systems immunology in io: a view from the parker...

Systems Immunology in IO: A View from the Parker Institute

Nikesh Kotecha, PhD

VP, Informatics

Disclosures

OUR MISSION

To accelerate the development of breakthrough immune

therapies to turn cancer into a curable disease.

A New Network

60+ Laboratorie

s

7 Research

Institutions

300+ Nation’s

Top

Researchers

40+ Industry

& Non-profit

Partners

April, 2016: Billionaire tech entrepreneur Sean Parker

announced a $250 million donation to establish the

Parker Institute for Cancer Immunotherapy to speed

research into innovative cancer treatments.

PROJECT-BASED

COLLABORATORS

Fortune Health

https://www.parkerici.org/person/ilan-zipkin-phd/



The Institute Leadership

Institute Leaders

JEFFREY

BLUESTONE, PhD

President + CEO

Parker Institute for

Cancer Immunotherapy

JEDD

WOLCHOK, MD, PhD

Memorial Sloan

Kettering Cancer Center

JAMES

ALLISON, PhD

MD Anderson

Cancer Center

CARL

JUNE, MD

The University of

Pennsylvania

LEWIS

LANIER, PhD

UCSF

ANTONI

RIBAS, MD, PhD

UCLA

CRYSTAL

MACKALL, MD

Stanford Medicine

SEAN

PARKER

Founder + Chairman

Parker Institute for

Cancer Immunotherapy

LAURIE

GLIMCHER, MD

Dana-Farber

Cancer Institute

Scientific Steering Committee Programmatic Collaborators

ELLIOTT SIGAL, MD, PhD

NEA

DAN LITTMAN, MD, PhD

NYU

NIR HACOHEN, PhD

Broad Institute

ELIZABETH JAFFEE, MD

Johns Hopkins

NINA BHARDWAJ, MD, PhD

Mount Sinai

LARRY TURKA, MD

Harvard Medical School

STEPHEN SHERWIN, MD

UCSF

JEFF HUBER

GRAIL

THOMAS DANIEL

Arch Venture Partners ROBERT

SCHREIBER, PhD

Washington University

PHIL

GREENBERG, MD

Fred Hutch

JIM

HEATH, PhD

Institute for Systems Biology

NINA

BHARDWAJ,

MD, PhD

Mount Sinai

STEPHEN

FORMAN, MD

City of Hope





Our Research Agenda

Large collaborative efforts focus on four areas with great potential.

Best-in-class T-cells Advance the next generation of T-cell therapies to provide targeted, safe,

long-lasting treatments to conquer cancer.

Immune Response Uncover why some patients respond to immunotherapy while others may not

to overcome cancer drug resistance.

Tumor Antigen Discovery Pinpoint novel cancer cell markers that will become the foundation for new

therapies and personalized treatments.

Tumor Microenvironment Discover how tumors impair immune cells, which will jumpstart the creation of

therapies that can fight hard-to-treat solid tumors.

INFORMATICS

Informatics is involved in all aspects of translational clinical research:

• Providing context through public data integration

• Evaluating new experimental technologies

• Supporting clinical trial design

• Sample tracking and data collection

• Clinical data entry and management

• Correlative assays data management

• Integrative data analysis

• Developing and scaling analytical methods

• Supporting computational efforts across the

consortium

I EXECUTION

I IDEATION

I ANALYSIS

Informatics supports PICI Clinical and Research studies with systems, programming, and biostats

o PICI Sponsored Clinical Studies

• Clinical Programming and Systems Setup for PICI trials (8+)

• Clinical Report Generation & Biostats

• Global library project

Towards a standardized set of CRFs/collection parameters to use across PICI studies

o PICI Sponsored Research Studies

• Programming and Systems Setup for Research Studies

Prospective collection underway for studying immune-related AEs after IO

• Novel deployment of REDCap environment

We are running it on a GCE environment that has others in the community excited.

