Peritumoral immune infiltrates in primary tumours are not associated with the presence of axillary lymph node metastasis in breast cancer: a retrospective cohort study

Tissue preparation and immunohistochemistry

This is a retrospective cohort study of 149 patients diagnosed with invasive BC of no special type in the Hospital de Tortosa Verge de la Cinta (HTVC), Spain, 75 of whom had ALN metastasis at diagnosis. The Ethics Committee of the Hospital Joan XXIII de Tarragona, Spain, approved the study (reference 22p/2011) and we followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Written informed consent was signed by all patients involved in the study, in accordance with Spanish law.

Two representative 2-mm tissue cylinders from the border of the tumour area of the biopsy were selected by a pathologist from the Pathology Department of the HTVC for the purpose of constructing tissue microarrays (TMAs). Ductal carcinoma in situ elements and tertiary lymphoid structures around the border of the tumour were excluded when selecting the areas from which cylinders were taken. Each TMA block contained 50 cylinders, giving 6 TMAs ((149 patients X 2 cylinders)/50 cylinders). Eleven slides were sectioned from each TMA in order to stain the 11 immune markers chosen for study (Fig. 1). TMA technology is of great value for analysing large numbers of cases, but it is clear that the degree of correlation between TMAs and whole-tissue sections is not ideal at the diagnostic level. Nevertheless, the use of TMAs with a large number of samples is widely considered to be adequate for research level (Pinder et al., 2013). In fact, as we mentioned in our previous report, in which we also used TMAs (López et al., 2020), a search using the terms “tissue microarray breast cancer immune” in PubMed identified more than 100 articles, highlighting the widespread use of TMAs in studies into the evaluation of the immune system (Pelekanou et al., 2018; Solinas et al., 2017). Moreover, Salgado explained in his guide for evaluating TILs that results in TMAs have proven concordance with those of other studies (Ali et al., 2014; Chavan, Ravindra & Prasad, 2017; Liu et al., 2014; Schalper et al., 2014), which makes them a good option for rapid evaluations (Salgado et al., 2015).

The selection of immune markers for the present work was based on the findings of other studies that have demonstrated the various immune cell populations evaluated (lymphocytes, macrophages and the DC) to be associated with BC patient outcome (De Melo Gagliato et al., 2017; Stovgaard et al., 2018; Zhao et al., 2017). Immune cells were immunohistochemically detected on each slide using the following primary antibodies: T helper lymphocytes (anti-CD4, clone 4B12, Dako, Santa Clara, CA, USA), cytotoxic T lymphocytes (anti-CD8, clone C8/144B, Dako, Santa Clara, CA, USA), natural killers (anti-CD57, clone NK1, Zymed, Thermo Fisher Scientific, Waltham, MA, USA), regulatory T cells (anti-FOXP3, clone 236A/E7, CNIO, Madrid, Spain), macrophages (anti-CD68, clone KP1, Dako, Santa Clara, CA, USA), follicular DC (anti-CD21, clone 1F8, Dako, Santa Clara, CA, USA), Langerhans DC (anti-CD1a, clone 010, Dako, Santa Clara, CA, USA), plasmacytoid DC (anti-CD123, clone 6H6, eBioscience, San Diego, CA, USA), interdigitant DC (anti-S100, polyclonal, Leica Microsystems GmbH, Wetzlar, Germany), LAMP3 DC+ (anti-CD208, polyclonal, Proteintech, Rosemont, IL, USA), mature DC (anti-CD83, clone1H4b, Leica Microsystems GmbH, Wetzlar, Germany). The ENDVISION™ FLEX method (Dako, Santa Clara, CA, USA) was applied to stain the slides, using the chromogen diaminobenzidine (DAB) as a substrate. The samples were counterstained with haematoxylin.

Slide digitization and image evaluation

We used whole-slide imaging (WSI), a method that is replacing the microscope for classical diagnosis in some centres (Pantanowitz et al., 2013), to analyse digital images. We obtained the images in TIFF format by scanning the 66 stained slides with an Aperio ScanScope XT scanner at 40X magnification at a resolution of 0.25 µm/pixel. We extracted each cylinder of the original WSI from each TMA as a single image using an automatic tool developed by members of our team (Roszkowiak & López, 2016). The tissue cylinder areas and the stained areas of immune markers in each image were evaluated using our own digital image analysis procedures (López et al., 2020), which enable the number and density (in µm²) of pixels of the positive-stained areas for each immune marker, and the area of each cylinder included in the TMAs, to be calculated (Callau et al., 2015). The concentration of each immune marker was calculated as the percentage of positive-stained areas of each immune marker relative to the whole area of the cylinder, as previously described (López et al., 2020).

