreases with increasing magnetic field; as a consequence, the effective correlation time for the electron–nuclear coupling, which includes correlation time contributions from rotation, chemical exchange, and electron relaxation, increases with increasing field. At higher field strengths, the nuclear spin-lattice relaxation rate constant generally drops to a very low value but the transverse relaxation rate becomes larger until the effective correlation time is limited by either rotational mobility of the metal center or the water exchange rate, which was measured in ProCA32-P40 using ¹⁷O NMR (k_ex= 5.44 × 10⁶ s⁻¹; Fig. 2F). The present data are unusual because the high-field r₁ remains larger than expected based on approximate models such as Solomon–Bloembergen–Morgan theory for first coordination sphere water exchange.

Several sources for the large high-field r₁ are possible: (i) unbound paramagnetic centers in solution (eliminated by chemical analysis), (ii) dipolar coupling of non-metal-protein–bound water molecules to the paramagnetic centers (19), and (iii) translational diffusion of water close to the paramagnetic centers in the protein interface region. Water molecules rigidly bound to the protein may contribute to relaxation by dipolar coupling to the metal centers; however, the contribution will disperse with the rotational correlation time of the protein. To maintain high-field relaxivity, there would have to be significant local motion in the bound site providing a shorter correlation time for the coupling (20). If there are many sites and a distribution of local correlation times, the magnetic field dependence may be weak and the high-field r₁ significant. Diffusive motion of water adjacent to the paramagnetic centers also contributes to the high-field relaxivity; however, the contribution is expected to be small for an unmodified protein surface because the correlation times are in the range of tens of picoseconds (19, 21). However, modification with PEG may slow this motion and amplify the contribution.

Robust Liver Preference, Significantly Improved Pharmacokinetics, Detection Limit, and Dose Efficiency of ProCA32-P40.

ProCA32-P40 has an elimination half-life time of 2.8 h in mouse compared with the value of ∼20 min for DTPA (Fig. 3A). It provides an excellent MRI enhancement of mouse blood vessel, liver, and kidney in T₁-weighted 3D gradient echo images at 4.7 T (Fig. 3B and Movie S1) even at 50 min after injection, whereas Eovist or Gd-DTPA failed to produce any pronounced enhancement at time points longer than 30 min. At 40–50 min after injection, ProCA32-P40 is mainly present in the bloodstream and extracellular extravascular spaces of liver and kidney (0.13 L/kg of volume distribution) with the highest MRI signal enhancement in these organs, and then returns to baseline at longer time as demonstrated by analyses from MRI and inductively coupled plasma optical emission spectrometry (ICP-OES) (Fig. 3 B–D). Results from both T₁-weighted 3D gradient echo imaging and ICP-OES suggest that ProCA32-P40 is partially excreted through the kidney with great enhancement of the bladder at 50 min (Fig. 3B) and via the liver based on the presence of Gd³⁺ in feces (Fig. S5C). Approximately 60% and 30% of ProCA32-P40 were excreted from kidney and liver, respectively, 4 d after injection (Fig. S5D). Intact ProCA32-P40 was detected by immunoblot in feces (24 h after injection) (Fig. S5E), suggesting that ProCA32-P40 does not break down in liver and kidney and Gd³⁺ and the protein form a complex during excretion. Strikingly, we are able to obtain MRI enhancement of mouse kidney at an injection concentration of 0.8 µmol/kg, 100 times lower than that of clinical injection dosage of Gd-DTPA. Injection of 1.6 µmol/kg Gd-DTPA fails to provide any significant kidney MRI enhancement at the same injection dosage (Fig. S5 F and G).

Fig. 3.

