Integrated, high-dimensional analysis of CD4 T cell epitope specificities and phenotypes reveals unexpected diversity in the response toMycobacterium tuberculosis

17 18 Immunity to Mycobacterium tuberculosis (Mtb), like many pathogens, is encoded jointly by the 19 antigen specificities and functions of responding CD4 T cells. However, these features span a 20 large two-dimensional possibility space – defined on one axis by the Mtb proteome, and on the 21 other by the T cell transcriptome – that exceeds the dimensionality of existing technologies. Here 22 we present an approach (“CRESTA”) that combines highly -multiplexed DNA-barcoded epitope 23 probes, single cell sequencing, and clonal analysis of T Cell Receptors (TCRs) to robustly detect 24 rare antigen-specific CD4 T cells across hundreds of epitopes simultaneously and reveal their 25 transcriptome-wide phenotypes. By comprehensively assaying known epitopes in Mtb-infected 26 participants, we reveal polyclonal and multi -epitope responses across a spectrum of 27 differentiation states, uncover previously -unobserved phenotypic diversity within and between 28 epitopes, and increase the total number of known Mtb epitope-mapped TCRα:βs by ~8-fold. We 29 expect CRESTA to enable high-dimensional analyses of CD4 T cell responses in various settings, 30 including infection, cancer, autoimmunity and allergy. 31 32 33 34

35 36 Helper T cell immunity resides in populations of CD4+ cells that have clonally expanded in 37 response to particular MHC class II -bound peptide antigens, and differentiated to acquire 38 specialized effector functions 1–4. Although these features together theoretically encode an 39 individual’s state of immunity, they are difficult to measure in an integrated and comprehensive 40 way because existing technologies do not scale well to the large diversity of possible CD4+ T cell 41 antigen specificities and functions. Traditional assays for the identification of antigen -specific T 42 cells include those that measure peptide -stimulated cytokine production (e.g., immunospot 43 assays and flow cytometric detection of intracellular cytokines5,6) or upregulated markers7, as well 44 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint as assays that use fluorescently - or isotopically-labeled peptide:MHC probes to directly detect 45 antigen-binding T cells 8,9. Although widely useful, these approaches have a limited ability to 46 multiplex across antigens and phenotypic markers, meaning that available sample volumes and 47 cell numbers quickly become limiting, typically precluding comprehensive analyses10. 48 49 More recently, DNA-barcoded peptide:MHC probes have been used to bypass the spectral limits 50 of cytometric approaches (which generally peak at 10 –50-plex), and have enabled CD8 T cell 51 responses to be resolved across a multiplexity of up to ~1000 distinct peptide:MHCs 11. While it 52 represents a major advance, this system has two key limitations: (i) by itself, it does not enable 53 the simultaneous capture of T cell phenotypic information and (ii) it has not yet been adapted to 54 the analysis of CD4 T cells. The latter likely reflects a combination of factors, including: (i) the 55 lower frequencies of epitope -specific CD4 T cells, (ii) the generally lower affinity of CD4 T cell 56 receptors (TCRs) for their MHC class II -restricted antigens, and (iii) greater challenges in 57 identifying MHC class II:peptide binding pairs and constructing the corresponding probes. Indeed, 58 despite considerable efforts to enable the multiplexed detection of CD4 T cell antigen -specificity 59 9,12–18, the greatest reported scale at which epitopes have been resolved simultaneously in a 60 primary sample using any approach is 6 -plex19. A powerful assay developed recently uses an 61 elegant cell interaction reporter system to enable the genome -scale discovery of CD4 T cell 62 antigens across 100,000s of candidate epitopes 20, however it requires substantial genetic 63 engineering of T cells and therefore does not enable the analysis of primary and/or polyclonal 64 samples (currently it has been limited to the mapping of TCRs). Overcoming the barriers to highly-65 multiplexed analysis of CD4 T cell epitopes in primary samples would represent an important 66 advance with widespread applications across the many settings in which helper T cell immunity 67 is implicated. 68 69 The opportunities that would be enabled by high-dimensional assays of CD4 T cell specificity and 70 phenotype are exemplified by Mycobacterium tuberculosis (Mtb), a pathogen responsible for 1.3 71 million global deaths in 2022, and for which an effective vaccine would have a major impact21. No 72 vaccine for Mtb has been approved since BCG (~100 years ago), although there are several 73 experimental formulations under development 22–24. A major challenge in the field has been 74 selecting which Mtb antigens (from among >4000 proteins) to include in a vaccine, and what T 75 cell functions it should elicit. Previous attempts to define protective immunity to Mtb have shown 76 mixed success and have focused on diversity in one dimension at a time – eg using proteins / 77 lysates to identify diverse CD4 T cell states (e.g. cytokine polyfunctionality25,26), or IFNγ production 78 in response to diverse sets of peptides predicted to bind Human Leukocyte Antigen (HLA) 79 proteins27,28. 80 81 More recently, sequence clustering approaches have been used to analyze the TCRs of T cells 82 enriched for Mtb specificity and reveal public TCR groups29,30, including groups that are positively 83 and negatively associated with disease progression 31. Importantly, these findings indicate that 84 particular antigen-specificities may be important in protection from Mtb. However, the analysis of 85 TCRs alone does not enable the efficient identification of the cognate Mtb epitopes (which have 86 been mapped for only a minority of TCR groups), nor does it capture the corresponding T cell 87 phenotypes. These two missing features – antigen-specificities and phenotypes – are critical 88 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint attributes if a correlate of protection is to be maximally actionable in guiding next -generation 89 vaccine design. 90 91 Here, we present an assay that can measure reactivity across 100s of HLA class II:peptide 92 antigen specificities simultaneously within a single PBMC sample, and associate each epitope -93 specific response with a transcriptome -wide T cell phenotype. We use this assay to generate a 94 comprehensive portrait of the CD4 T cell response to Mtb in infected participants – molecularly 95 resolved across 100s of epitopes and 1,000s of transcripts – and reveal previously undescribed 96 polyclonality, phenotypic diversity and TCR sequence features within this response. 97 98 99

100 101 Simultaneous, highly -multiplexed measurement of CD4 T cell epitope -specificity and 102 transcriptional state enabled by clonal analysis 103 104 To enable the integrated, high -dimensional analysis of CD4 T cell epitope -specificities and 105 transcriptional states, we adapted an approach developed previously for the highly -multiplexed 106 analysis of CD8 T cell epitope -specificities11. That assay involved generating DNA -barcoded 107 probes of MHC class I:peptide complexes multimerized onto dextran backbones, incubating T 108 cells with pools of probes, using fluorescence to sort probe -binding cells, and analyzing binding 109 by deep sequencing of the DNA barcodes. From this starting point, we introduced three main 110 modifications, each of which was critical to enable our analysis of CD4 T cell responses. First, in 111 place of MHC class I, we used MHC class II:peptide complexes, prepared by exchanging peptides 112 of interest into HLA reagents bearing peptides tethered by a protease-cleavable linker 32. Second, 113 we used single cell sequencing in place of bulk sequencing, to enable T cell transcriptional states 114 to be detected and linked to epitope specificity. Third, instead of cell sorting, we introduced an 115 antigen-specific clonal expansion step to overcome the low frequencies of circulating epitope -116 specific CD4 T cells, and then leveraged this expansion to boost analytical power by developing 117 a “pseudobulk” approach in which we aggregated single cell data at the clonal level. Together, 118 we refer to this workflow as the Clonally-Resolved Epitope-Specificity and Transcriptome Assay 119 (CRESTA) (Figure 1). 