Configuration is available at https://github.com/ParkerICI/redcap-k8s-templates

• Medidata • Veeva • Endpoint • Oracle Argus Safety • Kubernetes

https://github.com/ParkerICI/redcap-k8s-templates





Primary analysis pipelines in place for 15+ assays

Input from our investigators informs how we prioritize further development of these pipelines

Assays that can be

automatedPrototype stage:have run once or twice,

Advanced prototype:have run a few times,

Standard analysis:have run many times,

Mature analysis, mostly automated

WES

TCR

ATAC-seq

scATAC-seq

16S

RNA-seq

Nanostring

scRNA-seq

Assays that need a human in the loop

Prototype stage:have run once or twice,

Advanced prototype:have run a few times,

Standard analysis:have run many times,

Mature analysis,as automated as possible

Vectra

IMC

MIBI

CODEX

Cytokines/Luminex

Flow

CyTOF

Systems and infrastructure to support PICI trials and translational analyses (latest iteration)

Clinical & Research Ops Translational

Informatics Progress across PICI

o 30+ projects across 7 sites, 15+ data types, 10+ publications

o PICI worked with MD Anderson to identify TCR features in tumors associated with response to neoadjuvant nivolumab in high-risk resectable metastatic melanoma patients. o Amaria et al. Nature Medicine 2018

o PICI worked with UCLA to apply in-house CyTOF methodology, revealing differences in PD-1 expression patterns by anatomic site. o Davidson et al. Clinical Cancer Research 2018

o …and more

o Enabled novel work with PICI developed tools o https://github.com/parkerici

o “Deep” integrative analysis of multi-omic trials

https://github.com/parkerici

Outline of talk/examples of progressing IO

• End-to-end clinical trial analyses

• Approaches to think about (PD1) Resistance and Toxicities

• Approaches to think about neoantigen prediction (TESLA)

• Bringing data together to answer key questions

Pancreatic Study Combining Nivo CD40 and Chemo (PRINCE)

Exploring safety and efficacy of chemotherapy and immunotherapy combinations

for metastatic pancreatic cancer

o Chemotherapy + CD40 Ab + anti-PD-1:

• Standard of care chemotherapy

Nab-paclitaxel

Gemcitabine

• Immunotherapies

CD40 Antibody APX005M (Apexigen)

Nivolumab anti-PD-1 checkpoint inhibitor (Bristol-Myers Squibb)

PRINCE Overview

PRINCE Overview

Gemcitabine + nab-Paclitaxel + CD40 Ab APX005M +/- PD-1 Ab Nivolumab

Immunotherapy

Pancreatic cancer

Chemotherapy Blood

vessel Lymph

node

Tumor

1 Release of

cancer antigens

(cancer cell death)

2 Cancer antigen

presentation

(dendritic cells/APC)

3 Priming and

activation

(APCs + T-cells)

4 Trafficking of T-cells

to tumors (CTLs) 5 Infiltration of T-cells

into tumors

(CTLs, endothelial cells)

6 Recognition of

cancer cells

by T-cells

(CTLs, cancer cells)

7

Killing of

cancer cells Chemotherapy

CD40

Antibody (APX005M)

PD-1

Antibody (Nivolumab)

CD40 antibody

(APX005M)

PD-1 antibody

(Nivolumab)

Deep Immune Profil ing

Pa

tien

t S

am

ple

s

Biomarker

Samples

Clinical Metadata

Blood

Tumor

Germline WES

Tumor WES

Multiparameter

Imaging

Immune Profile

CyTOF/Flow

ctDNA: Mutant

KRAS

Neo-epitope

Prediction

HLA Determination

(MHC Class I and II)

TME Gene

Expression Signature

Tumor Genome/TMB

Extensive

Computational

Analysis

Harmonized methods of collection and processing at a central biorepository

RNAseq

(tumor & blood)

Cytokine

(serum)

Clinical Snapshot date: 05MAR19

Safety-evaluable Population

PRINCE Phase 1b: Promising Efficacy Signals

Cohort B1: Gem/NP/APX005M 0.1 mg/kg

Cohort B2: Gem/NP/APX005M 0.3 mg/kg

Cohort C1: Gem/NP/APX005M 0.1 mg/kg + nivo

Cohort C2: Gem/NP/APX005M 0.3 mg/kg + nivo

Overall Response Rate (CR or PR) = 46.7% (14/30)

