Research ArticleCancer

Pump-Probe Imaging Differentiates Melanoma from Melanocytic Nevi

See allHide authors and affiliations

Science Translational Medicine  23 Feb 2011:
Vol. 3, Issue 71, pp. 71ra15
DOI: 10.1126/scitranslmed.3001604


Melanoma diagnosis is clinically challenging: the accuracy of visual inspection by dermatologists is highly variable and heavily weighted toward false positives. Even the current gold standard of biopsy results in varying diagnoses among pathologists. We have developed a multiphoton technique (based on pump-probe spectroscopy) that directly determines the microscopic distribution of eumelanin and pheomelanin in pigmented lesions of human skin. Our initial results showed a marked difference in the chemical variety of melanin between nonmalignant nevi and melanoma, as well as a number of substantial architectural differences. We examined slices from 42 pigmented lesions and found that melanomas had an increased eumelanin content compared to nonmalignant nevi. When used as a diagnostic criterion, the ratio of eumelanin to pheomelanin captured all investigated melanomas but excluded three-quarters of dysplastic nevi and all benign dermal nevi. Additional evaluation of architectural and cytological features revealed by multiphoton imaging, including the maturation of melanocytes, presence of pigmented melanocytes in the dermis, number and location of melanocytic nests, and confluency of pigmented cells in the epidermis, further increased specificity, allowing rejection of more than half of the remaining false-positive results. We then adapted this multiphoton imaging technique to hematoxylin and eosin (H&E)–stained slides. By adding melanin chemical contrast to H&E-stained slides, pathologists will gain complementary information to increase the ease and accuracy of melanoma diagnosis.


Melanomas are among the most commonly occurring cancers, but they are clinically challenging to diagnose. Early detection remains difficult but is critical for successful treatment. Indeed, from 1991 to 2005, whereas overall cancer death rates decreased by 19% in men and 11% in women, the death rate for melanoma in the United States increased more than 5% (1). Presently, the best method for clinical evaluation of melanoma is visual inspection by simple magnification and epiluminescence (dermoscopy); however, this technique is less than reliable, with some studies rating its sensitivity at under 85% (2). The current standard for melanoma diagnosis remains biopsy and histopathology, but this too results in discordant conclusions because there is no one histological criterion for melanoma, instead a diagnosis must be made by subjectively weighing a series of separate indicators that may be present in atypical lesions as well. A recent study found a discordance rate of 14% among pathologists for melanoma diagnosis, which would imply that 214,000 to 643,000 cases annually in the United States would be diagnosed differently by another pathologist (3). Doctors must therefore err on the side of caution, which leads to an excess of false-positive diagnoses and increased medical costs, emotional trauma, and insurance difficulties from unnecessary surgeries, lymph node biopsies, and other treatments.

Melanoma presents a promising target for optical diagnosis both because suspicious lesions are accessible and because the two dominant types of melanin (eumelanin and pheomelanin) have a specific, intrinsic molecular contrast when imaged. Eumelanin is a brownish black pigment and has photoprotective and antioxidant properties. Pheomelanin is reddish yellow and exhibits phototoxic and pro-oxidant behavior (4). Because melanin carries information about the metabolism and location of melanocytes and melanogenesis in tissue, the distribution of eumelanin and pheomelanin could act as a marker for disease. Indeed, reflectance spectroscopy measurements on thousands of pigmented lesions in vivo found greater amounts of eumelanin than pheomelanin in malignant melanoma (5, 6). Unfortunately, reflectance spectroscopy methods are only capable of bulk measurements and are insensitive to crucial morphological features. In contrast, multiphoton scanning laser microscopy and intravital imaging has found broad use in the biomedical sciences (712). It provides inherent optical sectioning, even in highly scattering media such as tissue, with resolution sufficient to examine intracellular detail because nonlinear signals can be highly localized (1315). This technique allows the acquisition of three-dimensional images with submicrometer resolution in vivo, giving an “optical biopsy” with a minimally invasive procedure. However, traditional multiphoton detection methods use fluorescence or short-wavelength light generation, the signals of which are difficult to detect in pigmented lesions.

