Ethics statement

The research involving human participants was approved by the MD Anderson Institutional Review Board. All patients signed informed consent prior to participation (IRB protocol # LAB07-0950). All animal studies were approved by the MD Anderson Animal Care and Use Committee (ACUF protocol # 02-08-00833). This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health.

Patient population

Patients with histologically-confirmed invasive breast cancer were recruited for this study. All patients had greater than 1 cm tumor on physical exam or imaging prior to surgery. Fifty-five patients were consented and underwent breast-conserving surgery or mastectomy for local therapy with sentinel node biopsy with or without axillary lymph node dissection (ALND) for regional therapy; one patient had contralateral lymph node recurrence and had an ALND alone; one had bilateral breast cancer. Three patients underwent surgery in the presence of known distant metastases.

From the first fifty-five breast cancer patients enrolled in the study, eight patients were excluded based on lack of adequate tumor on gross pathology assessment, or hematoxylin and eosin (H&E) staining showed no viable tumor tissue in the donor specimen. Therefore, 48 viable tumors were implanted from 47 patients (one with bilateral breast cancer).

In vivo tumor implantation and histopathologic analysis

The tumor specimen was placed in sterile tissue-culture medium on ice and brought immediately to the animal facility. Tumor was implanted into five 6-week old female BALB/c nu/nu mice under isoflurane anesthesia, and all efforts were made to minimize suffering. Two ~0.3 cm skin incisions were made with subcutaneous pockets on the left and right mid-back. One tumor piece (~1–3 mm) was inserted into each pocket and the skin was closed. When tumor diameter reached 1.5 cm, mice were euthanized, tumors were excised, cut into ~1x1x1 mm fragments, and passaged to successive generations of 5 mice. At each passage, remaining tumor was snap frozen in liquid nitrogen and stored at –80°C. Samples of these tumors were fixed in formalin and embedded in paraffin blocks. All primary tumors and BCX samples were assessed by H&E staining for morphology. In selected samples immunohistochemistry for CD20, CD3, CD45, and cytokeratin was performed in the Department of Veterinary Medicine and Surgery and Department of Pathology.

In vitro growth of conditionally reprogrammed cells and reimplantation

A 1.5 cm diameter BCX-010 xenograft was excised from the mouse and finely minced in F-media as described in Liu et al. [23]. Tumor fragments were digested for 4 hr at 37°C in F-media using collagenase/hyaluronidase (Stemcell Technologies; 1X) with occasional light mixing. The slurry was filtered, pelleted, washed, and plated in F-media supplemented with 10 μM Y-27632 (Enzo). The media was changed after 24 hours. Once the culture reached ~70–90% confluence, the cells were split at a ~1:4 ratio. After 4 passages, the cells were collected and 1x10⁸ cells were subcutaneously injected with or without 50% Matrigel into the flank of athymic nude (MD Anderson) or CIEA NOG (Taconic) mice. STR fingerprinting was used to confirm the human/BCX-010 origin of the cultured cells and the single cell-based xenografts.

DNA isolation and DNA fingerprinting

Genomic DNA was isolated from frozen or FFPE tumor tissue using the QIAamp DNA Mini Kit (Qiagen). Short tandem repeat (STR) DNA fingerprinting was performed with the AmpFℓSTR Identifiler kit (Applied Biosystems) at the Characterized Cell Line Core Facility.

Polymerase chain reaction

Genomic DNA was amplified using Herculase II DNA polymerase (Agilent). Primer sequences were: Zfp42 (mouse) (5’-TGAGATTAGCCCCGAGACTGAG-3’ and 5’-CGTCCCCTTTGTCATGTACTCC-3’), HBB (5’-CCTGAGGAGAAGTCTGCCGTTA-3’ and 5’-GAACCTCTGGGTCCAAGGGTAG) [24], PTEN (5’-GGAATCCAGTGTTTCTTTTAAATACC-3’ and 5’-TCCAGGAAGAGGAAAGGAAAA-3’), and RB1 (5’-TCCCATGGATTCTGAATGTG-3’ and 5’-CGTTGTGCACATGTACCCTAGA-3’).

