Tese Carolina

Study of biomarkers in Multiple Myeloma: a statistical approach for longitudinal assessment of extracellular vesicles and its prognostic value

Mestrado em Bioestatística

Carolina Alexandre Festas Gomes Pestana Trabalho de projeto orientado por: Professora Doutora Lisete Sousa Professora Doutora Cristina João

25 de Julho de 2021

Index

4. Methodological Intro

8. Discussion

5. Survival Analysis

1. Multiple Myeloma

6. Longitudinal Modelling

9. Conclusion

2. Hypothesis

10. References

3. Study Design

7. Proteomic Analysis

1.

Multiple Myeloma

A bone marrow disease

This disease results from an unregulated proliferation of monoclonal plasma cells in the bone marrow, causing an overproduction of monoclonal immunoglobulins (M protein)

2nd

most frequent haematological cancer

(Kazandjian, 2016)

69

is the median age at diagnosis

(National Cancer Institute)

Source: https://juliesmyelomamoments.blogspot.com/2013/10/. Accessed in 13-03-2020

info

1.1

Symptoms and Diagnosis

CRAB

20%

does not present any symptoms

However...

Calcium elevationRenal failure Anemia Bony/lytic lesions

Extensive blood and urine tests

• Bone pain
• Weakness and Fatigue
• Weight loss, Kidney problems
• Fever and infections
• Blood clots

Biopsy

CT/MRI

Quantification of malignant plasma cells present in the bone marrow

Assessement of bone destruction and bone marrow infiltration

(American Cancer Society, Cancer.Net, Healthline)

1.2

Characterization

The final MM diagnosis is given if

≥ 1 features

Clonal bone marrow plasma cells >= 10%

BM clonal plasma cells ≥ 60%

Serum free ligh chain ratio ≥ 100

> 1 MRI focal lesion

AND

OR

OR

Biopsy-proven bone lesions

≥ 1 CRAB feature

(Rajkumar, 2018)

1.3

Pre-Symptomatic stages

Facts

20%

of MGUS patients progress to MM

...

M-Protein level (g/dL)

(Kyle et al., 2018)

...

59%

of SMM patients will eventually develop MM

(Kyle et al., 2007)

% of BM clonal plasma cells

SMM

MGUS

Monoclonal Gammopathy of Undetermined Significance

Smouldering MultipleMyeloma

(International Myeloma Foundation, 2019)

1.4

Extracellular Vesicles

IMMUNE RESPONSE

tUMOUR CELLS' SURVIVAL

CELL MIGRATION

TUMOUR CELL'S INVASION

REGULATION OF THE BONE MARROW MICROENVIRONMENT

INTERCELLULAR COMMUNICATION

Extracellular vesicles in multiple myeloma (MM). De Luca et al., 2019.

(Wang et al., 2016)

2.

Hypothesis

Questions

Will the variable EVs Cargo at the time of inclusion in the study discriminate between patients with better and worse prognosis?

Our main goal is to

demonstrate that extracellular vesicles can be used as liquid biopsies, particularly as disease biomarkers in the context of MM and as a tool to monitor patients over time

What variables could longitudinally characterize patients who, at each time point, present a worse prognosis?

Using the EVs Cargo variable as representative of extracellular vesicles

Which proteins present in these EVs are associated with the variables that have been proved to explain, over time, a worse prognosis? And do these proteins corroborate the results previously obtained?

3.

Study Design

Patients collected multiple samples during study time according to medical follow-up

MGUS, SMM or MM patients diagnosed or treated at Champalimaud Foundation

Observational and prospective cohort study

Adapted from: Durie et al. (2003); Kurtin & Faiman (2013)

3.1

Baseline overview

Number of patients by category

Diagnosis

Patients were enrolled in the study between

May 2016 - December 2019

R-ISS III

R-ISS II

Last known follow-up was reported on

naive

1 Line

2+ Lines

July 13th, 2020

Median follow-up time of

25.18 months

3.2

(some) Variables and Outcomes

Outcomes

In the first phase the outcome of interest was considered to be the death of the patient

Laboratory

Demographic andHealth-related

Subsequently, and in view of the study of the EVs Cargo variable, this became our response variable, with the High category the outcome of interest

EVs Cargo (µg/10^8 particles) Ig's serum level (IgG, IgA, IgM) sFLC's level (Kappa, Lambda) Lactate Dehydrogenase (LDH) β2 − M Albumin C-reactive Protein

Patient’s IDPatient’s age at sampling timeDiagnosis at time-pointR-ISS scoreTreatment lineFollow-up timeFollow-up status

4.

Methodological intro

Proteomic Analysis

Longitudinal Modelling

Survival Analysis

To understand which variables could explain their worse prognosis (EVs Cargo High) over time

To assess the influence of the variables considered as explanatory of the high EVs Cargo in EVs protein content

To study the association of the variable EVs Cargo with patients' overall survival

+info

5.