Recent advances in laser science have enabled tissue imaging based on many forms of nonfluorescent endogenous contrast, such as vibrational transitions, excited state absorption, and nonlinear refractive index changes (1622). Nonlinear spectroscopy techniques based on laser pulse train shaping developed by our laboratory have been previously used to image total melanin in tissue samples and microvasculature in vivo (23). Here, we focus instead on the chemical type and distribution of melanins, which reflect the metabolism and behavior of melanocytes. Currently, the only way to determine the composition and quantity of melanins in biological samples such as skin is through destructive chemical degradation and analysis (24). Here, we have extended pump-probe imaging to optically discriminate between eumelanin and pheomelanin and applied this approach to pigmented lesions of the skin, examining the biochemical and morphological changes that mark the transition from common melanocytic nevus to malignant melanoma. Immature melanocytes, junctional and dermal nests, and other morphological diagnostic indicators were imaged without the need for specialized stains or sample treatments. We discovered a marked chemical shift toward eumelanin in melanoma samples. Therefore, pump-probe imaging may provide a noninvasive way to improve melanoma diagnosis.


Distinguishing eumelanin and pheomelanin with pump-probe spectroscopy

Unstained slices from a set of 42 excised pigmented lesions were examined by pump-probe imaging and spectroscopy. Tissue slices were obtained from a variety of pigmented lesion types including benign dermal nevi (n = 6), blue nevi (n = 2), compound (n = 5) and dysplastic nevi (n = 12), melanoma in situ (n = 2), invasive primary melanoma (n = 7), melanoma arising in dysplastic nevi (n = 4), pigmented basal cell carcinoma (n = 2), and seborrheic keratoses (n = 2). The Breslow thickness of the invasive primary melanomas ranged from 0.32 to 1.98 mm, with an average of 0.78 mm. The slices were imaged with multiphoton scanning laser microscopy, with a laboratory-designed microscope using 0.75 mW of 720-nm pump light, 0.5 mW of 810-nm probe light, and a 40× 0.75 numerical aperture (NA) objective, and collecting the transmitted probe light (Fig. 1, A and B). Pump-probe images were compared to immunohistochemically MART-1 (melanoma-associated antigen recognized by T cells 1)–stained slices to confirm the source of signal as melanosomes (fig. S1).

Fig. 1

Tissue pump-probe signals. (A) Pump-probe image of a compound nevus at 0-fs interpulse delay. Under these conditions, all melanins demonstrated a strong negative signal (blue/green). (B) Interpulse delay increased to 300 fs. The blue layer along the stratum corneum is due to surgical ink (arrowhead). Regions containing eumelanin have positive signal (red/orange). (C) Fractional eumelanin concentration image: red regions are purely eumelanic, blue regions are purely pheomelanin. Melanocytes along the basal layer expressed a mix of melanins that lean toward pheomelanin (arrows), whereas large deposits of melanin in the dermis are mostly eumelanin (box 2). (D) Pump-probe time delay traces comparing tissue regions of interest 1 and 2 [white boxes in (A) and (B)] with pure solution melanins. Note how tissue melanins behaved similarly to pure solution melanins. Delay traces for all samples have been offset to zero at −1 ps to remove a small contribution from a long-lived (nanoseconds) component to best compare the transient signals. (E) The first three principal components found in tissue pump-probe signals (loadings plot). The first two components account for more than 98% of the variance. (F) Sepia eumelanin and synthetic pheomelanin delay traces when pumped at 720 nm and probed at 810 nm, fitted as a linear combination of components 1 and 2 obtained from PCA. Scale bars, 100 μm.

It was found that at zero delay between the pump and the probe pulses, both populations of tissue melanins have a negative signal (indicating increased probe pulse transmission when the pump pulse is present). This can be seen in the image of a compound nevus presented in Fig. 1A. When the interpulse delay was increased to 300 fs to remove the contribution of any coherent or instantaneous effects (such as sum frequency absorption or stimulated Raman scattering), some tissue regions demonstrated a positive signal from real excited states generated by the pump pulse (Fig. 1B, box 2) and some did not (Fig. 1B, box 1).

