Original Research
Correlation of MR Imaging and Histologic Findings in Mouse Melanoma' John 0. DeJordy, MD Yoram Salomon, PhD
Peter Bendel, PhD Ada Horowitz, MD Hadassa Degani, PhD
M2R melanoma tumors in male C57 black mice were used to correlate magnetic resonance (MR) images with the corresponding histologic slices and to determine if analysis of the achievable correlation can provide a basis for predicting gross histologic features with MR imaging alone. The MR imaging sections obtained at 4.7 T were each 680 Crn thick, with an in-plane resolution of 195 pm. The distribution of melanin within the histologic slices correlated well with the high-signal-intensity regions on the T1-weightedimages (TlWIs),while these regions had low signal intensity on the T2weighted images (T2WIs).providing evidence that melanin or melanin-associated paramagnetic species are responsible for the observed proton relaxation rate enhancement. Viable melanoma cells typically showed intermediate signal intensity on T2WIs. T1WIs, and proton-density images. Necrosis typically had high signal intensity on T2WIs, TlWIs, and proton-density images. Quantitation of the MR imaging results, followed by statistical analysis, demonstrated statistically significant differences between melanin-rich, viable-melanoma, and necrotic regions on MR images.
-
Index term.: Comparative studies High-resolution imaging Melanoma. 40.379 Neoplasms, MR. 40.379 * Tissue characterization JMRllS92; 2595-700 Abbreviations: ANOVA = analysis of variance, LSD = least significant difference. ROI = region of interest. TlWl = TI-weighted image. T2W = T2-weighted image.
I From the DepartmentsorChemical Physics (J.O.D.. P.B.. H.D.1 and The Weinnann Institute of Science. Hormone Research (J.O.D..Y.S.). 76100 Rehovot. Israel: and the Department of Patho1o.g. Sheba Medical Center. Tel-Hashomer, Israel [A.H.).Received June 22. 1992: revision requested August 17: revision received and accepted September 1. Y.S. supported by grants from Ihe Crown Endowment Fund for Immunologiral Research at the Weizmann Institute of Science and the Israeli Ministry of Health. H.D. supported by grants from the Grrmaii-Israeli Foundation for Srientifir Research and Development. Y.S. is the Charles and Tillie Luhin Professor of Hormone Research. Address reprint requests to H.1)
SMRI. 1992
MAGNETIC RESONANCE (MR)imaging has evolved into an important tool in clinical medicine; however, the application of MR imaging has largely been limited to tumor detection rather than tissue-specific diagnosis. In the vast majority of cases, the definitive diagnosis is still based on the microscopic evaluation of biopsied tissue. The potential of MR imaging as a tool for tissue-specific diagnosis is presently being investigated by many groups. Melanotic melanoma has been the subject of a number of these studies. In 1984, Sergeev and Murza claimed that proton relaxation times depended on the free radical content in human melanoma ( 1).Many of the early clinical studies concerned weal or choroidal melanomas of the eye (2-5); other studies have focused on melanomas in the central nervous system (6,7),cutaneous melanomas (8,9), comparisons of melanomas occurring at various sites ( 10).and melanomas in animal models ( 11).Most of these studies have shown that there is a unique enhancement of the proton relaxation rates in melanotic melanoma; however, there is no clear consensus regarding the source of this effect. These studies correlated the MR imaging appearance with the typical histologic characteristics of the tumors. The objective of the present study was to correlate MR images of mouse melanoma tumors with the specific, corresponding histologic slices. Using this MR imaging section to histologic slice correlation approach, we attempted to determine if gross histologic features can be defined with MR imaging alone and to provide evidence regarding relaxation rate enhancement in MR imaging of melanoma. 0
MATERIALS AND METHODS
Cell Culture and Tumor Model The M2R mouse cell line was developed from the B 16 mouse melanoma cell line, which was originally isolated from the C57 black mouse ( 12).M2R cells were cultured routinely a s monolayers in a 1:1 Ham F12 and Dulbecco-modified Eagle medium (Gibco,
695
Paisley, Scotland),supplemented with 10% heatinactivated horse serum ( 13).The M2R cells were harvested from monolayer plates with a rubber policeman. The cells were then counted, rinsed, and resuspended in isotonic saline ( 1 x lo7cells per milliliter). The tumor model was generated by subcutaneous injection of 1 x lo6 M2R cells into the right rear flank of 2-3-month-old male C57 black mice. The site on the flank was selected to minimize motion artifacts due to respiration.
