Age-associated genes in human mammary gland drive human breast cancer progression

The RSEM estimated raw read counts data were downloaded from TCGA Data Portal on July 25, 2014, and normalized by library sizes, log2 transformed, and further normalized by quantile normalization. Removal of 25% of genes with the lowest mean expressions resulted in subsequent analyses on 15,398 genes. In identifying genes associated with age (birth_days_to), the genes whose expression were considered to be correlated with age were selected by three criteria: (1) R² value among the highest 5%, (2) absolute value of slope for age among the highest 5%, and (3) adjusted P value for the slope among the lowest 5%. In identifying genes differentially expressed comparing post-menopausal normal against pre-menopausal normal, and comparing matched tumors against match normal samples, a combination of methods using moderate hierarchical t test (R/limma) and moderate fitting based on negative binomial distribution (R/DESeq2), with and without surrogate variable analysis for potential confounders (R/SVA) with cutoff as FDR < 0.05, were implemented. The common overlapping genes from four methods were determined as the final list for differentially expressed genes.

Two metrics were implemented to test the correlation between gene expression pattern of the 38 ABC genes and subtypes of breast cancer. The first metric was the sum of gene expression levels of 14 upregulated genes minus the sum of gene expression of 24 downregulated genes.

\( \mathrm{Score}=\sum \limits_{\mathrm{g}\;\mathrm{in}\kern0.17em \mathrm{genes}}\mathrm{Expression}\left(\mathrm{g}\right)\times \mathrm{Direction}\left(\mathrm{g}\right) \)

The second metric was Pearson’s correlation coefficient between expression levels of 38 genes with their expression direction, 1 for upregulated and − 1 for downregulated genes.

Both metrics were calculated for each patient then both metrics from all patients were examined against their subtypes.

Normalized log2 expression data from TCGA breast cancer cohort were rank-ordered by fold change between any two subtypes. The signature comprised two sets, one for 14 upregulated genes and the other one for 24 downregulated genes. Genes in the expression data were not collapsed but directly compared with the signature. Gene signatures were tested using default enrichment GSEA statistics with 1000 random permutations.

For relapse-free survival on patients with aggregated microarray data, kmplot.​com was used [5]. “Auto select best cutoff” was checked, and “only JetSet best probe set” was used to choose probeset for each queried gene. For the analysis on panel of genes, the weighted mean expression of all probeset was used to split patients.

For survival analysis of each individual gene, analyses were conducted on all patients and patients stratified by subtypes, in both TCGA RNASeq cohort and kmplot.​com microarray cohort. For analysis on a panel of genes, analyses were conducted on all patients.

The age information in unit of year of all donors in the GTEx database (www.​gtexportal.​org/​home/​) was applied and downloaded via dbGaP-controlled access. Among them, RNAseq gene counts data from 90 donors with normal breast tissue donation were extracted. This expression data was then normalized by library sizes, log2 transformed, and further normalized by quantile normalization. The 38 ABC genes identified from TCGA dataset were then queried in the normalized GTEx data, by studying the correlation between the gene expression and age. Multi-testing was addressed by Benjamini-Hochberg correction.

The human breast cancer cell lines (MDA-MB-231, MDA-MB-468, ZR-75-1, SK-BR-3, Hs 578T, and BT-474) were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA), and MCF-7 cells were obtained from the Michigan Cancer Foundation in 1994. They were cultured in McCoy’s 5A medium supplemented with sodium pyruvate, amino acids, vitamins, penicillin, streptomycin, sodium bicarbonate, and 10% fetal bovine serum (FBS) in a humidified atmosphere containing 5% CO₂ at 37 °C as previously described. These cells were authenticated in late 2016 with short tandem repeat assay by the DNA Lab in our university. They were routinely tested to be mycoplasma-free at least once a year with a luminescence-based mycoplasma detection kit from Lonza (Cat. # LT07-118).

Three siRNAs per gene were purchased from Sigma-Aldrich. A scrambled siRNA (Sigma-Aldrich) was used as a negative control. Dried siRNAs were reconstituted with DEPC-treated water (10 μM Stock). Lipofectamine RNAiMAX reagent (Invitrogen) was diluted by adding 0.3 to 25 μl basic medium in each well. siRNAs (0.75 μl at 10 μM) were diluted in 25 μl basic medium and then mixed with suspended cells to reach 50 nM working concentration. Diluted RNAiMAX and siRNAs were incubated for 20 min at 37 °C. Cell suspension in 100 μl containing 3000 MDA-MB-231 cells, 3000 MDA-MB-468 cells, 7000 MCF7 cells, 5000 ZR-75-1 cells, 7000 SK-BR-3 cells, or 9000 Hs 578T cells in culture medium with 15% FBS were added to pre-incubated transfection reagent. Cell proliferation was monitored, and confluency was analyzed every 3 h in IncuCyte for 1 week.