We then confirmed that the pump-probe signal arose from populations of eumelanin and pheomelanin. Figure 1D compares the pump-probe behavior of tissue with that of pure reference samples of eumelanin (Sepia officinalis–derived) and pheomelanin (synthetic). Some tissue regions showed a similar delay trace to synthetic pheomelanin, whereas others behaved similarly to eumelanin. Pheomelanin exhibited a negative signal, most likely from ground-state depletion. The negative pheomelanin signal was extremely short-lived, barely longer than the width of the pulse overlap. Eumelanin exhibited an excited state with a lifetime of a few picoseconds and a positive signal, indicating excited-state absorption. We also compared tissue and solution melanin responses at different wavelengths, as well as with red and black human hairs, and found behavior consistent with direct observation of eumelanin and pheomelanin in skin (fig. S2).

To quantify the melanin signal, we performed principal components analysis (PCA). We selected as a data set for PCA 137 different regions from a varied sample set including benign dermal nevi (n = 4), compound nevi (n = 5), dysplastic nevi (n = 5), melanoma in situ (n = 1), invasive primary melanoma (n = 3), pigmented basal cell carcinoma (n = 1), and seborrheic keratoses (n = 2) and collected data at 21 interpulse delays, from −1 to 4 ps. From these 137 independent decays, PCA revealed only two major components of pump-probe signals in the samples tested. These two components accounted for more than 98% of the total signal variance, indicating the skin slice images can be explained by a simple two-chromophore system (Fig. 1E). The components or eigenvectors generated by PCA are necessarily orthogonal; the delay traces of eumelanin and pheomelanin are not. Pure solution melanin traces of each melanin can be represented as a linear combination of the main components (Fig. 1F). Pheomelanin was closely related to pure PCA component 2, whereas eumelanin was a combination of PCA components 1 and 2. Therefore, among the 137 different regions in tissue from 21 separate patients containing a wide array of biological variability, the complicated time-dependent pump-probe signal almost completely resulted from only the two base melanin spectra. Thus, almost all of the variation in our tissue images could be decomposed into eumelanin and pheomelanin contributions.

On the basis of the reference solution melanin spectra, it was possible to roughly quantify the relative concentrations of eumelanin and pheomelanin in tissue samples. The reference solution delay traces were first represented as a linear combination of the first two principal components as determined from PCA by a least-squares fit regression analysis. This allowed us to calculate the ratio of components one and two for any given mixture of eumelanin and pheomelanin. The composition of melanin in each pixel was represented as a fractional melanin image. Figure 1C shows the fractional eumelanin content of a compound nevus. The epidermal pigment was a mix of melanins with a higher abundance of pheomelanin (represented as areas of blue and green, see box 1), with some isolated deposits of eumelanin in the dermis (orange to red, see box 2).

Chemical and morphological differences between malignant melanoma and benign lesions

Figure 2A shows an image spanning almost 2 mm of a melanoma, capturing the entire lesion. When imaged at 0-fs interpulse delay, all melanins should be visible and exhibit a negative signal. Therefore, an image at 0-fs delay was used to image the total melanin content. At the edges of the melanoma in this image, the melanin was distributed at the base of the epidermis, in what appears to be a single layer of cells (arrow). Toward the center of the image, melanized cells appeared deep in the dermis, which differs from their location in healthy tissue (fig. S3). The most pronounced feature of this lesion was found by increasing the interpulse delay to 300 fs and imaging again (Fig. 2B). At this delay, eumelanin should exhibit a positive signal, whereas pheomelanin should contribute almost no signal. Pheomelanin predominates at the edges of the image and in the epidermis; however, eumelanin is primarily expressed where pigmented cells were found deep in the dermis of the melanoma. We confirmed this by calculating the fractional eumelanin content of the lesion. At the basal layer (Fig. 2C, arrow), the mix of melanins tended toward pheomelanin, whereas in the dermis, eumelanin was the dominant pigment. Hematoxylin and eosin (H&E) staining (Fig. 2D) showed melanocytes in the dermis at the center of the lesion (arrow), but not along the margins.