MR Imaging The mice were anesthetized for the study with an intraperitoneal injection of sodium pentobarbital ( = 60 mg/kg). The MR images were obtained with a 4.7-T Biospec spectrometer (Bruker Medizintechnik, Rheinstetten, Germany) by using standard spin-echo sequences. A custom-built 4.5-cm-diameter imaging coil was used in the study. T1-weighted images (TlWIs)were acquired with a TR of 500 msec and a TE of 15 msec. For the TlWIs, 16 were averaged during a period of 35 minutes. T2-weighted images (T2WIs)and proton-density images were obtained with a TR of 2,400 msec and TEs of 68 and 17 msec, respectively. For the T2WIs and proton-density images, eight were averaged during a period of 83 minutes. The MR imaging sections were each 680 p n thick, with an in-plane resolution of 195 km (5-cm field of view and 256 X 256 matrix). Ten to 12 sections were selected for each tumor, with an intersection gap equal to approximately 20% of the section thickness. His tology After completion of imaging, the mice were killed by cervical dislocation. The skin was removed from the region of the tumor, and the whole animal was placed in fixative (75%picric acid, 20% formaldehyde, 5% glacial acetic acid).Before fixation, the tumors have a relatively fluid consistency that makes sectioning impractical. After 2-3 days, the tissue was removed from the fixative and the tumor was marked with stains (Davidson Marking System; Bradley Products, Bloomington, Minn) to help maintain orientation. The histologic plane corresponding to the central section of the MR imaging study was established by a single cut through the tumor. This plane was based on the position of the tumor in the spectrometer, sagittal and transverse images, and anatomic references. Once this initial cut was made, the tumor was removed from the rest of the tissue and processed for histologic examination. Three contiguous 10-km-thick histologic slices were obtained approximately every 100 km from either side of the initial cut. After processing, each of the histologic slice sets were stained with hematoxylin-eosin, a modified trichrome method, and the Prussian blue reaction. Hematoxylin-eosin was used to assess viability, necrosis, and pigmentation. The areas with substantial melanin appeared brown to black on the hematoxylineosin slices. The modified trichrome method was used to identify fibrotic regions, and the Prussian blue reaction was used to identify regions of substantial iron deposition. Note that iron in the form of hemoglobin does not stain positive with the Prussian 696
-
JMRl
-
November/December 1992
blue reaction. The orientation on the slides was maintained with the aid of the original tissue stains and with reference marks made on the slides at the time of slice mounting.
Data Analysis MR imaging section-histologic slice pairs were selected for further study on the basis of the degree of correlation of unique patterns and shapes between the pairs. A high degree of correlation of detailed patterns between the images and histologic slices provided some degree of assurance that a reasonable match had been achieved. To analyze the MR images, regions of interest (ROIs)were selected for six correlated pairs, and the area and average signal intensity were determined for each ROI with the image analyzing software (Bruker Medical Imaging). Each ROI selected for analysis had a minimum area of 50 pixels ( 1.9 mm2).The ROIs were selected on the basis of their relatively homogeneous appearance on all MR images of the same section and histologic slices. For example, if an ROI had homogeneously high signal intensity on the TlWI but consisted of high- and lowintensity areas on the T2WI or proton-density image, it did not meet the selection criteria. The signal intensity values were normalized to the intensity of muscle to eliminate differences arising from variations in the adjustment of the instrument. The signal intensities on the T2Ws and proton-density images (which were obtained from the second and first echo, respectively, of the same acquisition) were normalized to the intensity of muscle by assigning a value of one to the average intensity of muscle on the proton-density images. The intensities on the TlWIs (obtained in a separate acquisition) were normalized separately by assigning a value of one to the average intensity of muscle on the TlWIs. Within well-correlated section-slice pairs, high-powered microscopic evaluation of the ROIs in tumor slices served as the basis for histologic identification. Each ROI was categorized as viable melanoma cells, necrosis, melanin rich, or heterogeneous on the basis of microscopic evaluation of the histologic characteristics of that region. If the R 0 1 had a high melanin content, it was included in the melanin-rich group regardless of the degree of cell viability or necrosis. ROIs that did not have high melanin content and consisted of at least 80%necrosis or viable cells were grouped accordingly. The ROIs in the necrotic group represented frank necrosis with intercellular edema, typically with some melanin present. Regions with signs of early necrosis were not included in the necrosis group. The histologic information was then integrated with the information from the images to determine if any patterns could be appreciated. This initial evaluation was followed by further analysis of the data. A two-way analysis of variance (ANOVA)was used with two main factors: tissue type and image type. Initially, the data were statistically analyzed to determine if a significant correlation existed between image type and tissue type. After the existence of such a correlation was confirmed, a oneway ANOVA was performed for each image type. The ANOVA was followed by multiple comparisons of the tissue types in each type of image (the Duncan multi-
ple range test and the Fischer least significant difference [LSD]test) to determine if the tissue types could be divided into statistically different subgroups on each type of image section. Ultimately, conclusions were based on the information obtained for all three image types.
RESULTS The present study represents an effort to correlate the heterogeneity within a single MR imaging section with the corresponding histologic findings. By establishing a protocol to preserve the orientation of the tumor and by decreasing the MR imaging section thickness to less than 1 mm, we were able to achieve a good correlation between MR imaging and histologic findings. An example of an MR imaging-histology correlation pair, with high-magnification micrographs of selected regions, is presented in Figure 1. In our tumor model, the distribution of melanin within the histologic sections correlated well with hyperintense regions on TlWIs, regions which were hypointense on T2WIs. indicating shortening of T1 and T2. In addition, the Prussian blue reaction demonstrated the absence of any substantial deposition of free iron, indicating that shortening of T1 and T2 was not due to stainable iron deposits. In regons with heavy pigmentation, the effects of melanin appeared to dominate. Analysis of the data indicated that melanin-rich regions corresponded to the regions with the observed T1 and T2 shortening. These results support the view that melanin or melanin-associated paramagnetic species are responsible for this effect. Amelanotic to mildly pigmented viable regions typically showed intermediate signal intensity on TlWIs, T2WIs. and proton-density images, while frank necrosis typically had high signal intensity on all three types of images. Evaluation of the slides stained with the modified trichrome method indicated that there was no substantial fibrosis in these tumors other than in the capsule. In one correlation pair, a fresh hemorrhage was evident on the histologic slice (Fig 2). The region corresponding to the hemorrhage was hypointense on all three types of images, suggesting that in this region T2 relaxation enhancement dominates even at a TE of 15 msec. Figure 3 presents the mean values of the scaled signal intensities for melanin-rich, necrotic, and viablemelanoma ROIs. Statistical analysis of the data indicated that a significant correlation between image type and tissue type was present ( P = .0001).To determine if the tissue type could be distinguished on the basis of MR imagmg alone, one-way ANOVA was performed, followed by multiple comparison tests for each image type. For the normalized intensities of the proton-density images, one-way ANOVA gave a P value of ,0449, while the Fischer LSD test indicated that it was possible to distinguish necrosis from the other two tissue types. For the normalized intensities of the TlWIs, one-wayANOVA gave a P value of .OOO 1, while the Fischer LSD and Duncan tests indicated that it was possible to distinguish viable melanoma from the other two tissue types. Finally, for the normalized intensities of the T2Wls. one-way ANOVA gave a P value of .OOO 1, while the Fischer LSD and Duncan tests in0
dicated that it was possible to distinguish the three tissue types from one another. From this analysis, it appears that the T2WIs alone provide a means of distinguishing tissue types by means of MR imaging. In addition, the TlWIs enable clear differentiation between viable-melanoma regions and necrotic or melanin-rich regions. The proton-density images provided some insight into regions of frank necrosis, but the level of significance was equivocal. The combined information available from the three image types provides a basis for predicting certain aspects of gross histologic findings with MR imaging alone. The analysis of heterogeneous regions is difficult; however, the data obtained from homogeneous regions of amelanotic to slightly pigmented viable cells, regions of necrosis, and regions of h g h melanin concentration provide some insight for interpretation of heterogeneous areas.
