The role of FoxP3+ regulatory T cells and IDO+ immune and tumor cells in malignant melanoma – an immunohistochemical study

Background FoxP3+ Regulatory T cells (Tregs) and indoleamine-2,3-dioxygenase (IDO) participate in the formation of an immunosuppressive tumor microenvironment (TME) in malignant cutaneous melanoma (CM). Recent studies have reported that IDO expression correlates with poor prognosis and greater Breslow’s depth, but results concerning the role of FoxP3+ Tregs in CM have been controversial. Furthermore, the correlation between IDO and Tregs has not been substantially studied in CM, although IDO is known to be an important regulator of Tregs activity. Methods We investigated the associations of FoxP3+ Tregs, IDO+ tumor cells and IDO+ stromal immune cells with tumor stage, prognostic factors and survival in CM. FoxP3 and IDO were immunohistochemically stained from 29 benign and 29 dysplastic nevi, 18 in situ -melanomas, 48 superficial and 62 deep melanomas and 67 lymph node metastases (LNMs) of CM. The number of FoxP3+ Tregs and IDO+ stromal immune cells, and the coverage and intensity of IDO+ tumor cells were analysed. Results The number of FoxP3+ Tregs and IDO+ stromal immune cells were significantly higher in malignant melanomas compared with benign lesions. The increased expression of IDO in melanoma cells was associated with poor prognostic factors, such as recurrence, nodular growth pattern and increased mitotic count. Furthermore, the expression of IDO in melanoma cells was associated with reduced recurrence˗free survival. We further showed that there was a positive correlation between IDO+ tumor cells and FoxP3+ Tregs. Conclusions These results indicate that IDO is strongly involved in melanoma progression. FoxP3+ Tregs also seems to contribute to the immunosuppressive TME in CM, but their significance in melanoma progression remains unclear. The positive association of FoxP3+ Tregs with IDO+ melanoma cells, but not with IDO+ stromal immune cells, indicates a complex interaction between IDO and Tregs in CM, which demands further studies. Supplementary Information The online version contains supplementary material available at 10.1186/s12885-021-08385-4.

1 Outline of the image analysis method FoxP3+ Regulatory T cells were counted using a semi-automated computer vision program. The program uses a combination of image processing methods to identify and quantify FoxP3 positive cells in the input image (Fig. 1). Special care was taken to minimise the number of false positives with filters designed to reject FoxP3-expressing non-lymphocyte cells.

A brief description of the image analysis algorithm
The input image is first color vector filtered to selectively reject melanin containing areas while preserving the darker FoxP3 stain (Fig. 2). A grayscale image is then obtained by computing the minimum decomposition (Valuex,y = min (Rx,y,Gx,y,Bx,y)) of the color filtered image. Adaptive Otsu thresholding (6x6 sampling grid, 10% sample overlap) is performed to produce a (1-bit) mask of positively stained areas. The mask is then median filtered (5px radius) to reduce high frequency noise that could adversely affect subsequent processing steps and object detection. Contiguous areas in the mask are then labelled and sorted by size. Areas of appropriate size for a single lymphocyte are passed on to the cell feature filter stack. Areas larger than a single lymphocyte (presumed to be artefacts or clusters of cells) are segmented using a size-selective marker-based watershed transform and then passed through the cell feature filter.

The cell feature filter stack
The cell feature filter stack consists of consecutive filter stages designed to mimic the criteria used in manual counting. The filter stages work independently and reject cells that do not meet the criteria. For the purposes of this study, filters for size (min/max), roundness (isoperimeter quotient), aspect ratio, intensity relative to the background, intensity relative to other cells and proximity to image edges were used.

Validation of the image analysis method
Results from the automated image analysis were validated against a set of images counted independently by the authors. The output of the automated analysis was generally found to correlate very well with the manually counted set. Distinguishing artefacts and FoxP3-expressing tumour cells from FoxP3 positive lymphocytes proved to be a challenge, but good false-positive rejection rates were achieved with careful selection and tuning of filters. A further challenge was posed by the varying cell and background staining intensity between slides. The aforementioned factors were found to account for most of the errors in the automated analysis. However, with well selected image processing methods and carefully tuned settings a good level of accuracy was achieved. All results of the digital analysis were confirmed by the authors and corrections to the cell counts were made as needed.

Benefits and drawbacks of automated digital image analysis
One of the major benefits of digital image analysis in comparison to traditional counting methods is the vastly superior speed at which the analysis can be carried out. This enables the processing of much larger sets of data in a given amount of time. As a result, a greater number of samples can be analysed in more detail, as the sampling density and area can be increased. Digital image analysis algorithms are also objective by nature and produce consistent and repeatable results.
Drawbacks of digital image analysis include the difficult and time-intensive nature of developing and setting up an automated image analysis pipeline with currently available tools. Machine learning could be used to automate parts of the calibration and set up process and is a possible topic for further research.