Studies of the pharmacokinetics, biodistribution, and MR imaging capacity of ProCA32-P40. (A) Percentage of injection dose of Gd³⁺ in serum collected after injection of ProCA32-P40 (labeled as ProCA; n = 5), GdCl₃ (n = 5), and Gd-DTPA (n = 5) at different time points measured by ICP-OES at 342 nm. (B) Maximum intensity projection T₁-weighted MRI of mice before (Left) and 50 min after injection of ProCA32-P40 (Right). The enhancement in bladder is indicated by an arrow. (C) MRI signal intensity of liver (n = 5), kidney (n = 5), and blood (n = 5) in mice after injection of ProCA32-P40 and measurement at various time points using ImageJ. (D) Gd³⁺ concentration in blood, liver, kidney, and muscle after injection of ProCA32-P40 for 40–50 min measured by ICP-OES at 342 nm (n = 3). (E) T₁-weighted MRI of the mice liver before (Left Top, Left Bottom) and after injection of ProCA32-P40 (Right Top) or Eovist (Right Bottom). ProCA32-P40 generates a greatly enhanced liver image compared with that of Eovist in T₁-weighted MRI. (F) T₂-weighted MRI of mouse liver sections before (Left Top, Left Bottom) and after injection of ProCA32-P40 (Right Top) or Eovist (Right Bottom). ProCA32-P40 generates decreased liver intensity in T₂-weighted MRI, whereas Eovist does not cause any decrease in T₂ intensity in the same MRI sequence. [Scale bars: 5 mm (B, E, F).]

High-quality liver MR images have been acquired using both T₁- and T₂-weighted sequences. With 0.025 mmol/kg injection dose, ProCA32-P40 exhibits a 78 ± 15% increase in the T₁-weighted signal and a 22 ± 3% decrease in the T₂-weighted signal (Fig. 3 E and F) at 4.7 T, whereas Eovist has a 39 ± 7% signal increase in the T₁-weighted image and fails to provide any change in the liver signal in T₂-weighted MRI. Thus, ProCA32-P40 has significantly improved in vitro and in vivo detection sensitivity and dose efficiency for both T₁- and T₂-weighted MRI compared with Eovist. Furthermore, in contrast to the short imaging window for clinical MRI contrast agents, such as Eovist, ProCA32-P40 enhances the entire liver for at least 3 h, which substantially widens the MRI detection window for high-quality imaging of liver, allowing us to introduce T₁-, T₂-weighted and T₂/T₁ ratio imaging, leading to a further increase in sensitivity. Using fluorescence imaging and antibody staining, as well as ICP-OES to monitor ProCA32-P40 distribution (Fig. S5 H and I) (see Fig. 6H), we further show that ProCA32 largely distributes in the sinusoidal spaces of normal liver with a significantly longer retention time than Eovist or Multihance.

Fig. 6.

Histological confirmation of metastatic uveal melanoma in liver detected by MRI. (A) Immunohistochemical staining of metastatic melanoma in liver using S100 antibody. (B) Bright-field imaging of metastatic melanoma in liver. Melanin in uveal metastatic melanoma exhibits black color. (C and D) Immunofluorescence images of ProCA32-P40 (red) distribution between metastatic melanoma and liver tissue. ProCA32-P40 was mainly distributed in liver tissue. (E) Immunofluorescence images of ProCA32-P40 (red) in metastatic melanoma. ProCA32-P40 has limited distribution in tumor tissue. (F) CD31 (green) staining of blood vessels in metastatic melanoma tissue. (G) Colocalization of ProCA32-P40 (red) and CD31 (green) in metastatic melanoma tissue indicates that the distribution of ProCA32-P40 is restricted to the tumor blood vessel. (H) Immunofluorescence images of ProCA32-P40 (red) in normal liver tissue. ProCA32-P40 largely distributes in the sinusoidal space of normal liver within 1 h after tail vein injection of ProCA32-P40. (I) T₂/T₁ ratio MRI of metastatic melanoma (arrow) in liver after injection of ProCA32-P40. (J) H&E staining of metastatic melanoma (arrow) in the liver tissue. (K) The correlation between MRI and H&E staining. The tumor sizes detected in MRI (x axis) and H&E staining (y axis) show very good correlation (y = 1.04x + 0.05; R² = 0.92). [Scale bars: 100 µm (A–H) and 5 mm (I and J).]

In Vivo T₁-, T₂-Weighted and T₂/T₁ Ratio Imaging of Implanted and Metastatic Liver Tumors.

