GENOMICS-BASED IDENTIFICATION AND CHARACTERIZATION

OF RARE CELL TYPES

[01] Each reference cited in this disclosure is incorporated herein in its entirety.

TECHNICAL FIELD

[02] This disclosure relates generally to the identification and characterization of rare cell types.

BACKGROUND

[03] Circulating Tumor Cells (CTC) have been reported in patients with a wide variety of cancer types and stages of disease, and their detection and analyses holds great potential as a non-invasive approach to guide the diagnosis and treatment of solid tumors in the clinic.

However, significant technical hurdles continue to impede the development and adoption of CTC tests in routine medical practice. There exists a need for methods that provide highly sensitive and accurate detection and characterization of CTCs, as well as other rare cell types, in a liquid sample.

BRIEF DESCRIPTION OF THE DRAWINGS

[04] Figure 1 shows mutation abundance in a subset of wells from the specimen spiked with MCF7 cells diluted 1 : 10,000 before arraying into a microplate. Each bar in the graph represents data from one well. The figure shows high mutation signal in Well #4, indicating the presence of one or more MCF7 cells.

[05] Figure 2 shows the correlation of RNA abundance in MCF7-positive Well #4 and input MCF7 cells that were not added to blood.

[06] Figure 3 is a graph showing mutation frequency in individual microwells from specimens processed by density gradient and by density gradient plus CD45 depletion.

[07] Figure 4 is a graph showing mutation frequency in individual microwells from blood and from blood spiked with 100 MCF7 cells and processed by density gradient separation plus CD45 depletion. [08] Figure 5A and Figure 5B are graphs showing RNA abundance of selected tumor markers (Figure 5A, EPCAM; Figure 5B, CDH1) in NGS libraries. Data are shown for libraries generated from blood only, MCF7 only and blood samples spiked with varying levels of MCF7. Each input sample consisted of total RNA and gDNA as described in Example 5.

[09] Figures 6A-F. Spearman correlation plots of log2 expression showing detection of MCF7 RNA signature spike-in samples with low tumor content. Figure 6A, MCF7 only. Figure 6B, 8% MCF7. Figure 6C, 4% MCF7. Figure 6D, 2% MCF7. Figure 6E, 1% MCF7. Figure 6F, 0% MCF7.

DETAILED DESCRIPTION

[10] This disclosure provides methods that address key technical bottlenecks in each of the three pillars of rare cell analysis: enrichment, identification, and classification. These methods are described below with respect to CTCs, but can be readily applied to other rare cell types, such as circulating epithelial cells (CECs), stem cells, progenitor cells, and rare immune cells ( e.g ., PD1 ⁺ CD8 ⁺ IFNy- TIM3 ⁺ LAG3 ⁺ positive T cells), circulating endothelial cells

(CECs), white blood cells in emboli, cancer stem cells, activated or infected cells (e.g., activated or infected blood cells), and fetal cells.

[11] The disclosed methods can be used in liquid biopsies of fluids (e.g, blood, cerebrospinal fluid, urine) for a variety of purposes. These include, but are not limited to, screening for and diagnosing disease, identifying an appropriate therapy (e.g, as a companion diagnostic), monitoring a response to a therapy, and detecting drug resistance.

1. Enrichment

[12] It is often useful to enrich a biological sample for the presence of a rare cell type. For example, cells originating from solid tumors are an extremely rare component of blood even in patients with late-stage metastatic disease. Five milliliters of whole blood typically contain 25 billion erythrocytes, 1.5 billion platelets and 25 million leukocytes. CTCs, in contrast, can be present at 10 or fewer cells in the same volume of blood. To address this problem, current approaches to CTC analysis employ strategies to increase tumor representation relative to non tumor background cells.

[13] Conventional enrichment approaches for CTCs rely on biological properties and/or physical properties of CTCs. For example, antibody-based positive selection of cells expressing Epithelial Cell Adhesion Molecule (EPCAM), a surface protein present on many solid tumor cells, is frequently used to enrich the number of CTCs in a sample. This approach can result in cellular fractions with very high tumor content (90% or greater). This comes at a high cost, however, because the expression of EPCAM and similar markers can vary widely within a given patient due to tumor heterogeneity and across different cancer types. Tumor cells often escape capture, which introduces biases that can lower assay sensitivity dramatically. Another drawback of positive selection is that these methods are frequently optimized for single cell enrichment and, as a result, fail to account for clusters of 2 or more tumor cells. Clusters represent a class of CTCs that is increasingly recognized as an important driver of tumor progression and metastases.

