Heterogeneity of microsphere distribution in resected liver and tumour tissue following selective intrahepatic radiotherapy

Primary and metastatic liver tumours present a significant clinical problem, when radical surgery, the only curative option, is not possible because of extensive growth in connection with vital structures, even without extra-hepatic growth. Chemotherapy has limited efficacy [1]-[3], and external beam radiotherapy, including stereotactic techniques, is hampered by a normal liver tissue tolerance that is less than needed for tumour eradication [4],[5]. Other techniques used for local destruction of liver tumours include radiofrequency or ultrasound ablation and are moderately successful [6]-[8]; thus, new treatment modalities are urgently needed. Because liver malignancies are supplied mainly from the hepatic artery, in contrast to portal vein supply to the liver parenchyma [9], selective intra-arterial interventions have been used, including arterial chemoembolisation [10] and radioembolisation.

The evidence favouring one technique over the other is weak. However, it seems clear that with present radioembolisation techniques using yttrium-90-labelled (⁹⁰Y-labelled) microspheres selectively infused via the hepatic artery, tumour reduction can be achieved with relatively little radiation-induced hepatic damage in spite of mean liver doses far exceeding those tolerated in external beam therapy [9],[11],[12]. Knowledge of the physical and biological background is meagre; not least the explanation of the relatively high tolerance for ionising radiation of the normal liver tissue, but it is probably related to the low dose rate of radionuclide therapy and dose heterogeneity, allowing for regeneration from low-dose regions in the normal liver parenchyma.

The possible sparing effect of the low dose rate has been theoretically investigated by Cremonesi et al. [13]. They showed a minor sparing effect in regard to biologically effective doses (BEDs). In contrast, large non-uniform microsphere distributions have been observed in surgically removed tissue samples after radioembolisation [14]-[17]. However, no clear quantitative description of the non-uniformity has been presented from this limited number of studies and the main focus has been tumour tissue rather than normal liver parenchyma. Despite the unknown small-scale distribution of microspheres, Gulec et al. presented a detailed Monte Carlo modelling approach for performing small-scale dosimetry [18]. They used millions of closed-packed hexagonal lobules (radius 0.6 mm and length 1.5 mm) to model the liver parenchyma, and the microspheres were uniformly distributed in the portal tracts in the corners of the lobules. Despite the long range of ⁹⁰Y electrons (max range, 11 mm; mean range, 2.5 mm) [19], a more than twofold difference in the absorbed doses to the fine structures was obtained; the cross-irradiation would only partly smooth the dose distribution. Therefore, it can be anticipated that a variable distribution of microspheres in the portal tracts will create a primarily non-uniform dose distribution in the liver parenchyma.

To investigate the question of heterogeneity, here we evaluated activity distributions on biopsies taken from resected tissue samples pre-surgically treated with resin microspheres (SIR-Spheres®).

SPECT/CT studies from the ^99mTc-MAA infusions disclosed a pulmonary shunting of 3.0% and 3.5% for Pt 1 and Pt 2, respectively. The tumour-to-normal liver activity concentrations (TNCs), averaging voxel levels in the entire tumours as well as entire normal liver tissues, were calculated at 2.7 and 3.8 in Pt 1 and Pt 2, respectively. The CVs were 0.29 (normal) and 0.42 (tumour) for Pt 1 and 0.77 (normal) and 0.089 (tumour) for Pt 2.

From the pre-therapy studies, the prescribed activities to be delivered would result in expected average doses and BEDs to normal liver of 22 (BED = 29) Gy and 33 (BED = 49) Gy for Pt 1 and Pt 2, respectively. For practical reasons, the doses were averaged over both the portions of normal tissue planned to be resected (tumour margins) as well as the living liver tissue remaining in the patient after surgery. The expected (if unresected) doses to tumours (in accordance with TNC values reported above) were calculated to 59 (BED = 67) Gy and 125 (BED = 161) Gy for Pt 1 and Pt 2, respectively.

Based on the measured activity in biopsies, the average absorbed dose to normal liver parenchyma was calculated to 31 (BED = 45) Gy (620 Bq/mg) for Pt 1 and 52 (BED = 93) Gy (1,040 Bq/mg) for Pt 2. The corresponding calculation based on tumour biopsies resulted in 100 (BED = 123) Gy (2,000 Bq/mg) for Pt 1 and 150 (BED = 202) Gy (3,050 Bq/mg) for Pt 2. Thus, based on the limited sample volumes, the TNCs were 3.2 and 2.9 for Pt 1 and Pt 2, respectively.