Fluorescence in situ hybridization

FISH was performed at the Molecular Cytogenetics Core Facility. The all mouse centromere probe (Kreatech, Netherlands) or the human centromere 16 probe (Abbott Molecular) was applied and slides were stained with DAPI.

Copy number and somatic mutation determination

Genomic DNA from five of the patients (1, 2, 4, 5, and 6) and corresponding P3 BCXs were submitted to Affymetrix for molecular inversion probe (MIP) assay with copy number and somatic mutation determination. The Affymetrix platform interrogated 300,008 single nucleotide polymorphisms and 412 somatic mutations in 45 genes. The MIP assay was performed as previously described [25].

Single nucleotide polymorphism (SNP) genotyping was done at the Genomics Core Facility. Hot spot mutation testing was performed in 46 cancer related genes including PIK3CA on pre-treatment and on-treatment biopsies (at 12 weeks of chemotherapy) using the Ion Ampliseq Cancer Panel (Life Technologies) to assess hotspot mutations as previously described [26]

Next generation targeted exome sequencing was performed as previously described [27, 28]. For genomic profiling, DNA from mouse normal tissue, and six patients’ normal tissue or normal blood, primary tumors and corresponding BCXs: passage 0 (P0), P1, P4, P7, P10, P13, and P16 of BCX-006, -010, -011, and P0 and P1 of BCX-017, -022, -024 were evaluated as well as P6 of BCX-022, and P7 of BCX-024 [13, 15, 24–28].

Briefly library prep was performed using KAPA library prep kit (Kapa Biosystems, Inc), and equimolar amounts of DNA were pooled (8–12 samples per pool) for capture of 201 genes that are clinically relevant in cancer (S1 Table). The captured libraries were sequenced on a HiSeq 2000 (Illumina Inc.). In addition, whole exome sequencing was performed for normal tissue or blood, primary tumors and corresponding P0 and P1 for BCX-010, -017, -022, and -024. For data analysis, reads were aligned to human reference assembly hg19 using BWA software and duplicated reads were removed using samtools. Single nucleotide variants (SNVs) and small indels were called using VarScan2 [29].

Functional proteomics

Reverse phase protein array (RPPA) was performed as previously described [30] to compare the proteomic profile of P0 patient tumors and corresponding BCXs (P1, P2 and P3), in models BCX-001, 002, 004, 005, and 006. A logarithmic value reflecting the relative amount of each protein in each sample was generated. The median polish normalized RPPA data set consisted of 154 proteins.

Statistical analysis

Descriptive statistics were used to describe engraftment rates. BCX success rates in different tumor types and clinical scenarios were compared using Fisher’s Exact 2-tailed analysis. Linear mixed effects (LME) model was used to assess the differences in passage time among passage numbers. The LME model included fixed effect of passage number (1 vs. 2 vs. 3) and random effect of model.

For RPPA, unsupervised clustering was applied to all tumor samples across all of the proteins, using Spearman Correlation and complete linkage. Paired t-test was performed to compare changes in relative protein amounts between, P0, P1, and P3. The p values were fitted to a-beta-uniform mixture-BUM model to calculate the corresponding false discovery rate (FDR) [31–33]. FDR <0.2 was considered significant [34]. To compare protein expression levels across multiple generations, a g-statistic was used [35]. R packages including GeneCycle, q value and ClassComparison were employed [34].

Ability to generate patient-derived breast cancer xenografts is enhanced in TNBC and in chemoresistant disease

Forty-eight viable tumors were implanted from 47 patients: one patient had a bilateral tumor implanted as separate models. Implants were followed for 6 months after implantation. The resultant BCXs were designated MDA-BCX-001 to -055 (MD Anderson Breast Cancer Xenograft 001 to 055). At the time of diagnosis, the median age of patients was 55 years (range, 29–86). At presentation, 32 tumors were hormone receptor-positive (HR+), 13 were TNBC, and 3 were HER2-positive (HER2+). At presentation, average tumor size (± standard deviation) was 4.3±3.1 cm (range, 0.8–15). Twenty-four of the tumors (50%), came from patients who received NeoCT. Of the 23 patients who had NeoCT with intact primary tumors, 7 had progressive disease, 7 had a partial response, and 9 had stable disease. One patient had an axillary cross-metastasis and that responded to therapy.