Fig. 2

Malignant melanoma compared with a benign nevus. (A) Malignant melanoma imaged at 0-fs interpulse delay, showing melanin localized mainly to the basal layer at the edge of the lesion (arrow) and deep in the papillary dermis at the center. (B) Increasing the interpulse distance to 300 fs showed that the center of the melanoma contained far more eumelanin than the regions of normal tissue. (C) Fractional eumelanin content of the same melanoma. The arrow labels the basal layer. (D) H&E-stained slice of the same melanoma, melanocytes in the dermis labeled with an arrow. (E) Mosaic image of a benign nevus at 0-fs interpulse delay. Melanin was localized only to the basal layer of the epidermis (arrow). (F) Increasing the interpulse delay to 300 fs showed that the nevus had little eumelanin content. (G) Fractional eumelanin content image; this nevus was pheomelanic and uniformly pigmented. Arrows in (E) to (G) mark the base of the epidermis. (H) H&E-stained slice of the same nevus. Melanocytes were found in the dermis, which had no melanin pump-probe signal (arrow), indicating that they have matured and are of little clinical concern. Scale bars, 100 μm. Arrowheads denote the surface of the skin.

In contrast to the melanoma images, none of the benign nevi had these features. Figure 2E shows a benign nevus spanning 1.8 mm, with a neighboring slice stained with H&E in Fig. 2H. This benign nevus was mostly composed of a pheomelanin signal. Many melanocytes were present in the dermis, as shown in the H&E slide (arrow), but the melanin image revealed that these melanocytes were unpigmented, indicating that they have matured. Immature melanocytes are more likely to be observed in a cancerous lesion. In addition, the pump-probe image of Fig. 2E showed pigmented melanocytes distributed along the basal layer, where they would appear in normal tissue (arrow). In contrast, the pump-probe image contained no melanocytes in the superficial layers of the epidermis, indicating no pagetoid spread, which was confirmed in the H&E slide. There also were no junctional nests of melanocytes, which would appear as a clustering of pigmented cells at the basal layer. Normal skin was found to have similar features in pump-probe images as benign nevi (fig. S3). A lack of signal when the probe was delayed shows the lesion to be pheomelanic (Fig. 2, F and G).

The predominance of eumelanin in melanoma and a more balanced mix in nondiseased epidermal tissue was observed through all of the tissue slices examined. All nine melanoma samples contained at least 38% eumelanin (Fig. 3). This was calculated by performing a weighted average across the entire image, normalizing the percent eumelanin content in each pixel by the total melanin content of that pixel. Regions containing surgical ink were not considered. Dysplastic nevi showed a tendency to have higher percentages of pheomelanin than melanomas. Looking simply at raw melanin content, a threshold of 38% eumelanin captured all melanomas (n = 9) while excluding 75% of the dysplastic nevi (n = 12).

Fig. 3

Diagnostic utility of fractional eumelanin content. The average eumelanin content for primary melanomas and benign, compound, and dysplastic nevi was calculated by finding the eumelanin content of each pixel in an image and weighting by the total melanin content in each pixel. Results for 32 pigmented lesions are shown: benign dermal nevi (n = 6), compound nevi (n = 5), dysplastic nevi (n = 12), and melanoma [n = 9; including melanoma in situ (n = 2) and invasive primary melanoma (n = 7)]. Lesion types with fewer than five samples investigated, such as blue nevi, were excluded. This shows the trend of increasing eumelanin in melanomas compared to melanocytic nevi. A threshold of 38% eumelanin separates all the melanomas from three-quarters of the dysplastic nevi.

Another malignant melanoma is presented in Fig. 4. From the total melanin image (Fig. 4A), many pigmented cells in the dermis were visible. Nests abound in the junctional layer and the reticular dermis (arrows). We also saw that the nature of the melanin was highly heterogeneous after increasing the interpulse delay to 300 fs (Fig. 4B). The fractional eumelanin content image (Fig. 4C) shows that eumelanin accounted for most of the pigment, but several nests were composed of pheomelanic melanocytes (one nest of each type is highlighted with an arrow). H&E staining confirmed these structures as junctional and dermal melanocyte nests (Fig. 4D, arrows).

Fig. 4

Large nests of heterogeneous melanin in malignant melanoma. (A and B) Malignant melanoma imaged by pump probe at 0-fs (A) and 300-fs (B) interpulse delay. Arrows denote junctional and dermal nests. (C) The dermis was highly pigmented and composed of a heterogeneous mix of melanin. Arrows highlight nests of different melanin compositions. (D) H&E staining confirms the presence of melanocytic nests and immature melanocytes in the dermis. Scale bars, 100 μm. Arrowheads denote the surface of the skin.

