We obtained 211 samples from patients diagnosed between 1989 and 2009 (Table S1 in Supplementary Material) at the following institutions: Catalan Institute of Oncology (I.C.O.), Hospital Duran i Reynals and the Hospital Universitari de Bellvitge (L’Hospitalet de Llobregat, Spain); Consorci Hospitalari Parc Taulí (Sabadell, Spain); and I.C.O., Hospital Universitari Germans Trias i Pujol (Badalona, Spain). A final total of 208 patients were included in this study, whose clinical (Table S2 in Supplementary Material), pathological parameters and molecular subtype status were as follows (Table S3 in Supplementary Material): 40.3% Luminal A (ER-positive, Ki-67 < 14%); 18.5% Luminal B (ER-positive, Ki-67 ≥ 14%); 9.0% Luminal/Her2+ (ER-positive and Her2 overexpressed), 10.0% Her2-enriched (ER-negative and Her2 overexpressed), and 22.2% TN tumors (ER-negative, PR-negative, and normal expression of Her2). Some missing values from these variables discarded three patients, two from Luminal tumors and one from non-Luminal tumors.

For this work involving databases of human information we followed the Spanish National law on the protection of Personal Data “Ley 15/1999, 13 Dec.”

Tissue microarrays (TMAs) were prepared from three representative areas of the tumor, as described previously (27). Primary antibodies anti-GRP94 at 1/2,000 and anti-FN14 at 1/3,000 (Santa Cruz Biotechnology, Santa Cruz, CA, USA) were diluted in Dako Real™ Antibody Diluent Buffer (Dakocytomation, Dako, Denmark): Tris buffer, pH 7.2, 15 mM Na₃N. LSAB + System-HRP (Dakocytomation) was used for staining, including secondary antibodies in PBS, streptavidin conjugated to HRP in PBS, and liquid 3,3′ diaminobenzidine in chromogen solution.

The overexpression of GRP94 and FN14 was categorized as positive when strong expression was detected and negative when no or weak expression was detected, to avoid false positives (Figure 1). Morphologic diagnosis was performed with classical hematoxylin–eosin staining.

Frequencies of categorical variables were compared among groups using the χ²-test or Fisher’s exact test where appropriate. Brain metastasis-free survival (BrMFS) was estimated for each group using the Kaplan–Meier method and was compared among them using the Cox proportional hazards model, estimating their hazard ratio and 95% CI.

To evaluate the correlation between BrM and protein expression, immunostained samples were graded on a three-category scale as follows: negative, weak positive, and strong positive. The marker was classified as overexpressed only in strong positive samples to avoid false positives. Biomarker sensitivity and specificity, both singly and in combination, was assessed in both Luminal and non-Luminal patients. The biomarkers combinations were considered positive when at least one of them was positive and negative when all of them were negative.

A multivariate logistic regression analysis adjusted for covariates was carried out in both Luminal and non-Luminal groups to study in patients with BrM vs. NBrM the presence of biomarkers (GRP94 and FN14) in their primary tumor. The covariates used were as follows: age (≥50, 40–49, and <40), positive axillary nodes (0, 1–3, and ≥4), Her2 status (negative and positive), and presence of lung metastasis (no and yes), where the first category mentioned for each variable was the reference. We calculated the OR associated with the biomarker, its 95% CI and p-value.

In this analysis, the variable “triple negative (no, yes)” was not included as a covariate because all patients belonging to the Luminal group were “no” for this variable. Moreover, Her2 status was not included as a covariate when the combination (GRP94 + FN14 + Her2) was used as the biomarker.

Values were considered significant when p was less than 0.05. Software used: R Core Team (37).