DISCUSSION There has been a great deal of interest regarding the MR imaging appearance of melanotic melanoma. This interest has been fueled by the potential benefits from diagnosis of melanoma at extracutaneous sites. Uveal melanoma has been studied with MR imaging by a number of groups. Many of these investigators have reported that melanin appears to be responsible for T1 and T2 shortening, which results in increased signal intensity on TlWls and decreased signal intensity on T2WIs (2,3).These studies correlated the average melanin content of tumors with their appearance on MR images or with the variations in their relaxation measurements. Other investigators studying intracerebral melanoma with hemorrhage have suggested that various forms of iron rather than melanin may be responsible for this effect (7).More recently, it has been suggested that a variety of paramagnetic ions associated with melanin aggregates are largely responsible for this effect (14).In a previous study in which histologic slices were compared to MR imaging sections from melanoma tumors, the authors concluded that although there was a heterogeneous distribution of melanin in the histologic slices, this was not reflected on the images (15).As the authors suggested, the relatively low melanin content of their tumor model may have made it difficult, if not impossible, to demonstrate any effect due to melanin aggregates. In the present study, the M2R tumor model typically had melanin-rich regions. In addition, strict preservation of orientation and thin MR imaging sections enabled us to achieve a fairly high degree of correlation between MR imaging sections and histologic slices, despite comparing 10-bm-thick histologic slices with 680-~m-thick MR imaging sections. The success of this type of study depends on a number of factors including ( a )the signal-to-noise ratio, which limits section thickness and resolution; ( b )preservation of the MR imaging orientation in the histologic slices: and (c) the variation that occurs across the thickness of the MR imaging section. Moreover, gradient coil performance limited us to a minimum TE of 15 msec for the desired section thickness and degree of resolution. In MR microscopy, 10-k.m resolution has been achieved for small specimens
0
Volume 2
Number 6
JMRl
697
Figure 1. MR images and corresponding histologic slice of an M2R mouse melanoma tumor. Tumor cells were implanted in the flank of a male C57 black mouse. Note the detail of the leg muscle surrounded by tumor in the upper-left region of the MR images. This region serves as an internal marker to ensure proper correlation with the histologic slice. Three regions are indicated on the histologic slice: a = melanin-rich area, b = viable melanoma cells, c = area of frank necrosis. These regions are shown in greater detail on the corresponding micrographs (original magnification, x700). Note that the areas with high melanin content are hyperintense on the T 1W1 and hypointense on the T2W1. while the areas of frank necrosis appear hyperintense on each of the MR images. The bar represents 0.5 cm. HE = histologic slice stained with hematoqlin-eosin,PDWZ = proton-density image.
698
JMRl
November/December 1992
Flgure 2. MR images and the corresponding histologic slice of an M2R mouse melanoma tumor. Cells were implanted in the flank of a male C57 black mouse. Three regions are indicated on the histologic slice: a = viable melanoma cells, b = melaninrich area, and c = area of fresh hemorrhage. Note that, in contrast to melanin-rich areas, which are hyperintense on the TlWI and hypointense on the T2W1, fresh hemorrhage is hypointense on the T2WI and on the nominal T1WI and PDWI. The bar r e p resents 0.5 cm. HE = histologic slice stained with hematoxylin-eosin, PDWI = proton-density image.
