# Utilizing Drug Intelligence Science (DIS®) for tumor type selection and molecular characterization of HFB200603, a best-in-class B and T Lymphocyte Attenuator (BTLA) monoclonal antagonist

**Germain Margall, Hombline Poullain, Gabrielle Wong, Juying Li, Eladio Márquez, Spencer Hugget, Hani Alostaz, Joshua Whitener, Shaozhen Xie, Xi Lin, Olja Rapaic, William Hedrich, Jinping Gan, Jack Russella-Pollard, Liang Schweizer and Robert H.I. Andtbacka**  
HiFiBiO Inc, Cambridge, MA USA, Corresponding Author

## BACKGROUND  
## • HiFiBiO has developed an innovative approach to drug development  
through its Drug Intelligence Science (DIS®) platform, which leverages the  
power of multi-modal artificial intelligence (AI). By integrating both  
predictive and generative models, the DIS® platform is designed to:

# AI-READY CLINICAL DATABASE AND BIOMARKER DISCOVERY  
- Precisely match subjects with the most suitable therapies  
- Accelerate the drug development timeline  
- Improve the probability of success in clinical trials

• The DIS® platform has been applied to select tumor types most likely to  
respond to HFB200603, a best-in-class antagonistic monoclonal antibody  
targeting BTLA as monotherapy and in combination with tislelizumab (anti-  
PD-1) in an ongoing Phase 1 clinical trial (NCT05789069).

- BTLA is a co-inhibitory immune checkpoint molecule primarily  
expressed on B cells, T cells, and dendritic cells.

- Dual blockade of BTLA and PD-1 is expected to result in stronger  
activation of T effector cells.

• Here, we report progress on applying the DIS® platform to infer predictive  
signatures of clinical benefit, supporting indication and subject selection in  
ongoing and upcoming trials.

## Study Design & Tumor Sample Analysis  
## HFB200603 PHASE 1 TRIAL  
## TCGA MSS and MSI-low CRC tumors,  
by subtype (n = 287)

• This trial is evaluating HFB200603, a best-in-class BTLA antagonist, in 54 subjects with advanced,  
treatment-resistant solid tumors: 17 subjects received monotherapy HFB200603, and 37 subjects were  
treated with HFB200603 in combination with tislelizumab (anti-PD-1) every three weeks (Q3W).

• Bulk RNA-seq and multiplex immunofluorescence (mIF) data was generated from tumor samples  
collected at screening (baseline) and C2D8. Subject counts for each modality of tumor data are shown  
below.

| Indication | N subjects | Tumor bulk RNA-seq | Tumor mIF |  
| --- | --- | --- | --- |  
| ccRCC | 10 | 1 | 1 |  
| Gastric | 10 | 5 | 6 |  
| Melanoma | 5 | 0 | 0 |  
| CRC | 20 | 8 | 8 |  
| NSCLC | 9 | 2 | 3 |

## Dashboards for real-time clinical monitoring and subject portrait generation  
transition from data collection to actionable  
insights, facilitating clinical monitoring and  
expediting biomarker discovery.

## • Real-time Dashboards: Accelerate the

• **Deep Learning:** Generates precise subject  
features from HiFiBiO dashboards, feeding  
predictive AI models to improve decision-making and clinical outcomes.

• **Subject Portraits:** Integrate multi-modal  
biomarker and clinical data to create  
comprehensive profiles, enabling more  
personalized therapeutic approaches.

• **Augmented Subject Portraits:** Can be generated by integrating data from public databases. To illustrate this,  
we assigned MSI/MSS status to CRC tumors using deep learning foundation models trained on Hematoxylin  
and Eosin (H&E) images from CRC patients in public databases. These predictions were independently  
verified by an expert pathologist.