Historical ORR for Gem/NP: 23% Phase 2 ongoing

Tumor Immune Profi l ing with High -Dimensional Imaging

CD68 CD8 Ki67 PD-L1 FoxP3 panCK, DAPI

Imaging of baseline tumors with two technologies:

Vectra and 30-marker OHSU mIHC

Vectra tumor imaging with 3 panels reveals low overall immune infiltrate with

higher macrophages and low CD8 T Cells

Frac

tio

n o

f n

ucl

eate

d c

ells

Cell-free DNA in blood

• Plasma cfDNA concentration measured before mutation analysis

KRAS G12V, D, R mutations

• Droplet digital PCR (ddPCR) assay to detect KRAS mutations in cfDNA

• Measured the portion of cfDNA that is confirmed to contain one or more KRAS mutations

Circulating Tumor DNA: Mutant KRAS Early analysis suggests that KRAS fraction in cfDNA correlates with tumor diameter

in patients who have measurable KRAS mutations at baseline (67% of patients)

Timepoint

KR

AS

Var

ian

t A

llelic

Fra

ctio

n

PR SD

PR

PR PR PR

SD

Tumor Sum Diameter

KRAS Allelic Fraction

Tum

or Su

m D

iameter (cm

)

PR

B2: Gem/NP + APX005M 0.1mg/kg

C2: Gem/NP + APX005M 0.1mg/kg + Nivo

High-dimensional cytometry for prof i l ing of immune dynamics in blood Treg

CD4 Central memory

CD4 Naive

CD4 EM

CD8 Naive

CD8 Central memory

CD8 Effector memory

TEMRA

CyTOF Immune phenotyping

(37 markers)

Early observations of immune changes with treatment: - Activation of B cells during the course of therapy - Activation of Tregs at the early timepoints

FACS Symphony T cell panel

(28 markers)

Current work: Unbiased clustering analysis of cellular population dynamics

Memory B cells (CD19+ CD27+ CD38-)

Plasmablasts (CD19+ CD27-)

Treg-like > CD38+ (CD4+ FoxP3+)





• Bringing data together to answer key questions

The problem of the majority

Tang et. al. (2018) Nat Rev. Drug Discovery

2,250 active trials testing anti-PD1/PDL1 agents as of September 2018,

totaling about 380,900 patient volunteers

~247,585 patients will experience resistance in trials alone (not accounting for SOC patients)

PD-1

Total=100

Response

Resistance

~35%

~65%

PD-1 “Responders”

PD-1 “Non-Responders” PD-1 “Non-Responders”

~65%

Intrinsic & extrinsic factors: What’s known

Havel et. al. (2019) Nature Reviews

Important questions in breaking down & harmonizing biomarker studies of PD -1

Biomarker Clinical

Response

to PD-1

At what clinical timepoint(s) was this measured?

How was this measured?

At what clinical timepoint was this measured?

How was this measured?

PICI trials addressing the problem of PD -1 resistance

Spencer, Wells & LaValle (2019) Trends in Cancer

PICI trial name Challenge addressed in design

MAHLER

Melanoma Tx with Ipi/Nivo Enrolled at

Multiple Centers Melanoma

- Trial exclusively in patients who progressed on PD-1

- Prior PD-1 resistance type taken into account in analysis

- Longitudinal, multi-omic biomarker studies, standard processing

- Working with partners to build out reference dataset (expand

sample size)

PRINCE

Pancreatic Study Combining Nivo,

CD-40 and Chemo

- Novel combinations in Pancreatic cancer, a novel tumor type


MCGRAW

Melanoma CP and Gut Microbiome

Alteration With Microbiome

Intervention

- Novel biomarker (microbiome) as combination with PD-1

- Patients stratified by baseline microbiome (pro-resistance vs pro-

response)

- Longitudinal biomarker studies, standard processing

PORTER

Prostate Researching Translational

Endpoints Correlated to Response to

Inform Use of Novel Combos

- Novel combinations in castrate-resistance prostate cancer

(CRPC), a novel tumor type

- Platform trial design


We Ran A Workshop

“Translational Approaches to PD-1 Resistance Workshop”

• Brought pharma + non-profit science + academia together to discuss how PD1 resistance can be addressed through translational science.

