Following Institutional Review Board approval, samples were collected from 11 adult female patients ( Table 1 ) diagnosed with unilateral breast cancer-related lymphedema and treated at the Lymphedema Clinic in the Plastic and Reconstructive Surgery Service at Memorial Sloan Kettering Cancer Center in New York, NY, USA. Additional patient characteristics are included in Supplementary File 1 .

All animal procedures were performed in accordance with the protocol approved by the Institutional Animal Care and Use Committee at Memorial Sloan Kettering Cancer Center. Female C57BL/6J (wild type) mice were purchased from the Jackson Laboratories and experiments were started at age 8–10 weeks. All mice were maintained in a pathogen-free, temperature- and light-controlled environment and provided with a normal chow diet and freshwater.

The corresponding epitope sequence recognized by each unique T cell clone sequenced in the lymphedema samples was determined using a validated online TCR structural repertoire database (17). The database contains a curated collection of TCR structures from the Protein Data Bank and ranks the corresponding epitope sequence of each TCR based on the PAM30 scoring matrix. The PAM30 index is a bioinformatics metric that matches protein sequences based on the lowest number of amino acid substitutions. The lower the PAM30 index, the greater the similarity in sequence alignment. Using the amino acid construct of the TCR generated by the immunoSEQ™ analyzer for the top 5 unique clones detected in each lymphedema sample, the corresponding epitope sequence was determined using the TCR3d database.

The NCBI BLAST was used to search and identify the predictive antigen corresponding to each unique TCR sequenced in the lymphedema samples. BLAST identifies regions of similarity between biological sequences (18, 19). The program compares nucleotide or protein sequences to sequences in a curated database and organizes the queried amino acid sequence pairs based on the best alignment. The search was limited to human organisms (Homosapien sapiens). BLAST then generates a description of the predicted antigen based on matched sequences, producing significant alignments. It then ranks the antigen based on the percent identity (the maximum and total alignment score for the database sequence), the query coverage (the percentage of the query sequence that is covered by the database sequence alignment), and the expect value (E-value). The E-value is the number of expected hits with similar scores that could be found by chance. The lower the E-value, the more likely the query-pair alignment is significant.

Histology was conducted as described above. In brief, human tissue samples were embedded in paraffin and sectioned into 5-µm sections. The following anti-human antibodies were used for staining: CD4+ (1:1000; cat#AF379NA; R&D Systems), CD45RO (1:400; cat#MA511532; Thermo Fisher Scientific), and IR (1:1000; cat#AB137747; Abcam). Tissues were rested overnight with the appropriate primary antibody at 4°C, washed with PBS, and incubated with the corresponding secondary antibody conjugates (TRITC, Cy3, and FITC) at room temperature for 5 hours. A 4,6-diamino-2-phenmylindole (DAPI; cat# D4571, Molecular Probes) stain was performed to identify nucleated cells. Sections were scanned using a Mirax slide scanner (Zeiss), and ImageJ software (National Institutes of Health) was used to quantify CD4+/CD45RO+ and CD4+/CD45RO+/IR+ cells per DAPI+ cells in a 20-µM area.

Statistical analysis was performed using GraphPad Prism v8 (GraphPad Software). We used a two-tailed paired t-test to compare normally distributed paired data and a two-tailed unpaired t test to compare normally distributed unpaired data. Nonparametric testing with Wilcoxon signed-ranked test was used for data that were not normally distributed. Significance was defined as P<0.05.