Tumor from the surgical specimens (designated as “P0”) was implanted (Fig 1A) and mice with tumor growth at the site of implantation were considered to have successful engraftment. Tumors derived from 13 of the 48 specimens (27%) were successfully engrafted.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Generating and maintaining patient-derived breast cancer xenografts (BCXs) and their time to passage.

(A) After surgery, patient tumors (P0) were implanted into nude mice, creating a patient-derived BCX, passage 1 (P1). When tumors reached 1.5 cm diameter, they were harvested and implanted into 5 new mice (P2), and subsequent passages respectively. Patient tumors and BCXs were evaluated by STR and selected passages underwent molecular and histologic characterization. (B) Y-axis depicts time to reach 1.5 cm with each passage. The graph shows thirteen BCXs that were serially passage. (C) Time to passage (in months) at from implantation to first passage (P1), P1 to P2 (P2), and P1 to P3 (P3). Time to passage at P2 and P3 were compared to time to passage at from implantation to first passage (P1).

https://doi.org/10.1371/journal.pone.0136851.g001

BCXs that reached 1.5 cm in size were passaged to another set of mice (P2; Fig 1A). Time to tumor size of 1.5 cm diameter (i.e. passage from P1 to P2) was an average of 2.6 months (range 0.8–7 months; median 1.9 months) (Fig 1B). P1 tumors were passaged to P2 generation and then were transferred into the third passage (P3) in all 13 models. When the time from implantation to first passage (P1) was compared to the time from first passage to second passage (P1-P2 interval) or to the time from second passage to third passage (P2-P3 interval), was significantly longer than both latter passages (p<0.001 for both; Fig 1C). Table 1 demonstrates comparison of patient and tumor characteristics and success of BCX engraftment. BCX engraftment was higher in TNBC compared to HR+ tumors (7 of 13, 53.8% vs. 5 of 32, 15.6%, p = 0.02). One of three HER2+ tumors which had received NeoCT engrafted.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Tumor characteristics and engraftment rate.

https://doi.org/10.1371/journal.pone.0136851.t001

Engraftment was higher for tumors that had been exposed to NeoCT (10 of 24, 41.7% vs. 2 of 24, 8.3%, p = 0.02). Further, of the 7 patients who had had progressive disease on NeoCT, tumor implantation from 6 patients (85.7%) developed passagable BCXs; in contrast, tumors from 5 (29.4%) of 17 patients who had stable disease or partial response to NeoCT lead to passagable BCXs (p = 0.02).

Of note, two TNBC BCX models were derived from tumors not classified as TNBC prior to chemotherapy. For the BCX-011 model, the pre-NeoCT core biopsy was ER+, but the post-NeoCT and the emerging BCX-011 P1 samples were ER-. In addition, the pre-NeoCT core biopsy was HER2+ and ER+ for BCX-017, but the surgical sample after neoadjuvant chemotherapy was ER+, HER2- and the emerging BCX-017 P1 sample was TNBC. Notably this latter patient received paclitaxel with concomitant trastuzumab followed by 5-FU, epirubicin, and cyclophosphamide with concomitant trastuzumab.

At a median follow-up of 25 months (range, 1.8–39.4) after surgery, 12 of 44 patients with unilateral Stage II or III disease at surgery developed distant metastasis; the patient with bilateral tumors was disease-free at last follow up. For 5 patients, passagable tumors were derived prior to their development of distant metastases (Table 2). Among the 44 patients with unilateral localized disease, patients whose tumors developed BCXs had significantly lower recurrence-free survival (p = 0.015; Fig 2A) and distant recurrence-free survival (p = 0.004; Fig 2B) compared to patients whose tumors did not develop BCXs. Three patients with Stage IV disease underwent surgery; BCXs developed from the tumors of two patients. When all 47 patients are considered, patients whose tumors developed BCXs had significantly lower overall survival (p<0.001, Fig 2C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Disease progression and engrafted BCXs.

https://doi.org/10.1371/journal.pone.0136851.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Survival Outcomes in Patients Based on BCX development.