with the correspondingly small coils and fields of view used. The present study represents an effort to maximize resolution and the achievable correlation at the larger fields of view (5 cm) with which study of tumor morphology is practical. To explore the potential of MR imaging as a noninvasive method for specffic diagnosis, we must continue to strive to minimize MR imagmg section thickness and to maximize the accuracy of obtaining the correct orientation for the plane of the histologic slices. The hyperintensity of frank necrosis on TlWIs may appear puzzling, since the increased fluidity normally associated with necrotic regions is usually accompanied by a longer T1. A full explanation of this observation will require actual, spatially selective measurements of T1 and T2, which were not obtained at this stage of our investigation. Until the results of such measurements are available, we can only speculate about the several possible reasons for this effect. First, small amounts of melanin in these regions could contribute to shortening the T1 of necrotic water. Second, it should be kept in mind that, in general, the contributions to image signal intensities come
from all three MR properties (ie, T1, T2, and proton density). Images dominated by a certain type of contrast (eg, T1) are obtained by imaging under conditions that render the influence of other properties, such as T2, negligible. T2-dependent contrast is suppressed by choosing a short TE, ideally a value that is short relative to the shortest T2 in the sample. This condition may not have been completely fulfilled in our study even at a TE of 15 msec; in other words, T2 values may have contributed substantially to image intensity differences, even under nominally “Tlweighted” imaging conditions. Moreover, whatever the TE value, the elevated proton density observed on the corresponding proton-density images also tends to oppose the expected T 1-dependent effect. It should be noted that our results were obtained at the relatively high field strength of 4.7 T, and, strictly speaking, the quantitative aspects of this study (as reflected in Fig 3) pertain only to this field strength. However, the important qualitative conclusions, namely melanin-rich regions appearing hyperintense on TlWIs and hypointense on T2WIs and the hyperintensity of necrosis on T2WIs. are expected to also apVolume2
Number6
e
JMRl
e
699
experience, the foundation for MR "biopsy" methods and interpretation must be established before potential benefits can be fully exploited. 0 Acknowledgments: We thank Rachel Benjamin for excellent secretarial assistance, Dr Edna Schechtman for her expertise in performing the statistical analysis of the data, and Dr Michal Neeman for helpful comments on the manuscript.
1
References 1.
2.
I
2-
3.
1 T
4. 1-
5. ' 0
I
I
I
MELANIN
NECROSIS
VIABLE n=9
n=9
n5 .
Figure 3. Mean image signal intensities (with standard deviations) of melanin-rich, necrotic, and viable-tumor tissue types. In the top graph, normalized intensities were obtained by dividing the intensity values on the proton-density images (lined bars) and the T2WIs (black bars) (which were obtained from the first and second echo, respectively, of the same acquisition) by the average intensity of muscle on the proton-density images. In the bottom graph, the normalized intensity values on the TlWls (white bars), from a separate acquisition. were obtained by dividing the intensities by the average intensity of muscle on the T1Wls. Therefore, the normalized intensities in the top and bottom graphs are not comparable.
ply at the lower field strengths used for clinical imaging. Using the information obtained in this study, we can now identify certain gross histologic features of this tumor model in vivo on the basis of MR imaging alone. In addition, the high degree of correlation between the distribution of melanin within the histologic slices and the high-signal-intensity regions on the TlWIs, which had low signal intensity on the T2WIs. provides strong evidence that melanin or melaninassociated paramagnetic species are responsible for this effect. Whereas the field of microscopic histology rests comfortably on a well-developed foundation of
700
JMRl
November/December 1992
6. 7. 8.
9. 10. 11.
12.
13. 14.
15.
Sergeev AI, Murza LI. Dependence of proton magnetic relaxation times on free radicals content in human melanoma. Stud Biophys 1984; 103:139-142. Gomori J M , Grossman RI. Shields J A , Augsburger JJ, Joseph PM, DeSimeone D. Choroidal melanomas: correlation of NMR spectroscopy and MR imaging. Radiology 1986: 158:443-445. Mafee MF, Peyman GA, Grisolano JE, et al. Malignant uveal melanoma and simulating lesions: MR imaging evaluation. Radiology 1986; 160:773-780. Keizer RJW, Vielvoye GJ, de Wolff-RouendaalD. Nuclear magnetic resonance imaging of intraocular tumors. Am J Ophthalmol 1986; 102:438-44 1. Peyster RG, Augsburger JJ, Shields JA. Hershey BL. Eagle R J r , Haskin ME. Intraocular tumors: evaluation with MR imaging. Radiology 1988; 168:773-779. Atlas SW, Grossman RI, Gomori J M . et al. MR imaging of intracranial metastatic melanoma. J Comput Assist Tomogr 1987: 11:577-582. WoodrulTWW J r , Djang WT. McLendon RE, Heinz ER, Voorhees DR. Intracerebral malignant melanoma: highfield-strength MR imaging. Radiology 1987; 165:209-213. Schwaighofer BW, Fruehwald FXJ. Pohl-Mark1 H, Neuhold A, Wicke L, Landrum WL. MRI evaluation of pigmented skin tumors: preliminary study. Invest Radiol 1989: 24: 289-293. Zemtsov A. Lorig R, Bergfield WF. Bailin PL,Ng TC. Magnetic resonance imaging of cutaneous melanocytic lesions. Dermatol Surg Oncol 1989: 153854458. Marx FM. Colletti PM. Raval J K . Boswell WD J r , Zee C. Magnetic resonance imaging features in melanoma. Magn Reson Imaging 1990; 8:223-229. Atlas SW, Braffman BH, LoBrutto R, Elder DE. Herlyn D. Human malignant melanomas with varying degrees of melanin content in nude mice: MR imaging. histopatholog, and electron paramagnetic resonance. J Comput Assist Tomogr 1990: 14:547-554. Mather JP, Sat0 GH. The growth of mouse melanoma cells in hormone-supplemented, serum-free medium. Exp Cell Res 1979: 120:191-200. Gerst J, Sole J, Mather JP, Salomon Y. Regulation of adenylate cyclase by P-melanotropin in the M2R melanoma cell line. Mol Cell Endocrinol 1986: 46:137-147. Enochs WS, Hyslop WB, Bennett HF. Brown RD 111, Koenig SH, Swartz HM. Sources of the increased longitudinal relaxation rates observed in melanotic melanoma: an in vitro study of synthetic melanins. Invest Radiol 1989; 24:794803. Iyer SB, Goldberg HI, Benz CC. Magnetic resonance imaging of implanted melanomas before and after chemotherapy: relation to 31Pmagnetic resonance spectroscopy and tumor histology. Invest Radiol 1990; 25: 1076-1084.