120 121 We applied CRESTA to study the T cell response in HLA -typed ( Supplementary Table 1 ), 122 Quantiferon-positive, HIV -negative participants from a cohort in Western Kenya , sampled 123 following recent (≤ 3 months) household exposure to active pulmonary tuberculosis 33. To 124 represent the known antigen space, we used the Immune Epitope DataBase (IEDB) to 125 comprehensively identify a total of 206 Mtb peptide:HLA pairs known to generate human T cell 126 responses in the context of 4 class II restriction elements that were prevalent in this cohort (HLA-127 DQB1*06:02, HLA-DRB1*15:03, HLA-DRB1*11:01, or HLA -DRB5*01:01). We also included 32 128 peptide binders for a fifth prevalent HLA – HLA-DRB1*01:02 – identified in an in vitro peptide:HLA 129 binding assay. Finally, we selected a total of 85 additional peptide:HLA pairs as controls, which 130 included CLIP-tethered (uncleaved) versions of each HLA, Mtb epitopes restricted by additional 131 alleles (not expressed by the participants of interest), and IEDB epitopes from Influenza A virus 132 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint (see Supplementary Table 2 for a complete list of peptide:HLA pairs). We expanded antigen -133 specific T cells from 5 – 10 x 10 6 PBMCs for 7-9 days using a master pool of all peptides. We 134 then analyzed two aliquots of each sample: the first (“antigen-specificity aliquot”) was stained with 135 pooled DNA -barcoded peptide:HLA class II multimers corresponding to the participants’ HLA 136 types, the second (“gene expression aliquot”) was not stained but instead briefly restimulated with 137 PMA/ionomycin to enhance antigen-responsive gene expression. This dual aliquot strategy was 138 motivated by initial experiments (not shown) indicating that PMA/ionomycin greatly enhanced our 139 ability to resolve T cell states, but led to diminished multimer binding, likely due to downregulation 140 of the TCR34. Both aliquots of each sample were analyzed by single cell sequencing using the 141 10X Genomics Chromium platform to recover multimer barcode counts, transcriptome -wide 142 mRNA abundance data and paired, full -length TCRα:β sequences on 1,000s-10,000s of single 143 cells. We then informatically matched TCR sequences between the two aliquots to identify cells 144 from common clones and thereby unite antigen-specificity and gene expression data for each. 145 146 In a representative participant (ID:40059), we recovered data on a total of 28,129 cells –17,312 147 and 10,817 for the antigen -specificity and gene expression aliquots, respectively – the former 148 aliquot having been stained with a pool of 48 multimers (containing DRB1*15:03, DRB1*11:01 or 149 DQB1*6:02, all expressed by the participant). We reasoned that TCRα: β sequences could be 150 used to assign single cells into clonal families – each representing the progeny of a single Mtb-151 specific precursor – that each share a common epitope -specificity and differentiation state. To 152 test this hypothesis, for each aliquot we organized individual cells into clones based on shared 153 TCRα:β sequences, filtered on clones that contained ≥3 individual cells in each aliquot, and used 154 a Kruskal -Wallis test to quantify the extent to which signal from peptide:HLA multimers and 155 transcripts was driven by clonal identity ( Figure 2). This analysis included a total of 170 clones, 156 each of which contained 6 – 616 individual cells across the two aliquots (median=45) (Figure 2a). 157 158 Consistent with the hypothesis, we observed strong partitioning of multimer binding by clone. In 159 particular, we identified 10 of the 48 multimers to be binding in this participant (p -values ranging 160 from ~1e-5 to 1e -298), of which 3, 4 and 3 were restricted by DRB1*15:03, DRB1*11:01 and 161 DQB1*6:02, respectively (Figure 2b). To identify the individual binding clones for each of these 162 epitopes, we applied one -tailed Wilcoxon tests post -hoc to each binding multimer, in which we 163 compared the binding of each individual clone to the overall distribution (Figure 2c). This analysis 164 identified a total of 54 binding clones (32% of all clones), ranging from 1-19 per multimer, revealing 165 extensive polyclonality in the response to individual Mtb epitopes. Binding clones were non -166 overlapping between multimers, with the exception of 3 pairs of multimers that featured 167 overlapping peptides restricted by the same HLA and shared overlapping patterns of clonal 168 recognition (discussed further below), supporting the fidelity of the process. Moreover, the TCR 169 sequences of epitope -binding clones showed significant homology within epitopes and 170 recapitulated known public motifs, as discussed further below. 171 172 Similarly, we observed strong partitioning of transcript abundance across clonotypes: applied to 173 the same set of 170 clones, the Kruskal-Wallis test detected 155 genes at a Bonferroni-corrected 174 threshold of p<0.05 and fold -difference threshold of 10 ( Figure 2d ). These genes, which we 175 hereafter refer to as Clonal Differentiation Genes (CDGs), were strongly enriched for genes 176 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint implicated in helper T cell function, and prominently featured cytokines and chemokines known 177 to be produced by particular subsets, including IFNG, IL17A/F, IL1A, CCL3, CCL4, CCL5, GNLY, 178 CXCL8, CCL3L1, CCL4L2, CCL3 and IL22. We also observed patterns that were consistent with 179 the maintenance of distinct differentiation states across clones, for example the anti -correlated 180 expression of the Th1 chemokine CCL4 and the Th17 cytokine IL17A ( Figure 2e). Together, 181 these results establish CRESTA as a method for robustly resolving antigen -specific CD4 T cells 182 across many antigen-specificities simultaneously from a single PBMC sample, and associating 183 each with a transcriptome-wide gene expression state. 184 185 186 Comprehensive, hypothesis-free phenotyping of the Mtb-specific response using CRESTA 187 188 To perform a deep, multi-participant analysis of the epitope-specific CD4 T cell response to Mtb 189 infection, we performed CRESTA on a total of 5 LTBI individuals (ID:40059, analyzed above, and 190 4 additional participants: IDs: 30128, 30133, 30129, 30168) from the Kenyan cohort. We stained 191 cells from each participant with multiplexed probesets matching their HLA types (32 -173 192 multimers per participant, across 5 distinct HLA class II alleles), and recovered single cell 193 sequencing data on a total of 106,673 cells (4,400 -14,496 and 7,866-13,475 per participant for 194 the antigen-specificity and gene expression aliquots, respectively). In total, this yielded 1,496 195 evaluable clones (169–454 per participant). 196 197 Consistent with our observations above, epitope staining (analyzed as described in Figure 2b) 198 partitioned strongly by clonal identity and revealed polyclonal responses across all participants. 199 The highest dimensional staining was 173 -plex in participant ID:30168, in which we observed 200 robust binding across 15 distinct epitopes (Figure 3a), with a total of 126 epitope-binding clones. 201 Analysis of these profiles revealed 7 epitope clusters (each containing 1-5 epitopes) within which 202 there was extensive sharing of binding clones ( Figure 3b). All 7 clusters were internally HLA 203 matched and comprised sets of overlapping peptide sequences, indicative of a common core 204 peptide epitope in each (but with possible, clone -specific additional contributions from 205 flanking/polymorphic residues). Importantly, no clones were positive for epitopes from more than 206 1 cluster, indicating high assay specificity. 207 208 Across all 5 participants, we detected a total of 19 epitopes that showed significant binding to ≥1 209 clone in ≥1 participant (Figure 3c). Based on analysis of shared clones, these epitopes partitioned 210 into 11 clusters, each of which exclusively contained members sharing a common HLA and 211 overlapping peptide sequences, and we again found no examples of clones binding across 212 clusters. For simplicity, in all downstream analyses, we refer to these clusters as “epitopes”, even 213 though some comprise signal from several multimer probes that overlap a common core epitope. 214 In total, we detected 293 epitope-specific clones in the antigen specificity aliquot, ranging from 4–215 139 per participant and 4–93 per epitope. (Figure 3d). This number already represents nearly an 216 order of magnitude increase over the total number of TCRα: β sequence pairs mapped to 217 individual HLA II-restricted Mycobacterial epitopes in all prior studies to date (36 TCRs identified 218 in 5 studies from 2009-2023: IEDB, queried 6/29/2024). Finally, across the range of 32-173-plex 219 staining, we saw no evidence that the intensity of multimer probe binding was inversely correlated 220 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint with the number of probes pooled for any given participant ( Figure 3e), indicating that we have 221 yet to approach the limit of assay multiplexing. 