The inclusion criteria for this retrospective observational study included patients diagnosed with breast cancer, with FFPE tissue specimens available at Yantai Affiliated Hospital of Binzhou Medical University, Yantai, China, from 2008 to 2013 consecutively and non-selectively. The specimen must have more than 50% tumor tissue based on H&E staining. All the treatments were adjuvant treatment. Follow-up data were available for 143 patients out of 155 Luminal-like patients (92.3%), with the last follow-up on April 1, 2019. The endpoint of the overall survival analysis was defined as the time between breast cancer surgery and death or last follow-up (April 1, 2019). The median time to censoring was at 85 months, with a maximum at 132 months. All the procedures of the study, including sample collection and study protocol, were in accordance with the Declaration of Helsinki and were approved by the Medical Ethics Committee of Yantai Affiliated Hospital of Binzhou Medical University (approval #: 20191127001), with an informed consent form waiver for the archived specimens.

The general reagents for cell culture were purchased from Thermo Fisher Scientifics (Waltham, MA, USA). The chemicals used for protein expression were purchased from Takara, Inc. (Dalian, China), and the Nickel-His GraviTrap affinity column for protein purification was purchased from GE Healthcare. All the other chemicals were purchased from Sinopharm Chemicals (Beijing, China).

Rabbit anti-cyclinD1 monoclonal antibody (EP12) was purchased from ZSGB-BIO (Beijing, China). Horseradish peroxidase-labeled donkey anti-rabbit IgG secondary antibody was purchased from Jackson Immunoresearch lab (Pike West Grove, PA, USA). The Pierce bicinchoninic acid (BCA) protein quantification kit was purchased from Thermo Fisher Scientific, Inc. (Carlsbad, CA, USA). The recombinant human cyclinD1 protein was prepared in-house. The QDB plate was manufactured by Quanticision Diagnostics Inc., at RTP, NC, USA.

A DNA sequence corresponding to the 271-295AA of human cyclinD1 (NCBI #: NP_444284.1) was inserted into PET32a (+) expression vector and verified by DNA sequencing. The plasmid was expressed in BL21 (DE3) competent cells after the induction of isopropyl β-D-1-thiogalactopyranoside. Total bacterial protein lysate was extracted in binding buffer (20 mM sodium phosphate, 500 mM NaCl, 20 mM imidazole, pH 7.4) by ultrasonication. The recombinant cyclinD1 protein of bacterial lysate was purified by Ni-NTA column according to the instructions. Fractions containing cyclinD1 protein were pooled and concentrated on a Millipore ultrafiltration spin column. The protein concentration of cyclinD1 was determined using a BCA protein quantification kit, and the purity was determined by SDS-PAGE before subjecting to formalin fixation. The purified cyclinD1 protein was subject to fixation by incubating with nine volumes of 10% neutral buffered formalin solution at room temperature for 5 h before it was precipitated by nine volumes of cold acetone at -20°C for at least 2 h, followed by centrifugation for 10 mins at 12,000g to collect protein pellet. The pellet was washed once with ice cold acetone to remove residual formalin. The protein pellet was then dried at room temperature for 2 h before it was solubilized with a lysis buffer. The final protein concentration was determined using a BCA protein quantification kit.

The IHC analyses for both Ki67 and cyclinD1 were performed using a procedure described elsewhere (8), using SA38-08 antibody for cyclinD1 and SP6 for Ki67 from LBP (www.gzlbp.com) at 1:100 dilution, respectively.

Breast cancer FFPE specimens in 2 × 15-µm slices were de-paraffinized and solubilized with lysis buffer (50 mM HEPES, 137 mM NaCl, 5 mM EDTA, 1 mM MgCl₂, 10 mM Na₂P₂O₇, 1% TritonX-100, 10% glycerol). The MCF-7 cell pellet was collected and fixed in formalin for 30mins before solubilization with the same lysis buffer. The supernatants were collected after centrifugation for QDB analysis. The protein concentration was detected using a BCA protein quantification kit.

The QDB process was described previously, with minor modifications (9–12). In brief, FFPE tissue lysate (about 0.35 μg/unit) and MCF-7 cell lysate (about 0.08 μg/unit) were applied onto the QDB plate, at 2 μl/unit, in triplicate. Serially diluted cyclinD1 recombinant protein was included in each plate as protein standard for the calculation of absolute cyclinD1 protein levels. The loaded QDB plate was dried in the air for 4 h and then blocked with a blocking buffer (4% non-fat milk in tris-buffered saline–Tween 20 for 1 h. The plate was inserted into a 96-well microplate pre-filled with 100 μl of anti-cyclinD1 antibody (clone EP12, 1:500 in blocking buffer) and incubated overnight at 4°C. The plate was rinsed twice and washed 4 × 10 mins before it was incubated with a donkey anti-rabbit secondary antibody for 4 h at room temperature. The plate was then rinsed twice and washed 5 × 10 mins before being processed with enhanced chemiluminescence reagent and quantified with a Tecan Microplate reader. The chemiluminescence signals were converted into absolute cyclinD1 protein level in each sample using a protein standard. All results were averages of three independent experiments, with each sample in triplicate.