In Pt 2, different clustering patterns of the microspheres were found with light microscopy, with several clusters distributed through more than one serial section. The seven largest clusters, of the type shown in Figure 1B, ranged 22 to 59 microspheres when aligning serial adjacent slices. Of the 240 locations found with single or several microspheres, 154 were isolated single spheres (Figure 1A), i.e. at a distance >200 μm (7 sphere diameters) from the nearest neighbouring sphere. This distance was chosen as it was consistent with the largest portal tract found and therefore two single spheres would not occupy the same portal tract. Thus, 64% of all locations contained single isolated spheres, representing only 19% of the total number of spheres; 53% of all 793 spheres were located within any of the 19 largest clusters (7% of the locations), ranging 8 to 59 spheres per cluster (Figure 1B). The mean size of each location was 3.3 spheres, with a CV of 2.1 and SK of 4.9. The distribution of spheres is shown in Figure 2. The total activity found in the six biopsies, divided by their total mass, showed an activity concentration of 1,550 Bq/mg. The 793 spheres (50 Bq per sphere) found in the 60 sections with a total volume of 36 mm³ showed an activity concentration of 1,120 Bq/mg (28% lower).

Autoradiography

Images acquired through autoradiography, performed on resected tumour and normal liver parenchyma, are shown in Figure 3. The images indicate a heterogenic pattern, with several hot and cold spots of radioactivity and a generally higher activity concentration in tumour tissue.

Detector measurements

Tables 1, 2 and 3 show statistical data for the radioactivity distribution of Pt 1. No clear decreasing or increasing trends are observable for CVs or SKs; however, the CVs were consistently lower for the left lobe, showing a higher mean activity concentration.

	Small mass	Medium mass	Large mass	All
Mass (mg)	8 ≤ m ≤ 41	47 ≤ m ≤ 76	80 ≤ m ≤ 422	8 ≤ m ≤ 422
$\overline{m}$(mg)	20	63	210	100
n	11	11	12	34
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	1,200	3,300	9,200	4,400
$\overline{A}/\overline{m}$(Bq/mg)	3,100	2,600	2,200	2,200
CV	0.45	0.46	0.41	0.46
SK	−0.35	−0.52	−0.12	0.058

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

	Small mass	Medium mass	Large mass	All
Mass (mg)	12 ≤ m ≤ 62	63 ≤ m ≤ 125	128 ≤ m ≤ 391	12 ≤ m ≤ 391
$\overline{m}$(mg)	39	82	240	120
n	22	20	20	62
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	510	740	2,200	1,200
$\overline{A}/\overline{m}$(Bq/mg)	670	450	470	480
CV	0.63	0.67	0.57	0.65
SK	0.27	2.8	1.7	1.2

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

	Liver lobe
	Left		Right
Mass (mg)	15 ≤ m ≤ 26	54 ≤ m ≤ 104	49 ≤ m ≤ 73	76 ≤ m ≤ 90
$\overline{m}$(mg)	21	75	60	82
n	16	16	9	9
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	1,500	3,000	1,100	980
$\overline{A}/\overline{m}$(Bq/mg)	3,600	2,000	910	610
CV	0.49	0.75	0.66	0.51
SK	−0.46	0.82	1.0	0.64

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

Tables 4, 5 and 6 show the radioactivity distribution data for Pt 2. Tables 4 and 5 show CV and SK decreasing with increasing mass. A comparison of CV levels between the tables shows a higher CV for normal liver parenchyma at some distance from the tumour, compared to adjacent to the tumour, consistent with differing mean activity concentrations. Figure 4 shows histograms of the data presented in Table 4.

	Small mass	Medium mass	Large mass	All
Mass (mg)	6 ≤ m ≤ 14	15 ≤ m ≤ 39	40 ≤ m ≤ 91	6 ≤ m ≤ 91
$\overline{m}$(mg)	10	27	66	35
n	29	26	29	84
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	300	590	1,200	740
$\overline{A}/\overline{m}$(Bq/mg)	1,500	1,100	900	1,000
CV	1.4	1.0	0.63	1.3
SK	3.0	1.8	0.82	4.0

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

	Small mass	Medium mass	Large mass	All
Mass (mg)	5 ≤ m ≤ 14	16 ≤ m ≤ 41	43 ≤ m ≤ 102	5 ≤ m ≤ 102
$\overline{m}$(mg)	10	27	69	35
n	26	24	25	75
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	380	920	2,300	1,200
$\overline{A}/\overline{m}$(Bq/mg)	1,900	1,700	1,700	1,700
CV	0.81	0.58	0.45	0.64
SK	1.8	1.2	0.69	1.9

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

	Small mass	Medium mass	Large mass	All
Mass (mg)	5 ≤ m ≤ 11	12 ≤ m ≤ 37	38 ≤ m ≤ 95	5 ≤ m ≤ 95
$\overline{m}$(mg)	8.4	23	65	32
n	14	14	13	41
$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$	690	2,100	3,100	1,900
$\overline{A}/\overline{m}$(Bq/mg)	4,100	4,500	2,400	3,000
CV	0.50	0.71	0.36	0.65
SK	0.78	2.9	−1.3	2.8

CV, coefficient of variation;$\overline{m}$, mean mass of n tissue biopsies;$\overline{\mathrm{Spheres}}/\mathrm{biopsy}$, average number of spheres per biopsy;$\overline{A}/\overline{m}$, mean activity concentration; SK, skew of the distribution.