[14] Physical properties ( e.g ., tumor cell mass, tumor cell size, tumor cell shape) and negative selection of leukocyte-specific surface markers such as CD45, impart less bias during

enrichment, but they produce CTC fractions that contain much higher levels of non-tumor blood cells when compared to positive selection schemes, making downstream tumor identification and analysis technically problematic.

[15] Counterintuitively, in the disclosed methods, enrichment bias is mitigated by applying less stringency during the CTC enrichment. Because the disclosed methods do not depend on detecting one or two protein markers, CTCs— and CTC clusters— with more diverse molecular profiles and from more cancer types are recovered. Clusters can be distinguished from single CTCs microscopically. In embodiments of the disclosed methods, cell lysates from a biological sample are physically divided into individual wells of a microplate after enrichment. This permits analysis to be carried out in each well independently, which dramatically increases the signal-to-noise ratio in CTC-containing wells. For example, if a post-enrichment specimen containing one CTC and 10,000 non-tumor cells is dispensed into wells of a 96-well plate, there will be on average 104 total cells per well. Tumor representation, in this case, increases from 0.01% in the starting sample to 1% in a single CTC containing well.

2. Identification

[16] With conventional approaches, CTCs are distinguished from non-tumor cells by microscopic detection of a few protein markers using antibodies. Cells are commonly classified as CTCs if they are positive for EPCAM and cytokeratin proteins and negative for the leukocyte- specific protein CD45. As with enrichment, variations in expression due to tumor heterogeneity and cancer type limit the sensitivity and robustness of these markers for CTC identification. Moreover, they rely on subjective image analysis calls that can lead to misclassification of cells. These drawbacks can reduce the fidelity of assay performance and limit their application in the clinic. [17] The disclosed methods overcome these limitations by leveraging the exquisite specificity of cancer genomic alterations as identifying markers for CTCs; i.e., the methods use genomics for CTC identification. For example, cell samples ( e.g ., 100-200 total cells per well) arrayed in plates following an enrichment step can be assessed for one or more mutations in selected cancer driver genes using nucleic acid sequencing methods (e.g., next-generation sequencing, NGS). Wells in which a mutation is detected contain at least one CTC, while wells in which no mutations contain only non-tumor cells. Downstream CTC classification efforts can then be focused on mutation-positive CTC wells. See, for example, Example 3.

Construction of Sequencing Libraries

[18] In constructing an NGS sequencing library, a molecular barcode specific for each well is incorporated into genomic DNA constructs and/or cDNA constructs. The samples can then be pooled prior to sequencing because the identity of the barcode permits the identification of individual wells after sequencing. This allows CTC identification and CTC classification to be integrated during data generation. Moreover, by the choice of target sequences amplified by the primers, the content of the sequencing libraries is programmable. In some embodiments, a primer pool can comprise one or more primers specific for one or more biological markers (e.g, as a drug resistance marker, a tissue-specific marker, a drug response marker, a molecular subtyping marker).

[19] Sequencing libraries for use in the disclosed methods can be gDNA libraries, cDNA libraries, or libraries containing both gDNA and cDNA. In some embodiments the disclosed methods use a target RT-PCR sample preparation method in which gDNA and mRNA targets are co-amplified for analysis by NGS. Two types of primers provided in this disclosure are used: (1) universal NGS primers and (2) target-specific primers (TSPs).

[20] Lysed cells are reverse transcribed using an oligo d(T) primer to generate cDNA.

Multiplex PCR is then used to amplify a defined set of transcripts and gDNA targets using a pool of TSPs. Universal sites that facilitate NGS are added in two steps. First, a 15-16 nucleotide sequence representing a portion of the universal NGS site is added to the 5' terminus of oligo d(T) and TSPs during primer synthesis. These tail sequences are introduced into the library during the reverse transcription and multiplex PCR reactions. The 5' tails are then extended via a second round of PCR using primers containing the full-length universal NGS sequences. 3. Classification

[21] The scope and clinical utility of CTC profiling has been limited largely by the reliance on microscopy and high content imaging to characterize CTCs. Consequently, clinical tests have focused on simple metrics and biomarker readouts. CTC enumeration as a prognostic indicator or single protein markers to guide treatment in selected cancer types are a few examples of existing clinical applications.