Correlation of MR Imaging and Histologic Findings in Mouse Melanoma' John 0. DeJordy, MD Yoram Salomon, PhD
Peter Bendel, PhD Ada Horowitz, MD Hadassa Degani, PhD
M2R melanoma tumors in male C57 black mice were used to correlate magnetic resonance (MR) images with the corresponding histologic slices and to determine if analysis of the achievable correlation can provide a basis for predicting gross histologic features with MR imaging alone. The MR imaging sections obtained at 4.7 T were each 680 Crn thick, with an in-plane resolution of 195 pm. The distribution of melanin within the histologic slices correlated well with the high-signal-intensity regions on the T1-weightedimages (TlWIs),while these regions had low signal intensity on the T2weighted images (T2WIs).providing evidence that melanin or melanin-associated paramagnetic species are responsible for the observed proton relaxation rate enhancement. Viable melanoma cells typically showed intermediate signal intensity on T2WIs. T1WIs, and proton-density images. Necrosis typically had high signal intensity on T2WIs, TlWIs, and proton-density images. Quantitation of the MR imaging results, followed by statistical analysis, demonstrated statistically significant differences between melanin-rich, viable-melanoma, and necrotic regions on MR images.
-
Index term.: Comparative studies High-resolution imaging Melanoma. 40.379 Neoplasms, MR. 40.379 * Tissue characterization JMRllS92; 2595-700 Abbreviations: ANOVA = analysis of variance, LSD = least significant difference. ROI = region of interest. TlWl = TI-weighted image. T2W = T2-weighted image.
I From the DepartmentsorChemical Physics (J.O.D.. P.B.. H.D.1 and The Weinnann Institute of Science. Hormone Research (J.O.D..Y.S.). 76100 Rehovot. Israel: and the Department of Patho1o.g. Sheba Medical Center. Tel-Hashomer, Israel [A.H.).Received June 22. 1992: revision requested August 17: revision received and accepted September 1. Y.S. supported by grants from Ihe Crown Endowment Fund for Immunologiral Research at the Weizmann Institute of Science and the Israeli Ministry of Health. H.D. supported by grants from the Grrmaii-Israeli Foundation for Srientifir Research and Development. Y.S. is the Charles and Tillie Luhin Professor of Hormone Research. Address reprint requests to H.1)
SMRI. 1992
MAGNETIC RESONANCE (MR)imaging has evolved into an important tool in clinical medicine; however, the application of MR imaging has largely been limited to tumor detection rather than tissue-specific diagnosis. In the vast majority of cases, the definitive diagnosis is still based on the microscopic evaluation of biopsied tissue. The potential of MR imaging as a tool for tissue-specific diagnosis is presently being investigated by many groups. Melanotic melanoma has been the subject of a number of these studies. In 1984, Sergeev and Murza claimed that proton relaxation times depended on the free radical content in human melanoma ( 1).Many of the early clinical studies concerned weal or choroidal melanomas of the eye (2-5); other studies have focused on melanomas in the central nervous system (6,7),cutaneous melanomas (8,9), comparisons of melanomas occurring at various sites ( 10).and melanomas in animal models ( 11).Most of these studies have shown that there is a unique enhancement of the proton relaxation rates in melanotic melanoma; however, there is no clear consensus regarding the source of this effect. These studies correlated the MR imaging appearance with the typical histologic characteristics of the tumors. The objective of the present study was to correlate MR images of mouse melanoma tumors with the specific, corresponding histologic slices. Using this MR imaging section to histologic slice correlation approach, we attempted to determine if gross histologic features can be defined with MR imaging alone and to provide evidence regarding relaxation rate enhancement in MR imaging of melanoma. 0
MATERIALS AND METHODS
Cell Culture and Tumor Model The M2R mouse cell line was developed from the B 16 mouse melanoma cell line, which was originally isolated from the C57 black mouse ( 12).M2R cells were cultured routinely a s monolayers in a 1:1 Ham F12 and Dulbecco-modified Eagle medium (Gibco,
695
Paisley, Scotland),supplemented with 10% heatinactivated horse serum ( 13).The M2R cells were harvested from monolayer plates with a rubber policeman. The cells were then counted, rinsed, and resuspended in isotonic saline ( 1 x lo7cells per milliliter). The tumor model was generated by subcutaneous injection of 1 x lo6 M2R cells into the right rear flank of 2-3-month-old male C57 black mice. The site on the flank was selected to minimize motion artifacts due to respiration.