(A) Recurrence-free survival (in months) in patients whose tumors developed BCX versus those did not (no BCX). (B) Distant recurrence-free survival (in months) in patients whose tumors developed BCX versus those did not (no BCX). (C) Overall survival (in months) in patients whose tumors developed BCX versus those did not (no BCX).

https://doi.org/10.1371/journal.pone.0136851.g002

DNA fingerprinting demonstrated donor origin of patient-derived xenografts

STR DNA fingerprinting was used on all models that were serially transplantable to evaluate specific regions (loci) within nuclear DNA, to confirm that the xenografts were patient-derived. The STR profiles of the 13 passagable BCX models were all unique, showing no matches to known DNA fingerprints in American Type Culture Collection, the Deutsche Sammlung von Mikroorganismen und Zellkulturen, Cell Line Integrated Molecular Authentication, MD Anderson Cancer Center databases or to each other. The P0 genomic DNA from each patient resulted in a match with the corresponding BCX specimens for these 13 models (BCX-001, 002, 004, 005, 006, 010, 011, 017, 022, 024, 042, 051, and 055).

Histopathologic and molecular characterization identified murine origin tumors in two mice

A breast pathologist and a veterinary pathologist compared H&E stained slides of surgical specimens and BCXs. For BCX models designated successful BCX models, no histologic differences were found. The tumor derived from Patient 3 took the longest time (8.4 months) to develop a 1.5 cm mass. STR characterization of P0 demonstrated a unique human tumor. However, STR from P1 showed lack of amplification with seven of eight STR primer sets and discordance with P0 in the only one primer set that was successfully amplified. Mouse implanted with tumor from Patient 8 established a tumor; time from P0 to P1 and P1 to P2 took 5.2 and 4.5 months, respectively. However, tumor growth stopped at P2. STR from P1 showed lack of amplification. Therefore, STR results raised concern that both BCX-003 and -008 series xenografts may not be of human origin.

Although STR characterization of P0, P1 and P3 confirmed that BCX-001 series was indeed derived from Patient 1, histologic evaluation of late passages of the BCX-001 revealed morphologically undifferentiated tumors that lost the characteristics of epithelial/breast tumor cells, raising concerns about the fidelity of the model.

To determine whether the BCXs were of human origin, we performed PCR using two sets of primers designed to amplify a target sequence in which human (HBB) and murine (Zfp42) DNA could be distinguished (S2A Fig) [24]. We tested serial passages collected from BCX-001 and BCX-003 models as well as BCX-004 as a control. As expected, all three patient tumors (P0) showed no amplification of murine DNA, but amplification was seen with human primers (S2A Fig). In BCX-004, the mouse sequence was detected in P1 and P3, but at significantly less intensity than the human DNA product; this may be attributable to murine stroma and leukocytes within the xenograft. In contrast, for BCX-003, PCR of DNA from P1 and P3 xenografts demonstrated only amplification with murine DNA primers, suggesting the tumors were solely of murine origin. For BCX-001, there predominantly was amplification of human DNA at early passages including P4, but at P5, there was an abrupt loss of human DNA and amplification of murine DNA exclusively. Interestingly, this corresponded to an increase in growth rate (S2B Fig). The other model, BCX-008 P0 and P1 were tested similarly. Human sequence was detected in P0, and only mouse sequence detected in P1 (data not shown).

When we examined H&E stained slides of BCX-001, passages 1 to 4, showed well differentiated epithelial tumors, whereas passages 5 to 8 showed tumors composed of dense neoplastic cells with smaller round nuclei and minimal basophilic cytoplasm suggestive of spontaneous murine lymphoma (S2C Fig). However, immunohistochemistry for CD20, CD3, CD45 and cytokeratin were negative on P5, suggesting this is an undifferentiated tumor of unknown tissue origin. FISH using mouse and human centromere probes demonstrated presence of human and mouse DNA in P1 but only mouse DNA in the later passage, P6 (S2D Fig). For BCX-003, the H&E stained slide of P1 showed areas of normal mouse mammary glands adjacent to mammary adenocarcinoma (S2E Fig). FISH revealed only mouse DNA in P1 (S2E Fig). This confirms a spontaneous mammary adenocarcinoma of mouse origin rather than human.