222 223 Next, we used the clonally resolved expression analysis described above in Figure 2d, to identify 224 CDGs from the “gene expression aliquot” from each participant, and again observed these to be 225 strongly enriched for known CD4 T cell differentiation genes. For each CDG, we collapsed 226 individual cell expression values into per-clone medians, and then used UMAP to generate a 2 -227 dimensional representation of the CDG-wide expression profile of each clone. This allowed us to 228 cluster clones according to gene expression in a hypothesis -free fashion ( Figure 4a shows 229 participant ID:30133 as an example), and revealed distinct clusters whose expression profiles 230 corresponded to those of naive, Th1, Th2 and Th17 cells. Beyond canonical gene expression 231 patterns (i.e. IL2RA –, IFNG+, IL4+, IL17+, in naive, Th1, Th2, Th17 cells, respectively), 232 hypothesis-free correlation analysis across all CDGs revealed that these clusters were each 233 characterized by broad state -specific expression patterns, which included previously -described 234 genes (e.g. TNF, CCL3, CCL4, HOPX in Th1 clones; IL3, IL5, IL13, GATA3, IL17RB in Th2 235 clones; IL17F, IL22 RORC, CCL20, NR4A2 in Th17 clones) as well as genes not previously 236 associated with these T cell subsets (e.g. PTGS2 in Th2 clones; TGIF1 in Th17 clones) ( Figure 237 4b). We also observed clones with a Treg -like expression pattern, characterized by high 238 expression of FOXP3 and CTLA4, although these did not form a distinct cluster. 239 240 Among a total of 1,496 clones analyzed for gene expression across the 5 participants, 1,344 241 (90%) could be assigned a differentiation state; composed of Th1 (74%), Th2 (2%), Th17 (9%), 242 Treg (2%) or naive (4%) subsets. The average clone sizes across these diverse states differed 243 significantly, with Th1 clones being the largest (median = 10 cells), followed by Th17, Th2, Treg 244 and then naive clones (median = 3 cells) (Figure 4c). Of the 1,496 clones, we were able to assign 245 Mtb antigen specificity to 243 (16%), across the 11 epitopes described above in Figure 3 (50 of 246 the 293 clones described in Figure 3d were not evaluable because although they had ≥3 cells in 247 their respective antigen -specificity aliquot, they had <3 cells in the gene expression aliquot). 248 Comparing these epitope-mapped clones to the total population, we observed a marked further 249 skewing towards the Th1 state (97% of multimer -binding clones), at the expense of all other 250 subsets (Figure 4d). As expected, no naive clones were found to be multimer binding. For this 251 dataset, we therefore conclude that CD4 T cells specific for Mtb-specific epitopes are strongly 252 biased towards the Th1 subset. 253 254 255 256 257 258 259 Analysis of phenotypic heterogeneity within and between Mtb epitope-specific T cell responses 260 261 Despite the predominance of Th1 clones, we also detected rare Mtb-specific clones with Th17 262 (n=2) and Treg (n=2) phenotypes. A closer examination of these clones confirmed the expression 263 of a range of transcripts corresponding to their respective states, and revealed that these patterns 264 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint were broadly consistently across the individual constituent cells of each clone ( Figure 5a ). 265 Notably, each of these clones existed alongside others specific for the same Mtb peptide:HLA 266 epitope in the same participant, including the DRB1*15:03_HVSFVMAYPEMLAA 
(PE13) and 267 DRB1*01:02_NIAFASGFRAIN (Rv0293c) epitopes. Together, this represents the first evidence 268 of which we are aware that T cells recognizing the same Mtb epitope in the same individual can 269 adopt highly-divergent differentiation states. 270 271 We next tested the hypothesis that, even within a Th1 dominated response, T cells recognizing 272 different epitopes can have divergent gene expression programs. To address this in a systematic 273 way, we considered all cases in which we observed ≥2 distinct reactive epitopes within a 274 participant, each of which was recognized by ≥3 clones. These criteria focused the analysis to a 275 total of 230 clones across 9 epitopes in 3 participants. We began by identifying CDGs as above, 276 yielding a total of 81, 118 and 139 genes in participants ID:40059, ID:30129 and ID:30168, 277 respectively. Focusing on these subsets of genes, we then used Kruskal-Wallis tests to compare 278 the median gene expression of each clone across epitopes, measuring whether expression was 279 correlated with epitope specificity. We detected strong correlations between gene expression and 280 epitope specificity within all 3 participants, with p-values up to 1e-5 (Figure 5b). At thresholds of 281 p2, we identified 11, 17 and 27 epitope -linked genes in the 3 282 respective participants. These sets overlapped significantly: 5 genes were common to ≥2 283 participants – IFNG, CCL3L1, GNLY, CCL4, LTB – all of which have known roles in Th1 effector 284 function. 285 286 To visualize how clones specific for different epitopes cluster in gene expression space, we used 287 UMAP to render the CDG -wide expression profiles of each clone in 2 dimensions ( Figure 5c, 288 upper row). Rather than a random distribution of clones specific for different epitopes (indicated 289 in different colors) across this space, in each participant we observed clustering of clones 290 according to their epitope specificity, consistent with our Kruskal -Wallis analysis. Prominent 291 among these were clusters of: (i) DRB1*15:03_HVSFVMAYPEMLAA -specific clones in 292 participant ID:40059 enriched for GNLY expression (green dots in Figure 5c , upper left), (ii) 293 DRB1*11:01_VDLAKSLRIAAKIYS-specific clones in participant ID:30129 enriched for CCL3L1 294 expression (yellow dots in Figure 5c, upper middle), and (iii) DQB1*06:02_EQQWNFAGIEAAA-, 295 DRB1*15:03_AAVVRFQEAANKQK- and DRB1*15:03_HVSFVMAYPEMLAA -specific clones 296 with high, high and low IFNG expression, respectively (cyan, red and green dots in Figure 5c, 297 upper right). Together these results reveal that different epitopes can program diverse Th1 gene 298 expression states within the same Mtb response, and that these are characterized by the 299 differential expression of important effector genes. 300 301 302 Analysis of TCR sequence clustering within the anti -Mtb response across epitopes and 303 participants 304 305 An alternative approach, to the one described here, for the multiplexed detection of epitope -306 specific T cell responses has been the identification of clusters of homologous TCR sequences, 307 followed in some cases by screening assays to map their specificities 30. Applied to Mtb, this 308 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint approach has been used to identify reactivities associated with protection from disease 309 progression31. However, it remains unclear how TCRs identified by such homology -based 310 approaches map onto the overall epitope-specific response. 311 312 The identification here of an unprecedented number of TCR α:β pairs mapped across a panel of 313 Mtb peptide:HLA epitopes presented an opportunity to assess the degree of TCR sequence 314 homology within epitope-specific T cell responses to Mtb in a minimally-biased way. To enable 315 robust statistical power, we focused on the 8 HLA:peptide epitopes for which we detected ≥10 316 multimer-binding clones across participants – with the requirement that each clone have a single 317 α and single β chain (which excluded 19% of clones, for which we sequenced 1 or >2 chains), to 318 eliminate ambiguity about the chains involved in epitope binding – yielding a total of 206 TCRα:β 319 pairs. We then used TCRdist 35 to perform comprehensive pairwise sequence similarity 320 measurements within each epitope. To rigorously quantify significance, we constructed a null 321 distribution of >1e12 pairwise TCRdist measurements on a large set of TCR α:βs randomly -322 sampled from an unenriched repertoire. We used this distribution to transform TCRdist measures 323 into p -values and applied Benjamini –Hochberg adjustment for the number of pairwise 324 measurements performed for each epitope. 325 326 We detected significant TCR homologies for all 8 of the Mtb epitopes analyzed – evident both as 327 the formation of clusters at an adjusted threshold of p<0.1 ( Figure 6a ), as well as overall 328 deviations from the null p -value distribution ( Figure 6b). However, the extent of this clustering 329 differed markedly by epitope, encompassing up to 10/14 (71%) of TCRs specific for the esxH 330 epitope DRB1*11:01_HEANTMAMMARDTAE, and as few as 2/13 (15%) for the CFP-10 epitope 331 DQB1*06:02_ISTNIRQAGVQYSR. The PE13 epitope DRB1*15:03_HVSFVMAYPEMLAA, for 332 which we detected the largest number of clustered clones overall, was notable for a large and 333 unusually tight cluster of highly-homologous TCRs that was dominated by clones from two of the 334 three participants that reacted to that epitope. For these TCRs, we observed a precise match to 335 V/J segment usage ( α: TRAV25/27, TRAJ52/40, β: TRBV9) and CDR3 motifs ( α:CAG***S/TY, 336 β:CASSVAL*G) described previously30, further supporting the fidelity of our multiplexed epitope-337 specific assay (Figure 6c). Overall, across the 8 epitopes, 99/206 (48%) epitope -specific TCRs 338 were part of detectable homology clusters. Importantly, however, this degree of clustering was 339 highly-dependent on filtering for single epitope binding to focus the number of TCR comparisons. 