It is known that most liver diseases, including various types of cancers in the liver, such as HCC and metastases, have altered sinusoidal architecture and blood distribution (22). In addition, we demonstrate that hepatic metastases of uveal melanoma cells have significantly reduced hepatic distribution of both ProCA32-P40 and Eovist due to disruption of the sinusoid vascular network and lack of uptake by uveal melanoma cells (Fig. S5 H–J) (22). Moreover, liver metastases of uveal melanoma have significantly lower ProCA32-P40 distribution than that in surrounding liver tissue as revealed by ICP-OES (Fig. S5K) and fluorescence imaging (Fig. S5I). Such a unique property allows us to test our hypothesis that ProCA32-P40 has an enhanced capacity to detect small lesions, owing to its unique high relaxivities in both r₁ and r₂ and its differential distribution in liver lesions and normal liver tissue using both T₁- and T₂-weighted imaging, and combining them using ratio imaging in both implantation (Fig. 4) and metastasis (Fig. 5) models.

Fig. 4.

Imaging of implanted liver tumor using ProCA32-P40. (A) Cartoon illustration of detection of liver tumors by ProCA32-P40. Because of the different distribution of ProCA32-P40 between liver and tumor, tumors show hypointensity in T₁-weighted MRI and hyperintensity in T₂-weighted MRI. The MRI signal differences between liver and tumor were further improved in T₂/T₁ ratio imaging, which is generated by dividing the intensity of T₂-weighted MRI by the intensity of T₁-weighted MRI in each corresponding pixel. (B and C) T₁-weighted MRI of tumor implanted liver before (B) and after (C) injection of ProCA32-P40. (D and E) T₂-weighted MRI of tumor implanted liver before (D) and after (E) injection of ProCA32-P40. (F) Immunofluorescence staining shows that ProCA32-P40 (green) has a high distribution in liver and a low distribution in tumor. (G) The T₂/T₁ ratio imaging of the tumor implanted liver. Tumor locations in C, E, and G are pointed out by red arrows. (H) Signal-to-noise ratio (SNR) of tumor (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₁-weighted MRI. (I) SNR of tumor (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₂-weighted MRI. (J) SNR of tumor (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₂/T₁ ratio MRI. [Scale bars: 3 mm (B–E and G) and 100 µm (F).]

Fig. 5.

Detection of metastatic uveal melanoma in liver by ProCA32-P40 using T₁-weighted (A and D), T₂-weighted (B and E), and T₂/T₁ ratio imaging (F) using ProCA32-P40. Metastatic uveal melanoma cannot be detected in T₁- (A) or T₂-weighted MRI (B) without injection of ProCA32-P40. Metastatic uveal melanoma (indicated by arrows) with size larger than 0.24 mm (green arrow) can be detected by T₁-weighted (D), T₂-weighted (E), and T₂/T₁ ratio imaging (F) after injection of ProCA32-P40. (C) The tumors detected by MRI were confirmed by H&E staining. (G) SNR of metastatic uveal melanoma (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₁-weighted MRI. (H) SNR of metastatic uveal melanoma (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₂-weighted MRI. (I) SNR of metastatic uveal melanoma (n = 5) and liver (n = 5) after injection of ProCA32-P40 in T₂/T₁ ratio MRI. [Scale bars: 5 mm (A, B, and D–F) and 100 µm (C).]

Fig. 4 shows that the transplanted uveal melanoma in the liver appears brighter in T₂-weighted images due to specific signal reduction in surrounding normal liver tissue after i.v. injection of ProCA32-P40. On the other hand, the transplanted uveal melanoma in the liver appears darker in T₁-weighted imaging due to signal enhancement in surrounding normal liver tissue. Taking the ratio, T₂/T₁ effectively combines the contrasts between the tumor and normal liver as it decreases the liver signal and improves the homogeneity of liver signal.

