Quantification of contrast agent uptake in the hepatobiliary phase helps to differentiate hepatocellular carcinoma grade

This study aimed to assess the degree of differentiation of hepatocellular carcinoma (HCC) using Gd-EOB-DTPA-assisted magnetic resonance imaging (MRI) with T1 relaxometry. Thirty-three solitary HCC lesions were included in this retrospective study. This study's inclusion criteria were preoperative Gd-EOB-DTPA-assisted MRI of the liver and a histopathological evaluation after hepatic tumor resection. T1 maps of the liver were evaluated to determine the T1 relaxation time and reduction rate between the native phase and hepatobiliary phase (HBP) in liver lesions. These findings were correlated with the histopathologically determined degree of HCC differentiation (G1, well-differentiated; G2, moderately differentiated; G3, poorly differentiated). There was no significant difference between well-differentiated (950.2 ± 140.2 ms) and moderately/poorly differentiated (1009.4 ± 202.0 ms) HCCs in the native T1 maps. After contrast medium administration, a significant difference (p ≤ 0.001) in the mean T1 relaxation time in the HBP was found between well-differentiated (555.4 ± 140.2 ms) and moderately/poorly differentiated (750.9 ± 146.4 ms) HCCs. For well-differentiated HCCs, the reduction rate in the T1 time was significantly higher at 0.40 ± 0.15 than for moderately/poorly differentiated HCCs (0.25 ± 0.07; p = 0.006). In conclusion this study suggests that the uptake of Gd-EOB-DTPA in HCCs is correlated with tumor grade. Thus, Gd-EOB-DTPA-assisted T1 relaxometry can help to further differentiation of HCC.


Results
Overall, 33 HCC lesions were analyzed in this study. 14 HCC samples were well differentiated (G1), 14 were moderately differentiated (G2), and 5 were poorly differentiated (G3). Furthermore, tumors were categorized according to the specifications of the WHO (5th edition). The mean volume of the liver lesions was 170.94 ± 212.82 cm 3 . Lesion characteristics are shown in Table 1.
A twofold comparison of the growth pattern and immunostaining results of OATP1B3 showed that poorly differentiated HCCs had no positive cells upon immunostaining, and the immunoreactive score (IRS) was negative In healthy hepatocytes, the hepatobiliary contrast agent Gd-EOB-DTPA is distributed from sinusoidal spaces into the interstitium and consecutively into hepatocytes. Gd-EOB-DTPA sinusoidal clearance into the liver is controlled by the transport of organic anion-transporting polypeptides (OATPs). Absorbed Gd-EOB-DTPA passes from hepatocytes into the bile canaliculi and partially returns to the interstitium with the help of multiple resistance-associated proteins (MRP1); this intercompartmental transfer generates a ratio between the sinusoids, interstitium, hepatocytes, and bile canaliculi. In tumor cells, the expression of MRP is often maintained, whereas the expression of OATP1B1/B3 tends to decrease over the course of hepatocarcinogenesis 14 .  Table 2 shows the relationships between growth patterns and immunostaining results. Between plain and post-Gd-EOB-DTPA administration, there was a significant reduction in the T1 relaxation times (p = 0.001, p = 0.001, p = 0.043). There was no significant difference in the plain phase (G1, 950.2 ± 140.2 ms; G2, 959.5 ± 96.5 ms; G3 1149.3 ± 346.7 ms, p = 0.602). However, there was a significant difference in the mean T1 relaxation times for G1 lesions (555.4 ± 140.2 ms) compared to G2/G3 lesions (G2, 712.6 ± 84.4 ms, p = 0.015; G3, 858.0 ± 232.2 ms, p = 0.011) in the hepatobiliary phase 20 min after contrast administration (Fig. 2). In the pairwise comparison between G1, G2 and G3, there was a significant difference between G1 (0.40 ± 0.15) and G2 lesions (0.26 ± 0.07, p = 0.012). Comparing G1 to G3 (0.25 ± 0.07) lesions only a trend towards significancy (p = 0.065) was detected. There was no significant difference (p = 0.977) between G2 (0.26 ± 0.07) and G3 (0.25 ± 0.07) lesions (Fig. 3, Table 3). Pooling G2 and G3 into one group G2/G3, the T1 reduction rate was 0.25 ± 0.07, showing a significant difference (p = 0.006) to G1.