Survival Analysis

Why?

maximally selected rank statistics

Log-rank test

Cox regression Model

To understand the extent to which the EVs Cargo could be an indicator of patient's survival at the time of inclusion in the study and independent of the disease status

For > 2 groups

Extension of U-statistic

Model Formulation

Hypothesis

Hazard Ratio

To search for the optimal cut-off point which, by dividing the variable into two groups (low and high EVs Cargo), would be able to better define patients' prognosis

Hypothesis

Maximum of the standardized statistics

Test statistic

Hypothesis

(Hothorn & Lausen, 2003)

(Cox, 1972)

5.1

Exploratory Analysis Through Cox Models

Understanding the association between the survival time of patients under study and EVs Cargo

A traditional survival analysis based on Cox model was performed, fitting an univariable Cox model to the EVs Cargo variable

PH

5.2

Variable Stratification

An optimal cut-off value of

0.6 µg/10^8

was determined for the EVs Cargo variable

25 months survival probabilities of

97% and 84%

for EVs Cargo Low and EVs Cargo High, respectively

Approximately 7 out of 10 patients were classified as presenting EVs Cargo High

5.3

EVs Cargo as a Categorical Variable

KM estimates of the survival functions associated with high levels of EVs Cargo are those which present the worst prognosis, and these are associated to the diagnoses of SMM and MM

25 months survival probabilities of

92% and 69%

for MM diagnosis and EVs Cargo Low and EVs Cargo High, respectively

A traditional survival analysis based on Cox model was performed, fitting an univariable Cox model to the stratified EVs Cargo variable

6.

Longitudinal Modelling

Why?

Modelling Binary Responses in Longitudinal Studies

Random effects introduction

To account for the within-subject association, the GLMM Generalized Linear Mixed Models) implements random effects in the linear predictor:

To determine if common myeloma-related blood parameters can be explanatory of the potential predictive value of EVs Cargo on OS

Advantages

• Allows the dependence structure between observations to be considered;
• Missing response values are allowed provided they are Missing At Random (MAR)

where are assumed to be sampled independently from each other, following a distribution

6.1

Model Specificities

Note that:

Only patients diagnosed with SMM or MM were considered in this model (as they are those with more similarities between each other

Time

Chronological sequence in which samples were collected (number of times that a patient collected a sample)

The EVs Cargo cut-off point determined earlier was applied to the samples obtained throughout the study

Line

Number of previous therapeutic lines at sample collection time

Only samples with complete laboratory information were considered

88 samples

from 50 patients (9 SMM and 41 MM) were considered

6.2

Model Interpretation

Time and Line interaction

70% reduction

for each collected sample (fixing the treatment Line)

sFLC Lambda Levels

6 times increase

for patients with elevated sFLC Lambda when compared to patients with normal levels

IgA serum Levels

70 times increase

98% reduction

for patients with depletion of sFLC Lambda when compared to patients with normal levels

for patients with IgA depletion when compared to patients with normal levels

7.

Proteomic Analysis

Why?

PRE-ANALYSIS DATA MANAGEMENT

Limma method

Through its functions, one can assess the proteins who present a differential expression between some interest groups.

Key approach

To study the influence of IgA depletion as well as sFLC Lambda elevation/depletion in EVs protein content

• Two technical replicates were considered;

• The average expression between the technical replicates was computed;
• A logarithmic base 2 transformation was applied.

Use the entire data to shrink the observed sample variances towards a pooled estimate, leading to a more stable and robust inference when compared to other methods, such as the t-test.

Advantages

To evaluate the potential of exosomes as biomarkers

Deal with the data, as a whole, rather than performing piece-meal comparisons between pairs of treatments

7.1

Model Results' Validation - IgA

Note that:

This validation was based on a subset of 52 samples belonging to the longitudinal model built in the previous step (46 MM and 6 SMM), corresponding to 38 patients (33 MM and 5 SMM)

IgA Depleted vs. IgA Normal

3.1 times higher

IGHA1

Expression of

in the Normal level

is

Since there are patients with more than one sample, the existence of repeated measures was considered (through the introduction of a random factor associated to the patient)

and

2.6 times higher

IGHA2

Expression of

in the Normal level

is

when compared to the Depleted level

7.2

Model Results' Validation - sFLC Lambda

sFLC Lambda Depleted vs. sFLC Lambda Normal

The expression of

IGHM

IGLC2, IGLC3

are

and

2.4, 2.5 and 2.2 times higher

in the Normal level when compared to theDepleted level, respectively

sFLC Lambda Normal vs. sFLC Lambda Elevated

CALML3

is

The expression of

5.3 times higher

in the Elevated level when compared to theNormal level

7.3

Validation as Disease Biomarker

Note that:

82 proteins

with differential expression between Patients and HD

For this validation, in addition to the samples from MM and SMM patients previously considered, 13 samples from MGUS and 14 HD presenting complete laboratory information were included. In total, 51 patients (65 samples) and 14 HD (14 samples) were considered

8 proteins

24 proteins

in patients presented a |log2 FC| ≥ 2

The existence of repeated measures was also considered in this analysis

7.4

Enrichment Analysis

Proteins down-regulated in patients were mainly associated with

immune response, complement activation, phagocytosis and B cells signaling

Proteins up-regulated in patients were mainly associated with

endopeptidase activity and complement regulation processes

8.