340 When testing an oligoclonal scenario in which epitope -specific TCRs were diluted to 5% with 341 random TCRs – designed to simulate more traditional enrichment methods where TCRs specific 342 for >20 epitopes may be analyzed together – detectable clusters remained in only 5 of the 8 343 epitopes, and comprised just 24% of all TCRs. 344 345 Many of the observed clusters included TCRs from different participants, underscoring their public 346 nature, and indeed for 5 of the 7 epitopes that contained clones from >1 participant, we detected 347 no difference between the distributions of intra -participant v inter -participant TCR distances 348 (0.16<p<1 by Kolmogorov –Smirnov test), indicating that features recognizing the common 349 peptide:HLA epitope often predominate over any participant -to-participant differences in TCR 350 repertoires. A striking exception to this was the DRB1*15:03_HVSFVMAYPEMLAA epitope: 351 whereas the TCRs from participants 40059 and 30129 clustered indistinguishably (red v blue dots 352 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint in the left box of Figure 6a: p(Inter v Intra | 40059v30129) = 0.69), the TCRs from 30168 (green 353 dots) were an outlier (p(Inter v Intra | 30168v40059) = 0.001; p(Inter v Intra | 30168v30129) = 354 0.0005), driven by a dearth of 30168 epitope -specific TCRs in the epitope’s central homology 355 cluster, and a correspondingly greater frequency of non-clustered TCRs. Intriguingly, this outlier 356 participant (30168) is a homozygote for the restricting allele (DRB1*15:03), and in fact represents 357 the only case among the 7 DRB1-restricted epitopes for which the participant was homozygous. 358 359 Together, these findings indicate that public TCR sequence features arise in the response to all 360 or most Mtb epitopes: occasionally (e.g. for the PE13 epitope) these are prominent and can 361 characterize the majority of responding TCRs, however for most epitopes, sequence homologies 362 are relatively rare and likely only detectable when the analysis is focused on TCRs from epitope-363 specific cells (e.g. by filtering on probe binding). 364 365 366

367 368 In this study, we present a platform – the Clonally -Resolved Epitope -Specificity and 369 Transcriptome Assay (CRESTA) – that enables the integrated analysis of CD4 T cell specificities 370 and phenotypes at a dimensionality that greatly exceeds what was previously possible. By 371 applying this assay to study the response across 100s of HLA class II-restricted epitopes in Mtb-372 infected participants, we generate a portrait of CD4 T cell immunity to Mtb at unprecedented 373 breadth and resolution. This analysis reveals previously undescribed features including extensive 374 intra-epitope polyclonality and phenotypic heterogeneity, both within and between epitopes. In 375 the process, we also map 100s of new TCRα:β pairs to Mtb peptide:HLA epitopes, and describe 376 how public epitope -specific sequence features vary across epitopes and individuals. More 377 generally, since its assay targets are fully customizable, we expect CRESTA to be broadly 378 applicable beyond Mtb, and to enable similar insights in other research/disease settings in which 379 antigen-specific CD4 T cell responses are implicated. 380 381 Our finding that human Mtb epitope-specific T cells can occupy a range of differentiation states – 382 including Th17, Treg, and a range of Th1 substates – within the same host, and sometimes even 383 against the same epitope (Figures 4, 5) – illuminates layers of diversity in this response that were 384 not previously understood. Especially notable is the related observation that different epitope 385 specificities can elicit markedly different Th1 phenotypes within the same response (Figure 5b,c). 386 Together, these findings indicate that existing approaches for studying Mtb-specific T cell 387 responses – which are typically characterized by the use of antigen pools and the detection of a 388 limited number of phenotypic markers (e.g. IFN γ, TNF) 28,36 – capture only a subset of the 389 response. The finding that different Mtb antigens can elicit T cells with different 390 phenotypes/functions also offers a new class of mechanism to inform the interpretation of prior 391

that point to antigen-specific protection, including: (i) that different Mtb antigens are under 392 purifying v diversifying evolutionary selection pressure 37, (ii) that particular TCR clusters (i.e. 393 epitope-specific responses) are associated with protection v non -protection from disease 31. 394 Particularly intriguing is our finding that the same PE13 epitope that was associated with 395 protection in the latter study, can be uniquely associated with high expression of GNLY ( Figure 396 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 5c), a pore -forming effector found in cytotoxic granules that can kill Mtb bacteria at high 397 concentrations38,39. More generally, our findings invite future applications of CRESTA to larger 398 cohorts of Mtb-infected participants, including comparisons of progressors v controllers, to explore 399 more comprehensively which T cell epitope-specificities and phenotypes may be associated with 400 protection from disease. 401 402 Our unbiased, epitope -resolved analysis of >200 TCR α:β sequences identified by CRESTA 403 (Figure 5) yields several important insights. First is the finding that, when filtering stringently on 404 TCRα:β pairs specific for individual Mtb epitopes, some degree of homology is detectable for all 405 epitopes. This observation is consistent with the first-principles consideration that, although most 406 epitopes are likely capable of recognition by many possible TCR binding modes, each binding 407 sequence is likely to be surrounded by a cluster of similar sequences separated by amino acid 408 substitutions that preserve binding. Second, however, is the countervailing observation that, even 409 in our stringent setting (filtering on individual epitopes), these homologies encompass only a 410 minority of antigen -specific TCRs. Moreover, when extrapolating to the more typical scenario 411 where oligoclonal/polyclonal repertoires are analyzed, such homologies remained detectable in 412 only rare cases – implying a limit to the sensitivity of approaches that rely on TCR clustering alone. 413 Third, although the most common finding was indistinguishable TCR similarity patterns within v 414 between individuals, our observation of a striking exception – in which one participant showed 415 dramatically weaker TCR clustering for the PE13 epitope – represents the best evidence of which 416 we are aware that there can be large individual-specific differences in the TCR sequences raised 417 to a fixed peptide:HLA antigen. Conceivable interpretations include: (i) individual -to-individual 418 differences in the T cell repertoire, shaped by other HLAs, self-peptides and/or antigen exposures, 419 (ii) individual -to-individual differences in the milieu in which the epitope -specific T cells were 420 primed, potentially altering the affinity threshold for T cell activation, (iii) a gene-dose effect of the 421 restricting HLA, wherein homozygosity may increase the density of peptide:HLA complexes and 422 thereby decrease the affinity threshold for T cell activation. Finally, CRESTA’s ability to generate 423 large numbers of TCR α:β sequences mapped to individual peptide:HLA epitopes (as of June 424 2024, considering HLA class II:peptide -specific TCR α:β pairs in the IEDB, this study alone 425 generated ~8X more pairs for Mtb, and >10% of the total number of pairs identified across all 426 research fields), positions it as a powerful tool for TCR discovery, with the potential future 427 applications that include the development of TCR-based therapies, as well as the training of next-428 generation models for predicting antigen specificity from TCR sequences. 429 430 A cornerstone of CRESTA is clonal analysis, which we use to (i) make rare epitope-specific CD4 431 T cells detectable within limited sample volumes, and (ii) enable robust, multi -datapoint-based 432 inferences about their antigen specificity and phenotype from single cell sequencing data. At the 433 same time, the use of clonal expansion has the potential to limit the assay in several ways. First, 434 it is likely that the efficiency of such expansion varies according to the state of the precursor T 435 cells, meaning that highly proliferative subtypes may reach detectable levels before less 436 proliferative subtypes do, biasing the representation of cells analyzed. Indeed, within cells 437 expanded from Mtb-infected participants, we observe a significant correlation between clone size 438 and helper T cell state (Figure 4c). Nonetheless, CRESTA successfully detects cells across the 439 Th state spectrum, including numerous clones of the less proliferative Treg and Th2 subtypes 440 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint (Figure 4), and its sensitivity to less expanded clones is likely to increase as the throughput of 441 single cell sequencing technologies continues to grow. Importantly, the impact of such biases can 442 be controlled by comparing clones of interest against bystander clones within the same expanded 443 samples, as we exemplify in Figure 4d. It is also noteworthy that the sensitivity of common (non-444 expansion-based) T cell specificity assays is often also biased to particular phenotypes (eg 445 ELISpot, that depends on cytokine secretion). A second potential limitation of clonal expansion is 446 the possibility of introducing artefactual gene expression patterns. While we have not directly 447 quantified these effects, our observation of gene expression patterns that partition strongly 448 according to clonal identity within a mixed culture (Figures 2d,e) – and recapitulate the known Th 449 subsets upon unsupervised multi -gene analyses ( Figure 4a,b ) – indicates that initial 450 differentiation states persist during expansion to at least a substantial degree. In the future, any 451 remaining impact may be mitigated by reducing the degree of expansion (our observed clone size 452 distributions indicate that even an order of magnitude less expansion is unlikely to substantially 453 impact assay sensitivity), and/or by using other sample types (e.g. bronchoalveolar lavage) that 454 are more enriched for the relevant antigen-specificities. 