Singing modulates mood, stress, cortisol, cytokine and neuropeptide activity in cancer patients and carers

Deep investigation

Context

By 2016, the relationship between psychological state and immune function was well-established in the field of psychoneuroimmunology (PNI). Decades of research had shown that chronic stress suppresses immune activity, while positive affect and social connection tend to enhance it. What was far less understood was whether a structured, time-limited behavioural intervention — specifically singing — could produce measurable, multi-dimensional biological shifts in populations already under significant physiological stress: people living with cancer, caring for someone with cancer, or bereaved after losing them.

The immune system matters enormously in cancer. Beyond its role in fighting infection, immune surveillance is thought to be one of the body’s primary defences against tumour growth and recurrence. Cytokines — the chemical messengers that coordinate immune responses — are central to this process. Pro-inflammatory cytokines like IL-2 and TNF-α activate immune cells and trigger inflammatory responses. Anti-inflammatory cytokines like IL-4 regulate and dampen those responses. Growth factors like GM-CSF stimulate production of immune cells. When cortisol (the primary stress hormone) is chronically elevated, it suppresses this entire cytokine network — a phenomenon known as glucocorticoid immunosuppression. In cancer patients undergoing treatment, this suppression can have compounding effects on both immune function and psychological resilience.

Previous music and singing research had hinted at biological effects. Kreutz et al. (2004) showed choir singing increased salivary immunoglobulin A (sIgA). Beck et al. (2000) reported mood and cortisol changes in choristers. Fancourt’s own earlier work (2015) demonstrated that low-stress versus high-stress singing had contrasting glucocorticoid profiles. Animal studies showed music reduced stress-related immune suppression in cancer models. But no study had examined the full breadth of immune markers — cytokines, growth factors, soluble receptors, and neuropeptides — in a single human investigation of singing. Existing studies were also typically small (N < 30) and drawn from healthy adults.

This paper changes the scale of the question. A collaboration between the Royal College of Music, University College London, and Imperial College London, it enrolled 193 participants across five Tenovus Cancer Care choirs in Wales — cancer patients, active carers, and bereaved carers — and measured 13 simultaneous biological markers before and after a single 70-minute choir rehearsal. It is the largest sample size in this literature and the most comprehensive biomarker panel attempted in a singing intervention study.

For NeuroNest Hub, this paper is a cornerstone entry in the Vocal Sound & The Nervous System series. It provides the strongest human evidence that singing produces immune-level biological changes — not just mood shifts — and it does so in a population where those changes matter most. The scale, institutional credibility, and breadth of outcome measures make this the definitive reference point for the singing-immunity connection in this content pipeline.

---

Methodology deep-dive

• Design: Multicentre, single-arm, pre/post repeated-measures design. No control group. Authors describe it explicitly as a “preliminary study.” Participants served as their own controls.
• Subjects/Sample: N=193 recruited from five Tenovus Cancer Care choirs across South Wales. Groups: cancer patients (n=55), current carers (n=72), bereaved carers (n=66). Screened from 251 eligible; 58 excluded or declined. No demographic breakdown by cancer type or stage provided.
• Inclusion/Exclusion: Adults (18+) attending at least one choir session. Excluded: mouth cancer (affects saliva sampling), active chemotherapy/radiotherapy, immunosuppressive medications, pregnancy.
• Protocol: Single 70-minute choir rehearsal (19:00–20:10) at community venues. Led by a trained choir leader. Structure: vocal warm-up → group song-learning → full repertoire singing.
• Timing of samples: Saliva at 7:00 PM (pre-rehearsal) and ~8:15 PM (post-rehearsal). Background questionnaires completed the week prior.
• Measures — Biological: Salivary cortisol, beta-endorphin, oxytocin, and 10 cytokines/immune markers: IL-2, IL-4, IL-6, IL-17, IFN-γ, TNF-α, GM-CSF, MCP-1, sIL-2rα, sTNFr1. Multiplex assay at Aeirtec Laboratories. Inter-assay CV: 1.8–5.37%; intra-assay CV: 0.8–3.58%.
• Measures — Psychological: 12 visual analogue scales (VAS) for specific mood and stress dimensions; Warwick-Edinburgh Mental Wellbeing Scale (WEMWBS); Hospital Anxiety and Depression Scale (HADS); Connor-Davidson Resilience Scale (CD-RISC).
• Analysis: Repeated-measures ANOVA with Bonferroni correction. Time × Group interaction effects examined for all outcomes. Pearson correlations for psychobiological relationships.
• Blinding: Not applicable. No blinded control condition. Lab analysts not described as blinded.
• Power: Not formally stated. N=193 is adequate for large effects; subgroup analyses may be underpowered for medium effects.