[22] The disclosed methods use genomics to expand the size and complexity of CTC biomarkers to address a wide range of clinical applications. The ability of these methods to assess both DNA alterations and RNA abundance in a single assay enables applications such as tissue-of-origin classification, disease prognosis, tumor-subtyping to guide treatment, characterization of mechanisms of resistance, and monitoring of disease status and recurrence following treatment. The examples below demonstrate that the disclosed methods preserve the integrity of RNA abundance and mutational status when generated from low inputs (10-100 cells). Using the disclosed methods, RNA markers that are highly expressed in tumors and weakly expressed in leukocytes can be easily detected when tumor content is 1% or lower.

EXAMPLE 1. CTC Enrichment, Array Allocation, and Lysis

[23] Blood Collection. Whole blood was collected from into BCT tubes (Streck) and stored for 48 hours at room temperature until processing.

[24] Red Blood Cell (RBC) Lysis. Each blood sample was added to 40 mL of cold

Ammonium Chloride Solution (STEMCELL™ Technologies) in a 50 mL conical FALCON ^® tube, mixed by inverting the tube several times and incubated on ice for 10 minutes. Each tube was then centrifuged at 800 relative centrifugal force (RCF) for 10 minutes in a fixed rotor centrifuge at room temperature. Supernatant was decanted, and pellets were resuspended in 15 mL of Dulbecco’s Phosphate-Buffered Saline (DPBS) + 2% fetal bovine serum (FBS) before centrifuging at 200 RCF for 10 minutes. Cells were washed a second time and resuspended in 5 mL of DPBS + 2% FBS.

[25] Density Gradient Separation. Five mL of cell resuspension from the RBC lysis procedure were added to 5 mL of room temperature Wash Buffer (pluriSelect Life Science). Cell mixture was carefully layered on top of 3 mL of LYMPHOPREP™ density gradient medium (STEMCELL™ Technologies) and centrifuged at 800 RCF for 15 minutes at room temperature. The 9 mL top layer was carefully removed by pipetting and discarded. Ten mL of Wash Buffer was mixed with the remaining 4 mL bottom layer and centrifuged for 10 minutes at 300 RCF at room temperature. The supernatant was decanted, and the cell pellet was resuspended in 10 mL of Wash Buffer. The sample was centrifuged for 10 minutes at 300 RCF at room temperature, supernatant was decanted, and the cell pellet was resuspended in 1.0 mL of DPBS + 2% FBS. Cells were counted using a hemocytometer.

[26] CD45 Depletion. The 1.0 mL of cell mixture resulting from density gradient

centrifugation was centrifuged at 300 CFS for 10 minutes at room temperature. Cell pellet was reconstituted in 80 pL MACS ^® Buffer (Miltenyi Biotec), mixed with 20 pL of CD45

MicroBeads (Miltenyi Biotec) and incubated for 15 minutes on ice. The sample was mixed with 1.0 mL of cold MACS ^® Buffer and centrifuged for 10 minutes at 300 CFS at 4°C. The supernatant was completely removed and discarded. The cell pellet was resuspended in 500 pL of MACS ^® Buffer. An LS Column (Miltenyi Biotec) was placed in a MIDIMACS ^® Separator (Miltenyi Biotec) and washed with 3 mL of MACS ^® Buffer. Flowthrough was discarded. The 500 pL of cell suspension was added to the column. The column was washed 3X with 3 mL of MACS ^® Buffer. The 9.5 mLs of combined effluent containing the desired CD45-negative cells was centrifuged for 10 minutes at 300 CFS at room temperature. Supernatant was discarded and the cell pellet was resuspended in 1.0 mL of DPBS + 2% FBS each. The column containing the magnetically captured CD45-positive cells was discarded.

