Circulating neutrophils from patients with early breast cancer have distinct subtype-dependent phenotypes

Cancer mobilised neutrophils have been reported to undergo significant phenotypic changes in patients across a range of cancers [4]. However, most of the studies investigate circulating neutrophils in cancer patients with more advanced disease and following treatment interventions.

In a mouse model of breast cancer, mice harbouring spontaneous MMTV-PyMT breast tumours that induced neutrophilia [14], we found that systemically mobilised neutrophils showed reduced surface expression of L-selectin, (CD62L^low) in various tissues (Additional file 2: Figure S2). In this proof-of-concept study, we aimed to test if similar changes in circulating neutrophils could be detected in newly diagnosed patients with EBC. To do this, we compared neutrophils from blood samples of 44 treatment-naive patients with EBC (after biopsy, but prior to surgery or neoadjuvant chemotherapy) with those from 44 paired sex and age matched healthy volunteers (HVs) collected on the same day as each breast cancer sample. Moreover, to accommodate for any confounding factors such as the effect of adrenaline or cortisol released by patients experiencing stress in anticipation of treatment, we recruited an additional group of patients who were undergoing surgery for benign breast disease. Selected patients’ data and selection criteria are listed in Additional file 3: Table S1, Additional file 4: Table S2, Additional file 5: Table S3. The neutrophil gating strategy for flow cytometry is highlighted in Additional file 1.

We firstly analysed circulating neutrophils from EBC patients and paired HVs by FACS to assess whether a similar change in CD62L could be observed in patients. In line with the neutrophilia detected in the animal model of EBC [14], neutrophils were found to make up a higher proportion of total live cells in the blood (Fig. 1a). These data reinforce the notion that neutrophils are systemically responding to the presence of breast cancer. When we analysed the percentage of CD62L^low neutrophils in patients and compared to HVs, we found that, consistent with the murine HR negative MMTV-PyMT breast cancer model (Additional file 2: Figure S2), neutrophils from HR-negative patients had significantly elevated proportion of CD62L^low subpopulations compared to HVs (Fig. 1b). Conversely, HR-positive patients had significantly lower levels of CD62L^low subpopulations compared to matched HVs (Fig. 1c). These data indicate that perturbations in the surface expression of CD62L were detectable very early in cancer and are dependent on breast cancer’s hormonal receptor status (Fig. 1d, e).

As we could detect differences in markers associated with neutrophils aging via flow cytometry in circulating neutrophils of patients with EBC, we next assessed if a difference in their properties could be detected in isolated neutrophils ex vivo. Neutrophils are reported to have a short lifespan; however, this can be influenced in steady state by different tissue context [16] as well as by factors generated during inflammation [17, 18]. We therefore measured neutrophils’ lifespan, upon their isolation from the patients’ circulation. We calculated the half-life of neutrophils in culture for 32 h and found that neutrophils from patients with EBC (Additional file 6: figure S3a-b), regardless of subtype, had a longer half-life compared to paired HVs when cultured in their own plasma (Fig. 2a). Moreover, the fold increase in the half-life of neutrophils from EBC patients mildly correlated with the neutrophil to lymphocyte ratio (NLR) values (Additional file 6: figure S3c). In addition, when neutrophils from patients with EBC were cultured with plasma from the paired HVs we still observed a significant increase in half-life compared to neutrophils from HVs (Fig. 2b). Similarly, a difference was not observed when cancer patient-derived neutrophils were cultured in their own plasma or in HV plasma (Fig. 2c). This suggests that neutrophils are intrinsically primed to live longer when isolated from blood of cancer patients. Moreover, plasma from patients with EBC did not show any effect on neutrophils from HVs (Fig. 2d, e), reinforcing the idea that a change in neutrophil behaviour originates from a more complex in vivo priming.

Taken together, we could detect an intrinsic priming affecting neutrophils in EBC patients which display a prolonged half-life.