Figure 5 shows a dysplastic nevus. As opposed to the benign nevus, melanocytes formed junctional nests with bridging between rete ridges (Fig. 5A, arrow). Pigmented keratinocytes were seen in the upper epidermis (indicated by the melanin caps around the nuclei of the keratinocytes, which appear dark) (Fig. 5A, arrowhead), but there was not a pagetoid melanocyte spread. The pigment was mixed but weighted toward pheomelanin (Fig. 5C), which was typical of the studied dysplastic nevi. The lesion was relatively uniformly pigmented. The sparsely distributed melanin found in the dermis is likely excess melanin that had been scavenged by macrophages because free melanin is toxic. We can see that this melanin was different from the dominant species found in the basal layer; it showed eumelanin character (Fig. 5B, arrows).

Fig. 5

Dysplastic nevus with pheomelanic character and little dermal melanin. (A) Pump-probe image at 0-fs interpulse delay. The blue/green layer at the surface of the skin was due to surgical ink applied to denote margins. Melanin was distributed in pigmented cells along the basal layer (arrows) as well as epidermal keratinocytes (arrowheads). There are no dermal nests. (B) A lack of positive signal at 300 fs showed that the melanin was a mix, which leant toward pheomelanin. Red spots in the dermis corresponded to melanin in macrophages being decomposed (arrows). (C) Fractional eumelanin image confirms dominant pheomelanin character. (D) High-resolution image of pigmented cells at the basal layer showing melanin caps surrounding cellular nuclei (arrows) (60×, 0.9 NA, 0-fs interpulse delay). (E) H&E-stained slice confirming junctional, bridging nests. Scale bars, 100 μm.

Figure 6 compares three different types of pigmented lesions. The periphery of a borderline melanoma contained a mix of pheomelanin and eumelanin (Fig. 6A). However, nests of melanocytes located closer to the center of the lesion (white arrows) contained much more eumelanin than the surrounding epidermis or nests on the periphery (yellow arrow). In contrast, a strongly eumelanic dysplastic nevus could not be differentiated from melanomas on the basis of bulk eumelanin content alone (Fig. 6B). Imaging demonstrated that this nevus contained little chemical heterogeneity and that there were no ascending melanocytes or pigmented dermal nests. In addition, a seborrheic keratosis also contained a large amount of eumelanin (Fig. 6C). This benign lesion was uniformly pigmented, and melanocytes were arranged only along the basal layer (arrow).

Fig. 6

Borderline cases. (A) Fractional eumelanin image of a melanoma, showing chemical heterogeneity of nests. Nested melanocytes toward the center of the lesion appeared eumelanic (white arrows), but other epidermal pigmented cells and a nest on the periphery (yellow arrow) contained a mix of eumelanin and pheomelanin. (B) Fractional eumelanin image of a strongly eumelanic dysplastic nevus, showing uniform pigmentation. Nests (white arrows) contained little or no melanin. (C) Fractional eumelanin image of a seborrheic keratosis showing heavy but uniform eumelanin pigmentation. Pigmented melanocytes remain along the basal layer (arrow). (D to F) Corresponding H&E-stained slides of the lesions shown in (A) to (C). Scale bars, 100 μm. Arrowheads denote the surface of the skin.

Imaging eumelanin and pheomelanin on H&E-stained slides

Pump-probe imaging relies entirely on a new contrast, melanin composition. Without a connection with long-established architectural and cytological information from traditional H&E-stained slides, it would be difficult for physicians to integrate this signal with more traditional diagnostic methods. To overcome this problem, we have begun developing this technique for application with H&E-stained slides. We shifted the light source toward lower energies to reduce the response of H&E, which have absorbance maxima in the visible spectrum between 500 and 600 nm (fig. S4). When pumped at 740 nm and probed at 830 nm, eumelanin and pheomelanin continued to have opposite signal signs (Fig. 7A). However, hematoxylin (and to a lesser extent eosin) also responded at these wavelengths with a negative signal (Fig. 7B). Therefore, melanin was separated from hematoxylin by pumping at 830 nm and probing at 740 nm, where H&E stains have a small positive response and both melanins have a strong negative signal. By reversing the pump and probe wavelengths and imaging again, it was possible to build a composite image on H&E-stained slides (Fig. 7, C and D, and fig. S5).