Functional proteomics demonstrated alterations in PI3K signaling

Unsupervised hierarchical clustering of RPPA was used to evaluate the functional proteomic profile of five of the ten patient’s surgical specimens and corresponding serial transplanted in vivo tumors. 135 unique proteins and phosphoproteins were evaluated; 19 were done in duplicate, 2 in triplicate as part of the quality control process. Upon unsupervised clustering, patient tumors clustered together, separate from the BCXs, suggesting that there are significant differences between the primary tumor and the BCXs (Fig 3A). However, each BCX lineage clustered together demonstrated that their proteomic profile remains relatively stable for several in vivo passages.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Molecular differences between patients’ tumors and BCXs.

(A) Unsupervised clustering of proteomic profile of patient tumors and BCXs as determined by RPPA. Each protein tested represents a column: Red, high expression; green, low expression. Samples are listed on the right side. Left, cluster trees of sample groups; top, cluster trees of proteins. Each BCX model’s P1-3 generation clustered together, demonstrating relative stability of the proteomic profile once growth in mouse is established. However, all P0 generations clustered together, suggesting that differences between patient tumor-xenograft proteomic profiles is greater than inter-tumoral differences. (B) Selected proteins and phosphoproteins that are differentially expressed between patient tumors (P0) and the first-generation of BCXs passaged through nude mice (P1). Protein levels were compared between the two groups with RPPA; all shown have a FDR 0.1 or less. (C) Copy number analysis determined PTEN loss in MDA-BCX-002. Top panel shows the chromosome 10 ideogram and the PTEN gene. Deletions are plotted in red below the 0% baseline, and dark red indicates homozygous loss. The lowest portion of the top panel separates out P0 and P3. A heterozygous PTEN loss is detected in P0, and in the P3, the second PTEN allele is lost, resulting in a homozygous PTEN loss. Bottom panel shows the PTEN gene ideogram followed by the copy number aberration plot and the allele frequency plot for P0 and P3. A heterozygous PTEN loss is detected in P0 (single red line), and a homozygous PTEN loss is detected in P3 (double red line). Each blue dot corresponds to an individual probe on the array. The brown and purple lines mark the thresholds for loss of heterozygosity (LOH) and allelic imbalance regions, respectively. (D) PCR confirmed PTEN deletion in genomic DNA from Patient 2 tumor and BCX-002 P3. PTEN was undetectable in P3 but present in P0. RB1, another tumor suppressor gene, is detected in both samples and included for comparison. (E) PTEN loss demonstrated by next generation sequencing in BCX-024. P0 on the left, P1 on the right.

https://doi.org/10.1371/journal.pone.0136851.g003

When protein expression was compared between the original breast tumors (P0) versus those from the first BCX passage (P1), levels of 103 of 154 proteins were differentially expressed at a FDR of 0.2 (S2 Table). Eight of these proteins are displayed in Fig 3B: EGFR, merlin, YB1 S102, Cyclin E1, 4E-BP1 T37/46, mTOR S2448, PRAS40 T246 and S6 S240/244. Importantly, a number of these proteins are part of the same PI3K pathway indicative either of increased pathway activity or alternatively of decreased human stromal content. All patient tumor-BCX differences remained significant when the P0 tumors were compared to both P1 and P2 and the first three transplants (P1, P2 and P3).

Copy number analysis determined loss of PTEN in two BCX models

MIP genotyping assays detect genetic alterations including candidate point mutations between specimens and are particularly well designed for copy number analysis [36]. A total of 412 somatic mutations in 45 genes were analyzed in genomic DNA extracted from the surgical specimens from patients 1, 2, 4, 5, and 6 and compared to corresponding BCXs. A heterozygous loss of an allele of PTEN was identified in the P0 tumor derived from patient 2. In the corresponding P3 of BCX-002, the second PTEN allele was lost, resulting in a homozygous PTEN loss (Fig 3C). Consistent with this finding, when PCR was performed with primers to human PTEN, PTEN genomic DNA was detected from P0 DNA but not from DNA extracted from P3 BCXs (Fig 3D). However, another interpretation could be that human stroma PTEN is present in P0 and that the loss of human stroma allows a homozygous PTEN loss that was present to begin with to be identified.