MR Imaging The mice were anesthetized for the study with an intraperitoneal injection of sodium pentobarbital ( = 60 mg/kg). The MR images were obtained with a 4.7-T Biospec spectrometer (Bruker Medizintechnik, Rheinstetten, Germany) by using standard spin-echo sequences. A custom-built 4.5-cm-diameter imaging coil was used in the study. T1-weighted images (TlWIs)were acquired with a TR of 500 msec and a TE of 15 msec. For the TlWIs, 16 were averaged during a period of 35 minutes. T2-weighted images (T2WIs)and proton-density images were obtained with a TR of 2,400 msec and TEs of 68 and 17 msec, respectively. For the T2WIs and proton-density images, eight were averaged during a period of 83 minutes. The MR imaging sections were each 680 p n thick, with an in-plane resolution of 195 km (5-cm field of view and 256 X 256 matrix). Ten to 12 sections were selected for each tumor, with an intersection gap equal to approximately 20% of the section thickness. His tology After completion of imaging, the mice were killed by cervical dislocation. The skin was removed from the region of the tumor, and the whole animal was placed in fixative (75%picric acid, 20% formaldehyde, 5% glacial acetic acid).Before fixation, the tumors have a relatively fluid consistency that makes sectioning impractical. After 2-3 days, the tissue was removed from the fixative and the tumor was marked with stains (Davidson Marking System; Bradley Products, Bloomington, Minn) to help maintain orientation. The histologic plane corresponding to the central section of the MR imaging study was established by a single cut through the tumor. This plane was based on the position of the tumor in the spectrometer, sagittal and transverse images, and anatomic references. Once this initial cut was made, the tumor was removed from the rest of the tissue and processed for histologic examination. Three contiguous 10-km-thick histologic slices were obtained approximately every 100 km from either side of the initial cut. After processing, each of the histologic slice sets were stained with hematoxylin-eosin, a modified trichrome method, and the Prussian blue reaction. Hematoxylin-eosin was used to assess viability, necrosis, and pigmentation. The areas with substantial melanin appeared brown to black on the hematoxylineosin slices. The modified trichrome method was used to identify fibrotic regions, and the Prussian blue reaction was used to identify regions of substantial iron deposition. Note that iron in the form of hemoglobin does not stain positive with the Prussian 696
-
JMRl
-
November/December 1992
blue reaction. The orientation on the slides was maintained with the aid of the original tissue stains and with reference marks made on the slides at the time of slice mounting.
Data Analysis MR imaging section-histologic slice pairs were selected for further study on the basis of the degree of correlation of unique patterns and shapes between the pairs. A high degree of correlation of detailed patterns between the images and histologic slices provided some degree of assurance that a reasonable match had been achieved. To analyze the MR images, regions of interest (ROIs)were selected for six correlated pairs, and the area and average signal intensity were determined for each ROI with the image analyzing software (Bruker Medical Imaging). Each ROI selected for analysis had a minimum area of 50 pixels ( 1.9 mm2).The ROIs were selected on the basis of their relatively homogeneous appearance on all MR images of the same section and histologic slices. For example, if an ROI had homogeneously high signal intensity on the TlWI but consisted of high- and lowintensity areas on the T2WI or proton-density image, it did not meet the selection criteria. The signal intensity values were normalized to the intensity of muscle to eliminate differences arising from variations in the adjustment of the instrument. The signal intensities on the T2Ws and proton-density images (which were obtained from the second and first echo, respectively, of the same acquisition) were normalized to the intensity of muscle by assigning a value of one to the average intensity of muscle on the proton-density images. The intensities on the TlWIs (obtained in a separate acquisition) were normalized separately by assigning a value of one to the average intensity of muscle on the TlWIs. Within well-correlated section-slice pairs, high-powered microscopic evaluation of the ROIs in tumor slices served as the basis for histologic identification. Each ROI was categorized as viable melanoma cells, necrosis, melanin rich, or heterogeneous on the basis of microscopic evaluation of the histologic characteristics of that region. If the R 0 1 had a high melanin content, it was included in the melanin-rich group regardless of the degree of cell viability or necrosis. ROIs that did not have high melanin content and consisted of at least 80%necrosis or viable cells were grouped accordingly. The ROIs in the necrotic group represented frank necrosis with intercellular edema, typically with some melanin present. Regions with signs of early necrosis were not included in the necrosis group. The histologic information was then integrated with the information from the images to determine if any patterns could be appreciated. This initial evaluation was followed by further analysis of the data. A two-way analysis of variance (ANOVA)was used with two main factors: tissue type and image type. Initially, the data were statistically analyzed to determine if a significant correlation existed between image type and tissue type. After the existence of such a correlation was confirmed, a oneway ANOVA was performed for each image type. The ANOVA was followed by multiple comparisons of the tissue types in each type of image (the Duncan multi-
ple range test and the Fischer least significant difference [LSD]test) to determine if the tissue types could be divided into statistically different subgroups on each type of image section. Ultimately, conclusions were based on the information obtained for all three image types.
RESULTS The present study represents an effort to correlate the heterogeneity within a single MR imaging section with the corresponding histologic findings. By establishing a protocol to preserve the orientation of the tumor and by decreasing the MR imaging section thickness to less than 1 mm, we were able to achieve a good correlation between MR imaging and histologic findings. An example of an MR imaging-histology correlation pair, with high-magnification micrographs of selected regions, is presented in Figure 1. In our tumor model, the distribution of melanin within the histologic sections correlated well with hyperintense regions on TlWIs, regions which were hypointense on T2WIs. indicating shortening of T1 and T2. In addition, the Prussian blue reaction demonstrated the absence of any substantial deposition of free iron, indicating that shortening of T1 and T2 was not due to stainable iron deposits. In regons with heavy pigmentation, the effects of melanin appeared to dominate. Analysis of the data indicated that melanin-rich regions corresponded to the regions with the observed T1 and T2 shortening. These results support the view that melanin or melanin-associated paramagnetic species are responsible for this effect. Amelanotic to mildly pigmented viable regions typically showed intermediate signal intensity on TlWIs, T2WIs. and proton-density images, while frank necrosis typically had high signal intensity on all three types of images. Evaluation of the slides stained with the modified trichrome method indicated that there was no substantial fibrosis in these tumors other than in the capsule. In one correlation pair, a fresh hemorrhage was evident on the histologic slice (Fig 2). The region corresponding to the hemorrhage was hypointense on all three types of images, suggesting that in this region T2 relaxation enhancement dominates even at a TE of 15 msec. Figure 3 presents the mean values of the scaled signal intensities for melanin-rich, necrotic, and viablemelanoma ROIs. Statistical analysis of the data indicated that a significant correlation between image type and tissue type was present ( P = .0001).To determine if the tissue type could be distinguished on the basis of MR imagmg alone, one-way ANOVA was performed, followed by multiple comparison tests for each image type. For the normalized intensities of the proton-density images, one-way ANOVA gave a P value of ,0449, while the Fischer LSD test indicated that it was possible to distinguish necrosis from the other two tissue types. For the normalized intensities of the TlWIs, one-wayANOVA gave a P value of .OOO 1, while the Fischer LSD and Duncan tests indicated that it was possible to distinguish viable melanoma from the other two tissue types. Finally, for the normalized intensities of the T2Wls. one-way ANOVA gave a P value of .OOO 1, while the Fischer LSD and Duncan tests in0
dicated that it was possible to distinguish the three tissue types from one another. From this analysis, it appears that the T2WIs alone provide a means of distinguishing tissue types by means of MR imaging. In addition, the TlWIs enable clear differentiation between viable-melanoma regions and necrotic or melanin-rich regions. The proton-density images provided some insight into regions of frank necrosis, but the level of significance was equivocal. The combined information available from the three image types provides a basis for predicting certain aspects of gross histologic findings with MR imaging alone. The analysis of heterogeneous regions is difficult; however, the data obtained from homogeneous regions of amelanotic to slightly pigmented viable cells, regions of necrosis, and regions of h g h melanin concentration provide some insight for interpretation of heterogeneous areas.