Clinical and pathological variables

To determine which clinical and pathological variables, in addition to the peritumoral immune response, could also be associated with the presence of metastasis in the ALN at diagnosis, the following data were collected from the patients’ clinical records: age, tumour diameter, lymphovascular invasion (LVI), perineural invasion (PNI), histological grade, oestrogen receptor status (ER), progesterone receptor status (PR), HER–2 status, proliferation index (Ki67), menopausal status and molecular profile.

Statistical analysis

Differences in immune response marker concentrations between patients diagnosed with and without metastasis in their ALN were evaluated using the Mann–Whitney U test. The quantitative clinical and pathological variables in the two groups of patients were compared using the Mann–Whitney U test or Student’s unmatched samples t-test (age and tumour diameter). In order to identify disproportionate frequencies of combinations of categories of the clinical and pathological variables we performed chi-squared test or Fisher’s exact tests.

A univariate logistic regression analysis was carried out for each variable to evaluate its association with the presence or absence of ALNs with metastasis. Two multivariate logistic regression models were then developed to identify which of the clinical and pathological variables, and which of the immune cell populations present in the peritumoral area, were associated with the presence of ALN metastases at diagnosis. The Hosmer–Lemeshow test was used to estimate the goodness of fit of all the variables considered in the multivariate analyses. The area under the curve (AUC) and the receiver-operating characteristic (ROC) curve were also derived to estimate the sensitivity and specificity of each model. We can consider that the present study, which featured 75 events and yielded final multivariate models comprising three independent variables, had an adequate sample size for a reliable multivariate analysis as previously suggested (Peduzzi et al., 1996). In the first model, all variables with a significance of p ≤ 0.1 in the univariate logistic regression analyses were considered when deriving the multivariate model 1. In the second model, all the variables with a significance of p ≤ 0.3 were included. Each model was validated using two statistical techniques: (1) bootstrap simulation, carried out with IBM SPSS Statistics 21.0 (IBM, Armonk, NY, USA), based on 10,000 random samples; and (2) the multiple imputation method available in the IBM SPSS statistical application. This method replaces missing values of a specific variable by using linear regression to calculate values from others in the dataset. Failing to deal with missing data is a problem because it leads to a reduction in the statistical power of the model and can produce biased estimates.

Immune response and ALN metastasis

In BC, metastasis in ALNs plays a key role in spreading tumoral cells to other parts of the body (Ran et al., 2010); in fact, several types of intratumoral cells are known to be linked to the presence of metastasis in the ALN at diagnosis: cytotoxic and helper T lymphocytes (La Rocca et al., 2008; Matkowski et al., 2009), T regulatory lymphocytes (Gokmen-Polar et al., 2013; Miyan et al., 2016; Nakamura et al., 2009), macrophages (Jubb et al., 2010; Mansfield et al., 2012; Shiota et al., 2016) and dendritic cells (Mansfield et al., 2011; Treilleux et al., 2004). In addition, our group previously found the CD21+ follicular DC immune population in the intratumoral region to be associated with the presence of ALN metastasis. In the same study, we compared the immune populations of the non-metastatic ALN in patients diagnosed with or without ALN metastasis. We found that higher concentrations of CD68+ macrophages and of S100+ interdigitant DC in the non-metastatic ALNs were associated with the presence of ALN metastasis at diagnosis. On the other hand, higher concentrations of CD123+ plasmacytoid DC were found to be a factor protecting against ALN metastasis (López et al., 2020). Since the immune populations could be associated with ALN metastasis in locations other than the intratumoral region, we decided to evaluate the peritumoral area as well.