• Attendees from Merck, BMS, Genentech, Pfizer, Amgen, Regeneron; SITC, PICI, ISB; MDA/PICI, Yale, MSKCC/PICI

• Outcomes: core questions focus questions:

• How can data sharing be enabled between pharma?

• What should be the standard clinical definition of PD1 resistance?

• What are the molecular phenotypes of cancer resistance to immune killing?

What’s Coming – come join us!

• Molecular Phenotypes of Immune Resistance in Cancer • Working group based out of, but not restricted to, the workshop

• Initial goals

• Draft white paper/review on mechanisms of immune resistance (non-clinical)

• Retrospective, integrative analysis of gene expression and WES data from published cohorts

• Lead by SITC/TimIOs & ISB/UNC & PICI

• A follow up to “The Immune Landscape of Cancer”

• Actively seeking collaborators with large molecular data sets from IO trials!

What about toxicities?

PICI Launches the Autoimmunity and Cancer Program in Partnership with JDRF and Helmsley

CONSORTIUM:

PICI sites, academic labs, cancer centers, research hospitals, foundations, pharma and government institutions

GOALS:

Generate insight into the mechanisms behind irAEs following immune checkpoint inhibition

Determine overlap in mechanism with “classic” forms of autoimmune disease

Identify at-risk patients early to reduce the incidence and/or severity of such events

Understand irAE drug selectivity

Determine target antigen specificity

NERVOUS SYSTEM

Guillaine–Barré Syndrome

Myasthenia Gravis Encephalitis

THYROID

Hypothyroid

Hyperthyroid

HEART

Myocarditis

ADRENAL

Insuff iciency

GASTRO INTESTINAL

Colitis

SKIN

Vitiligo Psoriasis

PANCREAS

T1D

PITUITARY

Hypophysitis

LUNGS

Pneumonitis

LIVER

Autoimmune Hepatitis

RHEUMATOLOGIC

Vasculitis Arthritis

Autoimmune Events Resulting frOm Systemic Modulation by ImmunoTherapy (AEROSMITH)

• Aim: To collect clinical data and blood samples on patients before, at the time of, and after irAEs during checkpoint blockade therapy for cancer

• Goal: Enroll 1000+ patients prospectively and follow-up for 1.5yrs

• Progress: 200+ patients enrolled to date across 35 sites (April 2019)

Distribution of Cancer Types collected so far:

~50% melanoma + lung ~50% other

Distribution of AE Grades collected so far:

~50% are Grade 1





• Bringing data together across PICI to answer key questions

Tumor-specif ic neoepitopes are at t ract ive targets for cancer therapeutics

• Truly tumor-specific

o Novel peptides

o No pre-existing tolerance

o Immunogenic

• Safer therapeutic profile

o Minimal risk of autoimmunity

• Possibility of targeting multi-neoantigens could be a response to tumor heterogeneity and evolution

Hypothesis: if we can identify and target neoantigens with therapeutic agents, we can have personalized, safe, highly active therapies.

Yarchoan, & al. , 2017

Neoantigens can be targeted by therapeutic vaccines

Neoantigen Discovery Workflow

WES + RNA Sequencing

Variant Identification

Neoantigen identification

MHC I Binding prediction

Deep machine learning

Antigen processing models

Tumor cells

Normal

tissue

Ranked neoepitopes

• SNV • MNV • FS • Indel

Mass Spec.