We studied patient characteristics according to three different groups of progression patterns (Table S1 in Supplementary Material): brain metastases (BrM), with or without metastases (WoM) at other sites; non-brain distant metastases (NBrM), patients with metastasis in bones and/or liver and/or lungs and/or non-regional lymph nodes, but not in brain; and patients WoM. The distribution of breast cancer molecular subtypes changed across the three groups of patients. The main distinctive parameters that characterized the BrM group were age, whereby below 50 years old was significantly different (p = 0.001); hormone receptor negativity [both ER (p < 0.0001) and PR (p < 0.0001)], an attribute of tumors that developed BrM in contrast to tumors from NBrM and WoM patients; and Her2 positivity and a high Ki-67 index (p = 0.01 and p < 0.0001, respectively). Other parameters, such as tumor size (p = 0.001), HG (p < 0.0001), and lymph node involvement (p < 0.0001) were similar among BrM and NBrM patients, but different in WoM patients.

More than 50% of BrM patients had TN tumors. By contrast, only 20.6% of NBrM and 9.8% of WoM patients had TN tumors (χ²-test: p < 0.0001). Important differences were observed in the Her2-enriched subtype with regard to metastasis progression, occurring in 23.1% of BrM, 8.8% of NBrM, and 4.9% of WoM groups. Moreover, the distribution of Luminal/Her2+ tumors was different in patients according metastasis involvement, being 11.5% in BrM, 8.8% in NBrM, and 8.2% in WoM groups. Therefore, according to the clinical and pathological characteristics of breast cancer patients, we established two main groups due to limited size of our sample: the Luminal group (Luminal A, Luminal B, and Luminal/Her2+ subtypes), which encompassed 25% (13/52) of patients with BrM, 70.6% (24/34) with NBrM and 85.3% (104/122) of WoM patients; and the non-Luminal group (TN and Her2-enriched subtypes), encompassing 75% (39/52) of patients with BrM, 29.4% (10/34) with NBrM, and 14.7% (18/122) of WoM patients. The clinical and pathological parameters of the Luminal and non-Luminal groups are described in Supplementary Table S4 in Supplementary Material: tumor size, ≥21 mm in 32.61 vs. 55.56%, respectively (p = 0.002); HG 3, 32.12 and 81.25%, respectively (p < 0.0001); and regional lymph node involvement (affected ≥4), in 14.39 vs. 35.39%, respectively (p = 0.0008).

As expected, significant differences were related to the therapeutic approach adopted: 81.82% of Luminal patients vs. 58.21% of non-Luminal patients did not receive neoadjuvant chemotherapy (p = 0.0002); 83.22 vs. 55.88% (p < 0.0001) received conservative surgery; 44.76 vs. 18.03% (p = 0.001) did not receive adjuvant chemotherapy; and 85.92 vs. 6.06% (p < 0.0001) received tamoxifen or another antiestrogen as hormonal therapy, respectively. Moreover, clinical relapses, with the presence of local (8.39 vs. 19.12%, p = 0.024), regional (1.40 vs. 16.18%, p < 0.0001), or distant relapses (27.27 vs. 73.53%, p < 0.0001) were less frequent in patients from the Luminal group than in those in the non-Luminal group, respectively.

The incidence of BrM (Table 1) was significantly higher in the non-Luminal group than in the Luminal group [58.21% (39/67) vs. 9.22% (13/141), p < 0.0001]. Furthermore, patient outcomes corroborated the clinical and pathological characteristics, with an increase in metastasis-free survival when tumors were Luminal in both BrM and NBrWoM groups (Figures S1A,B in Supplementary Material). Moreover, the worst prognosis was found in patients with non-Luminal tumors (HR = 10.57, 95% CI: 5.60–19.96; and HR = 4.01, 95% CI: 2.52–6.38; respectively, p < 0.0001). These results indicate that the assessment of subtypes in our series provided an effective subclassification according to BrM progression in patients, similar to other reported series (11, 14).