Following the indications of an early perturbation detected by flow cytometry in circulating neutrophils from patients newly diagnosed breast cancer along with a prolongation of their lifespan, we aimed to assess if these phenotypic changes were reflected in appreciable overall signalling pathway activities. To do this we investigated their broader intracellular kinase network activation states using the Pamgene Kinome assay. To avoid unintentional activation due to antibody binding, circulating neutrophils were isolated from blood using a negative selection strategy. Total proteins were extracted from circulating neutrophils from EBC patients and their paired HVs and loaded onto kinome chips to functionally screen their kinases activity by quantitative measurements of phosphorylated target peptides specific to either serine/threonine or protein tyrosine kinases (STKs or PTKs respectively) (Additional file 7: Figure S4a and S4b). Specific peptides that increase or decrease output of phosphorylation can be linked to the kinase activity that typically recognised them. Collectively this analysis provides a functional readout of overall kinome activity in circulating neutrophils. Additional file 7: Figure S4c shows an example of the raw STK data represented as a heatmap for the HR-positive HER-2 negative group of patients. This heatmap represents the overview of phosphorylated peptides within neutrophils in patients and HVs samples prior to normalisation of the data. Only peptides which passed quality control were Log2 transformed, and combat corrected (normalised for batch and pair effect to account for experiments done on different days) prior to further analysis. The log fold change (LFC) of phosphorylated peptides between patients compared to paired HVs was calculated for each subtype of breast cancer (Additional file 7: Figure S4d, e). This showed there appears to be some variability between patients regarding the degree of phosphorylated peptides for individual patients compared to their paired HVs.

In order to identify global patterns in the data, we used Principal Component Analysis (PCA) to assess presence of clustering for the different subtypes of breast cancer compared to HVs for the two classes of kinases, STK and PTK family (Additional file 8: Figure S5). Overall, the results did not show obvious clustering of patient and HV groups, only in HR+ and triple negative patients some level of clustering could be observed compared to their paired HVs (Additional file 8: Figure S5b-c and S5f-g).

To get a better overview of the changes in kinase activity in the different breast cancer subtypes, we used Coral trees to plot all the kinases tested. Coral trees are used for visualization of the human kinome superfamily, representing both quantitative and qualitative data [19]. Using this analysis, the differences between the neutrophil kinome in all patients with a given breast cancer subtype compared to HVs became clearer (Fig. 3). Interestingly, the entire neutrophil kinome, and particularly the PTK family, appears to be more perturbed in patients with early HR-positive compared to HR-negative breast cancer (Fig. 3a). HR-negative disease is characterized by a milder increase in activity of the TK family compared to HVs while other kinase activity is overall reduced compared to the HVs (namely kinases within the CGMC family, AGC family and CK1 kinase) (Fig. 3b). Strikingly, the presence of HER-2 positive disease appears to revert this dominant TK activation increase and induced a reduction compared to HVs in both HR-positive and HR-negative patients (Fig. 3c, d). The rest of the kinome remains more activated compared to HVs.

Collectively these data show an early neutrophil cancer-specific priming in their overall intracellular kinome activity, that appears to be different accordingly to breast cancer subtype and deeply influenced by the presence of HER-2 positivity.

As shown in Fig. 3, the kinome activity signatures of circulating neutrophils from patients with breast cancer are perturbed compared to paired HVs and different accordingly to breast cancer subtype. However, there are differences between the EBC patients and HVs: only patients with cancer have previously undergone a biopsy and are under the stress of a surgery at the time of blood sampling. Therefore, to test if those differences are specific to cancer patients, we collected circulating neutrophils from patients that were diagnosed with benign breast disease after biopsy. Importantly, when analysing their kinome activity, while some perturbation could be observed, we did not detect alteration specifically in TK activity compared to paired HVs (Fig. 4a). Given this data suggested a potential breast cancer specific change, we tested if the changes in kinome activity in circulating neutrophils could retain the potential to predict characteristics of the disease in patients, by performing two types of predictive model analysis. The raw data used for the analysis are provided in Additional file 9: Supplementary File 1 and included the two variables measured for each kinase activity by comparing each patient to its paired HV measured using the Pamgene platform. The two variables indicate the degree of significance for each kinase for that given patient vs paired HVs (MFS) and the direction of the change, either up or down regulated compared to HVs (KS) respectively (Additional file 9).

