Circulating Cytokines in Metastatic Breast Cancer Patients Select Different Prognostic Groups and Patients Who Might Benefit from Treatment beyond Progression

Cancer induces immune suppression to overcome its recognition and eradication by the immune system. Cytokines are messengers able to modulate immune response or suppression. There is great interest in the evaluation of their changes during treatment in order to identify their relationship with clinical outcome. We evaluated 18 cytokines in breast cancer patients treated with eribulin before starting treatment (T0) and after four courses of therapy (T1). Longitudinal modifications were considered and cytokine clusters through PCA and HCPC correlated to patients’ outcomes were identified. Forty-one metastatic breast cancer patients and fifteen healthy volunteers were included. After clustering, we identified at T0 six patient clusters with different risk of relapse and death. At T1, only four clusters were identified, and three of them accounted for thirty-eight of forty-one patients, suggesting a possible role of treatment in reducing heterogeneity. The cluster with the best survival at T1 was characterized by low levels of IL-4, IL-6, IL-8, IL-10, CCL-2, CCL-4, and TGF-β. The cluster showing the worst survival encompassed high levels of IL-4, IL-6, IL-8, IL-10, CCL-2, and IFN-γ. A subgroup of patients with short progression-free survival (PFS) and long overall survival (OS) was comprised in the cluster characterized by low levels of CCL-2, IL-6, IL-8, IL-10, and IL-12 at T0. Our data support the prognostic significance of longitudinal serum cytokine analysis. This approach may help identify patients for whom early treatment stop avoids needless toxicity or might justify treatment beyond early progression. Further investigations are required to validate this hypothesis.


Introduction
The tumor microenvironment (TME) drives the dynamic interaction between tumor and immune cells, largely mediated by cytokines and chemokines that exert their biological actions even distantly moving through blood circulation, for instance supporting immunecell recruitment and cancer metastatization [1]. Group 1. Patients with both PFS and OS below the median; Group 2. Patients with PFS below the median and OS above the median; Group 3. Patients with PFS above the median and OS below the median; Group 4. Patients with both PFS and OS above the median.
Finally, we collected blood samples also from a group of healthy volunteers (HV).

Plasma Collection
For each patient, 12 milliliters of peripheral blood samples were stored in EDTA-treated Vacutainer (BD, Franklin Lakes, NJ, USA). Plasma samples were obtained by centrifugation step at 340× g for 10 min at room temperature (RT) and promptly stored at −80 • C until use.

Cytokine Measurement
Concentrations of all cytokines but IL-21 were determined using the Ella Simple Plex system (ProteinSimple™, San Jose, CA, USA) according to the manufacturer's instructions.
IL-21 was assessed with ELISA method (R&D System Minneapolis, MN, USA). Briefly, as previously described [13], a twofold dilution of each plasma sample was spun for 15 min at 1000× g and added to the Simple Plex cartridge. The cartridge was then inserted into the reactor and run for 90 min at RT. Concentrations were expressed as pg/mL.
All blood samples were tested centrally at the Translational Research Laboratory ARCO Foundation at S. Croce and Carle Teaching Hospital in Cuneo, Italy, and assessed in duplicate. The average of each duplicate was considered at each point.

Statistical Analyses
The exploratory nature of this translational study did not allow a priori sample size and statistical power calculation, therefore the sample size of the patient population and of the healthy volunteers was arbitrary established.
Differences in the median cytokine values were analyzed using a non-parametric Mann-Whitney U test and Wilcoxon signed-rank test for paired samples. Principal Component Analysis (PCA) and Hierarchical Clustering on Principal Components (HCPC) were performed on cluster subjects at T0 and T1 using circulating-cytokine concentrations previously normalized in z-score.
PFS and OS were estimated using Kaplan-Meyer method and relative hazard ratio (HR) was performed by the Cox model. PFS was defined as the time elapsed between the start of eribulin and progressive disease or death from any cause, whichever occurred first, or at the date of last follow-up for censored patients.
OS was defined as the time elapsed between the start of eribulin and death from any cause or the date of the time of the last follow-up for censored patients.
The Mann-Whitney U test and Wilcoxon signed-rank test were performed with Graph-Pad v.5. Kaplan-Meyer and the Cox model were performed with SPSS V.24. PCA and HCPC were computed with R v.3.5.3 by the FactoMiner R package.
In all tests, a p value equal or lower to 0.05 was regarded as significant. Bonferroni's correction was applied for the multiplicity test [24]. If not specified, a p-value is considered NS (not significant).

Healthy Volunteers
Blood samples were collected from 15 HV. Median age was 47 years (range 28-65), with two males and thirteen females.

