T Cell Receptor Repertoires Acquired via Routine Pap Testing May Help Refine Cervical Cancer and Precancer Risk Estimates

Abstract

Cervical cancer is the fourth most common cancer and fourth leading cause of cancer death among women worldwide. In low Human Development Index settings, it ranks second. Screening and surveillance involve the cytology-based Papanicolaou (Pap) test and testing for high-risk human papillomavirus (hrHPV). The Pap test has low sensitivity to detect precursor lesions, while a single hrHPV test cannot distinguish a persistent infection from one that the immune system will naturally clear. Furthermore, among women who are hrHPV-positive and progress to high-grade cervical lesions, testing cannot identify the ~20% who would progress to cancer if not treated. Thus, reliable detection and treatment of cancers and precancers requires routine screening followed by frequent surveillance among those with past abnormal or positive results. The consequence is overtreatment, with its associated risks and complications, in screened populations and an increased risk of cancer in under-screened populations. Methods to improve cervical cancer risk assessment, particularly assays to predict regression of precursor lesions or clearance of hrHPV infection, would benefit both populations. Here we show that women who have lower risk results on follow-up testing relative to index testing have evidence of enhanced T cell clonal expansion in the index cervical cytology sample compared to women who persist with higher risk results from index to follow-up. We further show that a machine learning classifier based on the index sample T cells predicts this transition to lower risk with 95% accuracy (19/20) by leave-one-out cross-validation. Using T cell receptor deep sequencing and machine learning, we identified a biophysicochemical motif in the complementarity-determining region 3 of T cell receptor β chains whose presence predicts this transition. While these results must still be tested on an independent cohort in a prospective study, they suggest that this approach could improve cervical cancer screening by helping distinguish women likely to spontaneously regress from those at elevated risk of progression to cancer. The advancement of such a strategy could reduce surveillance frequency and overtreatment in screened populations and improve the delivery of screening to under-screened populations.

Keywords: cervical cancer screening; cervical cancer surveillance; immune repertoire; machine learning; regression biomarker.

Figure 1

The timeline of prospective follow-up diagnosis collection. Residual, standard-of-care cytology samples were rescued at the time of scheduled discard, five weeks (35 days) after their collection and the completion of their use for clinical decision making. The associated cytology and hrHPV diagnoses constitute the patient’s reference diagnosis ( Table 1 ). These samples were used for HPV genotyping and TCR repertoire sequencing. The electronic health record was periodically checked for cytology, HPV, and pathology reports related to cervical cancer screening. Twenty-four women returned for follow-up screening 280-711 days after collection of the index sample ( Table 1 ). Seven of those women have returned for a second follow-up screening ( Table 1 ). The remaining women have not returned.

Figure 2

Model selection was conducted in earlier studies (45, 46, 50) to determine the best representation of TCR CDR3 sequences, such as the best k-mer size and the use of Atchley Factors rather than a DNA or AA sequence representation. The cervical cytology samples used in this study were not included in the model selection process. In this study, the previously identified best model was used, and the weights and bias term were fit by gradient descent in leave-on-out cross-validation.

Figure 3

(A) The age of women at the time the cervical sample was collected. Age was only available for 97 of the 103 women whose cervical sample was used for TCRB sequencing. Age is shown on the y-axis, and the result category, according to the cervical cytology and hrHPV test, is shown on the x-axis. A test for difference in the median value between result categories was conducted using a Kruskal-Wallis (KW) test followed by a Dunn’s post hoc multiple comparison test if the KW p value was ≤ 0.05. The Dunn’s multiple comparison-adjusted p values are reported when p ≤ 0.05. (B) The number of HPV types for which a sample tested positive using the Roche Linear Array. Only 96 of the 103 samples whose cervical sample was used for TCRB sequencing had sufficient DNA to also conduct HPV typing. The number of HPV types for which a sample tested positive is shown on the y-axis, and the result category, as defined in (A), is shown on the x-axis. Statistical tests were conducted and reported as in (A). (C) The proportion of samples in each result category, as defined in (A) that are positive for HPV16, HPV18, or both (left grouping) or are negative for both (right grouping), according to the Roche Linear Array. The proportion of samples is shown on the y-axis and indicated by the bar height. The result category is indicated by the bar color and legend. A Chi-square test of independence resulted in p < 0.0001.

Figure 4

(A) The number of HPV types for which a sample tested positive using the Roche Linear Array. Only 19 of the 24 samples used for TCRB sequencing and for which we obtained follow-up cytology and hrHPV test results had sufficient DNA to also conduct HPV typing. The number of HPV types for which a sample tested positive is shown on the y-axis, and risk as assessed by the follow-up cytology and hrHPV results is shown on the x-axis. A test for difference in the median value between the two groups was conducted using a Mann-Whitney test. p values are reported when p ≤ 0.05. (B) Age is shown on the y-axis. Other details are as in Panel (A). (C) The number of days between the first and second cytology and hrHPV test results is shown on the y-axis. Other details are as in Panel (A).

Figure 5

(A) The number of productive templates obtained for each sample is shown on the y-axis. Other details as in Figure 4A . (B) The number of productive rearrangements obtained for each sample is shown on the y-axis. Other details as in Panel (A). (C) The Fraction Productive of Cells Mass Estimate, referred to as the Estimated T cell Fraction, obtained for each sample is shown on the y-axis. Other details as in Panel (A). (D) The percentage of productive templates that correspond to a unique nucleotide sequence obtained for each sample is shown on the y-axis. Other details as in Panel (A). (E) The relative abundance of the largest clone obtained for each sample is shown on the y-axis. Other details as in Panel (A). (F) The Productive Clonality (1 – normalized Shannon Entropy) obtained for each sample is shown on the y-axis. Other details as in Panel (A).

Figure 6

(A) Model classification accuracy obtained during leave-one-out cross-validation in which all sequences from one sample were held out for validation. The y-axis shows the model-assigned probability that the follow-up sample would be associated with lower risk than the index sample. The x-axis enumerates patients, and the legend indicates how the index sample was classified for model training. (B) Illustration of the classifier weights after fitting the model to all 24 samples. For each of the five Atchley factors, the weights are shown for the four residue positions. The weight for 4-mer relative abundance is also shown. Positive weight values are shown pointing up, and negative weight values are shown pointing down. The length of the arrow corresponds to the weight’s magnitude. (C) The top-scoring 4-mer for each sample is shown in the context of the CDR3s that contain it.