MISSION® TRC2 pLKO.5-puro plasmids containing shRNAs against DYNLT3 and P4HA3 and MISSION® TRC2 pLKO.5-puro Empty Vector Control Plasmid were purchased from Sigma-Aldrich. MISSION® TRC2 pLKO.5-puro Empty Vector Control Plasmid was used as a control for these knockdowns. Plasmids harboring full-length CDS of ALX4 (HsCD00294887) and WDR86 (HsCD00436836) were purchased from DNASU Plasmid Repository (dnasu.​org). Full-length CDS of ALX4 and WDR86 were then cloned by PCR with primers. The plasmid pWPI (addgene: 12254) was inserted at PmeI site with linker harboring BamHI, MluI, and NdeI sites, denoted as pLenti-Control. The cloned CDS of ALX4 and WDR86 were then inserted into plasmid pLenti-Control linearized by BamHI and PmeI, to construct pLenti-ALX4 and pLenti-WDR86 correspondingly. The plasmid pLenti-Control was used as a control for the overexpression.

293T packaging cells were transfected lentiviral vectors and packing vectors to generate lentiviral shRNAs (DYNLT3, P4HA3, and control) or CDS (ALX4, WDR86, and control) to infect MDA-MB-231 and BT-474 cells. Knockdown and overexpression were then examined with quantitative RT-PCR and Western blot analyses.

Cells were lysed with QiaGen RLT Lysis buffer, and total RNAs were extracted using RNeasy Mini Kit (QiaGen: Cat No./ID: 74104) following the manufacturer’s protocol. cDNA was generated using random primers and M-MLV reverse transcriptase from Invitrogen (Grand Island, NY), and real-time PCR was performed using SYBR reagent (Invitrogen) as described before [6].

Whole cell lysates were obtained in Laemmli buffer containing protease inhibitors and processed as described previously [6]. Antibody to DYNLT3 (Abcam, catalog # ab121209) was diluted 1:200 in 3% milk, to P4HA3 (Abcam, catalog # ab101657) was diluted 1:500 in 3% milk, to ALX4 (Biologicals, catalog # NBP2-45490) was diluted 1:1000 in 3% milk, to WDR86 (Thermo Fisher, catalog # PA5-48389) was diluted 1:1000 in 3% milk, to GAPDH (Calbiochem, catalog # 80602-840) was diluted 1:2500 in 3% BSA, and to tubulin (Sigma, Catalog # T4026) was diluted 1:5000 in TBST.

Knockdown of DYNLT3 and P4HA3 or overexpression of ALX4 and WDR86 in MDA-MB-231 and BT-474, and corresponding knockdown or overexpression control cells were plated in at least triplicate wells at 2000 MDA-MB-231 cells or 5000 BT-474 cells per well into 96-well plates. In the following 1 week, MTT solution (50 μl, 2 mg/ml in PBS) was added to each well and allowed to incubate for 2 to 4 h in a cell culture incubator. After removing the culture medium, 100 μl DMSO was then added into each well with the plates gently agitated on a shaker. The absorbance was measured at 595 nm with a microplate reader (BioTek Instrument, Winooski, VT), which is proportional to the number of viable cells.

Cells were seeded at a density of 20,000 cells per well for MDA-MB-231 and 80,000 cells per well for BT-474 in serum-free medium in the upper 24-well Boyden chambers with 8-μm-size inserts (BD Biosciences, Durham, NC, USA). A serum-containing medium was added to the lower chamber. After allowing the cells to migrate for 16 to 18 h in cell culture incubator at 37 °C, the cells on the top of the membrane were removed and the migrated cells were fixed and stained with HEMA 3 STAIN SET from the Fisher Scientific Company. Picture of the stained cells on the insert was captured under a microscope at × 4 magnification for counting [7].

Soft agar assays were carried out as described previously [7]. Growth of cells in soft agar was determined by plating 10,000 cells in triplicate in 0.3% Noble agar in 6-well tissue culture plates. Three weeks after plating, soft agar plates were stained and counted as described previously [7].