A comparison of the reduction rate to the IRS (Fig. 4)  www.nature.com/scientificreports/ We performed ROC curve analyses to differentiate G1 from G2/G3 tumors based on the tumor reduction rate. The optimal cutoff value for the tumor reduction rate to differentiate G1 from G2/G3 tumors was 0.385 (AUC 0.76). This cutoff value resulted in a sensitivity of 50% and a specificity of 100% for this differentiation, with a positive predictive value of 100% and a negative predictive value of 73%.

Discussion
This study shows the possibility of distinguishing well-differentiated from moderately/poorly differentiated HCCs using Gd-EOB-DTPA-enhanced T1 maps. In our study, all lesions showed a reduction in the T1 relaxation time, indicating Gd-EOB-DTPA uptake. The uptake and excretion of Gd-EOB-DTPA are regulated by OATP-B1/-B3 and MRP2 13 . OATP is a multispecific transporter that is ubiquitously expressed throughout the body, with the subtypes OATP-B1 and OATP-B3 being expressed only in the liver 26 . Tsuboyama et al. 27 proposed a hypothetical mechanism: Gd-EOB-DTPA accumulation is dependent on its uptake via OATP-B1/-B3 and its biliary excretion via MRP2. Depending on the localization of MRP2, Gd-EOB-DTPA is either secreted into canaliculi and therefore excreted into the bile, showing minor enhancement in hepatocytes, or into pseudoglands with consecutive accumulation and therefore strong enhancement.
In advanced HCCs, it has been shown that OATP-B1 or -B3 expression is usually reduced 15 and that MRP2 expression is consistent or increased 16 ; hence, these results support our findings of less Gd-EOB-DTPA uptake in G2/G3-differentiated HCCs either due to less uptake or a higher biliary excretion rate.
Not only Gd-EOB-DTPA but also a variety of substances, including atorvastatin 28 and cefazolin 29 , and anticancer drugs, such as methotrexate 30 , rifampicin 31 , paclitaxel, and docetaxel 32 , are transported by OATP-B1 and OATP-B3. Therefore, the anticancer drug effect might be competitive and dependent on the degree of tumor differentiation correlating with the expression of these polypeptides.
Previous studies have shown that HCCs present themselves differently on Gd-EOB-DTPA-enhanced MRI depending on their grade 21,25,27,33-40 . Frericks et al. 39 and Schelhorn et al. 34 showed no correlation between tumor grade and the enhancement pattern; however, histopathological grading was performed only on biopsy samples instead of resected liver tissues. The evaluation was based on signal intensities compared to the surrounding liver parenchyma, and underlying cirrhosis and impaired Gd-EOB-DTPA uptake were not considered. However, Tsuboyama et al. 27 showed OATP-B3 overexpression in all stages of differentiation, linking strong lesion enhancement to the altered expression of MRP2 due to the accumulation of Gd-EOB-DTPA. They also reported that enhancement was more pronounced than that in the liver parenchyma regardless of liver function. As they analyzed only five lesions with strong enhancement, no statement could be made about the correlation between tumor grade and strong enhancement.
For quantitative assessments, some studies, including ours, used T1 maps and measured changes in the T1 relaxation times after contrast administration to determine Gd-EOB-DTPA uptake. In the pre-interventional analysis, image-based methods with the additional T1 maps are beneficial as they allow a superior, quantitative function evaluation, compared to SI-based evaluations 41 , by maintaining morphological information simultaneously, offering a "one stop shot" approach 42 . Mio et al. showed the potential of T1 mapping for quantitative evaluation of focal liver lesions with the shortest pre-and post-contrast T1 values for HCC 24 ; they concluded that T1 mapping is useful for differentiating HCC from hemangioma, metastatic tumor, and cysts, enabling accurate diagnosis.
Only a limited number of studies are available investigating the potential of plain T1 mapping to characterize tumors. Keller et al. showed the potential of plain T1 mapping as a reliable non-invasive method for quantification and characterization of primary liver tumors in a rabbit model 23 . In our study, G3 tumors had a slightly www.nature.com/scientificreports/ higher mean T1 time than G1 and G2 tumors. However, no significant difference was found between the different degrees of HCC differentiation in plain T1 maps (Table 3). Huang et al. 21 investigated several features, such as tumor size, margin, signal homogeneity, and their correlation to histological grade, showing that a larger diameter, a more irregular margin, the existence of vessels inside the lesions, and peritumoral hypointensity are signs of a lesser differentiated HCC; however, there was no correlation between the relative reduction in T1 and tumor grade. As they defined G1 and G2 lesions as low grade due to the small number of G1 lesions in their study (only eight) and compared them to G3 lesions (as medium grade), the findings are in some way consistent with those of Peng et al. 25 , who showed a correlation between G1 and G2 lesions as well as G1 and G3 lesions, but no correlation between G2 and G3 lesions. These findings are also consistent with our results. There was a significant difference in the amount of Gd-EOB-DTPA uptake between well-differentiated G1 and moderately/poorly differentiated G2 or G3 HCCs.