Since molecular subtypes in NBrM (tumors with bone, liver, lung, skin, etc.) and WoM had a similar clinic-pathological parameters distribution (Fisher’s exact test, p = 0.230; data not shown), we considered both as a single group (NBrWoM) to better compare the expression of BrM biomarkers between Luminal (N = 141) and non-Luminal (N = 67) tumors (see Table 2). FN14 was overexpressed in 14.5% (20/138) of Luminal tumors and in 23.1% (15/65) of non-Luminal tumors. Moreover, in Luminal tumors, FN14 expression was significantly different between the BrM and NBrWoM groups, at 53.8% (7/13) vs. 10.4% (13/125), respectively (Fisher’s exact test, p = 0.0005). FN14 expression in non-Luminal tumors differed between BrM and NBrWoM patients, being 31.6% (12/38) vs. 11.1% (3/27), respectively, although not significantly (Fisher’s exact test, p = 0.07). Therefore, we concluded that FN14 expression in Luminal breast tumors was associated with BrM progression. GRP94 was overexpressed in 47.5% (66/139) of Luminal and 50% (33/66) of non-Luminal tumors. Biomarkers expression in Luminal tumors was also significantly different between BrM and NBrWoM patients, being 76.9% (10/13) vs. 44.4% (56/126) (Fisher’s exact test; p = 0.04), whereas GRP94 expression in the non-Luminal group was similar for BrM, at 57.9% (22/38) of patients, and NBrWoM at 39.3% (11/28) (Fisher’s exact test; p = 0.21). These results highlight the intrinsic value of FN14 and GRP94 as organ-specific BrM biomarkers mainly in patients with tumors from the Luminal subtype.

Furthermore, we analyzed in tumors with BrM progression if the expression of BrM biomarkers were associated with Her2 overexpression. We found that FN14 expression was independent of Her2 status in patients with BrM from the Luminal group (Fisher’s exact test; p = 0.59). By contrast, FN14 expression in the non-Luminal group was associated in 63.6% of cases with the Her2-enriched subtype (Table S5 in Supplementary Material), differing from TN tumors, of which 18.5% were positive for FN14 (Fisher’s exact test, p = 0.017).

GRP94 expression in tumors from BrM affected patients was independent of Her2 status (Table S5 in Supplementary Material) in the Luminal (Fisher’s exact test; p = 1.00) and non-Luminal groups (Fisher’s exact test, p = 0.30).

First, we studied metastasis-free survival in our series according to molecular subtype (Figure S2 in Supplementary Material), using the Kaplan–Meier method and the Cox proportional hazards model. The TN subtype was used as a reference group for comparative purposes. We analyzed the distribution of subtypes in NBrM patients (Figure S2A in Supplementary Material). Differences between the TN and Her2-enriched subtypes were not statistically significant (HR = 0.81; 95% CI: 0.41–1.60; p = 0.53). By contrast, Luminal/Her2+ (HR = 0.41; 95% CI: 0.18–0.96; p = 0.039), Luminal B (HR = 0.33; 95% CI: 0.17–0.64; p = 0.0012), and Luminal A (HR = 0.16; 95% CI: 0.08–0.30; p < 0.0001) showed a significantly lower incidence of NBrM than TN breast carcinomas. Furthermore, in BrM patients, we found significant differences in BrMFS with regard to molecular subtype, using the TN subtype as the reference group (Figure S2B in Supplementary Material): Her2-enriched did not differ significantly from TN (HR = 0.79; 95% CI: 0.40–1.55; p = 0.49), but the remaining subtypes showed a significantly lower risk than TN: Luminal/Her2+: HR = 0.33; 95% CI: 0.13–0.80; p = 0.014. Luminal B: HR = 0.13; 95% CI: 0.05–0.33; p < 0.0001 and Luminal A: HR = 0.02; 95% CI: 0.005–0.09; p < 0.0001.

Next, we studied BrMFS in the Luminal group according to whether the tumor expressed FN14 or not (Figure 2A), and we found that overexpression of FN14 was associated with a reduction in BrMFS (HR = 8.25; 95% CI: 2.77–24.61; p = 0.00015). Although, different stratification of the non-Luminal group (Figure 2B) according to FN14 expression was not statistically significant (HR = 1.74; 95% CI: 0.87–3.47; p = 0.11).