Fig. 7

Melanin imaging in H&E-stained slices. (A) Pump-probe delay traces for Sepia eumelanin and synthetic pheomelanin at 740-nm pump/830-nm probe. Note the opposite signal between melanins. (B) Pump-probe traces for H&E. Hematoxylin concentration was 0.6% by weight, and eosin 1% by weight. (C) H&E-stained dysplastic nevus, imaged with pump probe at both 740-nm pump/830-nm probe and 830-nm pump/740-nm probe. The green channel is negative signal at 740-nm pump (pheomelanin), the red channel reflects positive signal at 740-nm pump (eumelanin), and the purple channel contains positive signal at 830-nm pump (H&E stains). The pigment was pheomelanic and localized to the basal layer of the epidermis (arrowheads), with some bridging junctional nests (arrow). Melanin surrounded the nuclei of melanocytes and keratinocytes (inset, arrows). (D) H&E-stained malignant melanoma imaged in the same fashion. Large deposits of melanin were found in junctional and dermal nests of melanocytes, including quantities of eumelanin (arrows). Scale bars, 100 μm.

H&E-stained slides allowed differentiation between melanocytes and pigmented keratinocytes based on features such as the presence of desmosomes, the quantity of cytoplasm, size of the nucleus, and morphology of the cell (25). In the imaged dysplastic nevus, pigmented melanocytes were found mostly along the basal layer in the epidermis, forming nests at the base of rete ridges (Fig. 7C, arrow). The lesion was predominantly pheomelanic. A high-resolution scan (Fig. 7C, inset) showed melanin in the cytoplasm surrounding nuclei stained by hematoxylin (arrows). The melanoma showed both dermal and junctional nests, which included the typical deposits of eumelanin (Fig. 7D, arrows).


Clinical imaging of pigmented lesions in the past has offered a number of challenges. There are several sources of intrinsic fluorescence and second-harmonic generation, but none are specifically applicable to melanoma diagnosis. Techniques such as fluorescence and Raman spectroscopy have been applied (2628), but lack critical information about tissue architecture and may be insensitive to changes occurring beneath the surface of the skin. Melanin fluorescence imaging in skin has been shown but has not yet found a clear biomarker for malignancy (29, 30). Photoacoustic tomography allows optical imaging of tissue at depths far greater than other methods and has been applied to melanomas (31), but relies on contrast from the linear absorption spectrum. Because eumelanin and pheomelanin have uniform and similar absorption spectra, fluorescence and linear absorption techniques have great difficulty discriminating between them. Structural imaging, such as reflectance confocal microscopy, has recently been developed to image pigmented lesions with cellular detail (32, 33). Although it has met with some success, this tool lacks specificity and molecular contrast and has not yet found widespread clinical adoption.

Our results create a compelling case for the development of pump-probe imaging as a basis for melanoma diagnosis. Melanin has been examined by pump-probe spectroscopy before by the Simon group (34, 35) and the Warren group where we observed pump-probe features unique to eumelanin and pheomelanin (36). Here, we extended and adapted this contrast to multiphoton scanning laser microscopy, identifying chemical and morphological changes in human melanoma samples. This has allowed us to probe the skin pigmentation changes that occur during malignancy and study the location and maturation of melanocytes in a variety of nevi and lesions. Moreover, compatibility of this method with H&E slides may enhance diagnostic protocols already in use and provide complementary information.

Bulk analysis of the eumelanin content of pigmented lesions (Fig. 3) showed that melanomas have higher amounts of eumelanin than melanocytic nevi. On the basis of this criterion alone, many false-positive diagnoses may be able to be rejected. Although the eumelanin/pheomelanin ratio is not sufficient to diagnose melanoma, it may greatly improve diagnostic accuracy in conjunction with current diagnostic techniques. Moreover, these data demonstrate that pigmentation changes are relevant to human disease and merit further investigation. Imaging pigment changes can lead to a better understanding of a lesion, as shown in Fig. 6. In this case, we show a eumelanic dysplastic nevus that was very uniformly pigmented, in contrast to the moderate to high degree of pigment heterogeneity found in all melanoma samples investigated. Quantifying these image parameters will be the next step in creating a robust basis for determination of disease.