Using our established mouse ocular melanoma metastasis model (23, 24), we showed that tail vein injection of ProCA32-P40 enables us to observe more than 10 hypointense lesions using T₁-weighted spin echo images (Fig. 5D and Fig. S6) and hyperintense lesions with sizes ranging from 0.24 to 2 mm in T₂-weighted fast spin echo images (Fig. 5E and Fig. S6). In contrast, these liver metastases of uveal melanoma were not detectable either without ProCA32-P40 or with injection of Eovist (Fig. S7). Furthermore, in the metastasis model, use of the T₂/T₁ ratio efficiently suppresses the liver signal, improves the homogeneity of the liver signal, and improves the contrast-to-noise ratio between liver and tumor from 3.58 ± 0.45 in T₁-weighted images and 4.33 ± 1.15 in T₂-weighted images to 5.05 ± 1.36 in T₂/T₁ ratio images (Fig. 5F and Fig. S6).

Histological Validation of Metastatic Lesions in Liver.

Next, we performed detailed histological analysis of the uveal melanoma biomarker S100 (Fig. 6A), bright-field imaging of melanin (Fig. 6B), immunofluorescence staining (Fig. 6 C–H), and H&E staining (Fig. 6J) to confirm the identity of the liver metastatic melanoma tumors detected by MRI. ProCA32-P40 staining in tumors colocalized with CD31 immunostaining of vascular channels, indicating that ProCA32-P40 is restricted to the distribution of blood vessels in the tumor (Fig. 6 E–G). Consistent with the differences in contrast between tumor and liver observed using MRI, immunofluorescence staining confirms that ProCA32-P40 is mainly distributed in the sinusoidal space of the liver and has an extremely low distribution in tumor tissue (Fig. 6 C and D). Such a low distribution in tumor is dramatically different from that of the liver (Fig. 6H). Furthermore, we did a correlation study between H&E staining and MRI. There is an excellent correlation (R² = 0.92) between noninvasive detection by MRI with ProCA32-P40 and H&E histological analysis of all stages of metastatic lesions of uveal melanoma of varying sizes (Fig. 6 I–K). Thus, ProCA32-P40 enables noninvasive early detection of lesions as small as 0.24 mm, or stage I, metastatic liver tumors, which are ∼100-fold smaller than the current detection threshold of >10–20 mm for stage II or later. It is also important to note that the combination of T₁-, T₂-weighted and ratio images significantly increased the confidence for detecting metastases, while reducing detection limits of liver lesions by 100-fold.

Discussion

These results demonstrate the feasibility of developing a protein MRI contrast agent, ProCA32, by protein design and modification with significantly improved r₁ and r₂. ProCA32 has unprecedented metal selectivity and improved blood, liver, and kidney pharmacokinetics. We have shown proof-of-concept that ProCA32 can be used to detect hepatic metastases of uveal melanoma at an early stage with high sensitivity.

To address a critical, unmet medical need, we developed a strategy in our creation of an MRI contrast agent using protein design and modification to increase relaxivity without compromising metal binding affinity/stability, selectivity, and pharmacokinetic properties that are dramatically different from previously reported approaches (25–32). Building on our previous success in improving relaxivity by designing a Gd³⁺-binding site into a scaffold protein CD2 and with a goal of improving metal binding affinity, we report a method to develop the protein contrast agent ProCA32 by converting the natural Ca²⁺-binding protein, parvalbumin, into a Gd³⁺-binding protein that is also PEGylated. We have simultaneously achieved significantly increased longitudinal and transverse relaxivities, Gd³⁺ binding affinity, and in vivo pharmacokinetics required for earlier detection of small lesions. ProCA32 exhibits a Gd³⁺ dissociation constant comparable to that of DTPA of 1.86 × 10⁻²¹ M, whereas wild-type parvalbumin was reported to have a Gd³⁺ affinity of 0.5 × 10⁻¹¹ M (33). ProCA32 achieves metal selectivity for Gd³⁺ over Ca²⁺, Mg²⁺, Zn²⁺, and other physiological metal ions that are >100- to 10¹¹-fold higher than DTPA. ProCA32 exhibits undetectable acute toxicity, cytotoxicity, and strong serum stability and low 14-d residual content in mouse bone. Taken together with its improved dose efficiency, ProCA32 is likely to significantly reduce the NSF risk associated with metal toxicity.