---

Sound protocol specifics

What was reported:

• 70-minute choir rehearsal, 7:00–8:10 PM
• Community venues (not a laboratory setting)
• Led by a trained choir leader
• Structure: vocal warm-up → song-learning → repertoire singing
• Group setting; five choirs, approximately 38 participants per choir on average

What was NOT reported (explicit gaps):

• Sound pressure level (dB SPL) — no acoustic measurement of rehearsal volume
• Frequency range — no vocal register, pitch range, or SATB/unison information
• Repertoire — no song titles, genre, or style
• Room acoustics — no reverberation, absorption, or venue dimensions
• Accompaniment — whether piano or recorded backing was used is not stated
• Posture — standing vs seated not specified
• Individual vocal output — no measurement of individual singing levels

From Dion’s sound engineering perspective: What’s striking here is that the study captures the output (biological change) without specifying the input signal chain at all. From an engineering standpoint, this is like measuring downstream changes in a mix without knowing the gain staging, EQ, or dynamics processing upstream. A choir warm-up and full repertoire singing are acoustically and physiologically distinct events: warm-up involves lower subglottic pressure, lighter phonation, and different airflow dynamics than open-chest choir singing. At ensemble level, you’re also dealing with constructive and destructive interference between voices, beat frequencies from slightly-detuned unisons, and room modes that return energy to individual singers via bone conduction and air pressure. The human vocal tract is a resonant tube instrument — its formants shift dramatically across registers, and the physical act of driving those resonances at ensemble amplitude is a biomechanical event as much as an expressive one. Any of these acoustic variables could be contributing mechanism — and none were measured. This is the protocol gap that most limits replicability.

Follow-up on gaps: No follow-up study from this group has subsequently addressed acoustic parameters. The closest related evidence comes from Weitzberg & Lundberg (2002) — Post #8 in this series — which provides a specific acoustic mechanism (resonant airflow and nitric oxide production during humming) for how vocal production may affect systemic physiology beyond the psychological pathway.

---

Key findings (beyond the headline)

• Cortisol reduced significantly across all three groups: 2.87 → 2.37 pg/mL (F(1,156) = 48.46, p < .001, f = 0.56 — medium-to-large effect). Consistent across patients, carers, and bereaved carers.
• Five cytokines increased significantly (all p < .001): GM-CSF (f = 0.45), IL-17 (f = 0.43), IL-2 (f = 0.48), IL-4 (f = 0.44), TNF-α (f = 0.42). Both soluble receptors sIL-2rα and sTNFr1 also increased significantly.
• Cortisol decrease directly correlated with cytokine increases: r = −.485 with IL-2, r = −.458 with IL-4, r = −.339 with sIL-2rα, r = −.326 with TNF-α (all p < .001). This psychobiological link is the mechanistic core of the paper.
• Mood effects were extraordinarily large: Aggregate mood F(1,187) = 162.10, f = 0.93 (p < .001). Individual items: energetic 48.9 → 73.4 (f = 0.87); happy 66.4 → 83.0 (f = 0.77).
• Anxiety dropped by more than half: VAS anxiety 25.9 → 11.5 (f = 0.61, p < .001). Stress VAS 28.7 → 12.4 (f = 0.69, p < .001).
• Oxytocin and beta-endorphin decreased (oxytocin: 4.91 → 3.75 pg/mL, f = 0.57; beta-endorphin: 4.63 → 4.04 pg/mL, f = 0.42 — both p < .001). Counterintuitive given their roles in social bonding — interpreted as generalised stress-system down-regulation.
• Between-group immune differences: Bereaved carers showed greatest IL-17 response and were the only group showing MCP-1 increase. Cancer patients showed no significant sTNFr1 change. Psychological outcomes did not differ between groups.
• IFN-γ and IL-6 approached but did not survive correction: p = .029 (f = 0.16) and p = .022 (f = 0.17) respectively — small effects below Bonferroni threshold.
• MCP-1 showed no overall change (p = .501) but increased specifically in bereaved carers, demonstrating within-aggregate heterogeneity.
• Mood improvement predicted lower pro-inflammatory response: Greater mood improvement associated with lower post-singing IL-6, IL-17, and MCP-1. Mood explained 10% of MCP-1 variance (p = .029), suggesting immune and psychological pathways are partially independent.