Animal experiments were conducted following appropriate guidelines. They were approved by the Institutional Animal Care and Use Committee and monitored by the Department of Laboratory Animal Resources at the University of Texas Health Science Center at San Antonio. Five million exponentially growing luciferase- and GFP-expressing BT-474 control or DYNLT3 knockdown, or P4HA3 knockdown cells in 0.1 ml volume were injected into the left inguinal mammary fat pad of ten 6-week-old female NOD scid gamma (NSG) mice purchased from The Jackson Laboratory. Similarly, nine 6-week-old female NSG mice were inoculated with 5 × 10⁶ exponentially growing luciferase- and GFP-expressing BT-474 control or ALX4 OE, or WDR86 OE cells. About 10 days after inoculation, tumor size was recorded twice a week. Tumor volumes were calculated as tumor volume (mm³) = length × width × width/2. For the detection and quantification of lung metastasis biofluorescence imaging with IVIS, mice were euthanized at the termination of the experiment. d-luciferin (200 μl at 15 mg/ml) was injected intraperitoneally. Ten minutes later, the lung tissues from each mouse were excised and transferred to wells of a 24-well plate and were immediately imaged to obtain photo flux. Living image software was used for analysis. Tumor tissues were then excised, weighted, and cut into pieces to be frozen at − 80 °C or fixed in 10% formalin for various analyses.

To identify genes with age-dependent expression, we extracted whole transcriptome profiling data of tumor-matched adjacent normal tissues from 82 female patients with age at diagnosis and menopausal status available in TCGA cohort (Supplementary Figure 1A). We applied simple linear regression to study the association between the gene expression level and age at diagnosis on the matched adjacent normal samples from all 82 patients, and from 54 post-menopausal patients to reduce the leverage effect of pre-menopausal patients with limited age range. We identified 210 upregulated and 98 downregulated genes in all patients (Supplementary Figure 1B), as well as 103 upregulated and 164 downregulated genes in post-menopausal patients only with the selection criteria defined in the “Materials and methods” section (Supplementary Figure 1C). The unions of these genes (258 upregulated (Supplementary Figure 1D) and 240 downregulated (Supplementary Figure 1E)) were considered as genes affected by age. Many of these genes were also influenced by menopausal status. Therefore, by comparing post-menopausal to pre-menopausal patients, we identified 493 upregulated (Supplementary Figure 1F) and 254 downregulated (Supplementary Figure 1G) genes altered by menopause. Exclusion of these menopause affected genes from those genes affected by age (258 upregulated and 240 downregulated) resulted in 148 upregulated (Supplementary Figure 1H) and 189 downregulated (Supplementary Figure 1I) genes during aging. Next, by comparing the matched tumors from the 82 patients against their matched adjacent normal samples, we identified 3356 upregulated (Supplementary Figure 1J) and 3124 downregulated (Supplementary Figure 1K) genes in breast cancer. The overlapping between these breast cancer-associated genes and age-associated genes resulted in 14 upregulated ABC genes (Supplementary Figure 1L) and 24 downregulated ABC genes (Supplementary Figure 1M), which are listed in Supplementary Table 1.

These 38 ABC genes clearly showed an age-dependent expression pattern in the 82 adjacent normal samples (Fig. 1a), while such pattern was not observed in those corresponding matched tumor samples (Supplementary Figure 2). Interestingly, most of these 38 genes were not altered in other cancers. Only CTHRC1 and LPAR5 (Fig. 1b) were upregulated and MAGI1 (Fig. 1c) was downregulated in lung adenocarcinoma, thyroid carcinoma, and kidney and renal papillary cell carcinoma in addition to invasive breast cancer when these genes were queried from TCGA database. Importantly, both the higher average expression of the 14 upregulated ABC genes (Fig. 1d) and the lower average expression of the 24 downregulated ABC genes (Fig. 1e) were significantly predictive for poor relapse-free survival in all breast cancer patients from KM plotter dataset [5].

Because the expression data in the TCGA dataset were derived from bulk tissue samples and are not cell type-specific, we queried the 38 ABC genes in The Human Protein Atlas (https://​www.​proteinatlas.​org/​) for their cell type-specific expression information in breast tissue based on immunohistochemistry (IHC). Interestingly, while eight genes were not detectable by IHC and ten genes have no IHC data, the expression of the remaining twenty genes is either only detectable or higher in breast epithelial cells in comparison with adipocytes suggesting that the 38 ABC genes are mostly expressed by the breast epithelial cells. We next examined the expression pattern of the 38 ABC genes among different subtypes of breast cancer. We developed two metrics: (1) the sum of these 38 ABC gene expression values (Fig. 1f) and (2) the Pearson correlation (Fig. 1g) of these 38 ABC gene expression values based on whether they were up- or downregulated. Both metrics indicated that luminal A and B and Her2 subtypes had higher expressions of the 14 upregulated genes and lower expressions of the 24 downregulated genes compared to triple-negative and normal-like subtypes. This observation is also reflected in the heatmap shown in Fig. 1h. Consistently, the signatures of 14 upregulated genes and 24 downregulated genes were more enriched in luminal A and B and Her2 than basal and normal-like subtypes (Fig. 1I). Strikingly, the majority of these 38 genes did not exhibit an increased number of genetic alterations except CTHRC1 and ETV3L (Supplementary Figure 3, generated from cBioPortal).

