Surface Profiling of Histological Specimens by using AFM
Prior to our AFM indentation experiments, we performed AFM imaging for morphological analysis of histological specimens (Fig. 2a-c and Supplementary Fig. S1). The AFM micrographs showed circular or elliptical craters (approximately 10 µm), which are presumed to be cell markers (Fig. 2d-f). We found that the AFM topographies notably differed among normal, benign nevus, and melanoma. We believed that the higher density and correspondingly smaller sizes of the craters had evolved during nevus development, but there were still some aligned crater structures in the nevus specimens. In contrast, the melanoma showed irregular crater shapes, sizes, densities, and even alignments. These structural irregularities could have been formed during the transformation to malignancy 27,28, and it can affect the changes in the physical properties of the samples.
For the quantitative analysis beyond that of appearance, we compared the histograms of height distribution from each AFM micrograph in Fig. 2g-i. The standard deviations (σ) of the histograms, fitted to Gaussian distributions, are shown in Fig. 2j. In addition, we attempted to calculate the surface roughness (Sq) from the AFM micrographs, but we found that both metrics (i.e., σ and Sq) were ineffective at discriminating among normal, benign nevus, and melanoma (Fig. 2k and Supplementary Fig. S2). The results of morphological analysis were similar because the histological specimens were finely sectioned at the same thickness (4 µm). Although the microscopic analysis (i.e., AFM imaging) showed higher-resolution images compared with macroscopic inspection (e.g., simple visual inspection using ABCDE criteria), we comprehensively demonstrated the vulnerability of AFM image analysis for melanoma detection using histological specimens.
Mechanical Properties of Histological Specimens that can Discriminate Between Benign Nevus and Melanoma
To overcome the vulnerability of AFM image analysis, we performed AFM indentation of the histological specimens in the vicinity of the epidermis (Fig. 3a-c). The mechanical properties of specimens can be quantified by using FD curve measurements, which we acquired from 100 independent locations throughout the entire lesion. To confirm whether the histological specimen is softer than the substrate (i.e., glass) or not, we evaluated the mechanical properties of the glass substrate. We observed that the elasticity of the substrate that supported the histological specimens was exceeded the usual range of the mechanical properties of histological specimens, indicating that the heterogeneous properties of the histological specimens are not influenced by the substrate (Supplementary Fig. S3). Therefore, we were convinced that our experimental setup and the following results are reasonable.
All the FD curves are presented from the contact point to the maximum indentation depth in the form of approach curves after the contact point. The displayed FD curves (Fig. 3d-f) is a set of the approach curves where the contact points of the FD curves fit in the origin (d = 0 µm, F(d) = 0 µN). The majority of FD curves from normal sample exhibited linear behavior during the indentation (Fig. 3d). The majority of FD curves from the benign sample showed relatively steeper slopes in the linear region than did the normal sample, implying that benign sample retains higher resistance to elastic deformation. It is noteworthy that a few of the FD curves in the benign sample revealed non-linear characteristics (Fig. 3e). In contrast, the FD curves for the melanoma specimen exhibited non-linear characteristics (Fig. 3f), suggesting that benign nevus is relatively harder than normal sample, whereas melanoma appears to combine softness and hardness. We will discuss the non-linear characteristics of FD curves in detail later in this article.
We conducted the stiffness mapping of the specimens; stiffness maps are derived from FD curves, and they show the deformation of samples. In our study, a representative stiffness map of a benign nevus revealed that overall stiffness was higher than that in the normal sample (Fig. 3g and h). In contrast, the stiffness map of the melanoma displayed randomly blended colors, indicating a mixture of soft and hard materials, unlike with normal or nevus sample (Fig. 3i). This suggests that the local elastic properties of melanoma are heterogeneous.
To quantify the extent of any heterogeneity, we calculated the elastic modulus (Ea) using the FD curve data. The average Ea in the normal sample was 401 ± 148 MPa, and that of the benign nevus was 575 ± 107 MPa. Here, we found two things: The elasticities of the benign nevus were generally 100 MPa higher than those of the normal sample, and the histograms of the elasticity distributions of the normal and benign samples followed Gaussian distributions with single peaks (Fig. 3j and 3k). In contrast, the elasticities of the melanoma were described by multiple Gaussian distributions (188 ± 78, 497 ± 110, and 787 ± 56 MPa; Fig. 3l). This result implies that the epidermis of melanoma consists of heterogeneous materials including soft matter. It is surprising because such mechanical characteristics are of tumor tissue with ECM 27,28,40. Perhaps, the role of ECM in the tumorigenic process is seemed to replace with some factors (e.g., melanin transfer and pigmentation) in melanoma development. Meanwhile, we checked whether the samples would retain similar trends in elastic properties even after long periods of time, and this is of practical clinical importance as well; our AFM indentation testing verified that the characteristics of the melanoma such as the multimodal Gaussian distribution survived for more than three months (Supplementary Fig. S4). Previous studies have found that using histological specimens has a strong advantage in overcoming the weakness of the time-dependent biodegradation of biopsied tissues 35,41.