---

What the authors didn’t say

• The control problem is understated. The authors acknowledge participants might have changed biologically "if they had simply rested for 70 minutes" but dismiss this with a general citation. The real comparison isn’t passive rest — it’s any 70-minute enjoyable social gathering. Social contact, anticipatory reward, physical movement, or relief from isolation could each independently produce cortisol reduction and secondary immune changes without a single note being sung.
• Self-selection is a systematic, not incidental, limitation. Participants were existing choir members. This sample almost certainly over-represents people who already respond positively to group music activity. Whether these effects replicate in socially withdrawn, musically disengaged, or acutely ill cancer patients is completely unknown.
• Acute immune changes ≠ clinically meaningful immune changes. The authors acknowledge this but it deserves emphasis: cytokine levels fluctuate hourly. A single-session perturbation in IL-2 or GM-CSF with no longitudinal follow-up cannot be interpreted as immune enhancement with therapeutic relevance in cancer. The paper establishes that change occurred — not that it matters.
• The anti-inflammatory framing is incomplete. Simultaneous increases in TNF-α (pro-inflammatory) and IL-4 (anti-inflammatory) represent a mixed immune signal. The authors frame this as “cytokine network activation” but do not discuss the functional implications of co-elevating opposing mediators in a cancer context.
• No adjustment for diurnal cortisol variation. The rehearsal ran 7:00–8:10 PM, during the naturally declining phase of the cortisol diurnal curve. Some fraction of the cortisol decrease may reflect normal evening physiology rather than the singing intervention. This confound is not addressed.
• The oxytocin/beta-endorphin decrease is not adequately explained. While the stress down-regulation hypothesis is plausible, other explanations (salivary measurement artefacts, ceiling effects in a high-stress population, peptide depletion from pre-choir anticipatory activation) are not considered.
• Medication confounds beyond exclusion criteria. Active chemo/radiotherapy/immunosuppressants were excluded, but many participants may have been on antidepressants, anxiolytics, or corticosteroids for non-cancer reasons — all of which affect HPA axis activity and cytokine profiles. This is not reported.
• Conflict of interest not formally declared. The study used Tenovus Cancer Care choirs, and the charity presumably had an interest in demonstrating the value of their singing programme. The paper does not include a conflict of interest statement.

---

Cross-references in NeuroNest Research Hub

• Post #7 — Bernardi et al. (2001): Rhythmic breathing during music and prayer. Establishes the respiratory-vagal mechanism for vocal activity — a parallel physiological pathway to this paper’s immune findings, operating through HRV and baroreflex rather than cytokines.
• Post #8 — Weitzberg & Lundberg (2002): Humming and nasal nitric oxide. Provides a specific acoustic mechanism (resonant airflow in the sinuses) by which vocal production could affect systemic physiology — partially filling the mechanism gap this paper leaves open.
• Narrative link: Posts #7 and #8 establish the respiratory and neurological mechanisms of vocal sound; this paper (Post #9) provides the largest-scale evidence for immune-level outcomes. Together they build a multi-pathway case: singing affects the vagal system, the nitric oxide system, and the cytokine network.
• Future entries to connect: Grape et al. (2003) — professional vs amateur singing and hormonal response; Kreutz et al. (2004) — choir singing vs listening and sIgA; any group music-making and immunity literature.

---

7-Dimension Scoring Table

Tier: Silver — High-priority Silver. The single-arm design prevents Gold classification, but biomarker breadth, sample size, and storytelling value make this a must-write post. Treat with Gold-level depth.