455 456 While it represents a major advance beyond what has been enabled in prior studies of CD4 T 457 cells, the application of CRESTA at a multiplexity of up to ~170 epitopes in this work does not yet 458 reach the scale that is needed for broad proteome-scale epitope discovery and characterization. 459 However, since: (i) we expect DNA barcoding and sequencing to be intrinsically highly -scalable 460 (>100,000s-plex), (ii) we saw no evidence for loss of binding signal as plexity increased up to 461 ~170 (Figure 3e), and (iii) analogous approaches for CD8 T cells been successfully implemented 462 up to ~1000-plex (even without the clonal analysis enhancement we describe here)11, we expect 463 CRESTA to scale well beyond the plexity that we have demonstrated here. At that point, new 464 bottlenecks to be overcome will be the efficient identification of candidate peptide:HLA pairs (e.g. 465 using in silico prediction and/or high -throughput binding screens), and the preparation of the 466 corresponding probes in large numbers (e.g. using microfluidic automation). Realizing these 467 advances could enable powerful new studies of CD4 T cell specificities and functions, and their 468 interactions, at a truly genome-wide scale in diverse disease states. 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint Figures and legends 485 486 487 488 489 Figure 1: Workflow of the Clonally-Resolved Epitope Specificity and Transcriptome Assay 490 (CRESTA). To enable the integrated, high-dimensional analysis of CD4 T cell epitope-specificities 491 and transcriptional states, we begin with the high -throughput assembly of hundreds of 492 peptide:MHC probes, each multimerized and DNA-barcoded using a streptavidin-bearing dextran 493 backbone construct (upper left). Pools of these probes are used to stain T cells clonally expanded 494 from PBMC samples using peptides (lower left, center), and then single cell sequencing is used 495 to read out epitope -level specificities and transcriptome -wide expression profiles ( upper right). 496 Finally, to interpret the resulting data, we apply a “pseudobulk” analysis that aggregates cells into 497 clones according to shared TCRα:β sequences (lower right), enabling inferences about epitope-498 specificity and transcriptional state that are robust to the noise inherent in single cell data. 499 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 500 501 Figure 2: Clonal analysis enables robust, high -dimensional analysis of CD4 T cell Mtb 502 epitope-specificity and transcriptome -wide gene expression, simultaneously. Peptide-503 expanded PBMCs from a LTBI participant (ID:40059) were stained with a pool of 48 DNA -504 barcoded peptide:MHC multimer probes corresponding to all Mtb T cell epitopes in IEDB known 505 to be restricted by either HLA-DRB1*15:03, HLA-DRB1*11:01 or HLA-DQB1*6:02, and analyzed 506 by single cell sequencing (“antigen-specificity aliquot”). A second aliquot that was unstained but 507 stimulated with PMA/ionomycin was also assayed (“gene expression aliquot”). A total of 28,129 508 cells across the 2 aliquots were collapsed into clones based on identical TCRα:β sequences, 170 509 of which contained ≥3 cells in each aliquot (≥6 total). (a) Shown are the distribution of cell numbers 510 in each of the 170 clones. (b) To identify which of the 48 epitopes were recognized, we applied a 511 Kruskal-Wallis test to data from the “antigen -specificity aliquot” to determine whether, for each 512 multimer probe, staining across cells partitioned in a clonally-restricted way (p-value, y-axis), and 513 to quantify the magnitude of such partitioning (fold-difference, x-axis). (c) For significant epitopes, 514 we next identified their particular binding clones using a Wilcoxon test to compare the distribution 515 of probe binding to each clone against the distribution for all other clones. Shown, by way of 516 example, are 2 of the significant epitopes identified in (b), with their respective significant clones 517 indicated in yellow. (d) To identify genes whose expression varies by clone (“clonal differentiation 518 genes / CDGs”), we applied the same clonally-resolved analysis described in (b), but this time to 519 gene expression values measured in the “gene expression aliquot”. (e) Examples of the clonally-520 resolved expression patterns of 2 CDGs with anti -correlated abundance, corresponding to the 521 Th17 (IL17A) and Th1 (CCL4) states, respectively. Vertical alignment of the 220 clones is 522 consistent between panels (a), (c) and (e). 523 524 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 525 526 Figure 3. CRESTA epitope detection is specific up to at least 173-plex and reveals highly-527 polyclonal, multi-epitope responses to Mtb. (a) LTBI+ participant ID:30168 was analyzed as 528 described in Figure 2a-c, this time using a pool of 173 DNA-barcoded multimer probes restricted 529 by HLA-DRB1*15:03, HLA-DRB5*01:01 or HLA-DQB1*6:02. Shown is a volcano plot revealing 530 the significant binding of 15 probes. (b) For the analysis in (a), boxplots show all 126 (of 454 total) 531 clones for which we detected ≥1 significant probe binding event (highlighted in yellow) across the 532 15 probes. Based on these binding profiles, the 15 probes cluster into 7 groups (demarcated by 533 horizontal boxes), whose members were uniformly HLA matched and contained overlapping 534 peptide sequences (in 2 cases these groups contained identical replicates). Notably, none of the 535 126 clones showed significant staining across >1 of these groups (100% specificity on this 536 measure). (c) The assay and analysis described in (a, b) was applied to a total of 5 participants 537 (IDs: 30128, 30133, 40059, 30129, 30168), which revealed a total of 19 unique probes that had 538 significant binding to ≥1 clone in ≥1 participant. Each of these probes is depicted as a node in a 539 force-directed graph that connects probes whose binding clones overlap. This analysis yielded a 540 total of 11 clusters (distinguished by color) which again correspond precisely to sets of probes 541 with shared HLA and overlapping peptide sequences. (d) Shown are the number of significant 542 binding clones for each of the 5 participants and 11 epitope clusters described in (c). Relative to 543 (c), peptide sequences are trimmed to show only the longest subsequence common to all 544 members of each cluster, and bold type indicates the HLA -binding core 9mer predicted by 545 netMHCIIpan. (e) For each binding probe across the 5 participants (colored as in (d), and plotted 546 by participant on the x -axis), shown is the intensity of probe staining (y -axis), which shows no 547 decline with increasing plexity. 