The molecular contrast we achieved with pump-probe spectroscopy hinges on the phase difference between eumelanin and pheomelanin when pumped below 750 nm. We have found that at wavelengths of 740 nm and shorter, eumelanin shifts from a ground-state depletion (bleaching) signal to excited-state absorption. This indicates that higher-energy pulses are able to access a transition not available at wavelengths of 750 nm and longer. Because melanins are disordered macromolecules consisting of a variety of constituents with overlapping absorption spectra, no sharp transitions were observable with linear absorbance spectroscopy. Instead, this wavelength likely corresponds to the transition energy of a major constituent of eumelanin, indolequinone. Previous work by the Meredith group using photopyroelectric spectroscopy on thin eumelanin films found an energy gap at 1.7 eV (729 nm) (37). This corresponds to an absorption of indolequinone, the two-electron oxidation product of 5,6-dihydroxyindole (DHI) and 5,6-dihydroxyindole-2-carboxylic acid (DHICA) (38).

A large quantity of pheomelanin was found in melanocytic nevi, with a shift to eumelanin dominance in melanoma. Because eumelanin is photoprotective and has antioxidant properties, whereas pheomelanin can act as a photosensitizer, it has been postulated that elevated amounts of pheomelanin would lead to increased damage from ultraviolet radiation and an increased risk of malignant transformation. Previous work shows that dysplastic nevi, which display atypical growth, do indeed seem to have increased pheomelanin content compared to normal skin. The Koerten group showed an increase in sulfur content in dysplastic nevi and melanoma cells, indicating elevated amounts of pheomelanin (39). Analysis of chemical degradation products from normal skin, common melanocytic nevi, and dysplastic nevi performed by Jimbow found elevated amounts of pheomelanin in dysplastic nevi compared to normal skin and common melanocytic nevi (40). However, the chemical identity of melanin in melanomas is less clear, and some evidence exists to show that eumelanin may in fact occur in increased concentration. Jimbow found that eumelanin metabolites in blood plasma correlated well with melanomas, but pheomelanin metabolites did not have diagnostic significance (41). Recent bulk reflectance spectroscopy measurements on pigmented lesions by Marchesini et al. in vivo found a shift toward eumelanin in malignant melanoma. Although red hair has been shown to correlate with greater incidence of melanoma (42), the Jimbow group found no corresponding increase in epidermal pheomelanin. Rees and colleagues suggest that pheomelanin concentration alone is insufficient to explain the greater melanoma risk of individuals with red hair (43). These results support our findings that dysplastic nevi have a strong pheomelanin character and suggest that as a lesion progresses to melanoma, the balance shifts toward eumelanin.

In addition to melanin, hemoglobin exhibits a pump-probe signal under these conditions. However, hemoglobin does not contribute meaningfully to these images or findings for several reasons. Hemoglobin’s response is far weaker than that of melanin, and red blood cells are both rare and spatially resolved from melanocytes and keratinocytes in our images. The pump-probe response of hemoglobin, as previously demonstrated by Fu et al., is opposite in sign to that of melanin and therefore easy to separate. PCA would have isolated any significant signal sources beyond melanins and surgical ink.

Pump-probe imaging is a good candidate for in vivo applications as well. We have already demonstrated the feasibility of in vivo imaging with pulse train shaping to image microvasculature and blood oxygenation in vivo (44), and the Xie group later adapted the method to image stimulated emission in biological samples (45). Pump-probe imaging may also be useful in the future in assessing surgical margins in anatomically sensitive areas such as the head and neck, where conservation of healthy tissue is important, or for imaging eumelanin and pheomelanin in ocular lesions.

One of the challenges of pump-probe imaging, at least in unstained tissue, is insensitivity to unpigmented cells. Mature or amelanotic melanocytes are not shown, as well as other cancerous or irregular growths. Our technique remains to be tested in amelanotic lesions. Although these lesions are not pigmented in bulk, melanocytes can still individually carry melanin, which would make them amenable to imaging. Molecular melanin contrast can also be combined with other techniques designed to give structural information, such as two-photon autofluorescence and reflectance confocal microscopy.

We also note that our methods are hardly restricted to thin biopsies. Nonlinear imaging can produce a three-dimensional image in tissue, so thick sections (50+μm) of biopsies could be examined to give a much more information about a lesion than the current 5-μm sections. Adaptation of this technique to in vivo imaging is also quite reasonable at the powers we use and may lead to an effective noninvasive clinical screening method; however, we see the greatest short-term potential in the possibility of enhancing existing clinical protocols by reducing false positives and the associated procedures.