[27] Array Allocation and Cell Lysis. Ten pL aliquots of the CD45-negative cell mixture were placed into each well of a 96-well microplate. Plates were centrifuged at 800 RCF for 10 minutes at room temperature to pellet cells. The supernatant was removed, and the cell pellets were resuspended in 5 pL of CELLS-TO-SIGNAL™ Lysis Buffer (AMBION ^® ), mixed by pipetting, and incubated at room temperature for 5 minutes. Lysates were frozen at -20°C until library construction.

Example 2. Library Construction

[28] This example describes preparation of a library for NGS sequencing from cell lysates prepared as described in Example 1.

[29] Reverse Transcription. Whole transcriptome cDNA was synthesized from cell lysates in each well by oligo dT priming in 20 pL reverse transcription reactions. Each reaction contained 20 units of SUPERSCRIPT™ IV Reverse Transcriptase (Life Technologies), IX

SUPERSCRIPT™ IV buffer, 5 mM DTT, 0.5 mM dNTP, and 2.5 pM custom-tailed oligo dT primer (SEQ ID NO:2). Lysates, dNTPs, and primer were heated at 65°C for 5 minutes then cooled on ice. The remaining components were added, and samples were incubated at 50°C for 10 minutes. The enzyme was inactivated at 80°C for 10 minutes. The samples were stored at 4°C until PCR.

[30] Multiplex Target Amplification (PCR #1). Following reverse transcription, 80 pL of PCR1 pre-mix was added to each 20 pL cDNA reaction. The final concentrations of PCR components after combining with cDNA reactions were 20 mM Tris-HCl (pH 8.5), 25 mM KC1, 4.5 mM MgCh, 0.2 mM dNTP, 4 units of PLATINUM™ Taq DNA Polymerase (Life

Technologies) ,and 50 nM of each oligo in the Multiplex Primer Pool #RNADNA_v2018_03 (SEQ ID NOS:2-54). The primer pool, described in more detail below, included 53 oligos targeting 29 mRNA transcripts and 12 genomic DNA sites. Only one transcript-specific primer was used per RNA target (sense-strand), because the oligo dT primer was used to tag the 3' terminus (antisense). Two primers were used per gDNA target site.

[31] Reactions were heated at 95°C for 5 minutes to denature templates. Amplification was carried out for 20 cycles of 95°C for 15s, 65°C for 90s, and 72°C for 30s. Reactions were held at 68°C for 5 minutes and then held at 4°C. Each sample was purified by mixing 100 pL sample with 180 pL of AMPure XP beads (Beckman Coulter) and incubated at room temperature for 10 minutes. Beads were captured by placing samples in MAGNESPHERE ^® Separation Stands (Promega) and washed 2X with 70% ethanol. Beads were air dried for 10 minutes and resuspended in 30 pL of water. Beads were captured, and supernatant containing purified PCR products were transferred to clean tubes.

[32] Universal PCR and Barcode Integration (PCR #2). Following purification, 25 pL of multiplex PCR products from each well were added to 25 pL of PCR2 pre-mix. The final concentrations of PCR components were 20 mM Tris-HCl (pH 8.5), 50 mM KC1, 1.5 mM MgCb, 0.2 mM dNTP, 2 units of PLATINUM™ Taq DNA Polymerase (Life Technologies), and 200 nM of each of two universal primers. A first primer that contains sites to facilitate bulk amplification and NGS on Illumina platforms, P5PM1 (SEQ ID NO:58), was used in every well. The second primer was one of a collection of 96 primers (P7-001 through P7-096 (SEQ ID NOS:58-153), each containing a unique molecular barcode sequence in addition to the universal sites used for bulk amplification and NGS on Illumina sequencing platforms. The respective molecular barcodes mark the identity of the well from which each genomic fragment is generated, so only one of these P7 variants was used per well or sample subset.

[33] Reactions were heated at 95°C for 5 minutes to denature templates. Amplification was carried out for 5 cycles of 95°C for 15s, 55°C for 30s, and 72°C for 30s, followed by 10 cycles of 95°C for 15s and 68°C for 30s. Reactions were incubated at 68°C for 5 minutes and then held at 4°C. Each sample was purified using AMPure XP beads (Beckman Coulter) as described for PCR #1.