DISCUSSION There has been a great deal of interest regarding the MR imaging appearance of melanotic melanoma. This interest has been fueled by the potential benefits from diagnosis of melanoma at extracutaneous sites. Uveal melanoma has been studied with MR imaging by a number of groups. Many of these investigators have reported that melanin appears to be responsible for T1 and T2 shortening, which results in increased signal intensity on TlWls and decreased signal intensity on T2WIs (2,3).These studies correlated the average melanin content of tumors with their appearance on MR images or with the variations in their relaxation measurements. Other investigators studying intracerebral melanoma with hemorrhage have suggested that various forms of iron rather than melanin may be responsible for this effect (7).More recently, it has been suggested that a variety of paramagnetic ions associated with melanin aggregates are largely responsible for this effect (14).In a previous study in which histologic slices were compared to MR imaging sections from melanoma tumors, the authors concluded that although there was a heterogeneous distribution of melanin in the histologic slices, this was not reflected on the images (15).As the authors suggested, the relatively low melanin content of their tumor model may have made it difficult, if not impossible, to demonstrate any effect due to melanin aggregates. In the present study, the M2R tumor model typically had melanin-rich regions. In addition, strict preservation of orientation and thin MR imaging sections enabled us to achieve a fairly high degree of correlation between MR imaging sections and histologic slices, despite comparing 10-bm-thick histologic slices with 680-~m-thick MR imaging sections. The success of this type of study depends on a number of factors including ( a )the signal-to-noise ratio, which limits section thickness and resolution; ( b )preservation of the MR imaging orientation in the histologic slices: and (c) the variation that occurs across the thickness of the MR imaging section. Moreover, gradient coil performance limited us to a minimum TE of 15 msec for the desired section thickness and degree of resolution. In MR microscopy, 10-k.m resolution has been achieved for small specimens
0
Volume 2
Number 6
JMRl
697
Figure 1. MR images and corresponding histologic slice of an M2R mouse melanoma tumor. Tumor cells were implanted in the flank of a male C57 black mouse. Note the detail of the leg muscle surrounded by tumor in the upper-left region of the MR images. This region serves as an internal marker to ensure proper correlation with the histologic slice. Three regions are indicated on the histologic slice: a = melanin-rich area, b = viable melanoma cells, c = area of frank necrosis. These regions are shown in greater detail on the corresponding micrographs (original magnification, x700). Note that the areas with high melanin content are hyperintense on the T 1W1 and hypointense on the T2W1. while the areas of frank necrosis appear hyperintense on each of the MR images. The bar represents 0.5 cm. HE = histologic slice stained with hematoqlin-eosin,PDWZ = proton-density image.
698
JMRl
November/December 1992
Flgure 2. MR images and the corresponding histologic slice of an M2R mouse melanoma tumor. Cells were implanted in the flank of a male C57 black mouse. Three regions are indicated on the histologic slice: a = viable melanoma cells, b = melaninrich area, and c = area of fresh hemorrhage. Note that, in contrast to melanin-rich areas, which are hyperintense on the TlWI and hypointense on the T2W1, fresh hemorrhage is hypointense on the T2WI and on the nominal T1WI and PDWI. The bar r e p resents 0.5 cm. HE = histologic slice stained with hematoxylin-eosin, PDWI = proton-density image.