To study whether these ABC genes are promoting or suppressing cancer progression, we developed a scheme to prioritize these genes to a few for experimental validation. Firstly, we conducted a large-scale siRNA screening on the 14 upregulated genes. For each upregulated gene, three siRNAs were transfected into six breast cancer cell lines representing all subtypes of breast cancer (basal, MDA-MB-468; luminal A, MCF7; luminal B, ZR-75-1; Her2, SK-BR-3; claudin-low, Hs578T). From this screening, we found that knockdowns of CLEC3A, CTHRC1, RNASE2, LPAR5, and LRRC15 inhibited the proliferation of all tested cell lines by at least one siRNA (Supplementary Figure 4). Knockdown of DYNLT3 was more effective in inhibiting cell proliferation of MCF7 and SK-BR-3 than ZR-75-1 and Hs578T with little effect on MDA-MB-468. Knockdown of P4HA3 showed a more suppressive effect on MDA-MB-468 and SK-BR-3 than on ZR-75-1 and Hs 578T with a limited effect on MCF7. Knockdown of other upregulated ABC genes including PRR13, EGLN3, HSD11B2, and MATN3 also inhibited the proliferation of various breast cancer cell lines (data not shown). Secondly, to validate the age-association of these genes, we conducted correlation tests between these gene expression and age using 90 samples from the GTEx cohort. We found that 22, 16, and 4 out of the 38 ABC genes showed FDR below 0.5, 0.3, and 0.1, respectively (Supplementary Table 2). The four genes with FDR below 0.1 were CLEC3A, DLK1, DYNLT3, and ETV3L. Because of the large variation in human data and our limited number of samples, we decided to use the cutoff of FDR below 0.5 in combination with other filters discussed here for selecting ABC genes for further study. Thirdly, we examined the predictive prognostic power of individual genes in relapse-free survival of all breast cancer together as well as each subtype separately in a large cohort of breast cancer patients [5]. We found that RNASE2, ARRDC2, PRX, KCNQ4, and DLK1 were significantly predictive in relapse-free survival of all breast cancer patients and in all subtypes whereas EGLN3, DYNLT3, TUBGCP6, HOXA6, and DCHS2 were predictive in all patients, and all subtypes except for those of HER2 subtype, and ENPP6, WDR86, ALX4, CLEC3A, P4HA3, PRR13, CTHRC1, and OTUD7A were predictive in all patients and in at least one subtype of breast cancer patients (data not shown). Using the above filters and reviewing the literature, we selected two upregulated (DYNLT3 and P4HA3, Fig. 2a) and two downregulated (ALX4 and WDR86, Fig. 3a) ABC genes for further functional studies because (1) their functional role in cancer was largely unknown while their expression is significantly upregulated in tumors (Fig. 2c, g) or downregulated (Fig. 3c, g); (2) knockdown of DYNLT3 and P4HA3 by siRNAs inhibited breast cancer cell proliferation (Supplementary Figure 4 and data below); (3) their age-dependent expression pattern was validated in the GTEx dataset (Figs. 2b, f and 3b, f); and (4) they were significantly predictive for relapse-free survival in all breast cancer patients or in patients with certain subtypes of breast cancer (Figs. 2b, h and 3d, h).