Presurgical determination of the degree of malignancy is essential for choosing the right therapy. According to current German guidelines, the treatment of choice for HCC in a cirrhotic liver is a liver transplant that also cures the underlying cirrhosis if the Milan criteria apply (meaning only one lesion less than five centimeters or three lesions less than 3 cm and no macrovascular invasion 43 ). As these strict criteria might exclude potential patients who could benefit from a liver transplant, it has been suggested that the degree of differentiation also plays a role in liver transplant selection criteria 44,45 , as it has been shown that the histopathological grade is a prognostic factor for the outcome 19,46 . www.nature.com/scientificreports/ Due to their poor prognosis, patients with poorly differentiated lesions are excluded from liver transplantation in some centers and treated noncuratively 47,48 ; therefore, the importance of determining the correct degree of differentiation is evident.
The histopathological grade is currently determined by biopsy, the most common procedure being needle core biopsy, which is associated with some complications, such as bleeding, infection, and tumor seeding 49 . Although its specificity can reach up to 100%, its sensitivity can be as low as 34.6% (preoperative biopsy rather than undergraded tumors). Therefore, it has been suggested for clinicians not to rely on the histopathological   Table 3. T1 relaxation times and reduction rates for well-(G1), moderately (G2) and poorly (G3) differentiated HCCs and their surrounding liver parenchyma. The values indicate the mean ± standard deviation.    www.nature.com/scientificreports/ grade alone for liver transplant eligibility 50 . Therefore, determining the degree of differentiation by combining radiological and histopathological features might result in greater certainty regarding the grade of HCC lesions. This study has several limitations, including the small number of included lesions. Due to the limited group size, only five G3 HCCs were included due to their rare occurrence 51 . More than half of the lesions had to be excluded for various reasons, such as those that were pretreated or less than 1 cm in size. As a consequence of the screening program for patients at risk for HCC, more lesions are detected at an early stage 52 ; therefore, they are too small to be diagnosed correctly. Due to organ shortages, HCCs are sometimes pretreated to bridge liver transplantation 53 , changing the morphology of the lesions. HCCs show high intratumor heterogeneity, which is a challenge for both pathologists and radiologists, as different degrees of differentiation might exist in one tumor 54 . This heterogeneity is seen in the high SD rates of T1 relaxation times. To reduce inaccuracies in assessing T1 relaxation times, we measured them three-dimensionally by segmentation.

Material and methods
Study group. This study was performed as a retrospective subgroup analysis on a prospective study to evaluate contrast uptake through T1 mapping. The local institutional ethics committee of the University Hospital Regensburg approved this retrospective analysis, and the study was performed following the relevant guidelines and regulations. The ethics committee waived informed consent (Ethics Committee, University of Regensburg, 93040 Regensburg, Germany).
Between January 2014 and May 2019, 56 patients with HCC underwent Gd-EOB-DTPA-enhanced MRI of the liver, including T1 mapping. This was due to the presence of a liver lesion that was, after resection, histopathologically confirmed as HCC. Lesions less than 1 cm in size (n = 6), those with infiltrative growth (n = 1), histopathologically confirmed mixed tumors (n = 3), previously treated lesions (n = 1), lesions that went repeat MRI before resection (n = 2) or lesions with technical errors (n = 10), e.g., image artifacts in the area of interest, were excluded. Thus, a total of 33 HCC lesions were included.
Imaging. MR images of the liver were obtained using a clinical whole-body 3 T system (MAGNETOM Skyra, Siemens Healthcare, Erlangen, Germany) with a combination of body-spine array coil elements (an 18-channel body matrix coil and a 32-channel spine matrix coil) for signal reception. Gd-EOB-DTPA (Primov-ist®, Bayer Vital GmbH, Leverkusen, Germany) was used as a hepatocytic contrast agent. All patients received a body weight adapted dose of Gd-EOB-DTPA (0.025 mmol/kg body weight) administered via bolus injection with a flow rate of 1 mL/s, flushed with 20 mL NaCL.