The results for the mechanical properties of all specimens are summarized in Table 1 and shown in Supplementary Fig. S6, S7, and S8. We found that the elasticities of all histological specimens could be separated into three segmented regions across the full range: the ranges were region I (0–300 MPa), region II (300–600 MPa), and region III (600–900 MPa). The different specimens’ Ea value ranges depended on the specimens’ origins; for instance, the Eas for all normal samples were in region II (300–600 MPa), exhibiting a single-mode Gaussian distribution, and the Eas for the benign nevus spread across regions II and III and displayed single-mode Gaussian distributions. However, the elasticities of the melanoma exhibited multimodal Gaussian distributions that ranged from regions I to III and mainly consisted of three peaks (occasionally there were two peaks). It should be noted that this classification is highly useful in testing the versatility of our approach with all epidermal lesions regardless of sample type.
Summary of mechanical analysis and histological examination of biopsied samples with patient information. The corresponding elasticity distributions are shown in Figure S6, S7, and S8).
Versatility of the Mechanical Specimen Signatures Regardless of Age, Sex, or Site
The epidermis is the outermost layer of the skin, which suggests that it can be easily deformed and that it has different mechanical properties depending on age, sex, and site 42,43. To clinically apply our methodology, it was essential to verify whether the mechanical signatures for discrimination would hold constant irrespective of skin tissue type. Thus, we classified the mechanical properties of the specimens according to age (1–81 years), sex (male or female), and site (thigh, inguinal, cheek, leg, arm, back, flank, abdomen, and elbow). Surprisingly, the specimen signatures varied based on normal, benign nevus, or melanoma samples regardless of age, sex, or site (Fig. 4a-i). The average Eas for the normal and benign nevus samples fell in region II or regions II and III — showing single-mode Gaussian distributions — whereas those of the melanoma mostly had multiple peaks and were distributed throughout all the elastic regions (I, II, and III). In detail, the Eas for the normal samples ranged from 370 to 521 MPa, all within region II. The benign nevus had higher Eas, ranging from 441 to 848 MPa, in regions II and III. In contrast, the melanoma displayed three peak Gaussian distributions, and each peak had a different average: E1a = 158–274 MPa (region I), E2a = 363–542 MPa (region II), and E3a = 606–893 MPa (region III).
Whether the melanoma cancer stage can distort the mechanical signatures is another important factor for determining the clinical applicability of our method. It is well-known that melanomas become darker, more distorted, and physically harder as the stage increases (Fig. 4j). Accordingly, we determined the cancer stage of each melanoma based on Clark level (see Methods) and then inspected all the melanoma by AFM indentation. Notably, all samples exhibited multiple Gaussian peaks; each Ea occupied its own elastic region (Fig. 4k). From our results, it is plausible that the mechanical signature of the multiple peaks survives in melanoma irrespective of the cancer stage, even in stage I, and the existence of these multiple peaks in stage I melanoma provides us with a strong advantage in early detection.
At this stage, we needed to scrutinize in detail how the melanoma cancer stage affected its mechanical characteristics. One particular point is that unlike with other cancers, melanoma’s mechanical properties appeared to vary significantly depending on the stage. Specifically, each of our melanoma samples showed a most prevalent peak (Pm) — one with the highest population — among its multiple peaks (Supplementary Fig. S9). We considered the Pm and its corresponding elastic region, I, II, or III, as important parameters. We conducted our statistical analyses of each category based on the two parameters above, and we present the results in Fig. 4l and Supplementary Fig. S9. We found no significant differences in the mechanical properties of melanoma by sex, age, or site. However, we found that the melanoma specimens tended to become harder as the cancer stage advanced. In particular, all early-stage (stage I) melanoma samples had their own Pm values, which were all in the region I, which implied that the specimens contained large amounts of soft material. However, in the late melanoma stages (IV and V), the Pm values ranged across all three elastic regions, I, II, and III. This finding indicates that at the late stage, hard material fills the samples in larger amounts than in the early stage. This phenomenon is consistent with the fact that melanin accumulates as the cancer stage advances 44. Taking these findings together, although there were subtle differences by category and cancer stage, we found it surprising that all the melanoma samples still retained their mechanical signature of multiple Gaussian peaks.
Non-linearity of FD Curves for Normal, Benign Nevus, and Melanoma
Meanwhile, we paid attention that the FD curves for normal and benign nevus samples exhibited linear behavior whereas the FD curves of melanoma showed non-linear characteristics (Fig. 3). To quantitatively compare the non-linear characteristics of each sample, we calculated the non-linearity from FD curves as follows. As shown in Fig. 5a, we draw a straight line (red-dashed) between the contact-point (Xmin, Ymin) and end-point (Xmax, Ymax) in the FD curve. We define this straight line as the ideal curve (i.e., perfectly linear line):