[34] Quantification and Pooling of Subset-Specific Genomic Libraries. Prior to sequencing, the molarity of libraries from each well were quantified by qPCR using the KAPA Library Quantification Kit for Illumina Platforms (Kapa Biosystems). Individual libraries were combined at equimolar concentrations to a final pool concentration of 10 nM.

[35] Target-Specific Primer Design: Primers were selected using Primer3 v0.4.0 (see the website bioinfo.ut.ee/primer3-0.4.0/). Custom design settings included primer length (18-27 nt), primer melting temperature (58-63°C), and product length (140-160 bp). Stringency was lowered for some parameters ( e.g ., primer length, Tm) with a few targets that failed standard design conditions. A human mispriming library was used to filter all primer designs. Default design parameters were used unless specified.

[36] For gDNA targets, forward and reverse primers were designed to amplify selected target sites, which included cell line mutation sites and TP53 coding exons. Primers were placed in adjacent intronic regions when possible. Input sequences were obtained from Human Dec. 2013 (GRCh38/hg38) Genome Assembly.

[37] For RNA targets, primers were designed to amplify 3' regions of selected mRNA transcripts. Transcript sequences extending up to 300 bp from annotated 3' termini were used as inputs for forward and reverse primer design. Only the forward (sense strand) primer was selected for inclusion in multiplex PCR assays, given that antisense first strand cDNA was carried out with oligo d(T) primers. Input sequences were obtained from NCBI Reference Sequence Database (RefSeq).

[38] Appropriate universal tail sequences were added to the 5' terminus of each TSP (and oligo d(T)) prior to synthesis.

[39] Primer sequences are provided in Table 1. Target names and coordinates are shown in Table 2. Table 1. Primer Sequences

Table 2. Target Names and Coordinates

EXAMPLE 3. Detection and Classification of MCF7 Cells in Blood

[40] Overview. Whole blood from healthy donors was collected in Cell-Free DNA BCT blood collection tubes (Streck). Five mL of whole blood from one healthy donor was combined with 500,000 MCF7 cells and five mL of whole blood from the same donor was processed without addition of MCF7 cells. Each specimen was mixed with 8 volumes of buffered ammonium chloride solution to selectively lyse erythrocytes. Density gradient centrifugation was then used to separate cells into three fractions. The top and middle layers containing platelets and a significant portion of leukocytes were discarded. The bottom 4 mL fraction containing granulocytes and tumor cells was retained and concentrated by centrifugation. Cell pellets were resuspended in PBS + 2% FBS and a dilution series was made in PBS. Diluted fractions were divided into 96-well plates. Cells were lysed and genomic libraries with selected RNA and gDNA targets were generated from each well. Library content was characterized using qPCR analysis. Quantification of an MCF7-specific point mutation in the PIK3CA gene was used to identify wells containing tumor. Quantification of selected RNA transcripts was carried out using diagnostic primers. The process is described in detail below.

[41] Addition of MCF7 Cells. Adherent cells from the MCF7 breast tumor cell line were cultured in 75 mm flasks containing DMEM + 10% FBS + IX Antibiotic- Antimycotic (Thermo Fisher Scientific). Cells were dissociated using TrypLE™ Select Enzyme (Thermo Fisher Scientific) and counted using a hemocytometer. A total of 500,000 cells in a volume of less than 200 pL was added to one of the two 5 mL blood specimens.

[42] Density Gradient Separation. Five mL of cell resuspension from the RBC lysis procedure (Example 1) were added to 5 mL of room temperature Wash Buffer (pluriSelect Life Science) and processed as described in Example 1.

[43] Dilution of Cell Mixture. Cells retained after density gradient separation were diluted in PBS. The ten-fold dilution series for each specimen ranged from 1 : 10 to 1 : 100,000.

[44] Array Allocation and Cell Lysis. Cell mixtures were divided and lysed as described in Example 1.

[45] Reverse Transcription. Reverse transcription was carried out as described in Example 2, above. [46] Multiplex PCR. Multiplex PCR was carried out as described in Example 2, above.

[47] qPCR analysis. The CASTPCR™ TAQMAN ^® Mutation Assay PIK3CA_763_mu (Life Technologies) was used to measure the abundance of a point mutation in MCF7 cells. Several diagnostic SYBR™ Green assays were run to quantify RNA targets and the PIK3CA genomic region amplified during library construction (primer sequences provided below).