Medgenome

From Personalis Website

Neoantigen discovery pipelines are becoming more diverse and complex

Genome Medicine20168:11

Speedy and accurate identification of neoantigen has been identified as a critical bottleneck for the delivery of neoantigen

promises

The need for benchmarking predict ion algori thms is becoming more pressing

Benchmarking predict ion a lgor i thms is a press ing need

TESLA : a community-based effort to optimizing neoepitope discovery

The Tumor neoEpitope SeLect ion All iance (TESLA) Program Goals

o TESLA is a community-based initiative that aims to support the field’s efforts to develop

safe and efficacious neoantigen-based therapeutics/vaccines for cancer by:

• Delineating the variation of neoepitope predictions in existing computational pipelines

• Generating high quality epitope validation sets that provide a basis to assess and

improve prediction pipelines

• Elucidating the key factors for accurate neo-epitope prediction

TESLA Par t i c i pat i ng G roups and Cont r i but ors

• 24 Academia/Non-Profits • 22 Pharma/Biotech

Advaxis Immunotherapies Agenus AMGEN Biontech BGI Genomics Bristol-Myers Squibb EpiVax Genentech Illumina ISA Pharmaceuticals MSD MedImmune Medgenome Neon Therapeutics Oncolmmunity Personalis Seven Bridges TEMPUS Vaccibody Yu Bio

Schematic of TESLA

Validation of the predicted peptides is central to algori thm improvement

o Validation aims at determining whether patient’s T cells are able to recognize the predicted

neoepitope.

• Focus on peptides binding to Class I MHC (pMHC)

o Goal: To validate predicted peptides in at least 2 assays guided by

• HLA restriction

• Availability of biological material

N. Bhardwaj, Mt Sinai A. Sette, LIAI

1. Peptide:MHC Binding 2. ex vivo stimulation

TESLA Functional Validation Methods

J. Heath, Caltech/ISB R. Schreiber, WUSTL P. Kvistborg, NKI

3. Tetramer detection (FACS) 4. NP-tetramer isolation

TESLA functional val idation methods

There is no “consensus” neoantigen identif icat ion pipeline.

Conclusions

• Only few steps were

commonly used across all

teams

• TESLA teams use a wide

variety of approaches, tools,

filters, etc. in their process.

• Most teams use 20-25

features for predictions

Manuscript in preparation Confidential – Do Not Post

• Overlap in ranked neoepitope are limited

• In the majority of the patients, reactive T cells could be detected against

some of the predicted neoepitopes

• Substantial variability exists between teams for predicted and ranked the

neoepitopes

o There are a set of teams that can consistently identify and rank highly peptide

MHC (pMHC) for which binding T-cells can be identified

o Filtering seems to play in important role in the quality of the results

• Pipelines are complex and diverse across teams

• Overall 600 pMHC were tested for T-cell binding in a tetramer-based

assays o 37 of these pMHC were found to have binding T-cells

o Majority of tested pMHC also have in-vitro MHC binding measurement

• in-depth analyses and elucidation of the key factors for accurate neo-

epitope prediction are underway

Summary





• Bringing data together across PICI to answer key questions

How we engage with sites and partners…

• Ask broad questions of interest to PICI and the field

• Find stakeholders who are interested in asking the question with us and can bring data and expertise to the table

• Align on a question of interest.

• Recent example:

“What features of the immune repertoire are associated with particular somatic tumor alterations, and how does this interaction shape the response to checkpoint inhibitor therapy?”

Bring PICI together to answer the question

PICI Informatics Brings Together:

MSK Melanoma:

• Clinical (R/NR)

• WES/TCR

MDA Melanoma:

• Clinical (R/NR)

• WES/TCR/RNA

MSK NCLC:

• Clinical (R/NR)

• WES/TCR

Tools from Broad

Institute:

• GATK, Mutect2,

Polysolver,

Oncotator,

Tools from

Stanford

Immunology:

GLIPH

(TCR Clustering)

• 1000s of unique samples

• Data from 13+ trials, 6+ published papers

• 300+ WES samples

• 1M+ TCR from 550 unique samples

• Bioinformatic methods for:

• Variant calling, HLA, HLA mutations. HLA copy

number, neoantigens, mutational signature ID,...