Tumour tissue analysis revealed no clear trend of decrease or increase with increasing biopsy volumes for CVs and SKs (Table 6).

In this study, surgical resection of tumours enabled a detailed investigation of microsphere distribution. To avoid haemorrhage and dissection difficulties along critical structures, surgery was performed before radiation-induced inflammatory reactions developed. Late surgery after inflammatory decline might have increased the risk for extensive fibroses.

The current autoradiogram results, showing irregular distributions and varying sizes of hot and cold spots of radioactivity, are in line with previous studies, mainly involving light microscopy evaluation of tissue samples collected months after injection [14]-[17]. However, the previous and current studies are of limited value for describing a full volumetric distribution of microspheres, because serially aligned sections have to be prepared and carefully analysed, which was done in limited volume in this study. We also applied a less time-consuming new approach for analysing clustering size and its distributions.

Our approach to studying the geometric pattern of non-uniformity relied on measurements of activity concentration in tissue samples of various sizes. With this approach, the SK and CV should decrease with increasing biopsy size for heterogenic patterns with larger basic elements than the biopsy sizes.

The difference between TNCs, comparing imaging and biopsy measurements, can be explained by the fact that the latter method excludes averaging over the tissue left in the patient after surgery and that tumour tissue dominated the resected volume of Pt, whereas resected normal liver parenchyma dominated for Pt 2.

In Pt 1, the mean activity concentration decreased with increasing masses, but CVs and SKs did not follow the expected trend. These results indicate that the mass groups probably were too wide and mean masses may have been too large. Mass homogeneity of each mass group thus appears statistically important; furthermore, the number of biopsies might have been too low.

The normal liver parenchyma consists of a very well structured distribution of more than one million lobules and associated portal tracts. Each lobule is associated with six portal tracts, but because of the hexagonal arrangement of lobules, three lobules share each portal tract. Therefore, there are no more than two portal tracts per lobule throughout the parenchyma. A lobule weighs around 2 mg, so there is about one portal tract per milligram of liver tissue [31]. Considering the portal tracts as the target site for microspheres, the number of target sites in the sample sizes from Pt 1 could range from 8 to 62 in the smallest sample groups to over 400 in the largest. To identify any spatial variation in cluster frequency, the sample sizes must include few targets with a low variation in mass.

The results from Pt 2 with many biopsies and smaller masses with a low variation in mass showed a clear trend consistent with the heterogeneity of normal liver parenchyma, with strictly decreasing CV and SK with increasing masses. The strong decreasing trend of the CV indicates that the basic element of the heterogenic pattern is larger than or similar in size to the biopsies in the group with the largest masses (about 70 mg). The decreasing positive SK indicates a rather frequent and repetitive presence of hot spots of radioactivity, i.e. microsphere clusters or regional gatherings. The small sample investigated by light microscopy showed a similar and slightly higher CV (2.1) and SK (4.9) than was identified by activity measurements on the sample group of origin (CV of 1.3 and SK of 4.0; Table 4). These findings seem to agree with the repetitive microscopic vascularisation structure of the liver. As only 8% of the total volume of the six biopsies selected for microscopic studies were in fact investigated in microscope, it is not surprising that the activity concentration differed with 28% comparing sphere counting in microscope with detector measurements on the same six biopsies.

Comparison of the two normal tissue regions showed a lower CV with higher activity concentration and similar masses. This finding indicated decreasing heterogeneity with increased activity concentration.

In contrast to normal liver parenchyma, tumour tissue architecture is unstructured, with scattered fibrotic and necrotic areas that obviously influence the statistical results. The current inconclusive results for tumour tissue are included only to highlight how our method, when applied correctly (e.g. for Pt 2), could distinguish a structured heterogenic distribution (normal liver parenchyma) from a more chaotic heterogenic distribution (tumour). These findings are in accordance with those of Tveit et al. who studied the regional perfusion ex vivo in human kidney specimens, including carcinomas. They found a homogeneous high perfusion in the well-structured renal cortex in contrast to the highly inhomogeneous perfusion in the tumour tissue [32]. The results for normal liver tissue are based on a comparatively large number of tissue biopsies: 96 from Pt 1 and 159 biopsies of Pt 2, supporting our conclusions.