Treatment Effects
Partial response (PR) was recorded in 10 patients (24.4%), and six patients (14.6%) achieved stable disease (SD), lasting 6 months or more, with clinical benefit rate of 39%.
According to the grouping system described above, we divided our population as follows: Due to the apparent contrasting results emerging from the analysis of the patients in group 2 (short PFS and long OS), we focused our attention on this cohort.
Detailed characteristics of patients allocated in group 2 are reported in Table 2. In particular, all patients in group 2 received further treatment after eribulin, but no one obtained treatment response.

Cytokine Profile in 15 HV and in 41 Patients
We determined the median values of the 18 cytokines in 15 HV and their values were compared to the values observed in 41 patients (Table 3).  We analyzed differences between HV and patients at T0. We found that, at T0, patients had higher values of IL-6, IL-8, IL-10, IL-21, TGF-β, VEGF, IFN-γ, CCL-2, CCL-4, and CXCL-10 and lower levels of IL-4 and IL-13 compared to HV.

Patients' Clusterization and Their Cytokine Profile at T0
The concentration of the 18 cytokines recorded before starting eribulin (T0) was used to cluster patients through PCA and HCPC methods.
Although many patients gathered in C2 T0 and C3 T0 , we observed that the population was dispersed among many clusters.
Patients in C2 T0 , mainly belonging to group 2, had significantly better OS compared to patients in C3 T0 (median OS 18.4 months and 8.9 months, respectively). Indeed, patients in C2 T0 had a significant mortality-risk reduction compared to all patients taken together (HR = 0.35, 95% C.I. 0.16 to 0.78, p = 0.01).

Patients' Clusterization and Their Cytokine Profile at T0
The concentration of the 18 cytokines recorded before starting eribulin (T0) was used to cluster patients through PCA and HCPC methods.
Patients in C2 T0 , mainly belonging to group 2, had significantly better OS compared to patients in C3 T0 (median OS 18.4 months and 8.9 months, respectively). Indeed, patients in C2 T0 had a significant mortality-risk reduction compared to all patients taken together (HR = 0.35, 95% C.I. 0.16 to 0.78, p = 0.01).
Moreover, we performed Cox multivariate analysis in order to evaluate the impact of previous lines of treatment received before eribulin (discriminating patients treated in the second or more-advanced lines), the metastatic site (visceral vs. bone/soft tissue), and Moreover, we performed Cox multivariate analysis in order to evaluate the impact of previous lines of treatment received before eribulin (discriminating patients treated in the second or more-advanced lines), the metastatic site (visceral vs. bone/soft tissue), and the number of metastatic sites in patients at C2 T0 . It seems that belonging to C2 T0 is an independent prognostic factor as it is the only one that remains significant. HR = 0.32 and p = 0.01 (Supplementary Materials: Figure S1).
HV had lower levels of IL-10 and CCL-4 compared to patients in C3 T0 and C4 T0 ; TGF-β level was lower in HV compared to patients in C2 T0 and C3 T0 and also IL-21 level was lower compared to patients in C3 T0 .
In addition, HV had higher levels of IL-13 compared to patients in C2 T0 and higher levels of TNF-α compared to patients in C4 T0 (Figure 3). β level was lower in HV compared to patients in C2 T0 and C3 T0 and also IL-21 level was lower compared to patients in C3 T0 .
In addition, HV had higher levels of IL-13 compared to patients in C2 T0 and higher levels of TNF-α compared to patients in C4 T0 (Figure 3).
OR PEER REVIEW 10 of 20 We investigated the longitudinal shifting of the cytokines from T0 to T1 of patients included in the clusters identified at T0 and we found that patients in C3 T0 had a significant increase in IL-2, IL-4, and VEGF (p = 0.034, p = 0.0006, and p = 0.01, respectively), We investigated the longitudinal shifting of the cytokines from T0 to T1 of patients included in the clusters identified at T0 and we found that patients in C3 T0 had a significant increase in IL-2, IL-4, and VEGF (p = 0.034, p = 0.0006, and p = 0.01, respectively), while patients in C2 T0 showed a significant longitudinal increase of IL-2 and IL-13 from the two time-points (p = 0.02 and p = 0.0005, respectively) ( Figure 4A,B).
Patients in HCPC at T1 were distributed as follow:    in C3 T1 . Patients in C2 T1 had lower median values of IL-6, IL-10, and IL-15 and higher median TGF-β values compared to those in C3 T1 (Figure 7).
No clear differences between patients in C4 T1 and each other cluster were observed in median cytokine values. However, C4 T1 accounted for three patients only.

Discussion
This exploratory study focuses on cytokine expression and their modification during treatment with eribulin, and the correlation between them and patients' outcomes. Albeit eribulin is a chemotherapy agent, it exerts some effects on circulating chemokines [8]. Therefore, eribulin allows the study of the correlation between cytokine modulation and patients' outcomes.