In addition, we analyzed BrMFS in the Luminal group according to GRP94 expression (Figure 2C) and observed that GRP94 was overexpressed in tumors from patients with reduced BrMFS (HR = 3.58; 95% CI: 0.98–13.11; p = 0.054—Wald test), although this finding did not reach statistical significance either. Meanwhile, BrMFS according to GRP94 expression in non-Luminal tumors was not significantly different (HR = 1.22; 95% CI: 0.64–2.32; p = 0.55) (Figure 2D).

Since overexpression of the Her2 gene is associated with a higher risk to develop BrM, we analyzed the sensitivity and specificity of FN14 and GRP94 expression to predict BrM and compared these parameters to the prediction given by Her2 in these patients (Table 3). In patients belonging to the Luminal group (low risk of BrM a priori), FN14 and Her2 showed more specificity (89.60 and 89.84%, respectively) than GRP94 (55.56%) to predict BrM progression. However, GRP94 expression showed more sensitivity (76.92%) than FN14 and Her2 (53.85 and 46.15%, respectively) to predict BrM involvement.

On the other hand, in patients with a higher risk of BrM, such as non-Luminal ones, FN14 (88.89%) was the most specific protein to discriminate tumors that developed BrM, followed by Her2 (67.86%) and GRP94 (60.71%). GRP94 sensitivity was again higher (57.89%) than that obtained with FN14 or Her2 (31.58 and 30.77%, respectively).

Expression of the biomarker combination FN14 + GRP94 improved BrM risk assessment in Luminal patients compared with non-Luminal ones (sensitivity: 84.62 and 68.42%, specificity: 50 and 57.14%, respectively). The addition of Her2 yielded 84.62 and 71.79% sensitivity, and 45.24 and 42.86% specificity in Luminal with regard to non-Luminal, respectively (Table 3). The combination of FN14 + GRP94 + Her2 increased sensitivity to predict BrM in both Luminal and non-Luminal cases, covering a wide range of patients.

The high specificity shown by FN14 and Her2 when evaluated singly was lost in combination, suggesting that both biomarkers stratified different patient subgroups of BrM risk.

To assess the usefulness of FN14 and GRP94 as independent risk factors of BrM, a multivariate analysis was performed, including age, axillary node involvement, Her2 status, and presence of lung metastasis as covariates in the analysis (Table 4).

Outcomes showed that patients belonging to the Luminal group (ER+) had a statistically significant higher risk to develop BrM if the tumor expressed FN14 compared with those with FN14 negative tumors (OR: 36.70; 95% CI: 3.65–368.25; p = 0.002). The likelihood of these patients to develop BrM increased 36.70-fold. By contrast, when tumors overexpressed GRP94, the risk to develop BrM (OR = 5.74; 95% CI: 0.87–37.66) remained non-significant (p = 0.069), either in combination with FN14 (p = 0.066) or Her2 (p = 0.071) positivity (OR = 7.10, 95% CI: 0.88–57.51; and OR = 5.80, 95% CI: 0.86–39.08, respectively).

In non-Luminal patients, only the combination of FN14 and GRP94 positivity was significantly predictive of BrM progression (95% CI: 1.19–13.65, p = 0.025), with a 4.04-fold likelihood to develop BrM (Table 4). These results are in consonance with our previous results reporting the usefulness of both biomarkers to predict BrM in TN breast cancer patients (27).

In summary, our study reveals that (1) A subset of breast cancer patients with a better prognosis a priori, such as estrogen-receptor positive with or without Her2 positivity, develop brain metastases if FN14 and/or GRP94 biomarkers are positive in their primary tumor. (2) FN14-positive status impairs the prognosis of breast cancer patients by shortening the length of BrMFS. (3) The likelihood to develop BrM in patients with FN14-positive tumors increases 36.70-fold (p = 0.002). (4) The combined assessment of biomarkers (FN14 + GRP94) shows a higher benefit in risk evaluation of BrM in patients with non-Luminal tumors.

Since stratification impairs statistical consistency, further multicentre studies with wider amount of patients are needed to reinforce the results.