MCF-7 cell lysate was used as internal control. The absolute cyclinD1 level of MCF-7 cell lysate was measured with the QDB method, and the results were documented. The overall experiment was judged as valid when the measured cyclinD1 level of MCF-7 cell lysate fell within ±20% of the documented value.

All statistical analyses were performed with either R version 3.6.2 or GraphPad Prism 7.0 software, using a two-sided statistical test. The results were reported as mean ± standard deviation (SD). A p-value of <0.05 was defined as statistically significant. All the missing values were treated as a new category. The cases lost to follow-up were not included in the analysis. Patients still alive at the last study follow-up were censored. The endpoint of overall survival analysis was defined as the time between breast surgery and death or last follow-up (April 1, 2019).

The correlation between cyclinD1 and Ki67 protein levels was analyzed with Spearman’s correlation coefficient analysis. Univariate Cox proportional hazard model-fitted overall survival was employed for hazard ratio (HR) and corresponding 95% confidence interval (CI) estimation. Multivariable Cox model was used to examine the association between Ki67 and/or cyclinD1 and OS adjusted for clinical factors, such as age, nodal positivity, pathological tumor size, tumor grade, and types of treatment. The Schoenfeld residuals test is used to test the proportional hazard assumption in the Cox model.

The cyclinD1 levels measured by the QDB method were dichotomized for OS analysis using an optimized cutoff value determined by the “surv_cutpoint” function of the “surviminer” R package.

cyclinD1 Harrell’s C-index was used as a measure of predictive accuracy of survival outcome. A C-index of 0.5 indicates an accuracy similar to random guessing and that of 1.0 indicates 100% predictive accuracy. We calculated unbiased estimates of the C-index for four combinations of variables to assess the contribution of different factors. Ki67 and cyclinD1 were used as categorical variables. The relationship between various subtypes or subgroups and OS of Luminal-like specimens was evaluated using the log-rank test and visualized by Kaplan–Meier survival curve analysis. The Schoenfeld residuals test is used to test the proportional hazard assumption in the Cox model.

The clinicopathological characteristics of all 143 Luminal-like FFPE specimens with follow-up data are provided in Table 1 . The flow chart of the specimens is shown in Figure 1 . These Luminal-like specimens were collected sequentially and non-selectively from a local hospital, as described in the “Materials and Methods” section, with the median time to censoring at 85 months and maximum at 132 months.

Total protein lysates were extracted from 2 × 15-μm FFPE tissue slices from each individual specimen. The linear ranges of both total tissue lysates and recombinant cyclinD1 protein were defined using the QDB method ( Supplementary Figure S1 ). The distribution of cyclinD1 levels was from 0.02 to 3.77 μmol/g, with mean at 0.31 ± 0.04 μmol/g ( Figure 2A ). The expression levels of cyclinD1 were weakly correlated with those of Ki67 when analyzed using Spearman’s correlation analysis (ρ = 0.30, p = 0.0003, n = 143) ( Figure 2B ).

The putative prognostic role of cyclinD1, alongside Ki67, for OS of Luminal-like breast tumors was explored first with univariate survival analysis ( Table 2 ). Both cyclinD1 and Ki67 were found to be independent prognostic factors for the OS of the patients, with HR for Ki67 at 1.17 (95% CI: 1.05–1.30, p = 0.0054) and for cyclinD1 at 2.33 (95% CI: 1.39–3.89, p = 0.0012), respectively. Meanwhile, the pathological node status and age were also found to be statistically significant.

These two biomarkers were further analyzed with multivariate Cox regression survival analysis ( Table 2 ). Unexpectedly, although both factors were considered biomarkers of cell proliferation, they were found to be independent from each other, with HR for Ki67 at 1.16 (95% CI: 1.01–1.33, p = 0.0357) and for cyclinD1 at 2.86 (95% CI: 1.46–5.60, p = 0.0022). The age and pathological node status were other independent prognostic factors in the same analysis.

The contributions of both cyclinD1 and Ki67 to the OS of these patients were also evaluated using C-index ( Supplementary Figure S2 ). The conventional clinical factors, including age, pathological tumor size, histological grade, and pathological node status, contributed collectively to the C-index at 0.72. The addition of Ki67 alone improved the C-index to 0.81, while the addition of cyclinD1 alone improved the C-index to 0.76. When both Ki67 and cyclinD1 were combined with conventional clinical factors, the C-index was improved to 0.82.