**MSI/MSS status assignment using digital pathology**

| Patient | Medical Record | Digital Pathology Prediction | Prediction Probabilities |  
| --- | --- | --- | --- |  
| Patient #1 | MSS | MSS | (93-100%) |  
| Patient #2 | MSS | MSS | (91-100%) |  
| Patient #3 | MSS | MSS | (99-100%) |  
| Patient #4 | NA | MSS | (87-100%) |  
| Patient #5 | NA | MSS | (93-100%) |  
| Patient #6 | NA | MSS | (99-100%) |  
| Patient #7 | NA | MSS | (76-100%) |  
| Patient #8 | NA | MSS | (97-100%) |

## DIS® INTEGRATES GENOMIC, PHENOTYPIC AND CLINICAL DATA TO DISCOVER PREDICTORS OF CLINICAL BENEFIT  
*Representative example images and predictions*

## Discovery of a Potentially Predictive Tumor Signature for MSS/MSI-low CRC subjects

• CRC subjects (N=8) progressed on 4-5 prior lines of therapy, had liver metastases, and were confirmed to be MSS/MSI-low using deep learning digital pathology.

• Bulk mRNA-seq of baseline tumor biopsies identified an expression profile associated with **changes in tumor size (BOR**). This profile was characterized by:  
  i) Overexpression of lymphocyte activity related genes, including *BTLA*, *CD274* (PD-L1), and *PDCD1* (PD1), alongside lower expression of inflammation markers *NFKB1* (NF-kB), *NFKBIA* (IkBa), and *TNFAIP3*, which were  
  associated with **clinical benefit (CB).**

ii) Overexpression of poor prognosis related genes (e.g., TGF-beta signaling, EMT) was associated with **progressive disease (PD).**

Notably, tumor inflammation signature (TIS) GEP score, predictive of responses to anti-PD-1, does not correlate with clinical benefit

**Tumor expression signatures in CRC patients (baseline, N = 8)**

**Gene Set Enrichment Analysis (GSEA)**

**Clinical benefit signature application to identify likely responders in a public dataset**  
**HFB200603 Ph1 trial CRC patients (n = 8)**

## HFB200603 DEMONSTRATES PROOF OF MECHANISM IN TUMOR TYPES PRIORITIZED USING DIS®

## DIS® identifies Phase I indications for predicted MoA

• *BTLA* and *PDCD1* (PD-1) are co-expressed in exhausted CD8+ T cells in anti-PD-(L)1-refractory tumors

• *BTLA* and its ligand *TNFRSF14* (HVEM) are found co-expressed in different tumor types in public  
databases, specifically clear cell renal carcinoma (ccRCC), melanoma, colorectal (CRC), gastric, and non  
small cell lung (NSCLC) cancers. These tumors generally express PD-L1.

**BTLA and PD-1 expression by CD8 T cells in tumors refractory to IO**  
(HiFiBiO Disease Cell Atlas)

## HFB200603 promotes on-mechanism proliferation of intratumoral BTLA+ T cells

**Intratumoral CD8+ T cell density in digitally**  **reconstructed tumor mIF slides**

**Volcano plot for on-treatment vs. baseline**  **comparison of mIF cell density features**

## SUMMARY & FUTURE DIRECTIONS

• HiFiBiO’s DIS® platform is being used to prioritize indications and subjects in ongoing Phase 1 clinical trials. This approach integrates AI and predictive modeling with real-time data from the ongoing Phase 1  
trial to enrich clinical enrollment for the tumor and subject populations most likely to respond to HFB200603.

• HFB200603 treatment, both as a monotherapy and in combination with tislelizumab, results in expansion of intratumoral BTLA+ CD8+ T cells.

• Powered by integration of tumor biopsy and near real-time clinical data, the DIS® platform enabled the generation of a preliminary clinical benefit prediction signature for HFB200603 based on RNA  
expression Phase 1 data from metastatic MSS/MSI-L CRC tumors.

• These results underscore HiFiBiO’s commitment to accelerate development timelines and improving the probability of success by prioritizing subject selection for the most suitable therapies.

## Acknowledgments and references

We extend our thanks to the patients, their families, and the investigators and staff members who made this trial possible.

1. Guinney et al. 2015. The consensus molecular subtypes of colorectal cancer. Nat Med 21:1350-6 (10.1038/nm.3967)  
2. *https://github.com/Sage-Bionetworks/CMSclassifier*