• TCR clustering

• Expertise in cancer immunotherapy:

• Renowned experts in cancer immunotherapy,

immunology, immunogenomics, …

Industry Partners

Funneling Data

300+ IO patients

Missing clinical data

Obviously incorrect

clinical data

Low TCR Quality

Low WES Quality

Final unified cohort

Singular patients incompatible with remaining cohort

TMB status is associated with TCR clonality in pretreatment.

Wells, et. al. 2019. In Preparation.

Bringing it all together: how does TMB shape TCR repertoire?

Summary:

• TMB High is associated with a “short, flat” repertoire

distribution

• TMB Low is associated with a “long, skewed” repertoire

distribution Wells, et. al. 2019. In Preparation.

Organize ourselves to answer broad questions working with PICI investigators + industry

oWhat are the genomic and immunologic features which predispose a patient to having an immune-related adverse event?

oWhich features of the pre-infusion product are associated with durable clinical response to cell therapy?

oWhat are the immune characteristics of the tumor microenvironment in castration-resistant prostate cancer, and how do they depend on previous treatment

oWhat are the molecular subtypes of cancer resistance to immune killing? How do these intersect with aPD1 therapy?

Data organized to enable the PICI community to ask further questions

• TCGA, published data, collaborator data, and PICI trial data.

Example question, answered in seconds:

“Which non-metastatic

head and neck cancer samples have

CNVs in PD-L1 and CD8 percentage > 15%?”

23,536 Samples

12,671 Subjects

33 Cancer Types

9 Technologies

286,954,730 Measurements

[email protected]

Thank you!

What about cell therapy?

Parker Institute: Next generation cell therapy

Novel CARs and vectors for clinical trials

NK cell evaluation and engineering

Endogenous T cell priming and therapeutics

CAR-T persistence and pediatric clinical trials

T cell trafficking and glioma targeting

Non-viral methods for T cell engineering

Novel cell therapy programs in GBM

CASSIAN YEE, MD

MD Anderson

CARL JUNE, MD

The University of

Pennsylvania

LEWIS LANIER, PhD

UCSF

HIDEHO OKADA, MD, PhD

UCSF

STEPHEN FORMAN, MD

City of Hope

CRYSTALL MACKALL, MD

Stanford Medicine

ALEXANDER MARSON, MD, PhD

UCSF

56

Efforts at PICI informatics are focused in 4 areas

Molecular Data Clinical Data

FLOW

FISH

SEQUENCING

MICROBIOME

IMAGING

CTMS

NOTES

REPORTS

SAMPLES/LIS

LABS

EMR

GENOMIC TESTS

DICOM

”Deep” Immune Profiling

From Tumor

From Blood

From both

GENETIC/

EPIGENETICS

WES

TCR

ATAC-seq

scATAC-seq

Methylation

PROTEIN

Cytokines

IHC (Vectra)

HD Imaging

HD FLow

GENE

EXPRESSION

Nano String

RNA Seq

scRNA-seq

GENE

EXPRESSION

Nano String

RNA Seq

scRNA-seq

MICROBIOME

Incorporate Prior Knowledge + Public Data

PRIOR KNOWLEDGE

PubMed

External databases

Past Sites’ Trials

Tech

Enthusiasts

Technology Evaluation & Standardization

Visionaries

Pragmatists

Conservatives

Skeptics

You can’t afford to ignore the outside world

Each sample is annotated with multiple molecular measurements (features)

Gene expression

(RNAseq)

Cytokines

(Luminex)

Cell populations

(CyTOF)

Responders

Non-responders

Categorical endpoint Continuous endpoint

Progression-free survival

No matter what, real-world data is incomplete

Features from all assays : 30,000 total

All

su

bje

cts

: 6

0 t

ota

l

Post-

treatm

ent.

P

re-t

reatm

ent

Number of patients

with all features: 8

The Core Questions

• How frequently do different mechanisms of immune evasion/suppression occur (in different cancers)?

• How do different mechanisms of immune evasion or suppression correlate (in different cancers)?

• How does aPD-(L)1 alter the tumor resistance

landscape?

• What are the subtypes of resistance to aPD-(L)1 therapy?

systems immunology in io: a view from the parker...

Documents