Discussion
This exploratory study focuses on cytokine expression and their modification during treatment with eribulin, and the correlation between them and patients' outcomes. Albeit eribulin is a chemotherapy agent, it exerts some effects on circulating chemokines [8]. Therefore, eribulin allows the study of the correlation between cytokine modulation and patients' outcomes.
We evaluated the differences among cytokines between patients before treatment (T0) and HV. Not surprisingly, we found that the median value of many cytokines was different in HV compared to patients. However, IL-2, IL-12, IL-15, TNF-α, (Th-1 cytokines), and CCL-22 did not change significantly between HV and patients. All these cytokines are linked to acute inflammation [17]. IFN-γ, differently from Th-1 cytokines, was higher in patients compared to HV. IFN-γ plays a crucial role in immune response. However, IFN-γ induces many pro-tumor reactions, including up-regulation of programmed death-ligand 1 (PD-L1), and Indoleamine 2,3-dioxygenase (IDO) affects T-cell immune response and, ultimately, has both pro-and antitumor properties depending on the concentration in the TME (reviewed in [25]). In line with the importance of pro-tumor effects of this cytokine, Jabeen et al. [26] observed reduction in IFN-γ in breast cancer patients responding to treatment with bevacizumab and chemotherapy.
The remaining cytokines, mostly significantly higher in patients, are related to Th-2 response and reflect the existing chronic inflammatory status [27].
After clustering patients through PCA and HCPC, we observed that the differences between HV and each cluster were not homogeneous. Lower levels of IL-6, IL-8, CCL-2, CXCL-10, and VEGF were observed when HV were compared to the patient clusters C2 T0 , C3 T0 , and C4 T0 . On the contrary, TGF-β level was significantly lower in HV compared to patients in C2 T0 and C3 T0 , and IL-21 was lower compared to C3 T0 . Only two cytokines were higher in HV and limited to the comparison to C2 T0 (IL-13) and C4 T0 (TNF-α). These differences among HV and each specific patient cluster underline that the clusters identify different TME and suggest the existence of multiple different immune-escape mechanisms even in a small series of patients such as our population.
The comparison among clusters at T0 revealed that the median OS of patients in C2 T0 was double compared to all other patients, with a 64% risk reduction in death. Patients in C2 T0 are characterized by a lower median level of many important immunosuppressive cytokines, such as IL-6, IL-8, IL-10, and CCL-2. Of note, patients in C2 T0 have better outcomes than patients in C3 T0 showing the highest value of the same cytokines.
The analysis of the longitudinal shift among cytokines from T0 to T1 revealed significant changes only in patients clustered in C2 T0 and C3 T0 . In particular, patients belonging to C2 T0 showed an increase in IL-13 at T1, while patients in C3 T0 exhibited IL-4 and VEGF increase. IL-2 grew at T1 in both clusters. IL-4 is reported as having an antiapoptotic effect [28] and its rise in C3 T0 justifies the poor outcome of this cluster. Curiously, IL-13 is the only cytokine which increased from T0 to T1 in each patient in C2 T0 . Apart from IL-2 rising, which may indicate an initial recovery of immune response, the increase in IL-13 is intriguing. At T0, IL-13 is the only cytokine significantly higher in HV compared to patients in C2 T0 . IL-13 is historically related to the pathophysiology of asthma and other common autoimmune diseases [29] that may have a long pre-clinical stage [30]. It might explain the high variation of IL-13 observed in HV. IL-13, structurally similar to IL-4, plays a role in tumor proliferation and metastatization and is considered a Th-2-derived protein [31]. However, no clear relationship between IL-13 levels and poor outcome in breast cancer patients has been described [32]. IL-13 is involved in inhibition of inflammatory cytokines [33], up-regulation of tumor associated macrophages (TAM), and myeloid derived suppressor cells (MDSC) [34] without affecting activated T cells [35,36]. Therefore, the increase in IL-13 can be considered a signal of reactivation of the immune response as well. Indeed, in patients in C2 T0 , IL-13 at T1 is similar to the values observed in HV, and may contribute to the good outcomes of these patients.
After four courses of eribulin (T1), PCA and HCPC identified only four clusters. This may suggest that treatment reduces the heterogeneity of the TME.
Among them, patients in C1 T1 showed the best outcome with a long median PFS and OS leading to a 50% and 63% risk reduction in disease progression and death, respectively.
PFS in the remaining clusters is very similar among them, while OS gradually decreases from C1 T1 to C4 T1 without overlapping the confidence intervals between C2 T1 and C3 T1 .
Patients in C1 T1 expressed very low median values of immune-suppressive cytokines, such as IL-4, IL-6, IL-8, CCL-4, and TGF-β. All of these are linked to major pro-tumor effects and poor survival [37][38][39][40]. Their low median values in C1 T1 may represent a positive factor and may contribute to the best outcomes observed in this cluster.
Cytokine expression in the two best clusters, C1 T1 and C2 T1 , is quite similar. However, significant differences exist in the median values of TGF-β, CCL-4, and CCL-22, that are higher in C2 T1 . TGF-β is a well-known negative prognostic factor, already demonstrated in many tumors [39,41]. CCL-4 and CCL-22 are also related to a poor prognosis [40,42] and may contribute to explaining the different outcomes of patients in C1 T1 and C2 T1 . Intriguingly, TGF-β was higher in C2 T1 than in C3 T1 , the latter showing a worst outcome compared to the former. However, many other Th-2 cytokines were higher in C3 T1 than in C2 T1 . This observation stresses the concept that focusing on a single cytokine is not adequate, as cytokines interact with each other as previously reported [1].
Even the limited magnitude of the different OS between C1 T1 and C2 T1 underlines that many factors may contribute to, attenuate or exacerbate, the effects related to a single cytokine.
Therefore, clustering patients using a panel of multiple cytokines analyzed by PCA and HCPC may offer a better way to understand the TME and the complex interplay among its many components, such as tumor cells, stroma, and immune cells, and may explain the context dependent effects of many proteins [43].
It would be interesting to link our data with the molecular subtypes and metastatic profiles of the patients. For example, Kawaguchi et al. [18] demonstrated a different cytokine signature between metastatic and non-metastatic breast cancer patients. Unfortunately, due to the limited number of patients and the many variables to be considered, this analysis would not have had adequate statistical power.
We grouped our population into four groups according to PFS and OS. Patients of all groups were dispersed among the clusters either at T0 or T1. Only patients of group 2, showing short PFS and long OS, were mainly allocated in cluster C2 T0 . At T1, four of these patients were included in C1 T1 and four in C2 T1 . Even if the former cluster showed the best behavior, the latter approached a C1 T1 OS curve that is clearly separated from the OS curves of the two remaining clusters. Therefore, eight of nine patients included in group 2 are included in the clusters with the best OS.
All nine patients representing group 2 received further treatments after eribulin, but no one achieved an objective response. Therefore, the longer OS cannot be attributed to the further treatments. Indeed, Cox analysis, performed using confounding factors such as the number of lines of therapy, number of metastases, and different sites of deposits, confirmed that belonging to the C2 T0 , which encompasses seven out of nine patients of group 2, was an independent prognostic factor.
Interestingly, Haddad et al. reported that a subgroup of patients with head and neck cancer, treated with nivolumab in the CheckMate 141 study, benefited from continuing treatment beyond progression, on investigator choice, if they met specific clinical characteristics such as investigator-assessed clinical benefit, no rapid disease progression, treatment tolerance, and stable performance status [44]. Continuing treatment beyond progression converted initial progression to PR in almost 25% of these patients.
In our series, patients of group 2 met all these characteristics. Therefore, we might hypothesize that treating these patients beyond early progression could improve their outcomes. Indeed, our study suggests that the empirical clinical characteristics used by Haddad et al. might correspond to specific clusters of circulating cytokines.

