8+ T cell receptor diversity analysis

Since the diversity of the T cell receptor library reflects the degree of clonal expansion in the T cell population, it is a valuable biological and potential biomarker for diseases.

Therefore, the assessment of the diversity of the T cell receptor library is very important, so what I want to share with you today is the assessment of the diversity of T cell receptors and the study of its relationship with the Cancer Immunology Research (IF:8.
728) published in January this year.
An article on the relationship between survival and response to anti-PD-1 treatment.

Improved T cell receptor diversity estimates associate with survival and response to anti-PD-1 therapy A heterodimer of beta or gamma and delta cell surface proteins, they give T cells extremely high sensitivity and specificity.

In cancer immunity and immunotherapy, the lineage analysis of the T cell receptor library is essential for biological discovery and the development of biomarkers, and the key to this analysis is to summarize the diversity of TCR distribution frequencies in mixed populations.
.

In addition, although the TCR diversity index is increasingly used in clinical trial research in oncology, its accuracy has not been verified in some published truth data sets.

Therefore, today the editor shares an article introducing the performance characteristics of RNA-seq data TCR spectrum analysis method.
The article reveals that insufficient sampling is the main source of diversity assessment bias.
The article also derived a model through statistical learning, which can be attenuated Deviations produce corrected diversity assessments.2.
Research data and methods 1.
TCR library generation: The author uses the synthetic TCR information generator (STIG) to generate one thousand TCR libraries, and uses STIG to generate 100 million individuals (ie virtual cells) from 5000 diverse clone types.
.

The readings in these populations match TCGA.

At the same time, STIG is provided with a basic quality score to match the same quality score and error rate as the TCGA data.

The median insertion length and standard deviation are also selected from the common range of TCGA sample sub-sampling.

The distribution of TCR clonotypes adopts the STIG chi-square distribution setting with 2 degrees of freedom.

The diversity of the model is within 99% of the TCGA sample.

Then the diversity and abundance distributions are matched and several rounds of library generation, library read comparison, model generation and diversity calculation are required.

In addition, the abundance of Reads also matches the distribution of TCGA data, where about 10% of STIG reads are aligned with the CDR3 region, and the remaining unaligned reads represent other regions of the TCR.

2.
TCGA analysis: study and analyze the primary tumor in TCGA, and process the sample according to sample annotation and repetition.

The final TCGA data used is the batch-corrected RNA-seq expression data and the TCR diversity index calculated from the MiXCR output.

3.
Diversity of the sub-sampling population: The initial sub-sampling is based on the truth counts generated by STIG before the readings are produced.

Then the author uses the R package vegan to evaluate the two parameters of Chao1 and Shannon entropy, and then uses Shannon entropy to compare logarithmic richness to calculate the uniformity.
The uniformity produces a result similar to "productive clonality", and the clonality is 1 minus uniformity.

In addition, the author calculated the d25, d50 and d75 (dXX) indices according to the description of VDJTools in GitHub.

4.
Comparison of MiXCR and TRUST: MiXCR, TRUST and TRUST 3.
0 Dockerfiles and sample commands can be found in https://github.
com/Benjamin-Vincent-Lab.

Since TRUST cannot align sequences, the author used Bowtie2 to align the reads before using TRUST for these analyses.

Morisita-Horn uses the vegdist function in the vegan R package to perform calculations and express it with similarity index.

The mismatch index is calculated by dividing the median mapped to the mismatch library by the number of mappings in the correct library.

This index greater than 1 means that matching CDR3 is more common in the mismatch library than the correct library.

Then use BLAST to perform a basic partial alignment search, and each sample uses the STIG truth library sequence as the database.

5.
Model generation: The author uses Monte Carlo cross-validation method to generate elastic network model.

These 1000 simulated TCR libraries are divided into discovery sets and verification sets.

The discovery set is repeatedly divided into three layers, two layers are used for training and one layer is used for testing.

Use library features (abundance, Chao1, d25 index, uniformity, inverse Simpson, richness, Shannon entropy) as predictor variables, and use true Shannon entropy as response variable to fit the elastic network regression model, and finally use cross-validation for evaluation .

3.
The main results and content of the research 1.
The fidelity of the diversity index In the first part of the article, the author first analyzed the fidelity of the diversity index.

The diversity of TCR receptors is usually calculated by assembling and counting sequences encoding CDR3 receptors (Figure 1).

It is becoming more and more common to use tissue RNA-seq data to assess this diversity (Figure 1B).

However, RNA-seq data sets usually have much lower CDR3 counts than TCR amplicon data sets, so there are more sub-sampling.

So the author first tried to understand the degree of influence of this sub-sampling on the diversity index.

The author started by assuming a population of 50 individuals and observed the variance in the diversity estimation using Shannon entropy and uniformity.

Both methods show that the more severe the sub-sampling, the greater the variance (Figure 2A).

Next, in order to evaluate whether sub-sampling affects the diversity index only at a smaller population size, the author simulated 1000 pools from 200 to 100 million individuals and used a common diversity index to evaluate the changes in diversity estimates, and then plotted the true The ratio of the value diversity index value to the sub-sampling value under each index (Figure 2B).

The author found that the error of diversity is not limited to Shannon entropy and uniformity, nor is it limited to undersampled samples.

It is also found that as the abundance increases, the calculated diversity index finally converges to an accurate estimate of the true diversity.

In addition, although previous studies have found that uniformity tends to overestimate diversity at low abundance, the authors found that this overestimation is more pronounced in samples with low initial diversity (Figure 2C).

