#66

Blood-based detection of early-stage colorectal cancer using multiomics and machine learning
Girish Putcha, Tzu-Yu Liu, Eric Ariazi, Marvin Bertin, Adam Drake, Michael Dzamba, Greg Hogan, Steven Kothen-Hill, Jeffrey Liao, Kang Li, Shivani Mahajan,
Krishnan K. Palaniappan, Poonam Sansanwal, John St. John, Peter Ulz, Nathan Wan, Hayley Warsinske, David Weinberg, Rui Yang and C. Jimmy Lin
Freenome, South San Francisco, CA
All correspondence to authors@freenome.com

Figure 2. AI-EMERGE® Study Design (NCT03688906) Figure 4. CRC sensitivity is high in distal and proximal tumors Figure 7. CRC sensitivity was higher for the multiomics blood test versus plasma CEA
BACKGROUND
Screening Cohort Case-control Cohort Specificity = 94% Specificity = 94%
• Approximately one-third of age-appropriate adults are not up to date with recommended colorectal n=3000 n=400
100 94% Stage I/II (n=17)
cancer (CRC) screening1 100 92% 91% 91% 91%
88% Distal (n=24) Stage III/IV (n=11)
Scheduled for screening colonoscopy Recent diagnosis of CRC or AA Proximal (n=8) 80 All CRC* (n=32)
• A non-invasive, blood-based screening test with high sensitivity and specificity in early-stage CRC 80
All CRC (n=32)
should improve adherence and ultimately reduce mortality; however, tests based on tumor-derived Specificity = 94%

Sensitivity
Sensitivity
60 60
biomarkers alone have limited sensitivity, especially in early-stage disease Average risk Average risk Average and 45%
high risk
40 40 31%
• Given the biological heterogeneity of CRC and its evolution over time, a multiomics approach wherein
non-tumor-derived signals complement tumor-derived signals is required for effective early detection 50-84 years of age 20 20 18%
50-84 yrs of age ≥18 yrs of age
(Figure 1)
0 0
Blood collected pre-colonoscopy, Blood collected post-colonoscopy, Multiomics Blood Test Plasma CEA Multiomics Blood Test
• Here we used a multiomics, machine learning platform to discover, refine, and combine tumor- and prior to bowel preparation prior to bowel preparation and treatment
non-tumor-derived signals to develop a blood test for the detection of early-stage CRC
*Four samples with unknown stage were tested. Two were correctly classified by CEA and 3 were correctly classified by the multiomics blood test.
Table 1. Clinical characteristics and demographics of subjects with CRC and • Seventy-five percent of tumors were distal and 25% were proximal
Figure 1. Biological signals change as cancer evolves2,3 colonoscopy-confirmed negative subjects • The multiomics blood test demonstrated 91% sensitivity and 94% specificity for CRC, whereas a
Figure 5. CRC sensitivity was higher for the multiomics blood test versus FIT CEA-only classifier achieved 31% sensitivity and 94% specificity, consistent with previous reports
Samples*
(n=574) (47% sensitivity and 80% specificity)8
Specificity = 96%
Colonoscopy-confirmed
CRC† negative subjects 100% 100% 100% Stage I/II (n=10)
100 Specificity = 96%
Cancer stage Pre-cancer 1 2 3 4
Gender: n (%)
n=32 n=539

80
Stage III/IV (n=2)
All CRC (n=12)
CONCLUSIONS
70% 67%
Non-tumor-derived Male 18 (56%) 292 (54%) • In our prospective, multi-center study (AI-EMERGE®) that included screening and case-control cohorts,