Limited data on microsphere distribution in patients are available. Some efforts have targeted dosimetric calculations, and studies mainly have been light microscopy investigations of tissue samples collected months after administration [15]-[17],[33],[34]. The reported results are consistent with ours, showing large heterogeneities in microsphere distribution arising from clustering effects and also heterogeneity patterns that are more prominent in normal liver tissue than in tumour tissue.

Gulec et al. simulated dose distributions in the context of liver tissue microanatomy and showed different mean absorbed doses to different tissue structures [18]. However, they noted that they never intended to consider the expected heterogeneities in the distribution of microspheres; thus, the assumed homogeneous distribution resulted in different mean doses between different structure types, but basically no variance within structure types. Applying our findings of heterogeneities on a large scale could be clinically valuable, comparing them to the size of individual liver lobules used as the fundamental unit in their simulations and showing a more realistic dose variance within specific structure types. Our findings suggest that some lobules may receive very few or no microspheres whereas others may receive large clusters or strings of spheres. Indeed, larger clusters of microspheres might be stuck within larger arterioles, limiting the mean absorbed dose to finer liver structures, such as within the liver lobules.

Recently, Walrand et al. simulated different arterial tree models, varying the parameters affecting the distribution of glass microspheres [35]. Their results showed heterogeneities and clustering of spheres due to stochastic and systematic variations of the blood flow. Their clusters tended to be much smaller. The majority of the spheres were found within clusters of two to three spheres, about 30% to 40% were found to be single spheres, 10% of the spheres were found within clusters of four to five spheres and only a few percent within cluster sizes of seven to ten spheres. The explanation for this could be that the 25 times lower specific activity per resin sphere, as compared to glass spheres, and thus a much higher number of resin spheres injected, tends to create larger clusters, for statistical reasons. This would be consistent with the results from our study, with the majority of the spheres trapped within cluster sizes of 8 to 59 spheres. It is also possible that the more extensive embolisation caused by higher numbers of resin spheres has a more dramatic effect on the arterial circulation, causing larger differences in blood flow and thus larger inter-regional variations in distributions of microspheres, making the mentioned model less applicable for resin spheres.

We studied heterogeneity on a scale slightly larger than the mean range of β⁻ electrons from ⁹⁰Y. Interestingly, the CVs for activity distributions are higher within limited regions, as measured on biopsies, than in entire tissue volumes, as measured with ^99mTc-MAA. Even though the ^99mTc-MAA gamma radiation is smeared out, the homogenising effect of crossfire between microspheres will not smooth heterogeneity at this scale (the aim in treating tumour tissue, but unwanted in normal liver parenchyma), as the mean range of the ⁹⁰Y beta radiation is only 2.5 mm. The statistical methods used here demonstrate activity heterogeneity at a therapeutically relevant geometric scale, which might be useful for protocol optimisation. In their modelling study, Cremonesi et al. showed that liver tissue might be spared by fractionated administration of microspheres [13]. In such calculations, however, the sparing effect arises only from a decreased dose rate. Our results indicate that the heterogeneity of microsphere trapping between fractions might heighten the sparing effect of fractionation protocols.

To obtain a more precise dosimetric distribution from our results, the heterogenic distribution of activity must be applied to a detailed 3-D activity distribution matrix model, convolved with a 3-D beta dose kernel matrix, to plot reliable dose-distribution histograms for input in response models such as the BED model.

Tissue activity distributions in normal tissue were heterogeneous on a relevant geometric scale considering the mean range of the ionising electrons. Given the similar and repetitive structure of the liver parenchyma within and between patients, the detected heterogeneities in normal liver parenchyma can partly explain the survival of normal tissue in regions receiving a substantially higher mean absorbed dose than could be tolerated with more homogeneous dose distributions resulting from external radiation techniques.

JH carried out the sample preparations, the detector measurements, the autoradiograms, the statistical calculations and the microscopic investigations; participated in the design and method development of the study and drafted the manuscript. MR carried out the surgeries, helped to draft the manuscript and revised it critically. RH participated in the design of the study, helped to draft the manuscript and revised it critically. JS coordinated and participated in the patient treatment and revised the manuscript critically. OH performed the radioembolisations and revised the manuscript critically. JM made interpretations related to microscopy and histology and revised the manuscript critically. PG supervised the imaging procedures for the patient and revised the manuscript critically. PB conceived of the study; participated in its design, method development and coordination; helped with the sample preparations; helped to draft the manuscript and revised it critically. All authors read and approved the final manuscript.