548 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 549 550 Figure 4: CRESTA reveals a spectrum of CD4 T cell differentiation states and localizes the 551 epitope-specific Mtb response predominantly, but not exclusively, within the Th1 552 compartment. (a) Shown, for representative LTBI participant (ID:30133; 217 clones), is a 553 clustering analysis in which each clone is represented as a circle (sized according to its number 554 of constituent cells) and represented in 2 dimensions using a UMAP projection of its median 555 expression of each Clonal Differentiation Gene (n=763 CDGs). Green circles (upper left) identify 556 multimer-binding clones, orange ovals (lower left) demarcate inferred T cell subsets, and blue-red 557 coloring represents the median clonal mRNA abundance for the genes indicated in the upper -558 right corner of each plot. (b) For the Th subset -specific cytokine genes IFN γ, IL4 and IL17A 559 (represented in purple), Pearson analysis across all clones and CDGs was used to identify 560 additional genes with correlated expression (represented in green). (c) Clones across 5 561 participants (n=1,496) were assigned to T-helper (Th) states Th1, Th2, Th17, Treg or naive states 562 (or unassigned); the distribution of cell numbers per clone across the 6 Th states is shown. (d) 563 The 1,496 clones described in (c) were further classified according to whether or not they bound 564 an Mtb peptide:HLA multimer. Shown are the distributions across Th states of all clones (upper) 565 or of multimer-binding clones (lower), which were compared by Fisher’s exact test. 566 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 567 Figure 5: CRESTA reveals phenotypic heterogeneity within and between the CD4 T cell 568 responses to individual Mtb epitopes. (a) Expression profiles of epitope-binding T cells for two 569 selected individuals/epitopes, showing genes representative of the Th1 (IFNG, CCL4), Th2 (IL4, 570 IL5, GATA3, CCL1), Th17 (IL17A, IL17F, IL22) and Treg (FOXP3, CTLA4) subsets. Each column 571 shows data from an individual epitope -specific clone (n=33 clones), which comprise 4 -347 572 individual cells, quantified in the barplot. Violin plots show the distribution of expression values for 573 cells in each clone and are shaded according to median expression values. (b) For all cases in 574 which we observed ≥2 distinct reactive epitopes each recognized by ≥3 clones within a participant 575 (total = 230 clones across 9 epitopes in 3 participants), we identified CDGs within each participant 576 and then applied Kruskal-Wallis tests to determine whether the expression of each CDG across 577 clones was correlated with epitope specificity. Highlighted are 5 unique genes that were significant 578 in ≥2 participants (using thresholds shown in the yellow box). (c) Gene expression states for the 579 clones described in (b) were rendered in 2 dimensions using UMAP across all CDGs, and colored 580 according to epitope specificity (dot plots, upper row). For each participant (column), expression 581 of a selected gene observed to have strong epitope -dependent expression is shown across all 582 clones (dot plots, center row ), and compared between epitopes with the lowest v highest 583 expression (violin plots, lower row). Violin plots are colored by epitope. 584 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 585 586 587 Figure 6: Analysis of 206 Mtb epitope-mapped TCRα: β pairs reveals that sequence 588 publicity is widespread, but varies in magnitude across epitopes and participants. (a) For 589 8 Mtb epitopes (columns) for which CRESTA identified ≥10 epitope -specific clones, TCRα:β 590 sequences were compared within and between participants using comprehensive pairwise 591 TCRdist measurements. Each TCR is shown as a single dot (colored to distinguish participants) 592 in a force -directed graph in which proximity indicates the degree of sequence similarity. Line 593 segments connect TCR pairs whose TCRdist scores are significant, based on a large set of 594 distances randomly-sampled from an unenriched repertoire, with Benjamini–Hochberg correction 595 for the total number of comparisons made for each epitope (FDR<0.1). The number of epitope -596 specific TCRs for each participant, and the total number of TCRs within significant clusters, are 597 shown at the top and bottom of each box, respectively. (b) Shown, for the analysis described in 598 (a), are the full distributions of TCRdist p -values, but this time unadjusted: x-axis = expected, y-599 axis = observed), for all comparisons (black), and comparisons within (pink) and between (gray) 600 participants. (c) Logos showing sequence features of the significantly -clustered TCRs for each 601 epitope shown in (a). V and J segments are shown for cases where >50% of clustered TCRs 602 contain the same segment. CDR3 letters are sized according to conservation/entropy, and 603 colored by amino acid properties (polar=red, non -polar=green, negatively charged=gold, 604 positively charged=blue, aromatic=purple). 605 606 607 608 609 610 611 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint

612 613 Study cohort, PBMC collection and processing 614 615 Household contacts (HHCs) of newly diagnosed active pulmonary TB cases were referred to the 616 Kenya Medical Research Institute (KEMRI) Clinical Research Center in Kisumu, Kenya, and their 617 demographic and medical history data were collected. HHCs were persons who shared the same 618 home residence as the index case for ≥5 nights during the 30 days prior to the date of TB 619 diagnosis of the index case, and were enrolled no more than 3 months (mean: 18 days; range: 620 1–77 days) after the index case began TB treatment. All participants provided written informed 621 consent to join the study and were recruited from two community -based health clinics located in 622 Kisumu City and Kombewa, Kisumu County. All enrolled individuals met the following inclusion 623 criteria: ≥ 13 years of age at the time of enrollment, positive QuantiFERON TB Gold in Tube (QFT) 624 result, seronegative for HIV antibodies, no previous history of diagnosis or treatment for active TB 625 disease or LTBI, normal chest X-ray, and not pregnant. All participants were presumed to be BCG 626 vaccinated due to the Kenyan policy of BCG vaccination at birth and high BCG coverage rates 627 throughout Kenya. All participants gave written informed consent for the study, which was 628 approved by the KEMRI/CDC Scientific and Ethics Review Unit and the Institutional Review Board 629 at Emory University, USA. 630 631 Blood samples were collected from participants in sodium heparin or lithium heparin Vacutainer 632 CPT Mononuclear Cell Preparation Tubes (BD Biosciences or Greiner Bio -One). PBMC were 633 isolated by density centrifugation, rested in complete media (RPMI 1640 containing L -glutamine 634 supplemented with 10% heat -inactivated fetal bovine serum (FBS), 1% PenStrep, 1% Hepes) 635 before counting. PBMC isolation was initiated <2 hours after the blood was drawn. Isolated PBMC 636 were cryopreserved in 90% heat -inactivated fetal calf serum/10% DMSO, and kept in LN2 (and 637 shipped on dry ice) until they were thawed for study at the TGen laboratory. 638 639 640 Cell culture and T cell expansion 641 642 PBMCs were thawed in a 37°C water bath and then washed and plated at 1.25 x 10 6 cell/mL in 643 24-well flat-bottomed plates in complete RPMI (RPMI-1640 with 10% AB human serum, 0.8 mM 644 sodium pyruvate, 0.8x non-essential amino acids, 80 U/mL penicillin-streptomycin, 0.4x HEPES, 645 200 mM L-glutamine, 0.07x 2-Mercaptoethanol; hereafter “cRPMI”) at 37oC with 5% CO2. On day 646 2, pools of peptides dissolved in DMSO (up to 579 peptides – which included all peptides from 647 multimers used for staining – each at a final concentration 0.47ug/mL; with a maximum final 648 DMSO concentration of 1.3%) were added. On day 3, taking care to not disturb the cells, 50% of 649 the media was exchanged. Cells were split as needed on days 4-5, after which 50% of the media 650 was again exchanged. Cultures were maintained for a total of 8 -10 days. Media included 651 recombinant human interleukin -2 (IL-2) at 230IU/mL – 1,025IU/mL (Biolegend), with the lower 652 concentrations used on days 1-2, followed by higher concentrations throughout the remainder of 653 the expansion. At the end of the culture, cells were harvested by collecting the supernatant, 654 treating wells with 2 mM EDTA in PBS for 2-4 minutes and adding the detached cell suspension 655 to the collected culture. After combining all wells from the same donor, cells were spun down at 656 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 300 x g for 5 min at room temperature, resuspended in storage media (9:1 cRPMI and DMSO) 657 and stored in liquid nitrogen until further use, or resuspended in cRPMI and used directly in the 658 CRESTA assay. 659 660 661 Production of HLA:peptide complexes 662 663 Peptides were custom ordered from Millipore -Sigma (PEPscreen, unpurified) and reconstituted 664 to 20 mg/mL in DMSO. CLIP peptide -tethered, biotinylated HLA monomers were obtained from 665 the NIH Tetramer Core Facility. To generate peptide-bound HLA monomers, we used a protocol 666 described previously32, consisting of (i) CLIP peptide cleavage, followed by (ii) peptide exchange. 667 CLIP peptide cleavage was performed by incubating the HLA monomer (at a final concentration 668 of 0.4 mg/mL with 3C protease (HRV-3C protease, Sigma Aldrich) in 3C cleavage buffer (0.05 M 669 Tris pH, 7.5, 150 mM NaCl) or thrombin protease in 10X Thrombin cleavage buffer (Thrombin 670 restriction grade, Millipore) overnight at room temperature. To perform peptide exchange, the 671 cleaved monomer was incubated at a final concentration of 0.2 mg/mL with individual peptides 672 (at a final concentration of 76-666 ug/mL) in 50 mM citrate buffer pH 5.2 containing 100 mM NaCl, 673 2 mM EDTA, 0.2x protease inhibitor (Promega), at reaction scales of 30 –1000 μL depending on 674 desired yield, at 30°C for 4 days. Following incubation, HLA:peptide complexes were cleared of 675 excess peptide and concentrated using dPBS-rinsed Amicon Ultra-0.5 Centrifugal 10 kDa filters 676 (spun twice at 14,000 x g for 15 min at 4°C). Finally, the cleared HLA:peptide monomer products 677 were quantified using a Nanodrop 1000 spectrophotometer reading at 280 nm, and stored in the 678 presence of 0.75x protease inhibitor cocktail (50X protease inhibitor, Promega) at -80°C for up to 679 12 months. 