ProCA32 exhibits several unique relaxation properties compared with previously reported contrast agents. ProCA32 has significantly increased r₁ and r₂ with per-Gd relaxivities about 10-fold greater than Eovist at 1.4 T, permitting acquisition of both positively and negatively enhanced liver images at physiologically achievable concentrations in the liver using T₁- and T₂-weighted pulse sequences, respectively. Combining the contrasts in T₁- and T₂-weighted images obtained by using the ratio, ProA32-P40 enables a 100-fold increase in its detection limit for identifying lesions compared with existing agents (Fig. 5). In addition, the injection dosage of ProCA32-P40 at 4.7 and 7 T is comparable with clinically used protein drugs, such as Avastin. Because relaxivity of ProCA32-P40 at medical-related field strength such as 1.5 T is 50% greater than at 4.7 T (Table S2), the required injection dose efficiency of our contrast agents is expected to be increased significantly at clinical field strength. Thus, ProCA32-P40 has great potential for clinical translation.

PEGylation of ProCA32 increases preferential liver distribution and improves the blood retention time to a half-life of 2.8 h in vivo, in addition to enhancing protein stability and improving pharmacokinetics with a resultant increase in dose efficiency. As shown in Fig. 6H, ProCA32-P40 is mainly distributed in the sinusoidal spaces of the liver, and hepatocyte uptake is absent within 1 h after injection. These improved in vivo properties substantially widen the MRI detection window and are essential for achieving high-quality imaging with high resolution. ProCA32-P40 has a high liver retention time with a blood half-life of 2.8 h, which allows the collection of both T₁- and T₂-weighted images and an associated significant change of liver signal after one injection of ProCA32-P40 (Fig. 3). Importantly, our imaging methodologies provide additional verification for signal detection by doubling detection confidence. On the other hand, iron oxide-based nanoparticles, such as super paramagnetic nanoparticles, generate negative contrast in liver T₂- or ${\text{T}}_2^{\ast }$-weighted MRI and accumulate in the sparely distributed Kuppfer cells of the reticuloendothelial system within a few minutes after injection, instead of in hepatocytes or the sinusoidal space.

ProCA32-P40 has the potential to fill a major gap in clinical MRI applications by enabling noninvasive and early detection of primary liver tumors and metastases, improving treatment effectiveness by targeted therapy, and guiding treatment and drug delivery. Because many liver diseases, including liver cancers such as HCC and liver metastasis from breast, brain, and colon cancer, as well as fibrosis and cirrhosis, have altered liver vascular architecture, it is expected that ProCA32-P40 can be widely used for the diagnosis of various types of diseases in the liver, including cirrhosis, fibrosis, hepatitis B virus infection, hepatitis C virus infection, alcoholism, fatty liver disease and other nonalcoholic liver disease. Improved detection sensitivity enables better delineation of tumor margins and may facilitate image-guided interventions and early intervention by providing an enhanced capacity to monitor progression in high-risk patients. To our knowledge, ProCA32 also exhibits the high per-Gd r₁ relaxivity of 18 mM⁻¹⋅s⁻¹ at high magnetic field, 7 T (Table S2) (17). Further understanding of the mechanism underlying such high relaxivities is expected to open a previously unidentified avenue to further broaden MRI applications at high magnetic fields with improved sensitivity and resolution.

Materials and Methods

Detailed descriptions of the expression, purification, PEGylation, and characterization of protein MRI contrast agents are provided in SI Materials and Methods. All animal procedures performed in this study complied with the Association for Research in Vision and Ophthalmology Statement for the Use of Animals in Ophthalmic and Vision Research and complied with an approved animal protocol from the Institutional Animal Care and Use Committee at Georgia State University and the University of Georgia. T₂/T₁ ratio images were generated using the following procedure: (i) T₁- and T₂-weighted images were acquired before and after injection of ProCA32-P40 in mice. (ii) A threshold was chosen to exclude areas of noise in the T₁- and T₂-weighted image. (iii) For each pixel with a signal intensity in the T₁-weighted image above the threshold, the corresponding pixel in the T₂-weighted image was divided by the signal intensity from the T₁-weighted image to generate new T₂/T₁ ratio image. Detailed methods, parameters, and procedures for in vivo MRI, histology analysis, biodistribution, and toxicity studies in the article are described in SI Materials and Methods.