Targeted exome sequencing using a 201 gene platform was performed in six models: BCX-006, -010, -011, -017, -022, and -024 [27]. Copy number based on targeted exome sequencing demonstrated loss of PTEN in P1 the BCX-024 model (Fig 3E and S3 Table). Loss of PTEN in BCX-024 P1 was also confirmed by whole exome sequencing (S4 Table).

Genomic sequencing demonstrated relative genomic stability of somatic mutations in BCX models

Whole exome sequencing was performed in the mouse host DNA as well as the matched P0-P1—normal host DNA in 4 models: BCX-010, -017, -022, and -024 (S4 and S5 Tables). Unsupervised clustering clustered each BCX P1 with the P0 parental tumor (Fig 4A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Analysis of mutation data.

(A) Clustering of whole exome sequencing mutation data based on the mutation status of the genes in P0 and P1 samples of four models. (B) Clustering of targeted exome sequencing data based on the mutation status of 201 genes in P0 and P1 samples of six models. (C) Venn diagram of mutations in P0 and P1 samples of six models on the targeted exome sequencing platform. (D) Venn diagram of high allelic frequency mutations (10% or higher) on targeted exome sequencing in P0 and P1 samples of six models. Allele frequency cutoff was 10%. (E) Genomic stability of PIK3CA mutation in BCX-006 model. Fine needle aspiration biopsy samples of before and after 12 weeks of neoadjuvant chemotherapy (including a rapalog) were sequenced by ion torrent. Patient’s P0 tumor and BCX-006 P1 tumor and subsequent passages were analyzed by targeted exome sequencing. PIK3CA H1047R allele frequencies are presented. (F) Conditionally reprogrammed cells (CRC) derived from BCX-010 were passaged four times in vitro and then injected into mice flanks. The cultured cells and CRC-derived xenografts maintained a mutation profile similar to the originating PDX.

https://doi.org/10.1371/journal.pone.0136851.g004

As described above, targeted exome sequencing was performed in 6 models: BCX-006, -010, -011, -017, -022, and -024. The mutations detected are listed in S6 Table. Upon unsupervised clustering each BCX clustered with its parental tumor (Fig 4B). Of 28 mutations detected on targeted exome sequencing in P0 tumors, 19 (68%) were preserved in P1 tumors (Fig 4C). Of 21 high allelic frequency mutations (allelic mutation frequency 10% or higher) detected in P1, 18 (86%) were also detected in the parental tumors (Fig 4D). In five of the models, serial passages were sequenced. With serial passaging, mutations were preserved, including potential driver mutations such as PIK3CA mutations, but additional alterations were observed in some models. In the BCX-006 model, pretreatment and on-treatment biopsies obtained during NeoCT with a rapalog-containing combination therapy regimen were also available. These demonstrated that the PIK3CA mutation was detected at a relatively high allelic frequency in the pretreatment and 12 week samples as well as in the P0 sample, and was preserved upon multiple BCX passages (Fig 4E). Although the allelic frequency increased in BCX passages, this was accompanied by an increase in mutant allelic frequency of other mutations as well, suggesting that this may be attributable to increase in tumor cellularity (S5 Table). Thus there did not appear to be selection of the mutation during therapy.

Somatic mutation profile is retained with in vitro culture and implantation

An experimental challenge is experimentally manipulating the PDX cells for functional target validation. Thus we also explored the possibility of developing conditionally reprogrammed cells (CRC) from PDXs and, and the effect of serial transplantation and in vivo re-implantation using the rapidly growing BCX-010 model. We cultured and expanded cells from BCX-010 [23]. After 4 passages, we injected these BCX-010 cultured cells into two strains of immune-deficient mice (nu/nu and NOG) with and without Matrigel. Rapidly growing tumors developed in all cases. STR fingerprinting confirmed identity of the CRC and the CRC-derived xenograft. The cultured cells as well as the CRC-derived xenograft maintained a mutation profile very similar to the originating PDX (Fig 4F).