To test the potential tumor-promoting roles of the two upregulated ABC genes DYNLT3 and P4HA3, we knocked down their expression in the basal-like breast cancer MDA-MB-231 cells and the luminal B breast cancer BT-474 cells. We chose the BT-474 cell line to represent luminal B subtype model because the 38 ABC genes were significantly enriched in the luminal B subtype. We also included MDA-MB-231 to contrast with BT-474. Both cell lines have tumorigenesis capacity in immune-deficient mice, which allows us to study their potential tumor-promoting role in vivo. Firstly, we confirmed the knockdown of these two genes in the two cell lines by Western blot (Supplementary Figure 5A-D). Next, we examined how their knockdown affected the malignant properties of the two cancer cell lines in vitro. We found that knockdown of either DYNLT3 or P4HA3 significantly reduced the tumor malignant properties in both cell lines, in terms of cell proliferation (Fig. 4a, b), migration (Fig. 4c, d), and anchorage-independent colony formation efficiency in soft agar (Fig. 4e, f). These data indicate that these two upregulated ABC genes are necessary for the malignant properties of the breast cancer cells. It is notable that the extent to which the knockdown of either gene inhibited proliferation of MDA-MB-231 was more moderate than that of BT-474. This could possibly be due to MDA-MB-231 being a faster-growing cell line than BT-474. Therefore, the moderate inhibition in MDA-MB-231 upon knockdown of either DYNLT3 or P4HA3 could be as large as in BT-474. Clearly, further studies are needed to elucidate the differences and similarities of the functions of the two genes in different subtypes of breast cancer cells.

We next examined the roles of DYNLT3 and P4HA3 in tumorigenesis in vivo. The control and DYNLT3 knockdown BT-474 cells expressing ectopic luciferase (Luc) and green fluorescent protein (GFP) were injected subcutaneously into female NOD scid gamma (NSG) mice bilaterally in the flank with 10 mice in each group. In contrast to the significant reduction of the malignant properties in vitro, we observed reduced tumor growth of DYNLT3 knockdown cells but with no statistical significance in comparison with the tumors formed the control cells in terms of tumor volume (Fig. 5a) and tumor weight (Fig. 5b). Consistently, the knockdown of DYNLT3 showed no effect on the extent of lung metastasis as measured with bioluminescence imaging and expressed as the total light flux of luciferase activity in the excised whole lung (Fig. 5c, Supplementary Figure 6). We confirmed the constitutive knockdown of DYNLT3 in the xenograft tumors (Supplementary Figure 7). This appears to indicate that the reduced malignancy due to DYNLT3 knockdown in vitro was somehow compensated under the in vivo condition. Next, to examine whether the reduced in vitro malignancy due to P4HA3 knockdown can be similarly compensated in vivo, we injected the control and P4HA3 knockdown BT-474/Luc-GFP cells subcutaneously in female NSG mice bilaterally in the flank with 10 mice in each group. In contrast to the knockdown of DYNLT3, the knockdown of P4HA3 significantly inhibited tumor progression and metastasis in vivo in terms of tumor volume (Fig. 5d), tumor weight (Fig. 5e), and total flux of luciferase activity in the lungs (Fig. 5f, Supplementary Figure 8). Similarly, we confirmed the constitutive knockdown of P4HA3 in the xenograft tumors (Supplementary Figure 9). Thus, the deficiency of P4HA3, which functions mainly as a mediator of collagen synthesis [8], consistently reduced tumor malignancy both in vitro and in vivo, suggesting that collagen synthesis pathways and extracellular matrix modification is an important regulator of tumor progression.

To test the function of the two downregulated ABC genes, ALX4 and WDR86, we ectopically introduced their cDNA via a lentiviral vector to increase their expression in the breast cancer cell lines. We found that overexpression of ALX4 moderately, but significantly, inhibited the growth of BT-474 cells (Fig. 6a), but not the growth of MDA-MB-231 cells (Fig. 6b). Nevertheless, the overexpression of ALX4 was found to significantly inhibit migration (Fig. 6c, d) and anchorage-independent colony-forming capacity (Fig. 6e, f) of both BT-474 and MDA-MB-231 cells. Conversely, knockdown of ALX4 in BT-474 cells with a siRNA significantly promoted cell migration in comparison with a control siRNA-transfected BT-474 cells (Supplementary Figure 10). On the other side, overexpression of WDR86 had no effect on these malignant properties of either cell line (Fig. 6a–f). These results suggest that ALX4, but not WDR86, plays a tumor-suppressive role in breast cancer cells, consistent with the reports that ALX4 is downregulated in various cancers [9‐11].

To confirm our findings from the in vitro assays, we next examined the effect of overexpression of ALX4 or WDR86 on breast cancer cell growth in a xenograft model in vivo. The control, ALX4-overexpressing, and WDR86-overexpressing BT-474/Luc-GFP cells were injected into NSG mice bilaterally with 9 mice in each group. We found that overexpression of ALX4 significantly inhibited the growth of tumors formed by BT-474 in vivo in terms of both tumor volume (Fig. 6g) and weight (Fig. 6h), supporting our conclusion that ALX4 is tumor-suppressive. In contrast, overexpression of WDR86 did not inhibit tumor progression (Fig. 6g, h), which is consistent with the in vitro data.