Conclusions
We are aware that our study may only generate hypotheses for further investigations. With this limitation, our data show the heterogeneity of the cancer-patient population even when we selected patients with the same tumor and clinical stage. However, considering circulating cytokines, we can group the population into six clusters with different outcomes. In addition, we can highlight that eribulin may reduce heterogeneity as witnessed by the decreased number of clusters after treatment.
Moreover, our data imply the possibility of using a low-invasive approach, such as serum analysis, to assess information regarding prognosis, and are supported by similar experiences from other authors [18,26,45]. This approach might be translated into clinical practice, in order to help physicians to identify cytokine clusters correlating to patients' outcomes.
In our opinion, the most relevant hypothesis is that some patients, despite early progression, might benefit from treatment beyond progression. If this theory should be confirmed, this approach might be rapidly translated into clinical practice.
In support of the hypothesis, we refer to the Checkmate141 study conducted in patients with a different solid tumor and treated with different drugs.
However, when the target of treatment is TME, such as for immunotherapy or for the immune off-target effects of conventional chemotherapy, we face three major situations: inflamed tumors, excluded tumors, or desert tumors, which are the same across all solid tumors, with different distribution in different primaries [46]. Therefore, differences among primary sites may be much less relevant.
For these reasons, we believe that this hypothesis deserves further investigation. Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.
Data Availability Statement: Data supporting results can be found at the ARCO Foundation laboratory at Santa Croce e Carle Teaching Hospital (Cuneo, Italy).