680 681 682 Generation of DNA-barcoded HLA:peptide probes 683 684 Barcoding DNA sequences compatible with the 10X Chromium Single Cell 5’ chemistry were 685 designed according to the 69 mer construct recommended by 10X Genomics (Surface Protein 686 Labeling Protocol CG000186), utilizing a 3’ Capture Sequence and containing 15 mer barcodes 687 from the 10X barcodes whitelist (Demonstrated Protocol, CG000193). These sequences were 688 purchased from IDT as 5’ biotinylated DNA oligos with standard desalting. For each desired 689 multimer probe, 1 μL of streptavidin-bearing dextran backbone (Klickmer APC or PE, Immunodex, 690 0.16 μM) was barcoded by incubating it with 1 μL of barcoding oligonucleotide (at 0.32 μM in 1x 691 TE) at 4°C for 30 min, in Lo-bind (Eppendorf) 96 well plate (stoichiometry = 1 dextran : 2 oligos). 692 After incubation, 2 μL of the desired HLA:peptide monomer (at 3.2 μM, prepared as above) was 693 added to the barcoded dextramer and incubated for 30 min at RT (stoichiometry = 1 dextran : 2 694 oligos : 20 HLA:peptide monomers). Binding reactions were quenched by adding 1 μL free D-695 biotin (5μM) in excess at 4°C for 30 min. These individual probe constructs (“multimers”) were 696 stored for up to 1 week at 4°C prior to cell staining. On the day of cell staining, the total volume 697 (5 μL) of each desired multimer (up to 173 – see Supplementary Table 2 for the probes used for 698 each donor) was pooled and concentrated using a Vivaspin2 column (#VS0241; Sartorius). The 699 column was pre-washed with 1x PBS, followed by 2 mL of barcode buffer (dPBS with 0.5% BSA 700 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint and 5 μg/mL Herring DNA (Promega)) and then stored at 4°C prior to loading with the probe pool. 701 Once the probe pool was added, the column was spun at 3,300 rpm for 10–20 minutes (until the 702 liquid level reached ~50-100 μl), and then inverted and spun at 2,000 x g for 5 minutes to recover 703 the pool, which was kept on ice until cells were ready for staining (up to 2 hours). 704 705 706 PMA/ionomycin stimulation (“gene expression aliquot”) 707 708 Expanded cells were thawed, washed with cRPMI, and then resuspended in cRPMI with 200 709 IU/mL IL-2 at 1.25 x 10 6 cell/mL in 24-well flat-bottomed plates and incubated at 37 oC with 5% 710 CO2 overnight. Cells were then stimulated with PMA and ionomycin at final concentrations of 0.04 711 μM and 0.67 μM, respectively (Biolegend, Catalog #423301), and then incubated for 1 -1.5hrs at 712 37oC. Stimulated cells were collected into 1 5mL conical tubes, including detachment from plates 713 with 2 mM EDTA in PBS for 2 -4 minutes, then centrifuged at 200 x g for 5 minutes. Cells were 714 washed a total of three times, in the following buffers, with spinning at 300 x g for 5 min at RT 715 between each: (i) 5 mL of cRPMI; (ii) 2.5 mL cRPMI + 2.5 mL EasySep Buffer (# 20144; StemCell); 716 and (iii) 5 mL of Loading Buffer (1x PBS + 0.04% BSA). Following the final wash, cells were 717 resuspended in 100μL of Loading Buffer and a 10 μL aliquot was taken for cell counting prior to 718 single cell partitioning. 719 720 DNA-barcoded peptide:HLA class II multimer staining (“antigen-specificity aliquot”) 721 722 Expanded cells were thawed and rested overnight as described above ( PMA/ionomycin 723 stimulation section). The next day, cells were collected into 15 mL conical tubes, including 724 detachment from plates with 2 mM EDTA in PBS for 2-4 minutes, then centrifuged at 200 x g for 725 5 minutes. Supernatant was discarded without disturbing the cell pellet, and cells were washed 2 726 times in cRPMI (300 x g for 5 min) and the pellet resuspended (via flicking) in 100 μL cRPMI. 727 Resuspended cells treated with 0.003 M of the protein kinase inhibitor Dasatinib (Axon Medchem, 728 VA) with gentle swirling at 37 oC, for 10 min with a loose cap, and then 20 min with a tight cap. 729 The multimer probe pool (50-100 uL total volume, prepared as above) was then added, resulting 730 in a final concentration of ~0.8 –1.1 nM for each multimer, and cells incubated for 30 min at RT 731 with a tight cap. After incubation, cells were spun in a 4 oC (300 x g for 5 min) for 3 washes; first 732 in 5 mL of cRPMI, then 2.5 mL cRPMI + 2.5 mL EaspSep Buffer (# 20144; StemCell) and finally 733 in 5mL Loading Buffer (1x PBS + 0.04% BSA). Following the last wash, cells were resuspended 734 in 100 μL Loading Buffer and a 10 μL aliquot was taken for cell counting prior to single cell 735 partitioning. 736 737 738 Generation and sequencing of single cell libraries 739 740 To generate single cell libraries, 10,000–40,000 cells per participant aliquot were loaded into the 741 Chromium Controller (10X Genomics), and processed according to the manufacturer’s 742 instructions (Chromium Next GEM Single Cell 5' Reagent Kits v2 protocol) to prepare libraries 743 corresponding to the feature/probe barcodes (antigen -specificity aliquot), 5′ gene expression 744 (gene expression aliquot), and VDJ-T repertoire (both aliquots). Fragment size and quantity was 745 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint assessed on the 4200 TapeStation (#G2991A; Agilent) using high sensitivity DNA tapes (#5067–746 4626; Agilent). Before sequencing, qPCR was performed after serial dilution of the libraries 747 (#KK4824 – 07960140001; Kapa Biosystems). Libraries were sequenced on Illumina 748 NextSeq1000 or NovaSeq instruments using the read configurations and PhiX loading 749 recommended by 10X Genomics. 750 751 752 Analysis of single cell sequencing data 753 754 Data processing and merging: Following Illumina BCL file conversion, fastq files were processed 755 through the CellRanger Multi Pipeline (10X Genomics) to perform demultiplexing, alignment, 756 filtering, barcode and UMI counting, and VDJ assembly. We then developed custom R code for 757 all downstream analytical steps. We first merged the filtered VDJ clonotype sequences (from the 758 filtered_contig_annotations.csv file) with the raw counts matrix of transcripts and/or probe 759 barcodes (raw_feature_bc_matrix), by matching on cell barcodes and retaining only those cells 760 associated with ≥1 filtered VDJ sequence. 761 762 Clonotype collapsing and shortlisting: Since some CellRanger VDJ-T clonotype calls contain the 763 same TCR chain sequence both alone and paired with other sequence(s) (likely attributable to a 764 combination of: (i) droplets containing >1 cell (multiplets), and (ii) the fact that TCR chain 765 sequences that are present do not always successfully amplify/assemble in every cell), we 766 developed a method to improve assignment of TCR chains to clonotypes that involved clustering 767 chains according to their correlated occurrence across cells (collapseclonotypes.R). This allowed 768 us to more accurately identify chains belonging to the same clone, by merging cells that were 769 assigned to different clonotypes but shared ≥1 chain, and by excluding cells that contained chains 770 from ≥1 cluster (as likely multiplets). For each aliquot (antigen-specificitiy or gene expression) we 771 then focused only on collapsed clones containing ≥3 cells (listclonotypes.R). 772 773 Normalization of multimer probe signal: We developed a method to normalize multimer -to-774 multimer and cell -to-cell differences in the multimer assay yield (normmaster.R). First, we 775 normalized the read counts for each multimer to a fixed depth such that the sums for all multimers 776 were the same (equal to the number of cells in the data matrix). For each multimer, we then 777 normalized in the other dimension (across cells) by dividing the resulting values for each cell by 778 the median value of all other cells. To reduce the impact of high-staining outliers (e.g. caused by 779 multimer aggregation) on visualization, for each multimer we applied the value of the cell at the 780 top 0.1 percentile to all cells with greater staining. Finally, for each multimer, we divided values 781 by the multimer max value, to scale values onto the [0,1] interval prior to visualization 782 (stackedplot.R) and statistical analysis. 783 784 Identification of staining multimers / clonotypes and Clonal Differentiation Genes: To identify 785 binding multimers and Clonal Differentiation Genes (CDGs), we applied Kruskal -Wallis tests to 786 quantify the influence of clonal identity on (i) the normalized multimer probe signal, or (ii) raw 787 reads, for each multimer or gene, respectively ( testclonalgenes.R). To identify the individual 788 binding clones for each of binding epitope, we then applied one-tailed Wilcoxon tests post-hoc to 789 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint each binding multimer, in which we compared the normalized multimer probe signals in each 790 individual clone against the signal of the central 20% of cells across all clones (testsigclones.R). 