Figure 1 Generation and inference of T cell receptor sequences Figure 2 The influence of sub-sampling on TCR diversity indicators 2.
Entropy modeling of low abundance samples After analyzing the fidelity of diversity indicators, the author wrote in the article The second part of the authors tried to derive a model to predict the entropy of the true TCR library when the abundance is low, aiming at the stability of Shannon entropy as a diversity index.
For this reason, the author uses elastic network regression to model the true Shannon entropy, and it can be observed The direct calculation of Shannon entropy shows a moderate correlation with the true value but large variance (Figure 3A).

In addition, the elastic network model has stronger correlation and smaller variance (Figure 3B).
Figure 3C shows the results of training and testing in different types of chains, while Figure 3D shows the results of training and testing in the same type of chains.

The author also found that when the abundance is high enough, the entropy value can be accurately estimated through direct calculation without model correction.

Then the author analyzed the accuracy of the entropy calculated by the sub-sampling library based on the directly calculated entropy based on the measured value (Figure 3E), and found that the Shannon entropy that needs to be measured to determine the abundance is within 5% of the true Shannon entropy.

Figure 3 Application of diversity index in the RNA-seq samples generated by 1000 STIG output by MiXCR 3.
Consistency of diversity index Then in the third part of the article, the author analyzed the consistency of diversity index.

Validating model-based modified diversity measurements on biological samples is a difficult task because there is no true value reference in this context.

In addition, there are multiple features that are considered to be related to diversity that can be tested for correlation with diversity estimation.

Previous studies have found that the diversity of TCR is inversely proportional to age, because the available TCR library is gradually lost over time.

The same diversity may be inversely proportional to tumor mutations, because the number of mutations is related to the number of neoantigens, which can be targeted by T cells.

Therefore, the authors hypothesized that a more accurate index of diversity would have a stronger negative correlation with age, SNV neoantigen and InDel neoantigen.

Therefore, the author analyzed the TCGA data and found that the diversity coefficient of the corrected TCR chain with age and neoantigens is closer to -1, and the variance is lower, which is significantly higher than other diversity indicators (Figure 4A) .

On the contrary, because the thymus is the source of T cells, the author believes that the diversity of thymus is higher than that of any other tissue type.
Therefore, the corrected diversity index of thymoma is better than other diversity indicators with low variance, and the coefficient is close to 1.
Significance is higher than any other indicators (Figure 4A).

In addition, reliable estimates of diversity should also tend to be internally consistent, providing similar predictions for a range of TCR abundances.

Therefore, in order to test the internal consistency of the diversity index, the author randomly divided all TCGA organizations into 3 quantiles according to the sample abundance, and used Cox proportional hazard regression to test the association between overall survival rate and TCR diversity index.

The author also performed average sampling within 3 abundance ranges as the fourth group.

The prediction accuracy is estimated as the difference between the upper and lower confidence intervals of the hazard ratio of each organization (Figure 4B, C).

It can be observed that an internally consistent metric should provide a narrow width that is consistent across the abundance range.
Within the abundance range of uniformity and model entropy, there is no significant difference in the width of the confidence interval, while Shannon entropy predicts between abundance groups.
There are significant differences.

In summary, these results show that the correlation between model-based diversity estimation and real biodiversity is improved and internal consistency is enhanced.

Figure 4 Calculate the diversity index on each TCR chain of the TCGA tumor sample and compare it with the modeled entropy measure 4.
Use the TCR model entropy to predict patient prognosis.
Because the author of the previous part found that the model-based correction provides more diversity than other indicators Better biodiversity relevance, so in this part the author then asked the question: Can this improved diversity index improve the survival prediction of cancer patients? The author found that even a perfect diversity measure may not Predict the survival rate of cancer.

That is to say, hypothetical reduction in diversity should predict better survival, because it indicates that the immune system is enhancing its effective response to tumors.

However, in the TCGA data, the authors found that higher correction diversity was associated with better survival rates.

In addition, the authors tested the relationship between tumor-associated T cell receptor diversity and survival in melanoma patients who received a monoclonal antibody that inhibits PD-1 before treatment.

The authors also observed an increase in TCR library diversity and monoclonal antibody treatment on another set of RNA-seq data sets generated using formalin-fixed paraffin-embedded (FFPE)-derived RNA related to immune checkpoint suppression tests.
The improvement in the survival of patients with melanoma is related (Figure 5A).

Similar results were also observed in TCGA samples and immune checkpoint suppression test related data sets (Figure 5B, C).

The author also observes that this effect is greatest when the model entropy of both the α chain and the β chain is high.

When the author compared the TCR diversity index based on patient responses, the author observed from the model entropy that the responders showed significantly higher diversity (Figure 5D).

It can also be observed that the non-progressive group and the responding group show similar differences, but not significant.
The non-progressive group shows a higher TCR model entropy than the progressive group (Figure 5E).

Next, the author analyzed TCGA and immune checkpoint suppression test related data set samples and quantitatively showed that the high-entropy model predicted the sampling percentage of TCR predicted survival.
The results are shown in Figure 5F.

In addition, the authors calculated the median hazard ratios of these samples (Figure 5G).

Both results indicate that samples treated with fewer immune checkpoint inhibitors have the same predictive power as untreated samples.

In summary, these results indicate that the model diversity measure in the article improves the response difference between the cancer genome atlas dataset and PD-1 inhibitory treatment of melanoma patients.

Figure 5 Using TCR model entropy to predict patient prognosis This is the main content of this article.
The author of the article will focus on the diversity of the T cell receptor library.

The author uses simulation data and TCGA data to evaluate diversity indicators from multiple perspectives, and obtains a corrected diversity evaluation model through statistical learning.
The author also analyzes the evaluation of T cell receptor diversity and survival and anti-PD-1 The relationship between treatment response.

Through this article, we can learn that a detailed and in-depth study of a valuable point may do a better job than a general analysis.

In 2021, follow us to become a better version of yourself