Discussion

Survival Analysis

Longitudinal modelling

proteomic analysis

High EVs Cargo level is related to some specific disease characteristics

EVs Cargo levels may be predictive of patient’s survival

The laboratory characteristics observed in the longitudinal model are corroborated at the protein level Patients’ EVs have distinct protein content compared to healthy donors

9.

Conclusion

Take home message

EVs Cargo may be predictive of patients' prognosis - prognostic biomarker

We demonstrated that

Extracellular Vesicles

(here represented by the variable EVs Cargo) found in the blood of patients can serve as

A worse prognosis over time may be explained by IgA depletion, increased sFLC Lambda and also by a shorter response time - mechanism for regular monitoring and follow-up of patients in clinical practice

Liquid Biopsies

10.

References

• American Cancer Society. (n.d.). Cancer Statistics Center - Myeloma. Retrieved 23-03-2021, from https://cancerstatisticscenter.cancer.org/?_ga=2.224251035.1208699148.1585581618-742406976.1582713354#!/cancer-site/Myeloma ;
• Cancer.Net. (2020). Multiple Myeloma. Retrieved 23-03-2021, from https://www.cancer.net/cancer-types/multiple-myeloma ;
• Cox, D. R. (1972). Regression models and life-tables. Journal of the Royal Statistical Society, 34(2), 187–202. doi: 10.1111/j.2517-6161.1972.tb00899.x ;
• Durie, B. G., Kyle, R. A., Belch, A., Bensinger, W., Blade, J., Boccadoro, M., … Van Ness, B. (2003) ;
• Myeloma management guidelines: a consensus report from the scientific advisors of the international myeloma foundation. Hematology Journal, 4(6), 379–398. doi: 10.1038/sj.thj.620031
• Healthline. (2019). Signs and Symptoms of Multiple Myeloma. Retrieved 23-03-2021, from https://www.healthline.com/health/cancer/multiple-myeloma-signs-symptoms ;
• Hothorn, T., & Lausen, B. (2003). On the exact distribution of maximally selected rank statistics. Computational Statistics & Data Analysis, 43(2), 121–137. doi: 10.1016/S0167-9473(02)00225-6 ;
• International Myeloma Foundation. (2019). What Are MGUS, Smoldering Myeloma, and MM? Retrieved 17-04-2020, from https://www.myeloma.org/what-are-mgus-smm-mm ;
• Kazandjian, D. (2016). Multiple myeloma epidemiology and survival, a unique malignancy. In Seminars in Oncology (Vol. 43, pp. 676–681). doi: 10.1053/j.seminoncol.2016.11.004 ;
• Kyle, R. A., Larson, D. R., Therneau, T. M., Dispenzieri, A., Kumar, S., Cerhan, J. R. & Rajkumar, S. V. (2018). Long-term follow-up of monoclonal gammopathy of undetermined significance. New England Journal of Medicine, 378(3), 241–249. doi: 10.1056/NEJMoa1709974 ;
• Kyle, R. A., Remstein, E. D., Therneau, T. M., Dispenzieri, A., Kurtin, P. J., Hodnefield, J. M., … others (2007). Clinical course and prognosis of smoldering (asymptomatic) multiple myeloma. New England Journal of Medicine, 356(25), 2582–2590. doi: 10.1056/NEJMoa070389 ;
• Kurtin, S., & Faiman, B. (2013). The changing landscape of multiple myeloma. Clinical Journal of Oncology Nursing, 17(6), 7–11 ;

• National Cancer Institute. (n.d.). Myeloma - Cancer Stat. Facts. Retrieved 23-03-2021, from https://seer.cancer.gov/statfacts/html/mulmy.html ;

• Palumbo, A., Avet-Loiseau, H., Oliva, S., Lokhorst, H. M., Goldschmidt, H., Rosinol, L., … Moreau, P. (2015). Revised international staging system for multiple myeloma: a report from international myeloma working group. Journal of Clinical Oncology, 33(26), 2863. doi: 10.1200/JCO.2015.61.2267 ;
• Rajkumar, S. V. (2018). Multiple myeloma: 2018 update on diagnosis, risk-stratification, and management. American Journal of Hematology, 93(8), 1091–1110. doi: 10.1002/ajh.25117 ;

• Wang, J., Faict, S., Maes, K., De Bruyne, E., Van Valckenborgh, E., Schots, R., … Menu, E. (2016). Extracellular vesicle cross-talk in the bone marrow microenvironment: implications in multiple myeloma. Oncotarget, 7(25), 38927. doi: 10.18632/oncotarget.7792 ;

Thanks for your attention