Peritumoral immune infiltrates in BC

Little is known about the peritumoral immune infiltrates, and, especially, their links with ALN status. Bordea et al. showed BC to be more aggressive and associated with an increased rate of sentinel lymph node metastasis in patients with peritumoral TILs (Bordea et al., 2012). However, they did not use H&E or Salgado’s criteria; the latter were published in 2015, and have since become established as the gold standard for TILs assessment (Salgado et al., 2015). Specific types of immune cells have barely been studied in the peritumoral areas of BC patient samples. CD4+ follicular helper T cells are associated with peritumoral tertiary lymphoid structures (TLS) (Gu-Trantien et al., 2013). High levels of expression of intratumoral CD8+ TILs are significantly associated with the overall survival (OS) of luminal B patients treated with anthracycline-based neoadjuvant chemotherapy, but the peritumoral fraction is not (Al-Saleh et al., 2017). Conversely, Vgenopoulou et al. investigated peritumoral CD8 and CD57 markers, and found increased numbers of CD8+ cells in ALN+ patients, but no difference in the abundance of the CD57 marker (Vgenopoulou et al., 2003). Liu et al. reported high levels of peritumoral FOXP3+ to be a predictor for chemotherapy of HER2-positive patients. In addition, the latter article is the only one, to our knowledge, to report an association between the immune response in the peritumoral area of the primary tumour and ALN metastasis, showing positive correlation between peritumoral FOXP3+ and positive nodal status (Liu et al., 2014). A study analysing CD68+ macrophage cells showed that these cells were more likely to be present in the intratumoral area than in the peritumoral area, although its correlation with ALN status was not studied. The study was also limited by its low number of specimens (Carpenco, 2019). Controversially, Heiskala et al. found CD68+ infiltrations to be more abundant in the peritumoral than the intratumoral area, but CD68 frequency was not correlated with ALN status (Heiskala et al., 2019). Next, considering DCs, Bell et al. reported immature CD1a+ to be retained intratumorally, and mature CD83+ DCs to be confined to peritumoral areas in patients with BC; nonetheless, the study was performed with only 32 samples and no statistical analyses were carried out (Bell et al., 1999). However, our study revealed no association between the immune infiltrates of the peritumoral area and the presence of ALN metastases at diagnosis, which is in line with the findings of the few studies that have investigated the relation of the immune populations in the peritumoral areas with ALN status.

Clinical and pathological factors associated to ALN metastasis

Most of the studies on the clinical, molecular and pathological/histological factors associated with ALN metastasis were carried out a long time ago (Ahlgren et al., 1994; Chadha et al., 1994; Noguchi et al., 1993). We found primary tumour size, histological grade and LVI to be histological factors significantly associated with the presence of ALN metastases at diagnosis in two of four validations. These findings are consistent with the well-established pathological characteristics associated with ALN metastasis reviewed by Patani, Dwek & Douek (2007) and described in subsequent publications (Reynders et al., 2014; Yoshihara et al., 2013). In the present study, model 1 had the best AUC when validated and so had the best predictive capability. The validated first model included the LVI, primary tumour size and histological grade as independent factors associated with having ALN metastasis, and showed very good sensitivity and specificity. Patani et al. highlighted several works showing the values of the sensitivity and specificity of the multivariate models using tumour size, histological grade and LVI (Patani, Dwek & Douek, 2007). In some of these studies, sensitivity was related to tumour size, or the histological grade was higher than in our case, but conversely they showed very low specificity (Barth, Craig & Silverstein, 1997; Reynders et al., 2014; Silverstein, Skinner & Lomis, 2001). On the other hand, several studies employing models using LVI were very specific but not at all sensitive (Chadha et al., 1994; Tan et al., 2005). Our model gave a very good balance between sensitivity and specificity; in fact, the AUC of the validated first model was almost 0.9, which implies that our model can correctly classify around 90% of patients, and shows that these variables are closely associated with the presence of ALN metastasis at diagnosis.

Furthermore, Tseng et al. (2014) and Grigoriadis et al. (2018) have demonstrated that other histological features may be associated with the presence of ALN with metastasis at diagnosis (lymphocytic lobulitis, size and number or location of germinal centres, etc.). These variables could also be used to evaluate BC patient outcome. In fact, another study showed that the presence of CD4+ T cells, localized in the germinal centres of peritumoral TLS found in extensively infiltrated neoplastic lesions, predicted better disease outcome among BC patients (Gu-Trantien & Willard-Gallo, 2013). Indeed, TLS have recently proven to be relevant in patients’ survival and immunotherapy response in other cancer types like sarcoma and melanoma (Cabrita et al., 2020; Helmink et al., 2020; Petitprez et al., 2020). Nevertheless, we did not study the TLS on this occasion, although in the future it would be worthwhile evaluating them in the context of the immune response.