TAQMAN ^® Mutation Assay:

Hs00000824_mu, PIK3CA_763_mu (Life Technologies Cat#4465804)

SYBR™ Green Exon 12 Assay:

Forward primer 5'-GACAAAGAACAGCTCAAAGCAA-3 ' (SEQ ID NO: 161)

Reverse primer 5'-CCTGTGACTCCATAGAAAATCT-3' (SEQ ID NO: 162)

[48] Each TAQMAN® reaction contained 2 pL of PCR reaction (diluted 1 : 100 in water), 5 pL of 2X Fast Advanced Master Mix (Life Technologies), 2.0 pL of water, and 1.0 pL of 10X Mutation Detection Assay Mix Hs00000824_mu, PIK3CA_763_mu (Thermo Fisher Scientific, catalog #4465804). Reactions were run in Fast Mode on a QuantStudio 5 Real-Time PCR instrument (Life Technologies).

[49] Each SYBR™ Green reaction contained 2 pL of PCR reaction (diluted 1 : 100 in water), 5 pL of 2X POWERUP™ SYBR™ Green Master Mix (Life Technologies), 2.5 pL of water, and 0.5 pL of 10 pM primer pair. Reactions were run in Fast Mode on a QuantStudio 5 Real-Time PCR instrument (Life Technologies).

[50] Data Analysis. All Ct values were converted to abundance using the following formula derived from standard curves:

Abundance = 10[(Ct-34.231)/-3.558]

[51] Normalized PIK3CA mutation frequency (NMF) was calculated using the following formula:

NMF = (PIK3 C A_763 mu Abund/PIK3CA Exon 12 Abund) x 10,000 [52] Detection of MCF7-specific mutation. Tumor-containing wells were identified by qPCR detection of PIK3CA c 1633G>A (chr.3 179218303 in GRCh38, COSMIC #763), a heterozygous missense mutation in MCF7 cells.

[53] Figure 1 shows mutation abundance in a subset of wells from the specimen spiked with MCF7 cells diluted 1 : 10,000 before arraying into a microplate. Each bar in the graph represents data from one well. The figure shows high mutation signal in Well #4, indicating the presence of one or more MCF7 cells. Two wells from the“blood only” specimen are also shown.

[54] Figure 2 shows the correlation of RNA abundance in MCF7-positive Well #4 and input MCF7 cells that were not added to blood.

EXAMPLE 4. Detection of MCF7 Cells in Blood

[55] Overview. Whole blood from healthy donors was collected in Cell-Free DNA BCT blood collection tubes (Streck). Five mL of whole blood from one healthy donor was combined with either 100 or 1000 MCF 7 cells. Specimens were mixed with 8 volumes of buffered ammonium chloride solution to selectively lyse erythrocytes. Density gradient centrifugation was then used to separate cells into three fractions. The top and middle layers containing platelets and a significant portion of leukocytes were discarded. The bottom 4 mL fraction containing granulocytes and tumor cells was retained. Each sample was mixed with anti-CD45 beads and bound cells were captured with a magnetic LS column (Miltenyi Biotec). CD45(-) cells, including MCF7, were collected in the column flow-through and concentrated by centrifugation. Cell pellets were resuspended in PBS + 2% FBS and the whole volume was divided equally into a 96-well plate without further dilution. Cells were lysed and genomic libraries with selected RNA and gDNA targets were generated from each well. Library content was characterized using qPCR analysis. Quantification of an MCF7- specific point mutation in the PIK3CA gene was used to identify wells containing tumor. The process is described in detail below.

[56] Addition of MCF7 Cells. Adherent cells from the MCF7 breast tumor cell line were cultured in 75 mm flasks containing DMEM + 10% FBS + IX Antibiotic- Antimycotic (Thermo Fisher Scientific). Cells were dissociated using TrypLE™ Select Enzyme (Thermo Fisher Scientific) and counted using a hemocytometer. The desired number of cells in a volume of less than 200 pL was added to each 5 mL blood specimen. [57] Density Gradient Separation. Five mL of a cell resuspension from RBC lysis procedure (Example 1) were added to 5 mL of room temperature Wash Buffer (pluriSelect Life Science). An additional blood sample containing 1000 MCF7 cells was processed with density gradient separation alone, serving as a reference to confirm the benefit of adding the CD45 depletion step.