Sensitivity
60
Relative (e.g., immune) Female 14 (44%) 247 (46%)
50% we achieved high sensitivity (94%) and high specificity (94%) for early-stage (I/II) colorectal
contribution 40
adenocarcinoma
Age: median (IQR) 62 (45-83) 59 (53-67)
of signal Tumor-derived
Stage: n (%) 20
• Our multiomics approach combines tumor- and non-tumor-derived (e.g., immune) signals from cfDNA,
(e.g., ctDNA)
1 11 (34%) N/A
epigenetic, and protein biomarkers to detect early-stage CRC
0
2 6 (19%) N/A Fecal Immunochemical Test Multiomics Blood Test • When compared to a leading stool-based FIT, our multiomics blood test demonstrated higher sensitivity
3 7 (22%) N/A for CRC (100% vs. 67%) at comparable specificity
• Non-tumor-derived signal complements tumor-derived signal to enable earlier detection of disease 4 4 (12.5%) N/A
Polymedco OC-Auto® FIT results are reported per manufacturer’s specification at a cutoff of 100 ng/ml5.
• When compared to other blood-based tests, our multiomics blood test demonstrated higher sensitivity for
Unknown 4 (12.5%) N/A • In AI-EMERGE®, stool collection was optional. Only 52% of subjects who provided a blood sample CRC vs plasma ctDNA (90% vs. 47%) or plasma CEA (91% vs 31%) at comparable or better specificity
provided stool for FIT
OBJECTIVE *Excludes analyte training set (n=17).
†
Excludes 3 non-adenocarcinoma cases: 1 squamous cell carcinoma (stage III) and 2 neuroendocrine tumors (stage II). • These results will be validated in a prospective, multi-center registrational study
• The multiomics blood test demonstrated 100% sensitivity and 96% specificity for CRC, whereas FIT achieved
Figure 3. Sensitivity was high in both early (I/II) and late (III/IV) stage CRC† for the 67% sensitivity and 96% specificity, consistent with previous reports (74% sensitivity and 95% specificity)6
• The objective of this study was to assess the performance of our multiomics blood test in prospectively
multiomics blood test
collected CRC samples and colonoscopy-confirmed negative controls by combining tumor- and
non-tumor-derived (e.g., immune) signals from cfDNA, epigenetic, and protein biomarkers Specificity = 94%
Figure 6. CRC sensitivity was higher for the multiomics blood test versus plasma ctDNA REFERENCES
94% 91% 91% Stage I/II (n=17)
1. White et al. MMWR Morb Mortal Wkly Rep. 2017
100 Specificity = 100%

METHODS 80
Stage III/IV (n=11)
All CRC* (n=32) 100 94% 91% 90%
Stage I/II (n=16)
2. Heitzer et al. Nat Rev Gen. 2018
3. Guinney et al. Nat Med. 2015
Stage III/IV (n=11)
Sensitivity

60 80 Specificity = 75% All CRC* (n=30)

• Samples from a statistically-driven subset of subjects enrolled in a multi-center prospective study 4. Wan et al. BMC Cancer. 2019
(AI-EMERGE®) including average-risk screening and case-control cohorts were included in this analysis

Sensitivity
40 60 55% 5. OC-Auto® FIT [package insert]. Cortland Manor, NY: Polymedco
(Figure 2) 47%
40
38% 6. Imperiale et al. NEJM. 2014
20
• Forty-three subjects with CRC and 548 colonoscopy-confirmed negative controls were analyzed across
7. AVENIO ctDNA Targeted Kit [package insert]. Pleasanton, CA: Roche Sequencing
assays, including 17 analyte training samples (Table 1) 0 20
Multiomics Blood Test 8. Gao et al. Sci Rep. 2018
• Plasma was analyzed by whole-genome sequencing, bisulfite sequencing, and protein 0
†
Only results for colorectal adenocarcinoma, which represents >95% of CRC, are shown. Three non-adenocarcinoma subtypes were tested: The single squamous cell
quantification methods carcinoma (stage III) was correctly classified but both neuroendocrine tumors (stage II) were misclassified.
Plasma ctDNA Multiomics Blood Test

• Four-fold cross-validation was done, and performance based on the model was reported
*4 samples with unknown stage were tested, 3 were classified correctly

• In early stage CRC (I/II), 16/17 samples were correctly classified, with a sensitivity of 94% and specificity
*Three samples with unknown stage were tested for both plasma ctDNA and the multiomics blood test, and 2 were classified correctly. Plasma ctDNA results are ACKNOWLEDGEMENTS
reported per manufacturer’s specification.7
• The multiomics blood test was compared to FIT, plasma ctDNA, and plasma CEA. The multiomics blood of 94% The authors gratefully acknowledge members of the Freenome Clinical Laboratory and Clinical Development
test performance was calculated for each comparison using only the samples for which paired data • In late stage CRC (III/IV), 10/11 samples were correctly classified, with a sensitivity of 91% and specificity • The multiomics blood test demonstrated 90% sensitivity and 100% specificity for CRC, whereas plasma teams, and Mitch Bailey, Anna Cunningham, Dan Steiger, Francesco Vallania and Irving Wang for intellectual
was available of 94% ctDNA achieved 47% sensitivity and 75% specificity and technical contributions.

Presented at the American Society of Clinical Oncology (ASCO) GI Symposium 2020, January 23-25, 2020, San Francisco, CA, USA