Patients with secondary lymphedema were recruited from the Lymphedema Clinic in the Plastic and Reconstructive Surgery Service at Memorial Sloan Kettering Cancer Center ( Table 1 , Supplementary File 1 ). For TCR sequencing, genomic DNA was isolated from full-thickness punch biopsies of normal and lymphedematous skin in patients with International Society of Lymphology (ISL) stage II lymphedema ( Figure 1A ). TCR sequencing of skin samples revealed 1.0–1.5 thousand unique sequences per paired sample, comprising the majority of clones sequenced ( Figures 1B, C ). The Simpson clonality index was used to measure the diversity of the T cell repertoires in a sample ( Figure 1D ). Measures closer to 0 indicate a perfectly diverse repertoire or no duplication of clones, whereas measures closer to 1 indicate a monoclonal population with one clone dominating a repertoire (Adaptive Biotechnologies). Oligoclonal populations fall within this range and are recognized as repertoires with significantly expanded clones, which may indicate a response to putative antigens.

In our study, T cells sequenced in lymphedema skin samples demonstrated an increased degree of oligoclonal expansion compared with clones detected in matched normal skin samples ( Figure 1D , p<0.0001). CDR3 length distribution is an additional method used to determine repertoire diversity. The average CDR3 lengths in this study did not differ significantly between normal and lymphedema samples. Additionally, both groups were observed to fit well within a Gaussian distribution curve (R²≈1; Figures 1E–G ). Variable (V) gene recombination was also identified in subsets of sequenced T cells. V gene usage appeared to be conserved between TCRs in normal and lymphedema samples. Among the 30 high-frequency expressed V genes (frequency >1.0%), the greatest variation was observed in TCRβV-20, although this was not found to be significant between normal and lymphedema samples ( Figure 1H ).

The Morisita overlap index (MOI) was used to assess the similarity of TCR repertoires between paired normal and lymphedema skin samples, as well as among patients with the disease ( Figure 2A ). The MOI ranges from zero to one; the closer the index is to one, the more similar the repertoires, and the closer the index is to zero, the more dissimilar the repertoires. Some patients demonstrated high TCR similarity between normal and lymphedema skin (MOI>0.5; patients 1 and 6), whereas some patients had very dissimilar repertoires (MOI<0.2; patients 2, 4, 5, 9, 10 and 11). No overlap was observed among patients in our cohort, suggesting that TCR repertoires in lymphedema are patient-specific. We then compared the MOI to L-Dex (ImpediMed), a lymphedema measuring system that compares fluid levels in the affected and non-affected limb (22), and volume differential ( Figures 2B, C ) and found no significant correlation with TCR similarity (p>0.05).

To better delineate clone distribution within a repertoire, we identified the frequencies of each donor’s top 5 clones, which were significantly expanded in the sample relative to other clones ( Figure 2D ). Notably, clones are defined by their unique amino acid rearrangements. Single clone expansion among top clones was commonly observed in lymphedema skin (>50% frequency, x-axis), whereas top clones in normal skin tended to have a more even distribution, arguing for a strong T cell response to certain antigens in lymphedema skin. A pair-wise scatter plot was similarly used to visualize the relative abundance of shared clones between normal and lymphedema samples ( Figure 2E ). Clone frequency in normal tissue alone was plotted on the y-axis, clone frequency in lymphedema tissue alone was plotted on the x-axis, and the clone frequency in both normal and lymphedema tissues were displayed as individually colored dots. We observed that lymphedema skin had the highest frequency of shared clones, suggesting that T cells encounter antigens in lymphedema skin that are absent in the contralateral normal limb ( Figures 2F–H ).

Specific TCR target antigens were predicted for top clones sequenced from lymphedema skin using a two-stage immunoinformatics approach. Briefly, the amino acid identity of the TCR was identified through immunosequencing, and a list of antigenic major histocompatibility complex-2 (MHC-II) epitopes was generated using a validated, online TCR structural database. Corresponding MHC epitopes were ranked by PAM30 score, with a lower score indicating fewer amino acid rearrangements of the input sequence or best match. A PAM30 score of less than 90 was used to determine the most accurate MHC-II antigen-TCR pair.