[58] Array Allocation and Cell Lysis. Cell mixtures were divided and lysed as described in Example 1.

[59] Reverse Transcription. Reverse transcription was carried out as described in Example 2, above.

[60] Multiplex PCR. Multiplex PCR was carried out as described in Example 2, above.

[61] qPCR analysis. The CASTPCR™ TAQMAN ^® Mutation Assay PIK3CA_763_mu (Life Technologies) was used to measure the abundance of a point mutation in MCF7 cells. Several diagnostic SYBR™ Green assays were run to quantify RNA targets and the PIK3CA genomic region amplified during library construction (primer sequences provided below).

TAQMAN ^® Mutation Assay:

Hs00000824_mu, PIK3CA_763_mu (Life Technologies Cat#4465804)

SYBR™ Green Exon 12 Assay:

Forward primer 5'-GACAAAGAACAGCTCAAAGCAA-3 ' (SEQ ID NO: 161)

Reverse primer 5'-CCTGTGACTCCATAGAAAATCT-3' (SEQ ID NO: 162)

[62] Each TAQMAN ^® reaction contained 2 pL of PCR reaction (diluted 1 : 100 in water), 5 pL of 2X Fast Advanced Master Mix (Life Technologies), 2.0 pL of water, and 1.0 pL of 10X Mutation Detection Assay Mix Hs00000824_mu, PIK3CA_763_mu (Thermo Fisher Scientific, catalog #4465804). Reactions were run in Fast Mode on a QuantStudio 5 Real-Time PCR instrument (Life Technologies).

[63] Each SYBR™ Green reaction contained 2 pL of PCR reaction (diluted 1 : 100 in water), 5 pL of 2X POWERUP™ SYBR™ Green Master Mix (Life Technologies), 2.5 pL of water, and 0.5 pL of 10 pM primer pair. Reactions were run in Fast Mode on a QuantStudio 5 Real-Time PCR instrument (Life Technologies). [64] Data Analysis. All Ct values were converted to abundance using the following formula derived from standard curves:

Abundance = 10[(Ct-34.231)/-3.558]

[65] Normalized PIK3CA mutation frequency (NMF) was calculated using the following formula:

NMF = (PIK3 C A_763 mu Abund/PIK3CA Exon 12 Abund) x 10,000

[66] Detection of MCF7-specific mutation. Tumor-containing wells were identified by qPCR detection of PIK3CA c 1633G>A (chr.3 179218303 in GRCh38, COSMIC #763), a heterozygous missense mutation in MCF7 cells. Data were normalized in all cells to the abundance of PIK3CA exon 12, which spans the region containing the MCF7 mutation. Data collected previously from individual wells of a“blood only” control sample were used to establish a baseline and detection threshold for analysis of MCF7 spiked test samples.

[67] Figure 3 shows mutation frequency in individual microwells from specimens processed by each enrichment method. Each bar in the graph represents data from one well. The dotted line represents the previously described detection threshold. For samples that were processed by density gradient alone, only 1/16 wells yielded mutation signal above background. Even then, the magnitude of the mutation signal in the positive well was near background. In contrast, samples processed by density gradient and CD45 depletion yielded many wells with mutation signal above background. Moreover, the magnitude of the signal in most positive wells was 10-fold or more above

background. In donor 1, 11/16 wells were positive as were 13/16 in donor 2 (75% of wells from both donors).