with the correspondingly small coils and fields of view used. The present study represents an effort to maximize resolution and the achievable correlation at the larger fields of view (5 cm) with which study of tumor morphology is practical. To explore the potential of MR imaging as a noninvasive method for specffic diagnosis, we must continue to strive to minimize MR imagmg section thickness and to maximize the accuracy of obtaining the correct orientation for the plane of the histologic slices. The hyperintensity of frank necrosis on TlWIs may appear puzzling, since the increased fluidity normally associated with necrotic regions is usually accompanied by a longer T1. A full explanation of this observation will require actual, spatially selective measurements of T1 and T2, which were not obtained at this stage of our investigation. Until the results of such measurements are available, we can only speculate about the several possible reasons for this effect. First, small amounts of melanin in these regions could contribute to shortening the T1 of necrotic water. Second, it should be kept in mind that, in general, the contributions to image signal intensities come
from all three MR properties (ie, T1, T2, and proton density). Images dominated by a certain type of contrast (eg, T1) are obtained by imaging under conditions that render the influence of other properties, such as T2, negligible. T2-dependent contrast is suppressed by choosing a short TE, ideally a value that is short relative to the shortest T2 in the sample. This condition may not have been completely fulfilled in our study even at a TE of 15 msec; in other words, T2 values may have contributed substantially to image intensity differences, even under nominally “Tlweighted” imaging conditions. Moreover, whatever the TE value, the elevated proton density observed on the corresponding proton-density images also tends to oppose the expected T 1-dependent effect. It should be noted that our results were obtained at the relatively high field strength of 4.7 T, and, strictly speaking, the quantitative aspects of this study (as reflected in Fig 3) pertain only to this field strength. However, the important qualitative conclusions, namely melanin-rich regions appearing hyperintense on TlWIs and hypointense on T2WIs and the hyperintensity of necrosis on T2WIs. are expected to also apVolume2
Number6
e
JMRl
e
699
experience, the foundation for MR "biopsy" methods and interpretation must be established before potential benefits can be fully exploited. 0 Acknowledgments: We thank Rachel Benjamin for excellent secretarial assistance, Dr Edna Schechtman for her expertise in performing the statistical analysis of the data, and Dr Michal Neeman for helpful comments on the manuscript.
1
References 1.
2.
I
2-
3.
1 T
4. 1-
5. ' 0
I
I
I
MELANIN
NECROSIS
VIABLE n=9
n=9
n5 .
Figure 3. Mean image signal intensities (with standard deviations) of melanin-rich, necrotic, and viable-tumor tissue types. In the top graph, normalized intensities were obtained by dividing the intensity values on the proton-density images (lined bars) and the T2WIs (black bars) (which were obtained from the first and second echo, respectively, of the same acquisition) by the average intensity of muscle on the proton-density images. In the bottom graph, the normalized intensity values on the TlWls (white bars), from a separate acquisition. were obtained by dividing the intensities by the average intensity of muscle on the T1Wls. Therefore, the normalized intensities in the top and bottom graphs are not comparable.
ply at the lower field strengths used for clinical imaging. Using the information obtained in this study, we can now identify certain gross histologic features of this tumor model in vivo on the basis of MR imaging alone. In addition, the high degree of correlation between the distribution of melanin within the histologic slices and the high-signal-intensity regions on the TlWIs, which had low signal intensity on the T2WIs. provides strong evidence that melanin or melaninassociated paramagnetic species are responsible for this effect. Whereas the field of microscopic histology rests comfortably on a well-developed foundation of
700
JMRl
November/December 1992
6. 7. 8.
9. 10. 11.
12.
13. 14.
15.
Sergeev AI, Murza LI. Dependence of proton magnetic relaxation times on free radicals content in human melanoma. Stud Biophys 1984; 103:139-142. Gomori J M , Grossman RI. Shields J A , Augsburger JJ, Joseph PM, DeSimeone D. Choroidal melanomas: correlation of NMR spectroscopy and MR imaging. Radiology 1986: 158:443-445. Mafee MF, Peyman GA, Grisolano JE, et al. Malignant uveal melanoma and simulating lesions: MR imaging evaluation. Radiology 1986; 160:773-780. Keizer RJW, Vielvoye GJ, de Wolff-RouendaalD. Nuclear magnetic resonance imaging of intraocular tumors. Am J Ophthalmol 1986; 102:438-44 1. Peyster RG, Augsburger JJ, Shields JA. Hershey BL. Eagle R J r , Haskin ME. Intraocular tumors: evaluation with MR imaging. Radiology 1988; 168:773-779. Atlas SW, Grossman RI, Gomori J M . et al. MR imaging of intracranial metastatic melanoma. J Comput Assist Tomogr 1987: 11:577-582. WoodrulTWW J r , Djang WT. McLendon RE, Heinz ER, Voorhees DR. Intracerebral malignant melanoma: highfield-strength MR imaging. Radiology 1987; 165:209-213. Schwaighofer BW, Fruehwald FXJ. Pohl-Mark1 H, Neuhold A, Wicke L, Landrum WL. MRI evaluation of pigmented skin tumors: preliminary study. Invest Radiol 1989: 24: 289-293. Zemtsov A. Lorig R, Bergfield WF. Bailin PL,Ng TC. Magnetic resonance imaging of cutaneous melanocytic lesions. Dermatol Surg Oncol 1989: 153854458. Marx FM. Colletti PM. Raval J K . Boswell WD J r , Zee C. Magnetic resonance imaging features in melanoma. Magn Reson Imaging 1990; 8:223-229. Atlas SW, Braffman BH, LoBrutto R, Elder DE. Herlyn D. Human malignant melanomas with varying degrees of melanin content in nude mice: MR imaging. histopatholog, and electron paramagnetic resonance. J Comput Assist Tomogr 1990: 14:547-554. Mather JP, Sat0 GH. The growth of mouse melanoma cells in hormone-supplemented, serum-free medium. Exp Cell Res 1979: 120:191-200. Gerst J, Sole J, Mather JP, Salomon Y. Regulation of adenylate cyclase by P-melanotropin in the M2R melanoma cell line. Mol Cell Endocrinol 1986: 46:137-147. Enochs WS, Hyslop WB, Bennett HF. Brown RD 111, Koenig SH, Swartz HM. Sources of the increased longitudinal relaxation rates observed in melanotic melanoma: an in vitro study of synthetic melanins. Invest Radiol 1989; 24:794803. Iyer SB, Goldberg HI, Benz CC. Magnetic resonance imaging of implanted melanomas before and after chemotherapy: relation to 31Pmagnetic resonance spectroscopy and tumor histology. Invest Radiol 1990; 25: 1076-1084.