Next, we attempted to apply these findings to improve the prognosis of Luminal-like tumors. An optimized cyclinD1 cutoff value was determined for OS analysis using the “surv_cutpoint” function of the “surviminer” R package at 0.44 μmol/g, and this value was used to separate these Luminal-like specimens into cyclinD1-high (C_h) and cyclinD1-low (C_l) groups. Meanwhile, these specimens were also separated into Ki67-high (K_h) and Ki67-low (K_l) groups using the 2.31 nmol/g defined in the previous study (8). In the accompanying manuscript III (submitted), the proposed 0.44 μmol/g cutoff was validated using an independent cohort from another hospital.

These cutoffs were used together to separate these Luminal-like specimens into four subgroups. The C_hK_h subgroup included specimens with both biomarkers above the respective cutoffs, while those in C_lK_l were below the respective cutoffs. C_hK_l referred to those with only cyclinD1 level above the corresponding cutoff, while the C_lK_h referred to those with only Ki67 level above the corresponding cutoff. The OS of these four subgroups was analyzed using Kaplan–Meier survival analysis ( Figure 3 ). We found that the C_lK_l subgroup (n = 65) had a significantly improved 10y SP at 90% (95% CI: 0.82–0.99), while the C_hK_h subgroup (n = 18) had the worst 8-year survival probability at 26% (95% CI: 0.09–0.78). The C_hK_l (n = 8) and C_lK_h (n = 52) subgroups had similar 10y SP at 75% (95% CI: 0.50–1.00) and 76% (95% CI: 0.65–0.89), respectively. The overall p <0.0001 from log-rank test. When compared with the C_lK_l subgroup, the HR of the C_hK_h subgroup increased by 9.02 folds ( Supplementary Figure S3 ).

Our data suggested that the combined use of Ki67 and cyclinD1 from the QDB measurement may significantly improve the prognosis of Luminal-like patients. However, until protein biomarkers are routinely quantitated objectively in daily clinical practice, IHC analysis remains the primary method of biomarker assessment. Meanwhile, numerous Luminal-like breast cancer patients face the clinical decision of chemotherapy or no chemotherapy every day. Therefore, we explored the feasibility of applying our finding into daily clinical use by improving the performance of the surrogate assay using IHC-based cyclinD1 scores.

cyclinD1 levels from QDB measurements were plotted against the corresponding IHC scores, with the proposed 0.44 μmol/g shown in the plot ( Supplementary Figure S4 ). We observed that the majority of the specimens with cyclinD1 scores ≤30% had their absolute levels <0.44 μmol/g. Thus, an IHC-based cyclinD1 score of 30% was used as putative cutoff to evaluate its impact on the performance of the surrogate assay.

The patients were stratified into Luminal A-like and Luminal B-like subtypes from the surrogate assay using an IHC-based Ki67 score of 14% and 20%, respectively. The stratification of Luminal A-like patients using cyclinD1 score of 30% as cutoff yielded no statistical difference using either a Ki67 score of 14% or 20% as cutoff. However, among the Luminal B-like subtype, we observed a significantly improved separation of those Ki67 <14% and cyclinD1 ≤ 30% from those of Ki67 ≥14% and/or cyclinD1 >30%, with p = 0.016 ( Supplementary Figure S5A ). We combined this subgroup (Ki67 <14% and cyclinD1 ≤ 30%) with the Luminal A-like subtype and named them modified Luminal A-like subtype. The rest of the Luminal B-like subtype (Ki67 <14 and cyclinD1 >30% or Ki67 ≥14%) were named as modified Luminal B-like subtype. We achieved 10y SP at 89% (n = 79) and 61% (n = 64) for the modified Luminal A subtype and B subtype, respectively, with p = 0.00061 from log-rank test. As comparison, the 10y SP for Luminal A-like subtype and Luminal B-like subtype was 88% (n = 61) and 68% (n = 82), respectively, with p = 0.031 from the surrogate assay using a Ki67 score of 14% as cutoff ( Figure 4A and Supplementary Table S1 ).

When the Ki67 score of 20% was used as cutoff, we were unable to achieve statistical significance between those of Ki67 <20% and cyclinD1 ≤30% from those of Ki67 ≥20% and/or cyclinD1 >30% among Luminal B-like patients ( Supplementary Figure S5B ). Nonetheless, we assigned again the modified Luminal A-like subtype to include both patients of Ki67 <20% and cyclinD1 ≤30% of Luminal B-like subtype with those of Luminal A-like subtype. The rest of the Luminal B-like subtype (Ki67 <20% and cyclinD1 >30% or Ki67 ≥20%) were assigned as modified Luminal B-like subtype. We calculated their 10y SP at 84% for modified Luminal A-like subtype (n = 91) and 64% for modified Luminal B-like subtype (n = 52), with p = 0.02 using log-rank test. As comparison, the 10y SPs were 84% (n = 70) and 70% (n = 73) for Luminal A- and B-like subtypes, respectively, from the surrogate assay using a Ki67 score of 20% as cutoff, with p = 0.10 from log-rank test ( Figure 4B and Supplementary Table S1 ).