T1 maps before and 20 min after contrast injection were generated from a prototypical T1-weighted volumeinterpolated breath-hold examination (VIBE) sequence (TR 5.79 ms, TE1 2.46 ms, TE2 3.69 ms) by using a prototypic technique based on a 3D spoiled-gradient echo sequence with variable flip angles (1°, 7° and 14°) and a voxel size of 3.6 mm × 2.5 mm × 4.7 mm interpolated to 1.3 mm × 1.3 mm × 3.0 mm 55,56 . In addition, a B1 map of the liver was acquired for each patient prior to the T1 relaxometry measurements to perform B1 correction to improve the homogeneity of the T1 maps 57 , and color-coded T1 maps were calculated inline. Using controlled aliasing in parallel imaging results in higher acceleration (CAIPIRINHA) as a parallel imaging technique with an acceleration factor of 4, the entire liver was covered during a single breath-hold (acquisition time 17 s).
Image processing. Open-source Horos imaging software (Horos Project, Annapolis, MD, USA) was used to assess the volume and mean T1 relaxation time three-dimensionally by manual segmentation of the lesions in the T1 maps before and after contrast medium administration. The HCC was segmented using a closed polygon selection. The boundary of the HCC was marked in every second slice. After outlining one-half of the slices, the missing ones were calculated by interpolation. The segmentation was subsequently transported to the plain phase and manually adjusted for different respiratory levels or patient movement. The "repulsor tool" was used to manually adjust the regions of interest (ROIs) with particular care to ensure that the lesion was completely captured. The volumes were calculated by multiplying surface and slice thickness.
An additional ROI was placed manually in the surrounding liver parenchyma (with identical sizes and locations in the non-contrast and post-contrast T1 HBP), excluding visible vessels and imaging artefacts.
The uptake of Gd-EOB-DTPA was determined by calculating the reduction rate in the T1 relaxation time between plain (T1 plain ) and Gd-EOB-DTPA-enhanced (T1 HBP ) MRI as follows: The T1 relaxation times and T1 reduction rates were then correlated with the histopathologically determined degree of HCC differentiation.
Histological assessment. Histopathological analysis. All tissue samples were obtained from routine therapeutic surgeries performed between 2014 and 2019. After fixation in neutral buffered formalin, all tissue specimens were embedded in paraffin. According to a standard protocol, 4 μm thick tissue sections were cut and stained with hematoxylin and eosin (HE). Histological and immunohistochemical analysis were performed by 2 experienced liver pathologists (A.S. and K.U.). The histological grade of HCC was determined according to the World Health Organization (WHO) criteria 51 . HCCs were subdivided into three groups according to their degree of malignancy: G1, well differentiated; G2, moderately differentiated; and G3, poorly differentiated. Evaluation of OATP1B3 immunostaining. First, all slides were screened to assess the minimum and maximum staining intensities in the tumor cells. Only membranous immunostaining on tumor cells was determined to be immunopositive. The intensity of membranous OATP1B3 immunostaining within HCC was evaluated and categorized as absent (0) (= no evidence of staining), weak (1+), moderate (2+), or strong (3+). Then, the percentage of positively stained tumor cells was grouped into five categories (0: no positive cells; 1: < 10% positive cells; 2: 10-50% positive cells; 3: 51-80% positive cells; and 4: > 80% positive cells). Subsequently, a semiquantitative score according the IRS was calculated as a product of multiplication between the positive cell proportion score (0-4) and the staining intensity score (0-3). The IRS ranged from 0 to 12 (Table 4).
Statistics. Statistical analyses were performed using IBM SPSS Statistics (version 26, Chicago, IL, USA), and all data are presented as mean ± standard deviation (SD) for normal distributed data and median [q1; q3] for non-normal distributed data. The Kruskal-Wallis test followed by post-hoc pairwise comparisons was used to analyze differences between the stages of differentiation (G1, G2 and G3). For comparison between G1 and the grouped G2/G3 lesions the non-parametric Mann-Whitney test was used. The non-parametric Wilcoxon signed rank test was used for comparisons between plain and HBP phase. Receiver operating characteristic (ROC) curve analyses were performed to differentiate between G1 and G2/G3 tumors, and the optimal cutoff was estimated according to the Youden indices. The estimated areas under the curve (AUCs) with corresponding 95% confidence intervals and the sensitivity, specificity, positive predictive and negative predictive value are reported. All tests were two-sided, and values of p < 0.05 indicated a significant difference in all statistical tests.