791 792 UMAP analysis, transcript correlation analysis and identification of T cell subsets: Beginning with 793 either all clones ( Figure 4) or all epitope -mapped clones ( Figure 5c), we identified significant 794 CDGs as above (with max fold -difference > 10 and Bonferroni -adjusted p-values < 0.05), and 795 then generated log 10(median+1) values (“clone medians”) for each gene in each clone. To the 796 resulting gene-by-clone table, we applied UMAP to generate a 2-dimensional rendering using the 797 R umap package. For the lineage defining genes IFNγ, IL4 and IL17A, we generated Pearson 798 correlation coefficients (r) across clone median values, against every other significant CDG. 799 Genes with r > 0.5 are depicted as a force-directed graph rendered using the R igraph package. 800 Using clone median values, we identified Th2, Th17 and Treg clones as IL4+GATA3+, 801 IL17A+RORC+, and FOXP3+CTLA4+, respectively. Among the remainder, we identified Th1 802 clones as IFNG+CCL4+, and finally, among the remainder, naives clones as IL2RA–. 803 804 Analysis of TCRα:β homologies using TCRdist: To analyze TCR homologies, we developed our 805 own computationally efficient implementation of the TCRdist metric 35 to quantify the degree of 806 sequence similarity between pairs of TCRα: β heterodimers. We focused on epitope -specific 807 clones with exactly 1 TCRα and 1 TCR β chain, and, within each epitope, performed TCRdist 808 measurements between all pairs of binding TCRs. We then mapped each TCRdist value to a p -809 value using a large distribution of TCRdists measured on a large, unenriched repertoire, as 810 follows. Beginning with 1e4 randomly -sampled TCRαs and TCR βs, we calculated all ~5e7 811 pairwise TCRdists for each chain type. We then calculated the frequencies of all possible α: β 812 chain TCRdist values by considering all combinations of α and β TCRdists, allowing us to 813 efficiently estimate frequencies down to ~1e-12. Epitope-specific sets of p-values were adjusted 814 using the Benjamini –Hochberg procedure. We visualized the CDR3 sequences of significantly 815 clustered TCRs using the ggseqlogo R package, after multiple sequence alignment. Positions in 816 which there was an alignment gap for the majority of CDR3s were excluded from display. 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint

834 835 836 1. Mosmann, T. R. & Coffman, R. L. TH1 and TH2 cells: different patterns of lymphokine 837 secretion lead to different functional properties. Annu. Rev. Immunol. 7, 145–173 (1989). 838 2. Zinkernagel, R. M. & Doherty, P. C. Restriction of in vitro T cell-mediated cytotoxicity in 839 lymphocytic choriomeningitis within a syngeneic or semiallogeneic system. Nature 248, 840 701–702 (1974). 841 3. Babbitt, B. P., Allen, P. M., Matsueda, G., Haber, E. & Unanue, E. R. Binding of 842 immunogenic peptides to Ia histocompatibility molecules. Nature 317, 359–361 (1985). 843 4. Davis, M. M. & Bjorkman, P. J. T-cell antigen receptor genes and T-cell recognition. Nature 844 334, 395–402 (1988). 845 5. Lovelace, P. & Maecker, H. T. Multiparameter Intracellular Cytokine Staining. Methods Mol. 846 Biol. 1678, 151–166 (2018). 847 6. Sidney, J., Peters, B. & Sette, A. Epitope prediction and identification- adaptive T cell 848 responses in humans. Semin. Immunol. 50, 101418 (2020). 849 7. Dan, J. M. et al. A Cytokine-Independent Approach To Identify Antigen-Specific Human 850 Germinal Center T Follicular Helper Cells and Rare Antigen-Specific CD4+ T Cells in Blood. 851 J. Immunol. 197, 983–993 (2016). 852 8. Altman, J. D. et al. Phenotypic analysis of antigen-specific T lymphocytes. Science 274, 853 94–96 (1996). 854 9. Newell, E. W., Klein, L. O., Yu, W. & Davis, M. M. Simultaneous detection of many T-cell 855 specificities using combinatorial tetramer staining. Nat. Methods 6, 497–499 (2009). 856 10. Bentzen, A. K. & Hadrup, S. R. Evolution of MHC-based technologies used for detection of 857 antigen-responsive T cells. Cancer Immunol. Immunother. 66, 657–666 (2017). 858 11. Bentzen, A. K. et al. Large-scale detection of antigen-specific T cells using peptide-MHC-I 859 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint multimers labeled with DNA barcodes. Nat. Biotechnol. 34, 1037–1045 (2016). 860 12. Magnin, M., Guillaume, P., Coukos, G., Harari, A. & Schmidt, J. High-throughput 861 identification of human antigen-specific CD8 and CD4 T cells using soluble pMHC 862 multimers. Methods Enzymol. 631, 21–42 (2020). 863 13. Lantz, O. & Teyton, L. Identification of T cell antigens in the 21st century, as difficult as 864 ever. Semin. Immunol. 60, 101659 (2022). 865 14. Rockinger, G. A. et al. Optimized combinatorial pMHC class II multimer labeling for 866 precision immune monitoring of tumor-specific CD4 T cells in patients. J Immunother 867 Cancer 8, (2020). 868 15. Yang, J. et al. Multiplex mapping of CD4 T cell epitopes using class II tetramers. Clin. 869 Immunol. 120, 21–32 (2006). 870 16. Davis, M. M., Altman, J. D. & Newell, E. W. Interrogating the repertoire: broadening the 871 scope of peptide-MHC multimer analysis. Nat. Rev. Immunol. 11, 551–558 (2011). 872 17. Ge, X. et al. Peptide-MHC cellular microarray with innovative data analysis system for 873 simultaneously detecting multiple CD4 T-cell responses. PLoS One 5, e11355 (2010). 874 18. Justesen, S., Harndahl, M., Lamberth, K., Nielsen, L.-L. B. & Buus, S. Functional 875 recombinant MHC class II molecules and high-throughput peptide-binding assays. 876 Immunome Res. 5, 2 (2009). 877 19. Uchtenhagen, H. et al. Efficient ex vivo analysis of CD4+ T-cell responses using 878 combinatorial HLA class II tetramer staining. Nat. Commun. 7, 12614 (2016). 879 20. Dezfulian, M. H. et al. TScan-II: A genome-scale platform for the de novo identification of 880 CD4 T cell epitopes. Cell vol. 186 5569–5586.e21 (2023). 881 21. Fletcher, H. A. & Schrager, L. TB vaccine development and the End TB Strategy: 882 importance and current status. Trans. R. Soc. Trop. Med. Hyg. 110, 212–218 (2016). 883 22. Hansen, S. G. et al. Prevention of tuberculosis in rhesus macaques by a cytomegalovirus-884 based vaccine. Nature medicine vol. 24 130–143 (2018). 885 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint 23. Schrager, L. K., Harris, R. C. & Vekemans, J. Research and development of new 886 tuberculosis vaccines: a review. F1000Res. 7, 1732 (2018). 887 24. Tait, D. R. et al. Final Analysis of a Trial of M72/AS01 Vaccine to Prevent Tuberculosis. N. 888 Engl. J. Med. 381, 2429–2439 (2019). 889 25. Lewinsohn, D. A., Lewinsohn, D. M. & Scriba, T. J. Polyfunctional CD4+ T Cells As Targets 890 for Tuberculosis Vaccination. Front. Immunol. 8, 1262 (2017). 891 26. Seder, R. A., Darrah, P. A. & Roederer, M. T-cell quality in memory and protection: 892 implications for vaccine design. Nat. Rev. Immunol. 8, 247–258 (2008). 893 27. Lindestam Arlehamn, C. S. et al. Memory T cells in latent Mycobacterium tuberculosis 894 infection are directed against three antigenic islands and largely contained in a 895 CXCR3+CCR6+ Th1 subset. PLoS Pathog. 9, e1003130 (2013). 896 28. Lindestam Arlehamn, C. S. et al. A Quantitative Analysis of Complexity of Human 897 Pathogen-Specific CD4 T Cell Responses in Healthy M. tuberculosis Infected South 898 Africans. PLoS Pathog. 12, e1005760 (2016). 899 29. Huang, H., Wang, C., Rubelt, F., Scriba, T. J. & Davis, M. M. Analyzing the Mycobacterium 900 tuberculosis immune response by T-cell receptor clustering with GLIPH2 and genome-wide 901 antigen screening. Nat. Biotechnol. 38, 1194–1202 (2020). 902 30. Glanville, J. et al. Identifying specificity groups in the T cell receptor repertoire. Nature 547, 903 94–98 (2017). 904 31. Musvosvi, M. et al. T cell receptor repertoires associated with control and disease 905 progression following Mycobacterium tuberculosis infection. Nat. Med. 29, 258–269 (2023). 906 32. Willis, R. A. et al. Production of Class II MHC Proteins in Lentiviral Vector-Transduced 907 HEK-293T Cells for Tetramer Staining Reagents. Curr Protoc 1, e36 (2021). 908 33. Ogongo, P. et al. Rare Variable Antigens induce predominant Th17 responses in human 909 infection. bioRxiv (2024) doi:10.1101/2024.03.05.583634. 910 34. Gallegos, A. M. et al. Control of T cell antigen reactivity via programmed TCR 911 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint downregulation. Nat. Immunol. 17, 379–386 (2016). 912 35. Dash, P. et al. Quantifiable predictive features define epitope-specific T cell receptor 913 repertoires. Nature 547, 89–93 (2017). 914 36. Lewinsohn, D. M. et al. Human Mycobacterium tuberculosis CD8 T Cell Antigens/Epitopes 915 Identified by a Proteomic Peptide Library. PLoS One 8, e67016 (2013). 916 37. Coscolla, M. et al. M. tuberculosis T Cell Epitope Analysis Reveals Paucity of Antigenic 917 Variation and Identifies Rare Variable TB Antigens. Cell Host Microbe 18, 538–548 (2015). 918 38. Dotiwala, F. & Lieberman, J. Granulysin: killer lymphocyte safeguard against microbes. 919 Curr. Opin. Immunol. 60, 19–29 (2019). 920 39. Stenger, S. et al. An antimicrobial activity of cytolytic T cells mediated by granulysin. 921 Science 282, 121–125 (1998). 922 923 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted November 8, 2024. ; https://doi.org/10.1101/2024.11.05.622086doi: bioRxiv preprint