Two machine learning algorithms were used for training the model according to the number of patients to compare in the two groups (Additional file 10: Supplementary File 2). Firstly, we compared the 29 samples from patients of all types with the 6 samples from benign lesions. In this case we use a Random Forest algorithm to train the model, which is better suited for imbalanced data (a much fewer patient in one group). Strikingly, randomly selected patients which were not used to train the model, could be predicted to belong to either the cancer or the benign group with a sensitivity of 82% and a specificity of 66% (Fig. 4b). Moreover, this model predicted that the kinase activity changes with the stronger weight were LCK and PKG1 (Additional file 11: Figure S6a). When looking at those two kinases in the Upstream Kinase Analysis (UKA) generated list of putative kinases based on peptides targeted for phosphorylation in patients with different breast cancer subtypes (Additional file 12: Figure S7), they are indeed among the kinases perturbed in all cancer patients but do not show changes in circulating neutrophils of benign lesions.

Secondly, given the marked effect of HER2 expression on PTK activity observed in patients of both breast cancer types compared to HV, we tested if HER-2 status could be predicted in cancer patients from the kinome perturbation in their circulating neutrophils. We compared the profiles of neutrophils isolated from both sub-types of breast cancer patients, either negative or positive for HER-2 mutation using Lightgbm algorithm to train the model since the number of patients in the two group are comparable (Additional file 10: Supplementary File 2). The results showed that we could identify the HER-2 status of patient with a specificity of 73% and a sensitivity of 57% (Fig. 4c). Interesting the model predicted that the kinase activity change with the stronger weight was p70S6K beta (Additional file 11: Figure S6b).

In conclusion, by training a predictive model with only the few patients analysed here, we could show that the differences in tumour-induced kinome activity of circulating neutrophils could have the potential to distinguish between EBC and benign lesions and it could detect the presence of HER-2 mutation in breast cancer.

Unless otherwise stated, statistical analyses were done using Prism Version 9 (GraphPad software). Error bars for column plots represent average ± Standard error of the Mean (SEM).

When analysing the difference in one variable between two experimental groups, paired or unpaired student T tests were carried out and p-values were obtained from this. One-way ANOVA was used for comparing the difference in one variable between two or more experimental groups. Two-way ANOVA was used to perform multiple comparisons or analyses between experimental groups. The significance level was set at p < 0.05 and the actual p values or symbols (*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001) were used. Biological replicates are represented as n values.

MMTV-PyMT+ mice spontaneously developed a primary tumour and increased neutrophil presence in the blood, bone marrow, spleen and liver. Neutrophil infiltration was quantified by flow cytometry by anti-Ly6G antibody (clone 1A8 from BioXcell).

Patients with EBC were selected based on HR status and HER2 status of their tumour. The term early breast cancer refers to the breast tumour being localised to the breast and includes patients with or without local lymph node involvement where the management is aimed for curative intent. The patient characteristics including age, menopausal status, medical conditions, medication, tumour size, tumour grade, HR status and HER2 status and lymph node status are shown Additional file 3: Table S1a. The paired HV characteristics are shown in Additional file 3: Table S1b. The characteristics for patients with benign breast disease and paired HVs are in Additional file 4: Table S2. Patients with medical conditions or using medication which were known to influence neutrophil phenotype or function were excluded from the study (full list of inclusion and exclusion criteria can be found in Additional file 5: Table S3).

Informed written consent was obtained from the patients and HVs on the day of blood taking itself, and samples were completely anonymised so participant identification could not be possible. The proposed human study was approved by Imperial Tissue bank committee and REC approval was obtained. It was also approved by the Crick ethical review committee for recruitment of HVs at the Crick. Material transfer agreements were signed by Imperial College and The Crick Institute to allow samples to be processed between the 2 sites. All work involving human participants was undertaken in accordance with Good Clinical Practice guidelines.

Biological variability between individuals and size of effect calculations was estimated based on previous research data from investigating human neutrophil subpopulations [24]. Patients with EBC were selected based on HR status and HER2 receptor status. A control group of patients with benign breast disease and their paired HVs were also recruited. With the help of the Bioinformatics department at The Crick Institute, we estimated that ideally 21 patients were required for each subgroup of breast cancer patients and the group with benign breast disease and 11 paired healthy control volunteers (HVs) for each group were needed to obtain a statistical power at > 80%, alpha = 0.05, to detect differences of 2.97 "units" in neutrophil subpopulations between groups (two-sided T test). The method of Schouten was used to estimate the sample sizes under the assumption of unequal variances, sample sizes and independent samples, along with approximation to the T test. Given the logistical challenges in recruiting patients during the COVID pandemic, we aimed to recruit 21 patients in each breast cancer subgroup but ended up recruiting fewer in each group.