To search and identify candidate antigens for sequenced TCRs, we used the National Center for Biotechnology Information (NCBI) Basic Local Alignment Search Tool (BLAST). Table 2 summarizes the representative antigens and the number of unique TCRs recognizing those antigens. Details on this search are provided in the methods section. Interestingly, human insulin was the most common antigen detected by frequently propagated clones in lymphedema and was identified by unique TCRs in 10 of the 11 samples sequenced. Other representative antigens, detected at lower frequencies, included Klebsiella bacterial antigen, gluten plant antigen, and the HIV viral antigen. Supplementary File 2 lists several parameters for this analysis, including predictive antigen, percent identity, and E-value. We chose to move forward with investigaing the role of insulin, since it was the most representative antigen detected of the patient cohort.

Immunofluorescence staining provided further evidence of significantly higher number of antigen-responsive T cells, as illustrated by the accumulation of activated T cells (CD4+/CD45RO+) in lymphedema skin compared with matched normal skin ( Figures 3A, B ). A significant number of these activated T cells are IR+ in lymphedema skin compared with matched normal skin ( Figure 3C ). Similarly, we observed higher frequencies of antigen-activated CD4+ T cell populations in lymphedema liposuction fluid (CD4+CD45RO) than in autologous blood on flow cytometry, supporting a local T cell response to antigens in lymphedema that is absent systemically ( Figure 3D ; Supplementary File 3 ). CD4+ T cells isolated from lymphedema liposuction fluid stimulated ex vivo with a peptide pool from human insulin also demonstrated increased frequencies of antigen-activated T cells (CD4+CD45RO+CD154+) compared with autologous T cells from blood in a patient with normal BMI ( Figure 3E ; *p<0.05). A second patient with an overweight BMI demonstrated reactive T cells both systemically and in lymphedematous tissue ( Figure 3F ; *p<0.05, **p<0.005; Supplementary File 4 ).

T cell oligoclonality, as determined by the Simpson clonality index, was similarly demonstrated in a tail lymphatic excision mouse model of lymphedema ( Figures 4A, B ). CDR3 length distribution exhibited a skewed pattern in the surgical group compared with the sham control, suggesting increased oligoclonality ( Figure 4C ). Comparison of TCRβV gene usage between the sham and surgery groups demonstrated significantly increased gene usage of TCRβV 19–01 and TCRβV 24–01 in the sham group ( Figure 4D ; p=0.05). Antigen-activated effector CD4+ T cells were examined in a popliteal lymph node dissection (PLND) model of lymphedema. Two factors influenced our decision to use the PLND model for this experiment. First, T cell yield from tail skin and associated draining lymph node (LN) sites (sacral nodes) was consistently low in our repeat experiments. Second, the PLND model improves T cell yield from draining inguinal LNs with increased cell viability, as it eliminates the need for additional digestion steps required in the tail skin model. Figure 4E illustrates our experimental design. In brief, CD4+/CD44+/CD62L- effector T cells isolated from draining inguinal LNs of the PLND model were stimulated in vitro with or without insulin. Effector T cells isolated from the PLND group demonstrated significant upregulation of CD154+, an antigen-specific marker, compared with effector cells either in the absence of insulin or effector cells stimulated with insulin in sham controls ( Figures 4F, G ; Supplementary File 5 ).

To test additional sources of insulin and the specificity of insulin as antigen, we designed an experiment to compare T cells from blood and T cells from lymphedematous tissue (lipoaspirate), incubated in the presence or absence of anti-insulin receptor (IR) antibody. We assessed the samples in triplicate per condition with flow cytometry and gated for activated memory T cells ( Figure 5A ). No significant differences were present between groups in stimulated T cells isolated from blood, either incubated without or with anti-IR antibody ( Figures 5B, C ). T cells from lipoaspirate from the same patient were found to be significantly responsive to B 9–23 peptide and whole insulin ( Figure 5D ; p<0.05), whereas these differences were no longer significant when incubated in the presence of anti-IR antibody ( Figure 5E ). These results suggest that T cells may be activated through both insulin receptor binding as well as via binding by insulin peptide-loaded MHC class II molecules. Taken together, a graphical abstract ( Graphical Abstract ) is provided.