Materials and Methods

Pump-probe spectroscopy was performed as previously described (46). The sample is illuminated with pulsed laser light of two different wavelengths. By modulating the pump light and measuring modulations of the same frequency that are transferred to the probe light, we can measure multiphoton processes such as excited absorption. Our light source was a titanium-sapphire mode-locked laser (Tsunami, Spectra-Physics), which typically produced pulses at either 810 or 830 nm. This laser synchronously pumped a tunable optical parametric oscillator (Opal, Spectra-Physics, and Mira-OPO, Coherent) tuned between 1440 and 1500 nm, which was frequency-doubled to 720 to 750 nm. Either beam (residual titanium-sapphire pump or OPO output) was modulated at 2 MHz by an acousto-optic modulator. The two beams were then combined on a dichroic mirror (DCXR760, Chroma) and directed either into a simple microscope (10×, 0.25 NA, Olympus) for spectroscopy measurements or into a laboratory-developed scanning laser microscope (40×, 0.75 NA, Olympus and 60×, 0.9 NA, Olympus) for imaging. Pulse durations were typically 140 fs at the sample. Less than 5-mW total incident optical power was used for solution melanin measurements. Images were taken with, at most, 0.75-mW power of the pump beam and 0.5-mW probe power when using the 40× objective and 0.4-mW power on beam when using the 60× objective (measured before tube lens and objective, power at the sample was decreased at least 10%). Imaging was performed in transmission mode. After the pump was spectrally filtered out, the probe light was collected onto an amplified photodiode (PDA55, Thorlabs). A lock-in amplifier (Stanford Research Systems 844, 100-μs time constant) was used to measure the magnitude and phase of the signal at 2 MHz. Quadrature detection was then digitized by a PCI-6110 data acquisition card (National Instruments) at 125 kilosamples per second. Rhodamine 6G in methanol was used as a standard sample to set the reference for absorption. By our convention, a positive response corresponds to an absorptive signal. Custom MATLAB software was used to process data files. Principal components statistical analysis was carried out with the Statistics toolbox for MATLAB.

Solution samples of melanins were prepared according to established methods at a concentration of 0.84 mg/ml. S. officinalis eumelanin was chosen because it is accepted to be comparable to human-derived eumelanins and melanosomes (47). Sepia melanin was purchased from Sigma-Aldrich. Synthetic pheomelanin was prepared with the method published by Ito (48). Both melanins were suspended in a 50 mM phosphate buffer. The melanin suspensions were then filtered with a 1-μm syringe filter to remove large particles. H&E was purchased from VWR.

Samples of skin biopsies were obtained from excised and fixed lesions mounted in paraffin. Thin (5 μm) contiguous slices were cut from the block. Pairs of neighboring slices were selected, where one slice remained unstained for pump-probe imaging and the other was treated with H&E stain. We operated under a protocol approved by the institutional review board. All samples were appropriately de-identified and randomly numbered. The type of lesion was determined on the basis of standard structural and cytological features (49, 50) (Supplementary Materials and Methods).

Supplementary Material

Materials and Methods

Fig. S1. MART-1 staining of melanoma.

Fig. S2. Melanin pump-probe spectroscopy at other wavelengths.

Fig. S3. Normal skin pump-probe images.

Fig. S4. Absorption spectra of hematoxylin and eosin.

Fig. S5. Pump-probe images of H&E-stained slices.



  • * Present address: U.S. Environmental Protection Agency, Research Triangle Park, NC 27711, USA.

  • Citation: T. E. Matthews, I. R. Piletic, M. A. Selim, M. J. Simpson, W. S. Warren, Pump-Probe Imaging Differentiates Melanoma from Melanocytic Nevi. Sci. Transl. Med. 3, 71ra15 (2011).

References and Notes

  1. Funding: This study was funded by the NIH (grant 5RC1CA145105) and Duke University. Author contributions: T.E.M. performed the experiments, analyzed the data, and wrote the paper. I.R.P. helped design the experiments and assisted with experiments and data analysis. M.A.S. provided tissue samples as well as expertise in pathology and dermatology. M.J.S. assisted with experiments. W.S.W. designed the experimental method and directed the study. Competing interests: The authors declare that they have no competing interests.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article