Pamgene’s Bionavigator software (version 2.3, 2020) was used for image quantification and statistical analysis. This included a quality check to exclude any arrays that showed clear visual defects. Data was log-transformed and normalised to take into consideration any variation between patient and HV kinase samples related to neutrophil isolation on different experiment days. PCA was performed on the transformed data. A prediction of kinases responsible for the peptide phosphorylation changes were calculated using the Upstream Kinase Analysis tool in Bionavigator software. The Median Final score is calculated by combining a sensitivity score (difference between patients and HV groups) and specificity score (a set of peptides to kinase relationship which is predicted from 6 databases that Pamgene uses). The Median Kinase Statistic represents the differences between the groups, with effect size (values) and direction > 0 is activation and < 0 is inhibition. Results were also visualised using a kinase phylogenetic tree created via the online Coral tool (http://​phanstiel-lab.​med.​unc.​edu/​).

Prediction Model The raw data for the prediction model are provided in Additional file 9 and the R code used in Additional file 10.

Code availability All analysis was performed using R version 4.1.3 and the and the R code used is provided in Additional file 10.

Human blood samples (40 ml) were collected tubes from HVs at the Crick Institute and Imperial College with informed consent. Blood was collected in heparinised tubes for neutrophil separation with Histopaque 1119 for the flow cytometry experiments for the first 8 patients and paired HVs (P01–P08). EDTA tubes were used for blood collection for neutrophil separation with the immunomagnetic neutrophil isolation kits (which was used for all Pamgene assays and flow cytometry experiments for patients P09–P44, benign patients and paired HVs). Blood was sent in an EDTA tube for Full Blood Count analysis at Charing Cross Hospital from patient P07 and paired HV onwards to calculate the NLR.

For the Kinome assay, neutrophil Isolation kits by Millentyi and Stem Cell technologies were used to isolate neutrophils from the blood which worked based on immunomagnetic negative selection. For patients P01–P14 and B01–B02 and the paired HVs, the Milltenyi neutrophil isolation kit was used (Catalogue number 130-104-434). However, the Stem Cell neutrophil isolation kit (Catalogue number 19666) was used for P15–P44 and B3–B9 and the respective paired HVs as this kit gave a much higher yield of neutrophils.

FACS analysis: Single cell suspensions from human blood (2.5 × 10⁶ cells/tube) were incubated with 5% human FcR Blocking Reagent (Miltenyi) for 10 min on ice and then incubated for 30 min over ice with various specified fluorescent surface marker antibodies (including CD16, CD66b and CD62L (clones CLB-gran 11.5 (catalogue 746199), G10F5 (catalogue 564679) and DREG-56 (catalogue 740982) respectively, from BD Biosciences) using a 1:200 dilution). The LSRFortessa cell analyser running FACSDiva software (BD Biosciences) and FlowJo software was used.

Pamgene kinase assay: Neutrophils lysates were generated using the Pamgene standard manufacturer protocol and were stored at − 80 °C. Lysate samples were placed on STK Pamchips® and PTK Pamchips® and were run separately using the standard manufacturer protocol. Additional file 7: figures S4a and S4b represent an outline of the various steps involved for STK and PTK assays respectively. Images are taken every 5 min to generate kinetic data in real time which was analysed using Pamgene™ Bionagivator software.

Blood was centrifuged at 300rcf for 10 min at 37 °C. The yellow supernatant was collected and centrifuged at 2000rcf for 10 min. The supernatant was collected for the functional use immediately or snap frozen and stored at -80 °C.

Human neutrophils (5 × 10⁴ cells/well) were seeded in a 96-well plate containing 10 mM HEPES and 3% plasma (obtained from the cancer patient or paired HV). Then 0.2 μM of Sytox Green (used to identify dead cells) and 4 μg/ml of Hoechst (used to identify DNA and cells) were added to each well. Cells were taken to the inverted Nikon wide-field microscope system and measurements were recorded per well every 30 min for 20–32 h using a 40 × objective. Quantification of Sytox positive cells compared with the total cells was performed using Fiji software. Half-life values represent the timepoint at which 50% of the cells within the field of view are Sytox positive.