[68] Figure 4 shows mutation frequency in individual microwells from specimens processed by density gradient separation plus CD45 depletion. Eight of the 96 wells from the“blood only” sample were tested, whereas 88 of the 96 wells from sample containing 100 MCF7 cells were analyzed. All eight of the“blood only” wells reported mutation signals below the detection threshold (red dotted line) as expected. Of the wells from the MCF7 spike-in sample, 21/88 (24%) reported mutation signal above the detection threshold. A summary of data is shown in Table 3. Table 3. Summary of MCF7 Spike-In Results for Samples Processed by CD45 Depletion

[69] These results demonstrate robust detection of 100 and 1000 tumor cells in 5 mL of whole blood. The magnitude of mutation signal observed in positive wells was high above that of negative wells and tracked with the number of MCF7 cells spiked. In both cases, however, the observed number of positive wells was lower than expected. Both spike-in levels were expected to yield mutation signal in every well tested (1.0 tumor cell per well for 100 cell spike-in and 10 tumor cells per well for 1000 cell spike- in). This difference could be due to a number of factors, such as 1) differential loss of tumor cells during processing, 2) uneven distribution of cells across the array, or 3) sensitivity limits of the qPCR assay used for detection. Using a digital DNA sequencing readout may substantially increase sensitivity of detection.

EXAMPLE 5. Evaluating CTC Library Content by Next Generation Sequencing

[70] The goal of this experiment was to assess RNA and DNA target content by NGS analysis of libraries constructed from mixtures of purified nucleic acids. In addition, a titration series of MCF7 tumor cell line spiked into healthy donor samples was conducted to obtain an initial assessment of both sensitivity and specificity of tumor detection.

[71] Procedures: NGS libraries were constructed using our previously described RT-PCR protocol (Example 2) and Multiplex Primer Pool #RNADNA_v2018_03 (SEQ ID NOS:2-54). Purified samples used in this study were obtained from BioChain Institute Inc. (Newark, CA). Sample mixture composition is shown below (units are genomic equivalents). Tumor Content: 0% 1% 2% 4% 8% 16% 32% 100%

MCF-7 genomic DNA 0 1 2 4 8 16 32 100

Normal female genomic DNA 100 99 98 96 92 84 68 0

MCF-7 total RNA 0 1 2 4 8 16 32 100

PBMC total RNA (healthy donor) 100 99 98 96 92 84 68 0

[72] Barcoded libraries from each microwell were pooled, purified using SPRIselect

paramagnetic beads (Beckman Coulter Inc., Indianapolis, IN) and quantified using KAPA Library Quantification Kits (Kapa Biosystems Ltd., Wilmington, MA). NGS was performed on the Illumina MiSeq System in paired end sequencing format (2 x 150 bp). Reads were aligned to genome and RNA reference sequences using Bowtie 2. Read counts for each transcript are normalized to the number of reads mapping to human actin beta (ACTB) mRNA (RefSeq NM 001101.3) sequences in each library.

[73] Results. We first assessed the relative proportion of NGS reads that mapped to RNA and gDNA target sites. Libraries constructed from a mixture of RNA and gDNA inputs did produce reads that aligned to both target classes. However, the number of reads mapping to gDNA targets were relatively low (avg. 1.69% of aligned reads, range 1.0-4.4%, n=12) compared to those mapping to RNA targets (avg. 96.1%, range 91.1-97.5%, n=12) even though inputs were 100 cell equivalents for both. Libraries built from gDNA only or RNA only produced reads with high alignment to each of the respective target classes (>99%) as expected. Minor protocol modifications can be made to modulate the ratio of gDNA:RNA target content in sequencing libraries, if desired.

[74] We then evaluated sequences mapping to RNA targets. Transcripts were analyzed at two levels. First, we looked at the expression of individual genes previously reported to be reliable tumor markers (EPCAM, Figure 5A; and CDH1, Figure 5B) in CTC assays. Our data confirm that 1) these markers are highly expressed in MCF7 tumor cells, 2) weakly expressed in blood cells, and 3) marker expression varied predictably across tumor titrations. Both tumor markers were robustly detected above blood only background levels even at 1% tumor content (i.e., 1 in 100 cells) and the observed marker expression tracked closely with the expected tumor fraction. [75] Second, we analyzed tumor RNA signatures over the MCF7 titration series. Twenty tumor markers, selected by comparison of 100% blood and 100% MCF7 libraries, were assessed at tumor spike-in levels ranging from 1% to 8% (Figures 6A-F). The MCF7 self-comparison (Figure 6A) demonstrates an extremely high level of reproducibility over a wide dynamic range. All transcripts shown here exhibited substantially higher expression in MCF7 than blood alone. The tumor signature was readily detected in blood samples containing as little as 1% MCF7 as observed with individual transcripts described in Figures 5A and 5B.